Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Peter Zijlstra | 391e43d | 2011-11-15 17:14:39 +0100 | [diff] [blame] | 2 | * kernel/sched/core.c |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3 | * |
| 4 | * Kernel scheduler and related syscalls |
| 5 | * |
| 6 | * Copyright (C) 1991-2002 Linus Torvalds |
| 7 | * |
| 8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and |
| 9 | * make semaphores SMP safe |
| 10 | * 1998-11-19 Implemented schedule_timeout() and related stuff |
| 11 | * by Andrea Arcangeli |
| 12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: |
| 13 | * hybrid priority-list and round-robin design with |
| 14 | * an array-switch method of distributing timeslices |
| 15 | * and per-CPU runqueues. Cleanups and useful suggestions |
| 16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. |
| 17 | * 2003-09-03 Interactivity tuning by Con Kolivas. |
| 18 | * 2004-04-02 Scheduler domains code by Nick Piggin |
Ingo Molnar | c31f2e8 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
| 20 | * fair scheduling design by Con Kolivas. |
| 21 | * 2007-05-05 Load balancing (smp-nice) and other improvements |
| 22 | * by Peter Williams |
| 23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith |
| 24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri |
Ingo Molnar | b913176 | 2008-01-25 21:08:19 +0100 | [diff] [blame] | 25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
| 26 | * Thomas Gleixner, Mike Kravetz |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | */ |
| 28 | |
| 29 | #include <linux/mm.h> |
| 30 | #include <linux/module.h> |
| 31 | #include <linux/nmi.h> |
| 32 | #include <linux/init.h> |
Ingo Molnar | dff06c1 | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 33 | #include <linux/uaccess.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | #include <linux/highmem.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | #include <asm/mmu_context.h> |
| 36 | #include <linux/interrupt.h> |
Randy.Dunlap | c59ede7 | 2006-01-11 12:17:46 -0800 | [diff] [blame] | 37 | #include <linux/capability.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 38 | #include <linux/completion.h> |
| 39 | #include <linux/kernel_stat.h> |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 40 | #include <linux/debug_locks.h> |
Ingo Molnar | cdd6c48 | 2009-09-21 12:02:48 +0200 | [diff] [blame] | 41 | #include <linux/perf_event.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 42 | #include <linux/security.h> |
| 43 | #include <linux/notifier.h> |
| 44 | #include <linux/profile.h> |
Nigel Cunningham | 7dfb710 | 2006-12-06 20:34:23 -0800 | [diff] [blame] | 45 | #include <linux/freezer.h> |
akpm@osdl.org | 198e2f1 | 2006-01-12 01:05:30 -0800 | [diff] [blame] | 46 | #include <linux/vmalloc.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 47 | #include <linux/blkdev.h> |
| 48 | #include <linux/delay.h> |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 49 | #include <linux/pid_namespace.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | #include <linux/smp.h> |
| 51 | #include <linux/threads.h> |
| 52 | #include <linux/timer.h> |
| 53 | #include <linux/rcupdate.h> |
| 54 | #include <linux/cpu.h> |
| 55 | #include <linux/cpuset.h> |
| 56 | #include <linux/percpu.h> |
Alexey Dobriyan | b5aadf7 | 2008-10-06 13:23:43 +0400 | [diff] [blame] | 57 | #include <linux/proc_fs.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | #include <linux/seq_file.h> |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 59 | #include <linux/sysctl.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 60 | #include <linux/syscalls.h> |
| 61 | #include <linux/times.h> |
Jay Lan | 8f0ab51 | 2006-09-30 23:28:59 -0700 | [diff] [blame] | 62 | #include <linux/tsacct_kern.h> |
bibo mao | c6fd91f | 2006-03-26 01:38:20 -0800 | [diff] [blame] | 63 | #include <linux/kprobes.h> |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 64 | #include <linux/delayacct.h> |
Ingo Molnar | dff06c1 | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 65 | #include <linux/unistd.h> |
Jens Axboe | f5ff842 | 2007-09-21 09:19:54 +0200 | [diff] [blame] | 66 | #include <linux/pagemap.h> |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 67 | #include <linux/hrtimer.h> |
Reynes Philippe | 30914a5 | 2008-03-17 16:19:05 -0700 | [diff] [blame] | 68 | #include <linux/tick.h> |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 69 | #include <linux/debugfs.h> |
| 70 | #include <linux/ctype.h> |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 71 | #include <linux/ftrace.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 72 | #include <linux/slab.h> |
Carsten Emde | f1c6f1a | 2011-10-26 23:14:16 +0200 | [diff] [blame] | 73 | #include <linux/init_task.h> |
Al Viro | 4040153 | 2012-02-13 03:58:52 +0000 | [diff] [blame] | 74 | #include <linux/binfmts.h> |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 75 | #include <linux/context_tracking.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 76 | |
David Howells | 96f951e | 2012-03-28 18:30:03 +0100 | [diff] [blame] | 77 | #include <asm/switch_to.h> |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 78 | #include <asm/tlb.h> |
Satyam Sharma | 838225b | 2007-10-24 18:23:50 +0200 | [diff] [blame] | 79 | #include <asm/irq_regs.h> |
Christian Borntraeger | db7e527 | 2012-01-11 08:58:16 +0100 | [diff] [blame] | 80 | #include <asm/mutex.h> |
Glauber Costa | e6e6685 | 2011-07-11 15:28:17 -0400 | [diff] [blame] | 81 | #ifdef CONFIG_PARAVIRT |
| 82 | #include <asm/paravirt.h> |
| 83 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 84 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 85 | #include "sched.h" |
Peter Zijlstra | 391e43d | 2011-11-15 17:14:39 +0100 | [diff] [blame] | 86 | #include "../workqueue_sched.h" |
Thomas Gleixner | 29d5e04 | 2012-04-20 13:05:45 +0000 | [diff] [blame] | 87 | #include "../smpboot.h" |
Gregory Haskins | 6e0534f | 2008-05-12 21:21:01 +0200 | [diff] [blame] | 88 | |
Steven Rostedt | a8d154b | 2009-04-10 09:36:00 -0400 | [diff] [blame] | 89 | #define CREATE_TRACE_POINTS |
Steven Rostedt | ad8d75f | 2009-04-14 19:39:12 -0400 | [diff] [blame] | 90 | #include <trace/events/sched.h> |
Steven Rostedt | a8d154b | 2009-04-10 09:36:00 -0400 | [diff] [blame] | 91 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 92 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
Paul Turner | 58088ad | 2011-07-21 09:43:31 -0700 | [diff] [blame] | 93 | { |
| 94 | unsigned long delta; |
| 95 | ktime_t soft, hard, now; |
| 96 | |
| 97 | for (;;) { |
| 98 | if (hrtimer_active(period_timer)) |
| 99 | break; |
| 100 | |
| 101 | now = hrtimer_cb_get_time(period_timer); |
| 102 | hrtimer_forward(period_timer, now, period); |
| 103 | |
| 104 | soft = hrtimer_get_softexpires(period_timer); |
| 105 | hard = hrtimer_get_expires(period_timer); |
| 106 | delta = ktime_to_ns(ktime_sub(hard, soft)); |
| 107 | __hrtimer_start_range_ns(period_timer, soft, delta, |
| 108 | HRTIMER_MODE_ABS_PINNED, 0); |
| 109 | } |
| 110 | } |
| 111 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 112 | DEFINE_MUTEX(sched_domains_mutex); |
| 113 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
Peter Zijlstra | dc61b1d | 2010-06-08 11:40:42 +0200 | [diff] [blame] | 114 | |
Peter Zijlstra | fe44d62 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 115 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
Venkatesh Pallipadi | 305e683 | 2010-10-04 17:03:21 -0700 | [diff] [blame] | 116 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 117 | void update_rq_clock(struct rq *rq) |
Peter Zijlstra | 3e51f33 | 2008-05-03 18:29:28 +0200 | [diff] [blame] | 118 | { |
Peter Zijlstra | fe44d62 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 119 | s64 delta; |
Venkatesh Pallipadi | 305e683 | 2010-10-04 17:03:21 -0700 | [diff] [blame] | 120 | |
Mike Galbraith | 61eadef | 2011-04-29 08:36:50 +0200 | [diff] [blame] | 121 | if (rq->skip_clock_update > 0) |
Mike Galbraith | f26f9af | 2010-12-08 11:05:42 +0100 | [diff] [blame] | 122 | return; |
Venkatesh Pallipadi | aa48380 | 2010-10-04 17:03:22 -0700 | [diff] [blame] | 123 | |
Peter Zijlstra | fe44d62 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 124 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
| 125 | rq->clock += delta; |
| 126 | update_rq_clock_task(rq, delta); |
Peter Zijlstra | 3e51f33 | 2008-05-03 18:29:28 +0200 | [diff] [blame] | 127 | } |
| 128 | |
Ingo Molnar | e436d80 | 2007-07-19 21:28:35 +0200 | [diff] [blame] | 129 | /* |
Ingo Molnar | bf5c91b | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 130 | * Debugging: various feature bits |
| 131 | */ |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 132 | |
| 133 | #define SCHED_FEAT(name, enabled) \ |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 134 | (1UL << __SCHED_FEAT_##name) * enabled | |
| 135 | |
| 136 | const_debug unsigned int sysctl_sched_features = |
Peter Zijlstra | 391e43d | 2011-11-15 17:14:39 +0100 | [diff] [blame] | 137 | #include "features.h" |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 138 | 0; |
| 139 | |
| 140 | #undef SCHED_FEAT |
| 141 | |
| 142 | #ifdef CONFIG_SCHED_DEBUG |
| 143 | #define SCHED_FEAT(name, enabled) \ |
| 144 | #name , |
| 145 | |
Hiroshi Shimamoto | 1292531 | 2012-05-25 15:41:54 +0900 | [diff] [blame] | 146 | static const char * const sched_feat_names[] = { |
Peter Zijlstra | 391e43d | 2011-11-15 17:14:39 +0100 | [diff] [blame] | 147 | #include "features.h" |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 148 | }; |
| 149 | |
| 150 | #undef SCHED_FEAT |
| 151 | |
Li Zefan | 34f3a81 | 2008-10-30 15:23:32 +0800 | [diff] [blame] | 152 | static int sched_feat_show(struct seq_file *m, void *v) |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 153 | { |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 154 | int i; |
| 155 | |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 156 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
Li Zefan | 34f3a81 | 2008-10-30 15:23:32 +0800 | [diff] [blame] | 157 | if (!(sysctl_sched_features & (1UL << i))) |
| 158 | seq_puts(m, "NO_"); |
| 159 | seq_printf(m, "%s ", sched_feat_names[i]); |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 160 | } |
Li Zefan | 34f3a81 | 2008-10-30 15:23:32 +0800 | [diff] [blame] | 161 | seq_puts(m, "\n"); |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 162 | |
Li Zefan | 34f3a81 | 2008-10-30 15:23:32 +0800 | [diff] [blame] | 163 | return 0; |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 164 | } |
| 165 | |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 166 | #ifdef HAVE_JUMP_LABEL |
| 167 | |
Ingo Molnar | c5905af | 2012-02-24 08:31:31 +0100 | [diff] [blame] | 168 | #define jump_label_key__true STATIC_KEY_INIT_TRUE |
| 169 | #define jump_label_key__false STATIC_KEY_INIT_FALSE |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 170 | |
| 171 | #define SCHED_FEAT(name, enabled) \ |
| 172 | jump_label_key__##enabled , |
| 173 | |
Ingo Molnar | c5905af | 2012-02-24 08:31:31 +0100 | [diff] [blame] | 174 | struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 175 | #include "features.h" |
| 176 | }; |
| 177 | |
| 178 | #undef SCHED_FEAT |
| 179 | |
| 180 | static void sched_feat_disable(int i) |
| 181 | { |
Ingo Molnar | c5905af | 2012-02-24 08:31:31 +0100 | [diff] [blame] | 182 | if (static_key_enabled(&sched_feat_keys[i])) |
| 183 | static_key_slow_dec(&sched_feat_keys[i]); |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 184 | } |
| 185 | |
| 186 | static void sched_feat_enable(int i) |
| 187 | { |
Ingo Molnar | c5905af | 2012-02-24 08:31:31 +0100 | [diff] [blame] | 188 | if (!static_key_enabled(&sched_feat_keys[i])) |
| 189 | static_key_slow_inc(&sched_feat_keys[i]); |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 190 | } |
| 191 | #else |
| 192 | static void sched_feat_disable(int i) { }; |
| 193 | static void sched_feat_enable(int i) { }; |
| 194 | #endif /* HAVE_JUMP_LABEL */ |
| 195 | |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 196 | static ssize_t |
| 197 | sched_feat_write(struct file *filp, const char __user *ubuf, |
| 198 | size_t cnt, loff_t *ppos) |
| 199 | { |
| 200 | char buf[64]; |
Mathieu Desnoyers | 7740191 | 2010-09-13 17:47:00 -0400 | [diff] [blame] | 201 | char *cmp; |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 202 | int neg = 0; |
| 203 | int i; |
| 204 | |
| 205 | if (cnt > 63) |
| 206 | cnt = 63; |
| 207 | |
| 208 | if (copy_from_user(&buf, ubuf, cnt)) |
| 209 | return -EFAULT; |
| 210 | |
| 211 | buf[cnt] = 0; |
Mathieu Desnoyers | 7740191 | 2010-09-13 17:47:00 -0400 | [diff] [blame] | 212 | cmp = strstrip(buf); |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 213 | |
Hillf Danton | 524429c | 2011-01-06 20:58:12 +0800 | [diff] [blame] | 214 | if (strncmp(cmp, "NO_", 3) == 0) { |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 215 | neg = 1; |
| 216 | cmp += 3; |
| 217 | } |
| 218 | |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 219 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
Mathieu Desnoyers | 7740191 | 2010-09-13 17:47:00 -0400 | [diff] [blame] | 220 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 221 | if (neg) { |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 222 | sysctl_sched_features &= ~(1UL << i); |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 223 | sched_feat_disable(i); |
| 224 | } else { |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 225 | sysctl_sched_features |= (1UL << i); |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 226 | sched_feat_enable(i); |
| 227 | } |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 228 | break; |
| 229 | } |
| 230 | } |
| 231 | |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 232 | if (i == __SCHED_FEAT_NR) |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 233 | return -EINVAL; |
| 234 | |
Jan Blunck | 4299472 | 2009-11-20 17:40:37 +0100 | [diff] [blame] | 235 | *ppos += cnt; |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 236 | |
| 237 | return cnt; |
| 238 | } |
| 239 | |
Li Zefan | 34f3a81 | 2008-10-30 15:23:32 +0800 | [diff] [blame] | 240 | static int sched_feat_open(struct inode *inode, struct file *filp) |
| 241 | { |
| 242 | return single_open(filp, sched_feat_show, NULL); |
| 243 | } |
| 244 | |
Alexey Dobriyan | 828c095 | 2009-10-01 15:43:56 -0700 | [diff] [blame] | 245 | static const struct file_operations sched_feat_fops = { |
Li Zefan | 34f3a81 | 2008-10-30 15:23:32 +0800 | [diff] [blame] | 246 | .open = sched_feat_open, |
| 247 | .write = sched_feat_write, |
| 248 | .read = seq_read, |
| 249 | .llseek = seq_lseek, |
| 250 | .release = single_release, |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 251 | }; |
| 252 | |
| 253 | static __init int sched_init_debug(void) |
| 254 | { |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 255 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
| 256 | &sched_feat_fops); |
| 257 | |
| 258 | return 0; |
| 259 | } |
| 260 | late_initcall(sched_init_debug); |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 261 | #endif /* CONFIG_SCHED_DEBUG */ |
Ingo Molnar | bf5c91b | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 262 | |
| 263 | /* |
Peter Zijlstra | b82d9fd | 2007-11-09 22:39:39 +0100 | [diff] [blame] | 264 | * Number of tasks to iterate in a single balance run. |
| 265 | * Limited because this is done with IRQs disabled. |
| 266 | */ |
| 267 | const_debug unsigned int sysctl_sched_nr_migrate = 32; |
| 268 | |
| 269 | /* |
Peter Zijlstra | e9e9250 | 2009-09-01 10:34:37 +0200 | [diff] [blame] | 270 | * period over which we average the RT time consumption, measured |
| 271 | * in ms. |
| 272 | * |
| 273 | * default: 1s |
| 274 | */ |
| 275 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; |
| 276 | |
| 277 | /* |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 278 | * period over which we measure -rt task cpu usage in us. |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 279 | * default: 1s |
| 280 | */ |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 281 | unsigned int sysctl_sched_rt_period = 1000000; |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 282 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 283 | __read_mostly int scheduler_running; |
Ingo Molnar | 6892b75 | 2008-02-13 14:02:36 +0100 | [diff] [blame] | 284 | |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 285 | /* |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 286 | * part of the period that we allow rt tasks to run in us. |
| 287 | * default: 0.95s |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 288 | */ |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 289 | int sysctl_sched_rt_runtime = 950000; |
| 290 | |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 291 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 292 | |
| 293 | /* |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 294 | * __task_rq_lock - lock the rq @p resides on. |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 295 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 296 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 297 | __acquires(rq->lock) |
| 298 | { |
Peter Zijlstra | 0970d29 | 2010-02-15 14:45:54 +0100 | [diff] [blame] | 299 | struct rq *rq; |
| 300 | |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 301 | lockdep_assert_held(&p->pi_lock); |
| 302 | |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 303 | for (;;) { |
Peter Zijlstra | 0970d29 | 2010-02-15 14:45:54 +0100 | [diff] [blame] | 304 | rq = task_rq(p); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 305 | raw_spin_lock(&rq->lock); |
Peter Zijlstra | 65cc8e4 | 2010-03-25 21:05:16 +0100 | [diff] [blame] | 306 | if (likely(rq == task_rq(p))) |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 307 | return rq; |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 308 | raw_spin_unlock(&rq->lock); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 309 | } |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 310 | } |
| 311 | |
| 312 | /* |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 313 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 314 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 315 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 316 | __acquires(p->pi_lock) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 317 | __acquires(rq->lock) |
| 318 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 319 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 320 | |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 321 | for (;;) { |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 322 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 323 | rq = task_rq(p); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 324 | raw_spin_lock(&rq->lock); |
Peter Zijlstra | 65cc8e4 | 2010-03-25 21:05:16 +0100 | [diff] [blame] | 325 | if (likely(rq == task_rq(p))) |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 326 | return rq; |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 327 | raw_spin_unlock(&rq->lock); |
| 328 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 329 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 330 | } |
| 331 | |
Alexey Dobriyan | a995744 | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 332 | static void __task_rq_unlock(struct rq *rq) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 333 | __releases(rq->lock) |
| 334 | { |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 335 | raw_spin_unlock(&rq->lock); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 336 | } |
| 337 | |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 338 | static inline void |
| 339 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 340 | __releases(rq->lock) |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 341 | __releases(p->pi_lock) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 342 | { |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 343 | raw_spin_unlock(&rq->lock); |
| 344 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 345 | } |
| 346 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 347 | /* |
Robert P. J. Day | cc2a73b | 2006-12-10 02:20:00 -0800 | [diff] [blame] | 348 | * this_rq_lock - lock this runqueue and disable interrupts. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 349 | */ |
Alexey Dobriyan | a995744 | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 350 | static struct rq *this_rq_lock(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 351 | __acquires(rq->lock) |
| 352 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 353 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 354 | |
| 355 | local_irq_disable(); |
| 356 | rq = this_rq(); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 357 | raw_spin_lock(&rq->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 358 | |
| 359 | return rq; |
| 360 | } |
| 361 | |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 362 | #ifdef CONFIG_SCHED_HRTICK |
| 363 | /* |
| 364 | * Use HR-timers to deliver accurate preemption points. |
| 365 | * |
| 366 | * Its all a bit involved since we cannot program an hrt while holding the |
| 367 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a |
| 368 | * reschedule event. |
| 369 | * |
| 370 | * When we get rescheduled we reprogram the hrtick_timer outside of the |
| 371 | * rq->lock. |
| 372 | */ |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 373 | |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 374 | static void hrtick_clear(struct rq *rq) |
| 375 | { |
| 376 | if (hrtimer_active(&rq->hrtick_timer)) |
| 377 | hrtimer_cancel(&rq->hrtick_timer); |
| 378 | } |
| 379 | |
| 380 | /* |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 381 | * High-resolution timer tick. |
| 382 | * Runs from hardirq context with interrupts disabled. |
| 383 | */ |
| 384 | static enum hrtimer_restart hrtick(struct hrtimer *timer) |
| 385 | { |
| 386 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); |
| 387 | |
| 388 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); |
| 389 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 390 | raw_spin_lock(&rq->lock); |
Peter Zijlstra | 3e51f33 | 2008-05-03 18:29:28 +0200 | [diff] [blame] | 391 | update_rq_clock(rq); |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 392 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 393 | raw_spin_unlock(&rq->lock); |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 394 | |
| 395 | return HRTIMER_NORESTART; |
| 396 | } |
| 397 | |
Rabin Vincent | 95e904c | 2008-05-11 05:55:33 +0530 | [diff] [blame] | 398 | #ifdef CONFIG_SMP |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 399 | /* |
| 400 | * called from hardirq (IPI) context |
| 401 | */ |
| 402 | static void __hrtick_start(void *arg) |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 403 | { |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 404 | struct rq *rq = arg; |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 405 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 406 | raw_spin_lock(&rq->lock); |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 407 | hrtimer_restart(&rq->hrtick_timer); |
| 408 | rq->hrtick_csd_pending = 0; |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 409 | raw_spin_unlock(&rq->lock); |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 410 | } |
| 411 | |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 412 | /* |
| 413 | * Called to set the hrtick timer state. |
| 414 | * |
| 415 | * called with rq->lock held and irqs disabled |
| 416 | */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 417 | void hrtick_start(struct rq *rq, u64 delay) |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 418 | { |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 419 | struct hrtimer *timer = &rq->hrtick_timer; |
| 420 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 421 | |
Arjan van de Ven | cc584b2 | 2008-09-01 15:02:30 -0700 | [diff] [blame] | 422 | hrtimer_set_expires(timer, time); |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 423 | |
| 424 | if (rq == this_rq()) { |
| 425 | hrtimer_restart(timer); |
| 426 | } else if (!rq->hrtick_csd_pending) { |
Peter Zijlstra | 6e27563 | 2009-02-25 13:59:48 +0100 | [diff] [blame] | 427 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 428 | rq->hrtick_csd_pending = 1; |
| 429 | } |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 430 | } |
| 431 | |
| 432 | static int |
| 433 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) |
| 434 | { |
| 435 | int cpu = (int)(long)hcpu; |
| 436 | |
| 437 | switch (action) { |
| 438 | case CPU_UP_CANCELED: |
| 439 | case CPU_UP_CANCELED_FROZEN: |
| 440 | case CPU_DOWN_PREPARE: |
| 441 | case CPU_DOWN_PREPARE_FROZEN: |
| 442 | case CPU_DEAD: |
| 443 | case CPU_DEAD_FROZEN: |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 444 | hrtick_clear(cpu_rq(cpu)); |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 445 | return NOTIFY_OK; |
| 446 | } |
| 447 | |
| 448 | return NOTIFY_DONE; |
| 449 | } |
| 450 | |
Rakib Mullick | fa74820 | 2008-09-22 14:55:45 -0700 | [diff] [blame] | 451 | static __init void init_hrtick(void) |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 452 | { |
| 453 | hotcpu_notifier(hotplug_hrtick, 0); |
| 454 | } |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 455 | #else |
| 456 | /* |
| 457 | * Called to set the hrtick timer state. |
| 458 | * |
| 459 | * called with rq->lock held and irqs disabled |
| 460 | */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 461 | void hrtick_start(struct rq *rq, u64 delay) |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 462 | { |
Peter Zijlstra | 7f1e2ca | 2009-03-13 12:21:27 +0100 | [diff] [blame] | 463 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
Arun R Bharadwaj | 5c33386 | 2009-04-16 12:14:37 +0530 | [diff] [blame] | 464 | HRTIMER_MODE_REL_PINNED, 0); |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 465 | } |
| 466 | |
Andrew Morton | 006c75f | 2008-09-22 14:55:46 -0700 | [diff] [blame] | 467 | static inline void init_hrtick(void) |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 468 | { |
| 469 | } |
Rabin Vincent | 95e904c | 2008-05-11 05:55:33 +0530 | [diff] [blame] | 470 | #endif /* CONFIG_SMP */ |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 471 | |
| 472 | static void init_rq_hrtick(struct rq *rq) |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 473 | { |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 474 | #ifdef CONFIG_SMP |
| 475 | rq->hrtick_csd_pending = 0; |
| 476 | |
| 477 | rq->hrtick_csd.flags = 0; |
| 478 | rq->hrtick_csd.func = __hrtick_start; |
| 479 | rq->hrtick_csd.info = rq; |
| 480 | #endif |
| 481 | |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 482 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
| 483 | rq->hrtick_timer.function = hrtick; |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 484 | } |
Andrew Morton | 006c75f | 2008-09-22 14:55:46 -0700 | [diff] [blame] | 485 | #else /* CONFIG_SCHED_HRTICK */ |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 486 | static inline void hrtick_clear(struct rq *rq) |
| 487 | { |
| 488 | } |
| 489 | |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 490 | static inline void init_rq_hrtick(struct rq *rq) |
| 491 | { |
| 492 | } |
| 493 | |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 494 | static inline void init_hrtick(void) |
| 495 | { |
| 496 | } |
Andrew Morton | 006c75f | 2008-09-22 14:55:46 -0700 | [diff] [blame] | 497 | #endif /* CONFIG_SCHED_HRTICK */ |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 498 | |
Ingo Molnar | 1b9f19c | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 499 | /* |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 500 | * resched_task - mark a task 'to be rescheduled now'. |
| 501 | * |
| 502 | * On UP this means the setting of the need_resched flag, on SMP it |
| 503 | * might also involve a cross-CPU call to trigger the scheduler on |
| 504 | * the target CPU. |
| 505 | */ |
| 506 | #ifdef CONFIG_SMP |
| 507 | |
| 508 | #ifndef tsk_is_polling |
Al Viro | 16a8016 | 2012-06-01 14:22:01 -0400 | [diff] [blame] | 509 | #define tsk_is_polling(t) 0 |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 510 | #endif |
| 511 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 512 | void resched_task(struct task_struct *p) |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 513 | { |
| 514 | int cpu; |
| 515 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 516 | assert_raw_spin_locked(&task_rq(p)->lock); |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 517 | |
Lai Jiangshan | 5ed0cec | 2009-03-06 19:40:20 +0800 | [diff] [blame] | 518 | if (test_tsk_need_resched(p)) |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 519 | return; |
| 520 | |
Lai Jiangshan | 5ed0cec | 2009-03-06 19:40:20 +0800 | [diff] [blame] | 521 | set_tsk_need_resched(p); |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 522 | |
| 523 | cpu = task_cpu(p); |
| 524 | if (cpu == smp_processor_id()) |
| 525 | return; |
| 526 | |
| 527 | /* NEED_RESCHED must be visible before we test polling */ |
| 528 | smp_mb(); |
| 529 | if (!tsk_is_polling(p)) |
| 530 | smp_send_reschedule(cpu); |
| 531 | } |
| 532 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 533 | void resched_cpu(int cpu) |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 534 | { |
| 535 | struct rq *rq = cpu_rq(cpu); |
| 536 | unsigned long flags; |
| 537 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 538 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 539 | return; |
| 540 | resched_task(cpu_curr(cpu)); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 541 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 542 | } |
Thomas Gleixner | 06d8308 | 2008-03-22 09:20:24 +0100 | [diff] [blame] | 543 | |
| 544 | #ifdef CONFIG_NO_HZ |
| 545 | /* |
Venkatesh Pallipadi | 83cd4fe | 2010-05-21 17:09:41 -0700 | [diff] [blame] | 546 | * In the semi idle case, use the nearest busy cpu for migrating timers |
| 547 | * from an idle cpu. This is good for power-savings. |
| 548 | * |
| 549 | * We don't do similar optimization for completely idle system, as |
| 550 | * selecting an idle cpu will add more delays to the timers than intended |
| 551 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). |
| 552 | */ |
| 553 | int get_nohz_timer_target(void) |
| 554 | { |
| 555 | int cpu = smp_processor_id(); |
| 556 | int i; |
| 557 | struct sched_domain *sd; |
| 558 | |
Peter Zijlstra | 057f3fa | 2011-04-18 11:24:34 +0200 | [diff] [blame] | 559 | rcu_read_lock(); |
Venkatesh Pallipadi | 83cd4fe | 2010-05-21 17:09:41 -0700 | [diff] [blame] | 560 | for_each_domain(cpu, sd) { |
Peter Zijlstra | 057f3fa | 2011-04-18 11:24:34 +0200 | [diff] [blame] | 561 | for_each_cpu(i, sched_domain_span(sd)) { |
| 562 | if (!idle_cpu(i)) { |
| 563 | cpu = i; |
| 564 | goto unlock; |
| 565 | } |
| 566 | } |
Venkatesh Pallipadi | 83cd4fe | 2010-05-21 17:09:41 -0700 | [diff] [blame] | 567 | } |
Peter Zijlstra | 057f3fa | 2011-04-18 11:24:34 +0200 | [diff] [blame] | 568 | unlock: |
| 569 | rcu_read_unlock(); |
Venkatesh Pallipadi | 83cd4fe | 2010-05-21 17:09:41 -0700 | [diff] [blame] | 570 | return cpu; |
| 571 | } |
| 572 | /* |
Thomas Gleixner | 06d8308 | 2008-03-22 09:20:24 +0100 | [diff] [blame] | 573 | * When add_timer_on() enqueues a timer into the timer wheel of an |
| 574 | * idle CPU then this timer might expire before the next timer event |
| 575 | * which is scheduled to wake up that CPU. In case of a completely |
| 576 | * idle system the next event might even be infinite time into the |
| 577 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and |
| 578 | * leaves the inner idle loop so the newly added timer is taken into |
| 579 | * account when the CPU goes back to idle and evaluates the timer |
| 580 | * wheel for the next timer event. |
| 581 | */ |
| 582 | void wake_up_idle_cpu(int cpu) |
| 583 | { |
| 584 | struct rq *rq = cpu_rq(cpu); |
| 585 | |
| 586 | if (cpu == smp_processor_id()) |
| 587 | return; |
| 588 | |
| 589 | /* |
| 590 | * This is safe, as this function is called with the timer |
| 591 | * wheel base lock of (cpu) held. When the CPU is on the way |
| 592 | * to idle and has not yet set rq->curr to idle then it will |
| 593 | * be serialized on the timer wheel base lock and take the new |
| 594 | * timer into account automatically. |
| 595 | */ |
| 596 | if (rq->curr != rq->idle) |
| 597 | return; |
| 598 | |
| 599 | /* |
| 600 | * We can set TIF_RESCHED on the idle task of the other CPU |
| 601 | * lockless. The worst case is that the other CPU runs the |
| 602 | * idle task through an additional NOOP schedule() |
| 603 | */ |
Lai Jiangshan | 5ed0cec | 2009-03-06 19:40:20 +0800 | [diff] [blame] | 604 | set_tsk_need_resched(rq->idle); |
Thomas Gleixner | 06d8308 | 2008-03-22 09:20:24 +0100 | [diff] [blame] | 605 | |
| 606 | /* NEED_RESCHED must be visible before we test polling */ |
| 607 | smp_mb(); |
| 608 | if (!tsk_is_polling(rq->idle)) |
| 609 | smp_send_reschedule(cpu); |
| 610 | } |
Mike Galbraith | 39c0cbe | 2010-03-11 17:17:13 +0100 | [diff] [blame] | 611 | |
Suresh Siddha | ca38062 | 2011-10-03 15:09:00 -0700 | [diff] [blame] | 612 | static inline bool got_nohz_idle_kick(void) |
| 613 | { |
Suresh Siddha | 1c792db | 2011-12-01 17:07:32 -0800 | [diff] [blame] | 614 | int cpu = smp_processor_id(); |
| 615 | return idle_cpu(cpu) && test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); |
Suresh Siddha | ca38062 | 2011-10-03 15:09:00 -0700 | [diff] [blame] | 616 | } |
| 617 | |
| 618 | #else /* CONFIG_NO_HZ */ |
| 619 | |
| 620 | static inline bool got_nohz_idle_kick(void) |
| 621 | { |
| 622 | return false; |
| 623 | } |
| 624 | |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 625 | #endif /* CONFIG_NO_HZ */ |
Thomas Gleixner | 06d8308 | 2008-03-22 09:20:24 +0100 | [diff] [blame] | 626 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 627 | void sched_avg_update(struct rq *rq) |
Peter Zijlstra | e9e9250 | 2009-09-01 10:34:37 +0200 | [diff] [blame] | 628 | { |
| 629 | s64 period = sched_avg_period(); |
| 630 | |
| 631 | while ((s64)(rq->clock - rq->age_stamp) > period) { |
Will Deacon | 0d98bb2 | 2010-05-24 12:11:43 -0700 | [diff] [blame] | 632 | /* |
| 633 | * Inline assembly required to prevent the compiler |
| 634 | * optimising this loop into a divmod call. |
| 635 | * See __iter_div_u64_rem() for another example of this. |
| 636 | */ |
| 637 | asm("" : "+rm" (rq->age_stamp)); |
Peter Zijlstra | e9e9250 | 2009-09-01 10:34:37 +0200 | [diff] [blame] | 638 | rq->age_stamp += period; |
| 639 | rq->rt_avg /= 2; |
| 640 | } |
| 641 | } |
| 642 | |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 643 | #else /* !CONFIG_SMP */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 644 | void resched_task(struct task_struct *p) |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 645 | { |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 646 | assert_raw_spin_locked(&task_rq(p)->lock); |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 647 | set_tsk_need_resched(p); |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 648 | } |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 649 | #endif /* CONFIG_SMP */ |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 650 | |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 651 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
| 652 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 653 | /* |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 654 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
| 655 | * node and @up when leaving it for the final time. |
| 656 | * |
| 657 | * Caller must hold rcu_lock or sufficient equivalent. |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 658 | */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 659 | int walk_tg_tree_from(struct task_group *from, |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 660 | tg_visitor down, tg_visitor up, void *data) |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 661 | { |
| 662 | struct task_group *parent, *child; |
| 663 | int ret; |
| 664 | |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 665 | parent = from; |
| 666 | |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 667 | down: |
| 668 | ret = (*down)(parent, data); |
| 669 | if (ret) |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 670 | goto out; |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 671 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
| 672 | parent = child; |
| 673 | goto down; |
| 674 | |
| 675 | up: |
| 676 | continue; |
| 677 | } |
| 678 | ret = (*up)(parent, data); |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 679 | if (ret || parent == from) |
| 680 | goto out; |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 681 | |
| 682 | child = parent; |
| 683 | parent = parent->parent; |
| 684 | if (parent) |
| 685 | goto up; |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 686 | out: |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 687 | return ret; |
| 688 | } |
| 689 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 690 | int tg_nop(struct task_group *tg, void *data) |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 691 | { |
| 692 | return 0; |
| 693 | } |
| 694 | #endif |
| 695 | |
Ingo Molnar | 45bf76d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 696 | static void set_load_weight(struct task_struct *p) |
| 697 | { |
Nikhil Rao | f05998d | 2011-05-18 10:09:38 -0700 | [diff] [blame] | 698 | int prio = p->static_prio - MAX_RT_PRIO; |
| 699 | struct load_weight *load = &p->se.load; |
| 700 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 701 | /* |
| 702 | * SCHED_IDLE tasks get minimal weight: |
| 703 | */ |
| 704 | if (p->policy == SCHED_IDLE) { |
Nikhil Rao | c8b2811 | 2011-05-18 14:37:48 -0700 | [diff] [blame] | 705 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
Nikhil Rao | f05998d | 2011-05-18 10:09:38 -0700 | [diff] [blame] | 706 | load->inv_weight = WMULT_IDLEPRIO; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 707 | return; |
| 708 | } |
| 709 | |
Nikhil Rao | c8b2811 | 2011-05-18 14:37:48 -0700 | [diff] [blame] | 710 | load->weight = scale_load(prio_to_weight[prio]); |
Nikhil Rao | f05998d | 2011-05-18 10:09:38 -0700 | [diff] [blame] | 711 | load->inv_weight = prio_to_wmult[prio]; |
Ingo Molnar | 45bf76d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 712 | } |
| 713 | |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 714 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
Gregory Haskins | 2087a1a | 2008-06-27 14:30:00 -0600 | [diff] [blame] | 715 | { |
Mike Galbraith | a64692a | 2010-03-11 17:16:20 +0100 | [diff] [blame] | 716 | update_rq_clock(rq); |
Ingo Molnar | 71f8bd4 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 717 | sched_info_queued(p); |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 718 | p->sched_class->enqueue_task(rq, p, flags); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 719 | } |
| 720 | |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 721 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 722 | { |
Mike Galbraith | a64692a | 2010-03-11 17:16:20 +0100 | [diff] [blame] | 723 | update_rq_clock(rq); |
Ankita Garg | 46ac22b | 2008-07-01 14:30:06 +0530 | [diff] [blame] | 724 | sched_info_dequeued(p); |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 725 | p->sched_class->dequeue_task(rq, p, flags); |
Ingo Molnar | 71f8bd4 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 726 | } |
| 727 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 728 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
Peter Zijlstra | 1e3c88b | 2009-12-17 17:00:43 +0100 | [diff] [blame] | 729 | { |
| 730 | if (task_contributes_to_load(p)) |
| 731 | rq->nr_uninterruptible--; |
| 732 | |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 733 | enqueue_task(rq, p, flags); |
Peter Zijlstra | 1e3c88b | 2009-12-17 17:00:43 +0100 | [diff] [blame] | 734 | } |
| 735 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 736 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
Peter Zijlstra | 1e3c88b | 2009-12-17 17:00:43 +0100 | [diff] [blame] | 737 | { |
| 738 | if (task_contributes_to_load(p)) |
| 739 | rq->nr_uninterruptible++; |
| 740 | |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 741 | dequeue_task(rq, p, flags); |
Peter Zijlstra | 1e3c88b | 2009-12-17 17:00:43 +0100 | [diff] [blame] | 742 | } |
| 743 | |
Peter Zijlstra | fe44d62 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 744 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
Venkatesh Pallipadi | aa48380 | 2010-10-04 17:03:22 -0700 | [diff] [blame] | 745 | { |
Glauber Costa | 095c0aa | 2011-07-11 15:28:18 -0400 | [diff] [blame] | 746 | /* |
| 747 | * In theory, the compile should just see 0 here, and optimize out the call |
| 748 | * to sched_rt_avg_update. But I don't trust it... |
| 749 | */ |
| 750 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
| 751 | s64 steal = 0, irq_delta = 0; |
| 752 | #endif |
| 753 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
Peter Zijlstra | 8e92c20 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 754 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
Peter Zijlstra | fe44d62 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 755 | |
| 756 | /* |
| 757 | * Since irq_time is only updated on {soft,}irq_exit, we might run into |
| 758 | * this case when a previous update_rq_clock() happened inside a |
| 759 | * {soft,}irq region. |
| 760 | * |
| 761 | * When this happens, we stop ->clock_task and only update the |
| 762 | * prev_irq_time stamp to account for the part that fit, so that a next |
| 763 | * update will consume the rest. This ensures ->clock_task is |
| 764 | * monotonic. |
| 765 | * |
| 766 | * It does however cause some slight miss-attribution of {soft,}irq |
| 767 | * time, a more accurate solution would be to update the irq_time using |
| 768 | * the current rq->clock timestamp, except that would require using |
| 769 | * atomic ops. |
| 770 | */ |
| 771 | if (irq_delta > delta) |
| 772 | irq_delta = delta; |
| 773 | |
| 774 | rq->prev_irq_time += irq_delta; |
| 775 | delta -= irq_delta; |
Glauber Costa | 095c0aa | 2011-07-11 15:28:18 -0400 | [diff] [blame] | 776 | #endif |
| 777 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING |
Ingo Molnar | c5905af | 2012-02-24 08:31:31 +0100 | [diff] [blame] | 778 | if (static_key_false((¶virt_steal_rq_enabled))) { |
Glauber Costa | 095c0aa | 2011-07-11 15:28:18 -0400 | [diff] [blame] | 779 | u64 st; |
| 780 | |
| 781 | steal = paravirt_steal_clock(cpu_of(rq)); |
| 782 | steal -= rq->prev_steal_time_rq; |
| 783 | |
| 784 | if (unlikely(steal > delta)) |
| 785 | steal = delta; |
| 786 | |
| 787 | st = steal_ticks(steal); |
| 788 | steal = st * TICK_NSEC; |
| 789 | |
| 790 | rq->prev_steal_time_rq += steal; |
| 791 | |
| 792 | delta -= steal; |
| 793 | } |
| 794 | #endif |
| 795 | |
Peter Zijlstra | fe44d62 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 796 | rq->clock_task += delta; |
| 797 | |
Glauber Costa | 095c0aa | 2011-07-11 15:28:18 -0400 | [diff] [blame] | 798 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
| 799 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) |
| 800 | sched_rt_avg_update(rq, irq_delta + steal); |
| 801 | #endif |
Venkatesh Pallipadi | aa48380 | 2010-10-04 17:03:22 -0700 | [diff] [blame] | 802 | } |
| 803 | |
Peter Zijlstra | 34f971f | 2010-09-22 13:53:15 +0200 | [diff] [blame] | 804 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
| 805 | { |
| 806 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; |
| 807 | struct task_struct *old_stop = cpu_rq(cpu)->stop; |
| 808 | |
| 809 | if (stop) { |
| 810 | /* |
| 811 | * Make it appear like a SCHED_FIFO task, its something |
| 812 | * userspace knows about and won't get confused about. |
| 813 | * |
| 814 | * Also, it will make PI more or less work without too |
| 815 | * much confusion -- but then, stop work should not |
| 816 | * rely on PI working anyway. |
| 817 | */ |
| 818 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); |
| 819 | |
| 820 | stop->sched_class = &stop_sched_class; |
| 821 | } |
| 822 | |
| 823 | cpu_rq(cpu)->stop = stop; |
| 824 | |
| 825 | if (old_stop) { |
| 826 | /* |
| 827 | * Reset it back to a normal scheduling class so that |
| 828 | * it can die in pieces. |
| 829 | */ |
| 830 | old_stop->sched_class = &rt_sched_class; |
| 831 | } |
| 832 | } |
| 833 | |
Peter Zijlstra | 1e3c88b | 2009-12-17 17:00:43 +0100 | [diff] [blame] | 834 | /* |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 835 | * __normal_prio - return the priority that is based on the static prio |
Ingo Molnar | 71f8bd4 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 836 | */ |
Ingo Molnar | 1453118 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 837 | static inline int __normal_prio(struct task_struct *p) |
| 838 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 839 | return p->static_prio; |
Ingo Molnar | 1453118 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 840 | } |
| 841 | |
| 842 | /* |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 843 | * Calculate the expected normal priority: i.e. priority |
| 844 | * without taking RT-inheritance into account. Might be |
| 845 | * boosted by interactivity modifiers. Changes upon fork, |
| 846 | * setprio syscalls, and whenever the interactivity |
| 847 | * estimator recalculates. |
| 848 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 849 | static inline int normal_prio(struct task_struct *p) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 850 | { |
| 851 | int prio; |
| 852 | |
Ingo Molnar | e05606d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 853 | if (task_has_rt_policy(p)) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 854 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
| 855 | else |
| 856 | prio = __normal_prio(p); |
| 857 | return prio; |
| 858 | } |
| 859 | |
| 860 | /* |
| 861 | * Calculate the current priority, i.e. the priority |
| 862 | * taken into account by the scheduler. This value might |
| 863 | * be boosted by RT tasks, or might be boosted by |
| 864 | * interactivity modifiers. Will be RT if the task got |
| 865 | * RT-boosted. If not then it returns p->normal_prio. |
| 866 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 867 | static int effective_prio(struct task_struct *p) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 868 | { |
| 869 | p->normal_prio = normal_prio(p); |
| 870 | /* |
| 871 | * If we are RT tasks or we were boosted to RT priority, |
| 872 | * keep the priority unchanged. Otherwise, update priority |
| 873 | * to the normal priority: |
| 874 | */ |
| 875 | if (!rt_prio(p->prio)) |
| 876 | return p->normal_prio; |
| 877 | return p->prio; |
| 878 | } |
| 879 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 880 | /** |
| 881 | * task_curr - is this task currently executing on a CPU? |
| 882 | * @p: the task in question. |
| 883 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 884 | inline int task_curr(const struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 885 | { |
| 886 | return cpu_curr(task_cpu(p)) == p; |
| 887 | } |
| 888 | |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 889 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
| 890 | const struct sched_class *prev_class, |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 891 | int oldprio) |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 892 | { |
| 893 | if (prev_class != p->sched_class) { |
| 894 | if (prev_class->switched_from) |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 895 | prev_class->switched_from(rq, p); |
| 896 | p->sched_class->switched_to(rq, p); |
| 897 | } else if (oldprio != p->prio) |
| 898 | p->sched_class->prio_changed(rq, p, oldprio); |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 899 | } |
| 900 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 901 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
Peter Zijlstra | 1e5a740 | 2010-10-31 12:37:04 +0100 | [diff] [blame] | 902 | { |
| 903 | const struct sched_class *class; |
| 904 | |
| 905 | if (p->sched_class == rq->curr->sched_class) { |
| 906 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
| 907 | } else { |
| 908 | for_each_class(class) { |
| 909 | if (class == rq->curr->sched_class) |
| 910 | break; |
| 911 | if (class == p->sched_class) { |
| 912 | resched_task(rq->curr); |
| 913 | break; |
| 914 | } |
| 915 | } |
| 916 | } |
| 917 | |
| 918 | /* |
| 919 | * A queue event has occurred, and we're going to schedule. In |
| 920 | * this case, we can save a useless back to back clock update. |
| 921 | */ |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 922 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
Peter Zijlstra | 1e5a740 | 2010-10-31 12:37:04 +0100 | [diff] [blame] | 923 | rq->skip_clock_update = 1; |
| 924 | } |
| 925 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 926 | #ifdef CONFIG_SMP |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 927 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
Ingo Molnar | c65cc87 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 928 | { |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 929 | #ifdef CONFIG_SCHED_DEBUG |
| 930 | /* |
| 931 | * We should never call set_task_cpu() on a blocked task, |
| 932 | * ttwu() will sort out the placement. |
| 933 | */ |
Peter Zijlstra | 077614e | 2009-12-17 13:16:31 +0100 | [diff] [blame] | 934 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
| 935 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 936 | |
| 937 | #ifdef CONFIG_LOCKDEP |
Peter Zijlstra | 6c6c54e | 2011-06-03 17:37:07 +0200 | [diff] [blame] | 938 | /* |
| 939 | * The caller should hold either p->pi_lock or rq->lock, when changing |
| 940 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. |
| 941 | * |
| 942 | * sched_move_task() holds both and thus holding either pins the cgroup, |
Peter Zijlstra | 8323f26 | 2012-06-22 13:36:05 +0200 | [diff] [blame] | 943 | * see task_group(). |
Peter Zijlstra | 6c6c54e | 2011-06-03 17:37:07 +0200 | [diff] [blame] | 944 | * |
| 945 | * Furthermore, all task_rq users should acquire both locks, see |
| 946 | * task_rq_lock(). |
| 947 | */ |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 948 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
| 949 | lockdep_is_held(&task_rq(p)->lock))); |
| 950 | #endif |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 951 | #endif |
| 952 | |
Mathieu Desnoyers | de1d728 | 2009-05-05 16:49:59 +0800 | [diff] [blame] | 953 | trace_sched_migrate_task(p, new_cpu); |
Peter Zijlstra | cbc34ed | 2008-12-10 08:08:22 +0100 | [diff] [blame] | 954 | |
Peter Zijlstra | 0c69774 | 2009-12-22 15:43:19 +0100 | [diff] [blame] | 955 | if (task_cpu(p) != new_cpu) { |
Paul Turner | 0a74bef | 2012-10-04 13:18:30 +0200 | [diff] [blame] | 956 | if (p->sched_class->migrate_task_rq) |
| 957 | p->sched_class->migrate_task_rq(p, new_cpu); |
Peter Zijlstra | 0c69774 | 2009-12-22 15:43:19 +0100 | [diff] [blame] | 958 | p->se.nr_migrations++; |
Peter Zijlstra | a8b0ca1 | 2011-06-27 14:41:57 +0200 | [diff] [blame] | 959 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
Peter Zijlstra | 0c69774 | 2009-12-22 15:43:19 +0100 | [diff] [blame] | 960 | } |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 961 | |
| 962 | __set_task_cpu(p, new_cpu); |
Ingo Molnar | c65cc87 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 963 | } |
| 964 | |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 965 | struct migration_arg { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 966 | struct task_struct *task; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 967 | int dest_cpu; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 968 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 969 | |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 970 | static int migration_cpu_stop(void *data); |
| 971 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 972 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 973 | * wait_task_inactive - wait for a thread to unschedule. |
| 974 | * |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 975 | * If @match_state is nonzero, it's the @p->state value just checked and |
| 976 | * not expected to change. If it changes, i.e. @p might have woken up, |
| 977 | * then return zero. When we succeed in waiting for @p to be off its CPU, |
| 978 | * we return a positive number (its total switch count). If a second call |
| 979 | * a short while later returns the same number, the caller can be sure that |
| 980 | * @p has remained unscheduled the whole time. |
| 981 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 982 | * The caller must ensure that the task *will* unschedule sometime soon, |
| 983 | * else this function might spin for a *long* time. This function can't |
| 984 | * be called with interrupts off, or it may introduce deadlock with |
| 985 | * smp_call_function() if an IPI is sent by the same process we are |
| 986 | * waiting to become inactive. |
| 987 | */ |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 988 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 989 | { |
| 990 | unsigned long flags; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 991 | int running, on_rq; |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 992 | unsigned long ncsw; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 993 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 994 | |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 995 | for (;;) { |
| 996 | /* |
| 997 | * We do the initial early heuristics without holding |
| 998 | * any task-queue locks at all. We'll only try to get |
| 999 | * the runqueue lock when things look like they will |
| 1000 | * work out! |
| 1001 | */ |
| 1002 | rq = task_rq(p); |
Linus Torvalds | fa490cf | 2007-06-18 09:34:40 -0700 | [diff] [blame] | 1003 | |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1004 | /* |
| 1005 | * If the task is actively running on another CPU |
| 1006 | * still, just relax and busy-wait without holding |
| 1007 | * any locks. |
| 1008 | * |
| 1009 | * NOTE! Since we don't hold any locks, it's not |
| 1010 | * even sure that "rq" stays as the right runqueue! |
| 1011 | * But we don't care, since "task_running()" will |
| 1012 | * return false if the runqueue has changed and p |
| 1013 | * is actually now running somewhere else! |
| 1014 | */ |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 1015 | while (task_running(rq, p)) { |
| 1016 | if (match_state && unlikely(p->state != match_state)) |
| 1017 | return 0; |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1018 | cpu_relax(); |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 1019 | } |
Linus Torvalds | fa490cf | 2007-06-18 09:34:40 -0700 | [diff] [blame] | 1020 | |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1021 | /* |
| 1022 | * Ok, time to look more closely! We need the rq |
| 1023 | * lock now, to be *sure*. If we're wrong, we'll |
| 1024 | * just go back and repeat. |
| 1025 | */ |
| 1026 | rq = task_rq_lock(p, &flags); |
Peter Zijlstra | 27a9da6 | 2010-05-04 20:36:56 +0200 | [diff] [blame] | 1027 | trace_sched_wait_task(p); |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1028 | running = task_running(rq, p); |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 1029 | on_rq = p->on_rq; |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 1030 | ncsw = 0; |
Oleg Nesterov | f31e11d | 2008-08-20 16:54:44 -0700 | [diff] [blame] | 1031 | if (!match_state || p->state == match_state) |
Oleg Nesterov | 93dcf55 | 2008-08-20 16:54:44 -0700 | [diff] [blame] | 1032 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 1033 | task_rq_unlock(rq, p, &flags); |
Linus Torvalds | fa490cf | 2007-06-18 09:34:40 -0700 | [diff] [blame] | 1034 | |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1035 | /* |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 1036 | * If it changed from the expected state, bail out now. |
| 1037 | */ |
| 1038 | if (unlikely(!ncsw)) |
| 1039 | break; |
| 1040 | |
| 1041 | /* |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1042 | * Was it really running after all now that we |
| 1043 | * checked with the proper locks actually held? |
| 1044 | * |
| 1045 | * Oops. Go back and try again.. |
| 1046 | */ |
| 1047 | if (unlikely(running)) { |
| 1048 | cpu_relax(); |
| 1049 | continue; |
| 1050 | } |
| 1051 | |
| 1052 | /* |
| 1053 | * It's not enough that it's not actively running, |
| 1054 | * it must be off the runqueue _entirely_, and not |
| 1055 | * preempted! |
| 1056 | * |
Luis Henriques | 80dd99b | 2009-03-16 19:58:09 +0000 | [diff] [blame] | 1057 | * So if it was still runnable (but just not actively |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1058 | * running right now), it's preempted, and we should |
| 1059 | * yield - it could be a while. |
| 1060 | */ |
| 1061 | if (unlikely(on_rq)) { |
Thomas Gleixner | 8eb90c3 | 2011-02-23 23:52:21 +0000 | [diff] [blame] | 1062 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
| 1063 | |
| 1064 | set_current_state(TASK_UNINTERRUPTIBLE); |
| 1065 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1066 | continue; |
| 1067 | } |
| 1068 | |
| 1069 | /* |
| 1070 | * Ahh, all good. It wasn't running, and it wasn't |
| 1071 | * runnable, which means that it will never become |
| 1072 | * running in the future either. We're all done! |
| 1073 | */ |
| 1074 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1075 | } |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 1076 | |
| 1077 | return ncsw; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1078 | } |
| 1079 | |
| 1080 | /*** |
| 1081 | * kick_process - kick a running thread to enter/exit the kernel |
| 1082 | * @p: the to-be-kicked thread |
| 1083 | * |
| 1084 | * Cause a process which is running on another CPU to enter |
| 1085 | * kernel-mode, without any delay. (to get signals handled.) |
| 1086 | * |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 1087 | * NOTE: this function doesn't have to take the runqueue lock, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1088 | * because all it wants to ensure is that the remote task enters |
| 1089 | * the kernel. If the IPI races and the task has been migrated |
| 1090 | * to another CPU then no harm is done and the purpose has been |
| 1091 | * achieved as well. |
| 1092 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1093 | void kick_process(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1094 | { |
| 1095 | int cpu; |
| 1096 | |
| 1097 | preempt_disable(); |
| 1098 | cpu = task_cpu(p); |
| 1099 | if ((cpu != smp_processor_id()) && task_curr(p)) |
| 1100 | smp_send_reschedule(cpu); |
| 1101 | preempt_enable(); |
| 1102 | } |
Rusty Russell | b43e352 | 2009-06-12 22:27:00 -0600 | [diff] [blame] | 1103 | EXPORT_SYMBOL_GPL(kick_process); |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1104 | #endif /* CONFIG_SMP */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1105 | |
Peter Zijlstra | 970b13b | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 1106 | #ifdef CONFIG_SMP |
Oleg Nesterov | 30da688 | 2010-03-15 10:10:19 +0100 | [diff] [blame] | 1107 | /* |
Peter Zijlstra | 013fdb8 | 2011-04-05 17:23:45 +0200 | [diff] [blame] | 1108 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
Oleg Nesterov | 30da688 | 2010-03-15 10:10:19 +0100 | [diff] [blame] | 1109 | */ |
Peter Zijlstra | 5da9a0f | 2009-12-16 18:04:38 +0100 | [diff] [blame] | 1110 | static int select_fallback_rq(int cpu, struct task_struct *p) |
| 1111 | { |
Peter Zijlstra | 5da9a0f | 2009-12-16 18:04:38 +0100 | [diff] [blame] | 1112 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); |
Peter Zijlstra | 2baab4e | 2012-03-20 15:57:01 +0100 | [diff] [blame] | 1113 | enum { cpuset, possible, fail } state = cpuset; |
| 1114 | int dest_cpu; |
Peter Zijlstra | 5da9a0f | 2009-12-16 18:04:38 +0100 | [diff] [blame] | 1115 | |
| 1116 | /* Look for allowed, online CPU in same node. */ |
Srivatsa S. Bhat | e3831ed | 2012-03-30 19:40:28 +0530 | [diff] [blame] | 1117 | for_each_cpu(dest_cpu, nodemask) { |
Peter Zijlstra | 2baab4e | 2012-03-20 15:57:01 +0100 | [diff] [blame] | 1118 | if (!cpu_online(dest_cpu)) |
| 1119 | continue; |
| 1120 | if (!cpu_active(dest_cpu)) |
| 1121 | continue; |
Peter Zijlstra | fa17b50 | 2011-06-16 12:23:22 +0200 | [diff] [blame] | 1122 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
Peter Zijlstra | 5da9a0f | 2009-12-16 18:04:38 +0100 | [diff] [blame] | 1123 | return dest_cpu; |
Peter Zijlstra | 2baab4e | 2012-03-20 15:57:01 +0100 | [diff] [blame] | 1124 | } |
Peter Zijlstra | 5da9a0f | 2009-12-16 18:04:38 +0100 | [diff] [blame] | 1125 | |
Peter Zijlstra | 2baab4e | 2012-03-20 15:57:01 +0100 | [diff] [blame] | 1126 | for (;;) { |
| 1127 | /* Any allowed, online CPU? */ |
Srivatsa S. Bhat | e3831ed | 2012-03-30 19:40:28 +0530 | [diff] [blame] | 1128 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
Peter Zijlstra | 2baab4e | 2012-03-20 15:57:01 +0100 | [diff] [blame] | 1129 | if (!cpu_online(dest_cpu)) |
| 1130 | continue; |
| 1131 | if (!cpu_active(dest_cpu)) |
| 1132 | continue; |
| 1133 | goto out; |
| 1134 | } |
Peter Zijlstra | 5da9a0f | 2009-12-16 18:04:38 +0100 | [diff] [blame] | 1135 | |
Peter Zijlstra | 2baab4e | 2012-03-20 15:57:01 +0100 | [diff] [blame] | 1136 | switch (state) { |
| 1137 | case cpuset: |
| 1138 | /* No more Mr. Nice Guy. */ |
| 1139 | cpuset_cpus_allowed_fallback(p); |
| 1140 | state = possible; |
| 1141 | break; |
| 1142 | |
| 1143 | case possible: |
| 1144 | do_set_cpus_allowed(p, cpu_possible_mask); |
| 1145 | state = fail; |
| 1146 | break; |
| 1147 | |
| 1148 | case fail: |
| 1149 | BUG(); |
| 1150 | break; |
| 1151 | } |
| 1152 | } |
| 1153 | |
| 1154 | out: |
| 1155 | if (state != cpuset) { |
| 1156 | /* |
| 1157 | * Don't tell them about moving exiting tasks or |
| 1158 | * kernel threads (both mm NULL), since they never |
| 1159 | * leave kernel. |
| 1160 | */ |
| 1161 | if (p->mm && printk_ratelimit()) { |
| 1162 | printk_sched("process %d (%s) no longer affine to cpu%d\n", |
| 1163 | task_pid_nr(p), p->comm, cpu); |
| 1164 | } |
Peter Zijlstra | 5da9a0f | 2009-12-16 18:04:38 +0100 | [diff] [blame] | 1165 | } |
| 1166 | |
| 1167 | return dest_cpu; |
| 1168 | } |
| 1169 | |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 1170 | /* |
Peter Zijlstra | 013fdb8 | 2011-04-05 17:23:45 +0200 | [diff] [blame] | 1171 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 1172 | */ |
Peter Zijlstra | 970b13b | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 1173 | static inline |
Peter Zijlstra | 7608dec | 2011-04-05 17:23:46 +0200 | [diff] [blame] | 1174 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
Peter Zijlstra | 970b13b | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 1175 | { |
Peter Zijlstra | 7608dec | 2011-04-05 17:23:46 +0200 | [diff] [blame] | 1176 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 1177 | |
| 1178 | /* |
| 1179 | * In order not to call set_task_cpu() on a blocking task we need |
| 1180 | * to rely on ttwu() to place the task on a valid ->cpus_allowed |
| 1181 | * cpu. |
| 1182 | * |
| 1183 | * Since this is common to all placement strategies, this lives here. |
| 1184 | * |
| 1185 | * [ this allows ->select_task() to simply return task_cpu(p) and |
| 1186 | * not worry about this generic constraint ] |
| 1187 | */ |
Peter Zijlstra | fa17b50 | 2011-06-16 12:23:22 +0200 | [diff] [blame] | 1188 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
Peter Zijlstra | 70f1120 | 2009-12-20 17:36:27 +0100 | [diff] [blame] | 1189 | !cpu_online(cpu))) |
Peter Zijlstra | 5da9a0f | 2009-12-16 18:04:38 +0100 | [diff] [blame] | 1190 | cpu = select_fallback_rq(task_cpu(p), p); |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 1191 | |
| 1192 | return cpu; |
Peter Zijlstra | 970b13b | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 1193 | } |
Mike Galbraith | 09a40af | 2010-04-15 07:29:59 +0200 | [diff] [blame] | 1194 | |
| 1195 | static void update_avg(u64 *avg, u64 sample) |
| 1196 | { |
| 1197 | s64 diff = sample - *avg; |
| 1198 | *avg += diff >> 3; |
| 1199 | } |
Peter Zijlstra | 970b13b | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 1200 | #endif |
| 1201 | |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1202 | static void |
Peter Zijlstra | b84cb5d | 2011-04-05 17:23:55 +0200 | [diff] [blame] | 1203 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1204 | { |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1205 | #ifdef CONFIG_SCHEDSTATS |
Peter Zijlstra | b84cb5d | 2011-04-05 17:23:55 +0200 | [diff] [blame] | 1206 | struct rq *rq = this_rq(); |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1207 | |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1208 | #ifdef CONFIG_SMP |
| 1209 | int this_cpu = smp_processor_id(); |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1210 | |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1211 | if (cpu == this_cpu) { |
| 1212 | schedstat_inc(rq, ttwu_local); |
| 1213 | schedstat_inc(p, se.statistics.nr_wakeups_local); |
| 1214 | } else { |
| 1215 | struct sched_domain *sd; |
| 1216 | |
| 1217 | schedstat_inc(p, se.statistics.nr_wakeups_remote); |
Peter Zijlstra | 057f3fa | 2011-04-18 11:24:34 +0200 | [diff] [blame] | 1218 | rcu_read_lock(); |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1219 | for_each_domain(this_cpu, sd) { |
| 1220 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
| 1221 | schedstat_inc(sd, ttwu_wake_remote); |
| 1222 | break; |
| 1223 | } |
| 1224 | } |
Peter Zijlstra | 057f3fa | 2011-04-18 11:24:34 +0200 | [diff] [blame] | 1225 | rcu_read_unlock(); |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1226 | } |
Peter Zijlstra | f339b9d | 2011-05-31 10:49:20 +0200 | [diff] [blame] | 1227 | |
| 1228 | if (wake_flags & WF_MIGRATED) |
| 1229 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); |
| 1230 | |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1231 | #endif /* CONFIG_SMP */ |
| 1232 | |
| 1233 | schedstat_inc(rq, ttwu_count); |
| 1234 | schedstat_inc(p, se.statistics.nr_wakeups); |
| 1235 | |
| 1236 | if (wake_flags & WF_SYNC) |
| 1237 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
| 1238 | |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1239 | #endif /* CONFIG_SCHEDSTATS */ |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1240 | } |
| 1241 | |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1242 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1243 | { |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1244 | activate_task(rq, p, en_flags); |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 1245 | p->on_rq = 1; |
Peter Zijlstra | c2f7115 | 2011-04-13 13:28:56 +0200 | [diff] [blame] | 1246 | |
| 1247 | /* if a worker is waking up, notify workqueue */ |
| 1248 | if (p->flags & PF_WQ_WORKER) |
| 1249 | wq_worker_waking_up(p, cpu_of(rq)); |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1250 | } |
| 1251 | |
Peter Zijlstra | 23f41ee | 2011-04-05 17:23:56 +0200 | [diff] [blame] | 1252 | /* |
| 1253 | * Mark the task runnable and perform wakeup-preemption. |
| 1254 | */ |
Peter Zijlstra | 8936338 | 2011-04-05 17:23:42 +0200 | [diff] [blame] | 1255 | static void |
Peter Zijlstra | 23f41ee | 2011-04-05 17:23:56 +0200 | [diff] [blame] | 1256 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1257 | { |
Peter Zijlstra | 8936338 | 2011-04-05 17:23:42 +0200 | [diff] [blame] | 1258 | trace_sched_wakeup(p, true); |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1259 | check_preempt_curr(rq, p, wake_flags); |
| 1260 | |
| 1261 | p->state = TASK_RUNNING; |
| 1262 | #ifdef CONFIG_SMP |
| 1263 | if (p->sched_class->task_woken) |
| 1264 | p->sched_class->task_woken(rq, p); |
| 1265 | |
Steven Rostedt | e69c634 | 2010-12-06 17:10:31 -0500 | [diff] [blame] | 1266 | if (rq->idle_stamp) { |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1267 | u64 delta = rq->clock - rq->idle_stamp; |
| 1268 | u64 max = 2*sysctl_sched_migration_cost; |
| 1269 | |
| 1270 | if (delta > max) |
| 1271 | rq->avg_idle = max; |
| 1272 | else |
| 1273 | update_avg(&rq->avg_idle, delta); |
| 1274 | rq->idle_stamp = 0; |
| 1275 | } |
| 1276 | #endif |
| 1277 | } |
| 1278 | |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1279 | static void |
| 1280 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) |
| 1281 | { |
| 1282 | #ifdef CONFIG_SMP |
| 1283 | if (p->sched_contributes_to_load) |
| 1284 | rq->nr_uninterruptible--; |
| 1285 | #endif |
| 1286 | |
| 1287 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); |
| 1288 | ttwu_do_wakeup(rq, p, wake_flags); |
| 1289 | } |
| 1290 | |
| 1291 | /* |
| 1292 | * Called in case the task @p isn't fully descheduled from its runqueue, |
| 1293 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, |
| 1294 | * since all we need to do is flip p->state to TASK_RUNNING, since |
| 1295 | * the task is still ->on_rq. |
| 1296 | */ |
| 1297 | static int ttwu_remote(struct task_struct *p, int wake_flags) |
| 1298 | { |
| 1299 | struct rq *rq; |
| 1300 | int ret = 0; |
| 1301 | |
| 1302 | rq = __task_rq_lock(p); |
| 1303 | if (p->on_rq) { |
| 1304 | ttwu_do_wakeup(rq, p, wake_flags); |
| 1305 | ret = 1; |
| 1306 | } |
| 1307 | __task_rq_unlock(rq); |
| 1308 | |
| 1309 | return ret; |
| 1310 | } |
| 1311 | |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1312 | #ifdef CONFIG_SMP |
Peter Zijlstra | fa14ff4 | 2011-09-12 13:06:17 +0200 | [diff] [blame] | 1313 | static void sched_ttwu_pending(void) |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1314 | { |
| 1315 | struct rq *rq = this_rq(); |
Peter Zijlstra | fa14ff4 | 2011-09-12 13:06:17 +0200 | [diff] [blame] | 1316 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
| 1317 | struct task_struct *p; |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1318 | |
| 1319 | raw_spin_lock(&rq->lock); |
| 1320 | |
Peter Zijlstra | fa14ff4 | 2011-09-12 13:06:17 +0200 | [diff] [blame] | 1321 | while (llist) { |
| 1322 | p = llist_entry(llist, struct task_struct, wake_entry); |
| 1323 | llist = llist_next(llist); |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1324 | ttwu_do_activate(rq, p, 0); |
| 1325 | } |
| 1326 | |
| 1327 | raw_spin_unlock(&rq->lock); |
| 1328 | } |
| 1329 | |
| 1330 | void scheduler_ipi(void) |
| 1331 | { |
Suresh Siddha | ca38062 | 2011-10-03 15:09:00 -0700 | [diff] [blame] | 1332 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
Peter Zijlstra | c5d753a | 2011-07-19 15:07:25 -0700 | [diff] [blame] | 1333 | return; |
| 1334 | |
| 1335 | /* |
| 1336 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since |
| 1337 | * traditionally all their work was done from the interrupt return |
| 1338 | * path. Now that we actually do some work, we need to make sure |
| 1339 | * we do call them. |
| 1340 | * |
| 1341 | * Some archs already do call them, luckily irq_enter/exit nest |
| 1342 | * properly. |
| 1343 | * |
| 1344 | * Arguably we should visit all archs and update all handlers, |
| 1345 | * however a fair share of IPIs are still resched only so this would |
| 1346 | * somewhat pessimize the simple resched case. |
| 1347 | */ |
| 1348 | irq_enter(); |
Peter Zijlstra | fa14ff4 | 2011-09-12 13:06:17 +0200 | [diff] [blame] | 1349 | sched_ttwu_pending(); |
Suresh Siddha | ca38062 | 2011-10-03 15:09:00 -0700 | [diff] [blame] | 1350 | |
| 1351 | /* |
| 1352 | * Check if someone kicked us for doing the nohz idle load balance. |
| 1353 | */ |
Suresh Siddha | 6eb57e0 | 2011-10-03 15:09:01 -0700 | [diff] [blame] | 1354 | if (unlikely(got_nohz_idle_kick() && !need_resched())) { |
| 1355 | this_rq()->idle_balance = 1; |
Suresh Siddha | ca38062 | 2011-10-03 15:09:00 -0700 | [diff] [blame] | 1356 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
Suresh Siddha | 6eb57e0 | 2011-10-03 15:09:01 -0700 | [diff] [blame] | 1357 | } |
Peter Zijlstra | c5d753a | 2011-07-19 15:07:25 -0700 | [diff] [blame] | 1358 | irq_exit(); |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1359 | } |
| 1360 | |
| 1361 | static void ttwu_queue_remote(struct task_struct *p, int cpu) |
| 1362 | { |
Peter Zijlstra | fa14ff4 | 2011-09-12 13:06:17 +0200 | [diff] [blame] | 1363 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1364 | smp_send_reschedule(cpu); |
| 1365 | } |
Peter Zijlstra | d6aa8f8 | 2011-05-26 14:21:33 +0200 | [diff] [blame] | 1366 | |
Peter Zijlstra | 39be350 | 2012-01-26 12:44:34 +0100 | [diff] [blame] | 1367 | bool cpus_share_cache(int this_cpu, int that_cpu) |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 1368 | { |
| 1369 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); |
| 1370 | } |
Peter Zijlstra | d6aa8f8 | 2011-05-26 14:21:33 +0200 | [diff] [blame] | 1371 | #endif /* CONFIG_SMP */ |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1372 | |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1373 | static void ttwu_queue(struct task_struct *p, int cpu) |
| 1374 | { |
| 1375 | struct rq *rq = cpu_rq(cpu); |
| 1376 | |
Daniel Hellstrom | 17d9f31 | 2011-05-20 04:01:10 +0000 | [diff] [blame] | 1377 | #if defined(CONFIG_SMP) |
Peter Zijlstra | 39be350 | 2012-01-26 12:44:34 +0100 | [diff] [blame] | 1378 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
Peter Zijlstra | f01114c | 2011-05-31 12:26:55 +0200 | [diff] [blame] | 1379 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1380 | ttwu_queue_remote(p, cpu); |
| 1381 | return; |
| 1382 | } |
| 1383 | #endif |
| 1384 | |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1385 | raw_spin_lock(&rq->lock); |
| 1386 | ttwu_do_activate(rq, p, 0); |
| 1387 | raw_spin_unlock(&rq->lock); |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1388 | } |
| 1389 | |
| 1390 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1391 | * try_to_wake_up - wake up a thread |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1392 | * @p: the thread to be awakened |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1393 | * @state: the mask of task states that can be woken |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1394 | * @wake_flags: wake modifier flags (WF_*) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1395 | * |
| 1396 | * Put it on the run-queue if it's not already there. The "current" |
| 1397 | * thread is always on the run-queue (except when the actual |
| 1398 | * re-schedule is in progress), and as such you're allowed to do |
| 1399 | * the simpler "current->state = TASK_RUNNING" to mark yourself |
| 1400 | * runnable without the overhead of this. |
| 1401 | * |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1402 | * Returns %true if @p was woken up, %false if it was already running |
| 1403 | * or @state didn't match @p's state. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1404 | */ |
Peter Zijlstra | e4a52bc | 2011-04-05 17:23:54 +0200 | [diff] [blame] | 1405 | static int |
| 1406 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1407 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1408 | unsigned long flags; |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1409 | int cpu, success = 0; |
Peter Zijlstra | 2398f2c | 2008-06-27 13:41:35 +0200 | [diff] [blame] | 1410 | |
Linus Torvalds | 04e2f17 | 2008-02-23 18:05:03 -0800 | [diff] [blame] | 1411 | smp_wmb(); |
Peter Zijlstra | 013fdb8 | 2011-04-05 17:23:45 +0200 | [diff] [blame] | 1412 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
Peter Zijlstra | e9c8431 | 2009-09-15 14:43:03 +0200 | [diff] [blame] | 1413 | if (!(p->state & state)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1414 | goto out; |
| 1415 | |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1416 | success = 1; /* we're going to change ->state */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1417 | cpu = task_cpu(p); |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1418 | |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1419 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
| 1420 | goto stat; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1421 | |
| 1422 | #ifdef CONFIG_SMP |
Peter Zijlstra | e9c8431 | 2009-09-15 14:43:03 +0200 | [diff] [blame] | 1423 | /* |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1424 | * If the owning (remote) cpu is still in the middle of schedule() with |
| 1425 | * this task as prev, wait until its done referencing the task. |
Peter Zijlstra | e9c8431 | 2009-09-15 14:43:03 +0200 | [diff] [blame] | 1426 | */ |
Peter Zijlstra | f3e9478 | 2012-09-12 11:22:00 +0200 | [diff] [blame] | 1427 | while (p->on_cpu) |
Peter Zijlstra | e4a52bc | 2011-04-05 17:23:54 +0200 | [diff] [blame] | 1428 | cpu_relax(); |
Peter Zijlstra | e4a52bc | 2011-04-05 17:23:54 +0200 | [diff] [blame] | 1429 | /* |
| 1430 | * Pairs with the smp_wmb() in finish_lock_switch(). |
| 1431 | */ |
| 1432 | smp_rmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1433 | |
Peter Zijlstra | a8e4f2e | 2011-04-05 17:23:49 +0200 | [diff] [blame] | 1434 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
Peter Zijlstra | e9c8431 | 2009-09-15 14:43:03 +0200 | [diff] [blame] | 1435 | p->state = TASK_WAKING; |
Peter Zijlstra | efbbd05 | 2009-12-16 18:04:40 +0100 | [diff] [blame] | 1436 | |
Peter Zijlstra | e4a52bc | 2011-04-05 17:23:54 +0200 | [diff] [blame] | 1437 | if (p->sched_class->task_waking) |
Peter Zijlstra | 74f8e4b | 2011-04-05 17:23:47 +0200 | [diff] [blame] | 1438 | p->sched_class->task_waking(p); |
Peter Zijlstra | ab19cb2 | 2009-11-27 15:44:43 +0100 | [diff] [blame] | 1439 | |
Peter Zijlstra | 7608dec | 2011-04-05 17:23:46 +0200 | [diff] [blame] | 1440 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
Peter Zijlstra | f339b9d | 2011-05-31 10:49:20 +0200 | [diff] [blame] | 1441 | if (task_cpu(p) != cpu) { |
| 1442 | wake_flags |= WF_MIGRATED; |
Mike Galbraith | f5dc375 | 2009-10-09 08:35:03 +0200 | [diff] [blame] | 1443 | set_task_cpu(p, cpu); |
Peter Zijlstra | f339b9d | 2011-05-31 10:49:20 +0200 | [diff] [blame] | 1444 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1445 | #endif /* CONFIG_SMP */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1446 | |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1447 | ttwu_queue(p, cpu); |
| 1448 | stat: |
Peter Zijlstra | b84cb5d | 2011-04-05 17:23:55 +0200 | [diff] [blame] | 1449 | ttwu_stat(p, cpu, wake_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1450 | out: |
Peter Zijlstra | 013fdb8 | 2011-04-05 17:23:45 +0200 | [diff] [blame] | 1451 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1452 | |
| 1453 | return success; |
| 1454 | } |
| 1455 | |
David Howells | 50fa610 | 2009-04-28 15:01:38 +0100 | [diff] [blame] | 1456 | /** |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 1457 | * try_to_wake_up_local - try to wake up a local task with rq lock held |
| 1458 | * @p: the thread to be awakened |
| 1459 | * |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 1460 | * Put @p on the run-queue if it's not already there. The caller must |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 1461 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 1462 | * the current task. |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 1463 | */ |
| 1464 | static void try_to_wake_up_local(struct task_struct *p) |
| 1465 | { |
| 1466 | struct rq *rq = task_rq(p); |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 1467 | |
| 1468 | BUG_ON(rq != this_rq()); |
| 1469 | BUG_ON(p == current); |
| 1470 | lockdep_assert_held(&rq->lock); |
| 1471 | |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 1472 | if (!raw_spin_trylock(&p->pi_lock)) { |
| 1473 | raw_spin_unlock(&rq->lock); |
| 1474 | raw_spin_lock(&p->pi_lock); |
| 1475 | raw_spin_lock(&rq->lock); |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 1476 | } |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 1477 | |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 1478 | if (!(p->state & TASK_NORMAL)) |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 1479 | goto out; |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 1480 | |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 1481 | if (!p->on_rq) |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1482 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
| 1483 | |
Peter Zijlstra | 23f41ee | 2011-04-05 17:23:56 +0200 | [diff] [blame] | 1484 | ttwu_do_wakeup(rq, p, 0); |
Peter Zijlstra | b84cb5d | 2011-04-05 17:23:55 +0200 | [diff] [blame] | 1485 | ttwu_stat(p, smp_processor_id(), 0); |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 1486 | out: |
| 1487 | raw_spin_unlock(&p->pi_lock); |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 1488 | } |
| 1489 | |
| 1490 | /** |
David Howells | 50fa610 | 2009-04-28 15:01:38 +0100 | [diff] [blame] | 1491 | * wake_up_process - Wake up a specific process |
| 1492 | * @p: The process to be woken up. |
| 1493 | * |
| 1494 | * Attempt to wake up the nominated process and move it to the set of runnable |
| 1495 | * processes. Returns 1 if the process was woken up, 0 if it was already |
| 1496 | * running. |
| 1497 | * |
| 1498 | * It may be assumed that this function implies a write memory barrier before |
| 1499 | * changing the task state if and only if any tasks are woken up. |
| 1500 | */ |
Harvey Harrison | 7ad5b3a | 2008-02-08 04:19:53 -0800 | [diff] [blame] | 1501 | int wake_up_process(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1502 | { |
Matthew Wilcox | d9514f6 | 2007-12-06 11:07:07 -0500 | [diff] [blame] | 1503 | return try_to_wake_up(p, TASK_ALL, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1504 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1505 | EXPORT_SYMBOL(wake_up_process); |
| 1506 | |
Harvey Harrison | 7ad5b3a | 2008-02-08 04:19:53 -0800 | [diff] [blame] | 1507 | int wake_up_state(struct task_struct *p, unsigned int state) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1508 | { |
| 1509 | return try_to_wake_up(p, state, 0); |
| 1510 | } |
| 1511 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1512 | /* |
| 1513 | * Perform scheduler related setup for a newly forked process p. |
| 1514 | * p is forked by current. |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1515 | * |
| 1516 | * __sched_fork() is basic setup used by init_idle() too: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1517 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1518 | static void __sched_fork(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1519 | { |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 1520 | p->on_rq = 0; |
| 1521 | |
| 1522 | p->se.on_rq = 0; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1523 | p->se.exec_start = 0; |
| 1524 | p->se.sum_exec_runtime = 0; |
Ingo Molnar | f6cf891 | 2007-08-28 12:53:24 +0200 | [diff] [blame] | 1525 | p->se.prev_sum_exec_runtime = 0; |
Ingo Molnar | 6c594c2 | 2008-12-14 12:34:15 +0100 | [diff] [blame] | 1526 | p->se.nr_migrations = 0; |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 1527 | p->se.vruntime = 0; |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 1528 | INIT_LIST_HEAD(&p->se.group_node); |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 1529 | |
Paul Turner | f4e26b1 | 2012-10-04 13:18:32 +0200 | [diff] [blame] | 1530 | /* |
| 1531 | * Load-tracking only depends on SMP, FAIR_GROUP_SCHED dependency below may be |
| 1532 | * removed when useful for applications beyond shares distribution (e.g. |
| 1533 | * load-balance). |
| 1534 | */ |
| 1535 | #if defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED) |
Paul Turner | 9d85f21 | 2012-10-04 13:18:29 +0200 | [diff] [blame] | 1536 | p->se.avg.runnable_avg_period = 0; |
| 1537 | p->se.avg.runnable_avg_sum = 0; |
| 1538 | #endif |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 1539 | #ifdef CONFIG_SCHEDSTATS |
Lucas De Marchi | 41acab8 | 2010-03-10 23:37:45 -0300 | [diff] [blame] | 1540 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 1541 | #endif |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1542 | |
Peter Zijlstra | fa71706 | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 1543 | INIT_LIST_HEAD(&p->rt.run_list); |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1544 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1545 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
| 1546 | INIT_HLIST_HEAD(&p->preempt_notifiers); |
| 1547 | #endif |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1548 | } |
| 1549 | |
| 1550 | /* |
| 1551 | * fork()/clone()-time setup: |
| 1552 | */ |
Samir Bellabes | 3e51e3e | 2011-05-11 18:18:05 +0200 | [diff] [blame] | 1553 | void sched_fork(struct task_struct *p) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1554 | { |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 1555 | unsigned long flags; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1556 | int cpu = get_cpu(); |
| 1557 | |
| 1558 | __sched_fork(p); |
Peter Zijlstra | 06b83b5 | 2009-12-16 18:04:35 +0100 | [diff] [blame] | 1559 | /* |
Peter Zijlstra | 0017d73 | 2010-03-24 18:34:10 +0100 | [diff] [blame] | 1560 | * We mark the process as running here. This guarantees that |
Peter Zijlstra | 06b83b5 | 2009-12-16 18:04:35 +0100 | [diff] [blame] | 1561 | * nobody will actually run it, and a signal or other external |
| 1562 | * event cannot wake it up and insert it on the runqueue either. |
| 1563 | */ |
Peter Zijlstra | 0017d73 | 2010-03-24 18:34:10 +0100 | [diff] [blame] | 1564 | p->state = TASK_RUNNING; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1565 | |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 1566 | /* |
Mike Galbraith | c350a04 | 2011-07-27 17:14:55 +0200 | [diff] [blame] | 1567 | * Make sure we do not leak PI boosting priority to the child. |
| 1568 | */ |
| 1569 | p->prio = current->normal_prio; |
| 1570 | |
| 1571 | /* |
Mike Galbraith | b9dc29e | 2009-06-17 10:46:01 +0200 | [diff] [blame] | 1572 | * Revert to default priority/policy on fork if requested. |
| 1573 | */ |
| 1574 | if (unlikely(p->sched_reset_on_fork)) { |
Mike Galbraith | c350a04 | 2011-07-27 17:14:55 +0200 | [diff] [blame] | 1575 | if (task_has_rt_policy(p)) { |
Mike Galbraith | b9dc29e | 2009-06-17 10:46:01 +0200 | [diff] [blame] | 1576 | p->policy = SCHED_NORMAL; |
Mike Galbraith | 6c697bdf | 2009-06-17 10:48:02 +0200 | [diff] [blame] | 1577 | p->static_prio = NICE_TO_PRIO(0); |
Mike Galbraith | c350a04 | 2011-07-27 17:14:55 +0200 | [diff] [blame] | 1578 | p->rt_priority = 0; |
| 1579 | } else if (PRIO_TO_NICE(p->static_prio) < 0) |
| 1580 | p->static_prio = NICE_TO_PRIO(0); |
| 1581 | |
| 1582 | p->prio = p->normal_prio = __normal_prio(p); |
| 1583 | set_load_weight(p); |
Mike Galbraith | 6c697bdf | 2009-06-17 10:48:02 +0200 | [diff] [blame] | 1584 | |
Mike Galbraith | b9dc29e | 2009-06-17 10:46:01 +0200 | [diff] [blame] | 1585 | /* |
| 1586 | * We don't need the reset flag anymore after the fork. It has |
| 1587 | * fulfilled its duty: |
| 1588 | */ |
| 1589 | p->sched_reset_on_fork = 0; |
| 1590 | } |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 1591 | |
Hiroshi Shimamoto | 2ddbf95 | 2007-10-15 17:00:11 +0200 | [diff] [blame] | 1592 | if (!rt_prio(p->prio)) |
| 1593 | p->sched_class = &fair_sched_class; |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 1594 | |
Peter Zijlstra | cd29fe6 | 2009-11-27 17:32:46 +0100 | [diff] [blame] | 1595 | if (p->sched_class->task_fork) |
| 1596 | p->sched_class->task_fork(p); |
| 1597 | |
Peter Zijlstra | 8695159 | 2010-06-22 11:44:53 +0200 | [diff] [blame] | 1598 | /* |
| 1599 | * The child is not yet in the pid-hash so no cgroup attach races, |
| 1600 | * and the cgroup is pinned to this child due to cgroup_fork() |
| 1601 | * is ran before sched_fork(). |
| 1602 | * |
| 1603 | * Silence PROVE_RCU. |
| 1604 | */ |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 1605 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
Peter Zijlstra | 5f3edc1 | 2009-09-10 13:42:00 +0200 | [diff] [blame] | 1606 | set_task_cpu(p, cpu); |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 1607 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
Peter Zijlstra | 5f3edc1 | 2009-09-10 13:42:00 +0200 | [diff] [blame] | 1608 | |
Chandra Seetharaman | 52f17b6 | 2006-07-14 00:24:38 -0700 | [diff] [blame] | 1609 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1610 | if (likely(sched_info_on())) |
Chandra Seetharaman | 52f17b6 | 2006-07-14 00:24:38 -0700 | [diff] [blame] | 1611 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1612 | #endif |
Peter Zijlstra | 3ca7a44 | 2011-04-05 17:23:40 +0200 | [diff] [blame] | 1613 | #if defined(CONFIG_SMP) |
| 1614 | p->on_cpu = 0; |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1615 | #endif |
Frederic Weisbecker | bdd4e85 | 2011-06-08 01:13:27 +0200 | [diff] [blame] | 1616 | #ifdef CONFIG_PREEMPT_COUNT |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1617 | /* Want to start with kernel preemption disabled. */ |
Al Viro | a1261f5 | 2005-11-13 16:06:55 -0800 | [diff] [blame] | 1618 | task_thread_info(p)->preempt_count = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1619 | #endif |
Dario Faggioli | 806c09a | 2010-11-30 19:51:33 +0100 | [diff] [blame] | 1620 | #ifdef CONFIG_SMP |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1621 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
Dario Faggioli | 806c09a | 2010-11-30 19:51:33 +0100 | [diff] [blame] | 1622 | #endif |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1623 | |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1624 | put_cpu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1625 | } |
| 1626 | |
| 1627 | /* |
| 1628 | * wake_up_new_task - wake up a newly created task for the first time. |
| 1629 | * |
| 1630 | * This function will do some initial scheduler statistics housekeeping |
| 1631 | * that must be done for every newly created context, then puts the task |
| 1632 | * on the runqueue and wakes it. |
| 1633 | */ |
Samir Bellabes | 3e51e3e | 2011-05-11 18:18:05 +0200 | [diff] [blame] | 1634 | void wake_up_new_task(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1635 | { |
| 1636 | unsigned long flags; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1637 | struct rq *rq; |
Peter Zijlstra | fabf318 | 2010-01-21 21:04:57 +0100 | [diff] [blame] | 1638 | |
Peter Zijlstra | ab2515c | 2011-04-05 17:23:52 +0200 | [diff] [blame] | 1639 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
Peter Zijlstra | fabf318 | 2010-01-21 21:04:57 +0100 | [diff] [blame] | 1640 | #ifdef CONFIG_SMP |
| 1641 | /* |
| 1642 | * Fork balancing, do it here and not earlier because: |
| 1643 | * - cpus_allowed can change in the fork path |
| 1644 | * - any previously selected cpu might disappear through hotplug |
Peter Zijlstra | fabf318 | 2010-01-21 21:04:57 +0100 | [diff] [blame] | 1645 | */ |
Peter Zijlstra | ab2515c | 2011-04-05 17:23:52 +0200 | [diff] [blame] | 1646 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
Peter Zijlstra | fabf318 | 2010-01-21 21:04:57 +0100 | [diff] [blame] | 1647 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1648 | |
Peter Zijlstra | ab2515c | 2011-04-05 17:23:52 +0200 | [diff] [blame] | 1649 | rq = __task_rq_lock(p); |
Peter Zijlstra | cd29fe6 | 2009-11-27 17:32:46 +0100 | [diff] [blame] | 1650 | activate_task(rq, p, 0); |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 1651 | p->on_rq = 1; |
Peter Zijlstra | 8936338 | 2011-04-05 17:23:42 +0200 | [diff] [blame] | 1652 | trace_sched_wakeup_new(p, true); |
Peter Zijlstra | a7558e0 | 2009-09-14 20:02:34 +0200 | [diff] [blame] | 1653 | check_preempt_curr(rq, p, WF_FORK); |
Steven Rostedt | 9a897c5 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1654 | #ifdef CONFIG_SMP |
Peter Zijlstra | efbbd05 | 2009-12-16 18:04:40 +0100 | [diff] [blame] | 1655 | if (p->sched_class->task_woken) |
| 1656 | p->sched_class->task_woken(rq, p); |
Steven Rostedt | 9a897c5 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1657 | #endif |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 1658 | task_rq_unlock(rq, p, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1659 | } |
| 1660 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1661 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
| 1662 | |
| 1663 | /** |
Luis Henriques | 80dd99b | 2009-03-16 19:58:09 +0000 | [diff] [blame] | 1664 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
Randy Dunlap | 421cee2 | 2007-07-31 00:37:50 -0700 | [diff] [blame] | 1665 | * @notifier: notifier struct to register |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1666 | */ |
| 1667 | void preempt_notifier_register(struct preempt_notifier *notifier) |
| 1668 | { |
| 1669 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
| 1670 | } |
| 1671 | EXPORT_SYMBOL_GPL(preempt_notifier_register); |
| 1672 | |
| 1673 | /** |
| 1674 | * preempt_notifier_unregister - no longer interested in preemption notifications |
Randy Dunlap | 421cee2 | 2007-07-31 00:37:50 -0700 | [diff] [blame] | 1675 | * @notifier: notifier struct to unregister |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1676 | * |
| 1677 | * This is safe to call from within a preemption notifier. |
| 1678 | */ |
| 1679 | void preempt_notifier_unregister(struct preempt_notifier *notifier) |
| 1680 | { |
| 1681 | hlist_del(¬ifier->link); |
| 1682 | } |
| 1683 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); |
| 1684 | |
| 1685 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
| 1686 | { |
| 1687 | struct preempt_notifier *notifier; |
| 1688 | struct hlist_node *node; |
| 1689 | |
| 1690 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) |
| 1691 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
| 1692 | } |
| 1693 | |
| 1694 | static void |
| 1695 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
| 1696 | struct task_struct *next) |
| 1697 | { |
| 1698 | struct preempt_notifier *notifier; |
| 1699 | struct hlist_node *node; |
| 1700 | |
| 1701 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) |
| 1702 | notifier->ops->sched_out(notifier, next); |
| 1703 | } |
| 1704 | |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 1705 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1706 | |
| 1707 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
| 1708 | { |
| 1709 | } |
| 1710 | |
| 1711 | static void |
| 1712 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
| 1713 | struct task_struct *next) |
| 1714 | { |
| 1715 | } |
| 1716 | |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 1717 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1718 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1719 | /** |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1720 | * prepare_task_switch - prepare to switch tasks |
| 1721 | * @rq: the runqueue preparing to switch |
Randy Dunlap | 421cee2 | 2007-07-31 00:37:50 -0700 | [diff] [blame] | 1722 | * @prev: the current task that is being switched out |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1723 | * @next: the task we are going to switch to. |
| 1724 | * |
| 1725 | * This is called with the rq lock held and interrupts off. It must |
| 1726 | * be paired with a subsequent finish_task_switch after the context |
| 1727 | * switch. |
| 1728 | * |
| 1729 | * prepare_task_switch sets up locking and calls architecture specific |
| 1730 | * hooks. |
| 1731 | */ |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1732 | static inline void |
| 1733 | prepare_task_switch(struct rq *rq, struct task_struct *prev, |
| 1734 | struct task_struct *next) |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1735 | { |
Andrew Vagin | 895dd92 | 2012-07-12 14:14:29 +0400 | [diff] [blame] | 1736 | trace_sched_switch(prev, next); |
Peter Zijlstra | fe4b04f | 2011-02-02 13:19:09 +0100 | [diff] [blame] | 1737 | sched_info_switch(prev, next); |
| 1738 | perf_event_task_sched_out(prev, next); |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1739 | fire_sched_out_preempt_notifiers(prev, next); |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1740 | prepare_lock_switch(rq, next); |
| 1741 | prepare_arch_switch(next); |
| 1742 | } |
| 1743 | |
| 1744 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1745 | * finish_task_switch - clean up after a task-switch |
Jeff Garzik | 344baba | 2005-09-07 01:15:17 -0400 | [diff] [blame] | 1746 | * @rq: runqueue associated with task-switch |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1747 | * @prev: the thread we just switched away from. |
| 1748 | * |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1749 | * finish_task_switch must be called after the context switch, paired |
| 1750 | * with a prepare_task_switch call before the context switch. |
| 1751 | * finish_task_switch will reconcile locking set up by prepare_task_switch, |
| 1752 | * and do any other architecture-specific cleanup actions. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1753 | * |
| 1754 | * Note that we may have delayed dropping an mm in context_switch(). If |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 1755 | * so, we finish that here outside of the runqueue lock. (Doing it |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1756 | * with the lock held can cause deadlocks; see schedule() for |
| 1757 | * details.) |
| 1758 | */ |
Alexey Dobriyan | a995744 | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 1759 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1760 | __releases(rq->lock) |
| 1761 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1762 | struct mm_struct *mm = rq->prev_mm; |
Oleg Nesterov | 55a101f | 2006-09-29 02:01:10 -0700 | [diff] [blame] | 1763 | long prev_state; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1764 | |
| 1765 | rq->prev_mm = NULL; |
| 1766 | |
| 1767 | /* |
| 1768 | * A task struct has one reference for the use as "current". |
Oleg Nesterov | c394cc9 | 2006-09-29 02:01:11 -0700 | [diff] [blame] | 1769 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
Oleg Nesterov | 55a101f | 2006-09-29 02:01:10 -0700 | [diff] [blame] | 1770 | * schedule one last time. The schedule call will never return, and |
| 1771 | * the scheduled task must drop that reference. |
Oleg Nesterov | c394cc9 | 2006-09-29 02:01:11 -0700 | [diff] [blame] | 1772 | * The test for TASK_DEAD must occur while the runqueue locks are |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1773 | * still held, otherwise prev could be scheduled on another cpu, die |
| 1774 | * there before we look at prev->state, and then the reference would |
| 1775 | * be dropped twice. |
| 1776 | * Manfred Spraul <manfred@colorfullife.com> |
| 1777 | */ |
Oleg Nesterov | 55a101f | 2006-09-29 02:01:10 -0700 | [diff] [blame] | 1778 | prev_state = prev->state; |
Frederic Weisbecker | bf9fae9 | 2012-09-08 15:23:11 +0200 | [diff] [blame] | 1779 | vtime_task_switch(prev); |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1780 | finish_arch_switch(prev); |
Stephane Eranian | a8d757e | 2011-08-25 15:58:03 +0200 | [diff] [blame] | 1781 | perf_event_task_sched_in(prev, current); |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1782 | finish_lock_switch(rq, prev); |
Catalin Marinas | 01f23e1 | 2011-11-27 21:43:10 +0000 | [diff] [blame] | 1783 | finish_arch_post_lock_switch(); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1784 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1785 | fire_sched_in_preempt_notifiers(current); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1786 | if (mm) |
| 1787 | mmdrop(mm); |
Oleg Nesterov | c394cc9 | 2006-09-29 02:01:11 -0700 | [diff] [blame] | 1788 | if (unlikely(prev_state == TASK_DEAD)) { |
bibo mao | c6fd91f | 2006-03-26 01:38:20 -0800 | [diff] [blame] | 1789 | /* |
| 1790 | * Remove function-return probe instances associated with this |
| 1791 | * task and put them back on the free list. |
Ingo Molnar | 9761eea | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 1792 | */ |
bibo mao | c6fd91f | 2006-03-26 01:38:20 -0800 | [diff] [blame] | 1793 | kprobe_flush_task(prev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1794 | put_task_struct(prev); |
bibo mao | c6fd91f | 2006-03-26 01:38:20 -0800 | [diff] [blame] | 1795 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1796 | } |
| 1797 | |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 1798 | #ifdef CONFIG_SMP |
| 1799 | |
| 1800 | /* assumes rq->lock is held */ |
| 1801 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) |
| 1802 | { |
| 1803 | if (prev->sched_class->pre_schedule) |
| 1804 | prev->sched_class->pre_schedule(rq, prev); |
| 1805 | } |
| 1806 | |
| 1807 | /* rq->lock is NOT held, but preemption is disabled */ |
| 1808 | static inline void post_schedule(struct rq *rq) |
| 1809 | { |
| 1810 | if (rq->post_schedule) { |
| 1811 | unsigned long flags; |
| 1812 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 1813 | raw_spin_lock_irqsave(&rq->lock, flags); |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 1814 | if (rq->curr->sched_class->post_schedule) |
| 1815 | rq->curr->sched_class->post_schedule(rq); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 1816 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 1817 | |
| 1818 | rq->post_schedule = 0; |
| 1819 | } |
| 1820 | } |
| 1821 | |
| 1822 | #else |
| 1823 | |
| 1824 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
| 1825 | { |
| 1826 | } |
| 1827 | |
| 1828 | static inline void post_schedule(struct rq *rq) |
| 1829 | { |
| 1830 | } |
| 1831 | |
| 1832 | #endif |
| 1833 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1834 | /** |
| 1835 | * schedule_tail - first thing a freshly forked thread must call. |
| 1836 | * @prev: the thread we just switched away from. |
| 1837 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1838 | asmlinkage void schedule_tail(struct task_struct *prev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1839 | __releases(rq->lock) |
| 1840 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1841 | struct rq *rq = this_rq(); |
| 1842 | |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1843 | finish_task_switch(rq, prev); |
Steven Rostedt | da19ab5 | 2009-07-29 00:21:22 -0400 | [diff] [blame] | 1844 | |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 1845 | /* |
| 1846 | * FIXME: do we need to worry about rq being invalidated by the |
| 1847 | * task_switch? |
| 1848 | */ |
| 1849 | post_schedule(rq); |
Steven Rostedt | da19ab5 | 2009-07-29 00:21:22 -0400 | [diff] [blame] | 1850 | |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1851 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
| 1852 | /* In this case, finish_task_switch does not reenable preemption */ |
| 1853 | preempt_enable(); |
| 1854 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1855 | if (current->set_child_tid) |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 1856 | put_user(task_pid_vnr(current), current->set_child_tid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1857 | } |
| 1858 | |
| 1859 | /* |
| 1860 | * context_switch - switch to the new MM and the new |
| 1861 | * thread's register state. |
| 1862 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1863 | static inline void |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1864 | context_switch(struct rq *rq, struct task_struct *prev, |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1865 | struct task_struct *next) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1866 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1867 | struct mm_struct *mm, *oldmm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1868 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1869 | prepare_task_switch(rq, prev, next); |
Peter Zijlstra | fe4b04f | 2011-02-02 13:19:09 +0100 | [diff] [blame] | 1870 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1871 | mm = next->mm; |
| 1872 | oldmm = prev->active_mm; |
Zachary Amsden | 9226d12 | 2007-02-13 13:26:21 +0100 | [diff] [blame] | 1873 | /* |
| 1874 | * For paravirt, this is coupled with an exit in switch_to to |
| 1875 | * combine the page table reload and the switch backend into |
| 1876 | * one hypercall. |
| 1877 | */ |
Jeremy Fitzhardinge | 224101e | 2009-02-18 11:18:57 -0800 | [diff] [blame] | 1878 | arch_start_context_switch(prev); |
Zachary Amsden | 9226d12 | 2007-02-13 13:26:21 +0100 | [diff] [blame] | 1879 | |
Heiko Carstens | 31915ab | 2010-09-16 14:42:25 +0200 | [diff] [blame] | 1880 | if (!mm) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1881 | next->active_mm = oldmm; |
| 1882 | atomic_inc(&oldmm->mm_count); |
| 1883 | enter_lazy_tlb(oldmm, next); |
| 1884 | } else |
| 1885 | switch_mm(oldmm, mm, next); |
| 1886 | |
Heiko Carstens | 31915ab | 2010-09-16 14:42:25 +0200 | [diff] [blame] | 1887 | if (!prev->mm) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1888 | prev->active_mm = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1889 | rq->prev_mm = oldmm; |
| 1890 | } |
Ingo Molnar | 3a5f5e4 | 2006-07-14 00:24:27 -0700 | [diff] [blame] | 1891 | /* |
| 1892 | * Since the runqueue lock will be released by the next |
| 1893 | * task (which is an invalid locking op but in the case |
| 1894 | * of the scheduler it's an obvious special-case), so we |
| 1895 | * do an early lockdep release here: |
| 1896 | */ |
| 1897 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
Ingo Molnar | 8a25d5d | 2006-07-03 00:24:54 -0700 | [diff] [blame] | 1898 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
Ingo Molnar | 3a5f5e4 | 2006-07-14 00:24:27 -0700 | [diff] [blame] | 1899 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1900 | |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 1901 | context_tracking_task_switch(prev, next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1902 | /* Here we just switch the register state and the stack. */ |
| 1903 | switch_to(prev, next, prev); |
| 1904 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1905 | barrier(); |
| 1906 | /* |
| 1907 | * this_rq must be evaluated again because prev may have moved |
| 1908 | * CPUs since it called schedule(), thus the 'rq' on its stack |
| 1909 | * frame will be invalid. |
| 1910 | */ |
| 1911 | finish_task_switch(this_rq(), prev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1912 | } |
| 1913 | |
| 1914 | /* |
| 1915 | * nr_running, nr_uninterruptible and nr_context_switches: |
| 1916 | * |
| 1917 | * externally visible scheduler statistics: current number of runnable |
| 1918 | * threads, current number of uninterruptible-sleeping threads, total |
| 1919 | * number of context switches performed since bootup. |
| 1920 | */ |
| 1921 | unsigned long nr_running(void) |
| 1922 | { |
| 1923 | unsigned long i, sum = 0; |
| 1924 | |
| 1925 | for_each_online_cpu(i) |
| 1926 | sum += cpu_rq(i)->nr_running; |
| 1927 | |
| 1928 | return sum; |
| 1929 | } |
| 1930 | |
| 1931 | unsigned long nr_uninterruptible(void) |
| 1932 | { |
| 1933 | unsigned long i, sum = 0; |
| 1934 | |
KAMEZAWA Hiroyuki | 0a94502 | 2006-03-28 01:56:37 -0800 | [diff] [blame] | 1935 | for_each_possible_cpu(i) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1936 | sum += cpu_rq(i)->nr_uninterruptible; |
| 1937 | |
| 1938 | /* |
| 1939 | * Since we read the counters lockless, it might be slightly |
| 1940 | * inaccurate. Do not allow it to go below zero though: |
| 1941 | */ |
| 1942 | if (unlikely((long)sum < 0)) |
| 1943 | sum = 0; |
| 1944 | |
| 1945 | return sum; |
| 1946 | } |
| 1947 | |
| 1948 | unsigned long long nr_context_switches(void) |
| 1949 | { |
Steven Rostedt | cc94abf | 2006-06-27 02:54:31 -0700 | [diff] [blame] | 1950 | int i; |
| 1951 | unsigned long long sum = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1952 | |
KAMEZAWA Hiroyuki | 0a94502 | 2006-03-28 01:56:37 -0800 | [diff] [blame] | 1953 | for_each_possible_cpu(i) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1954 | sum += cpu_rq(i)->nr_switches; |
| 1955 | |
| 1956 | return sum; |
| 1957 | } |
| 1958 | |
| 1959 | unsigned long nr_iowait(void) |
| 1960 | { |
| 1961 | unsigned long i, sum = 0; |
| 1962 | |
KAMEZAWA Hiroyuki | 0a94502 | 2006-03-28 01:56:37 -0800 | [diff] [blame] | 1963 | for_each_possible_cpu(i) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1964 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
| 1965 | |
| 1966 | return sum; |
| 1967 | } |
| 1968 | |
Peter Zijlstra | 8c215bd | 2010-07-01 09:07:17 +0200 | [diff] [blame] | 1969 | unsigned long nr_iowait_cpu(int cpu) |
Arjan van de Ven | 69d2587 | 2009-09-21 17:04:08 -0700 | [diff] [blame] | 1970 | { |
Peter Zijlstra | 8c215bd | 2010-07-01 09:07:17 +0200 | [diff] [blame] | 1971 | struct rq *this = cpu_rq(cpu); |
Arjan van de Ven | 69d2587 | 2009-09-21 17:04:08 -0700 | [diff] [blame] | 1972 | return atomic_read(&this->nr_iowait); |
| 1973 | } |
| 1974 | |
| 1975 | unsigned long this_cpu_load(void) |
| 1976 | { |
| 1977 | struct rq *this = this_rq(); |
| 1978 | return this->cpu_load[0]; |
| 1979 | } |
| 1980 | |
| 1981 | |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 1982 | /* |
| 1983 | * Global load-average calculations |
| 1984 | * |
| 1985 | * We take a distributed and async approach to calculating the global load-avg |
| 1986 | * in order to minimize overhead. |
| 1987 | * |
| 1988 | * The global load average is an exponentially decaying average of nr_running + |
| 1989 | * nr_uninterruptible. |
| 1990 | * |
| 1991 | * Once every LOAD_FREQ: |
| 1992 | * |
| 1993 | * nr_active = 0; |
| 1994 | * for_each_possible_cpu(cpu) |
| 1995 | * nr_active += cpu_of(cpu)->nr_running + cpu_of(cpu)->nr_uninterruptible; |
| 1996 | * |
| 1997 | * avenrun[n] = avenrun[0] * exp_n + nr_active * (1 - exp_n) |
| 1998 | * |
| 1999 | * Due to a number of reasons the above turns in the mess below: |
| 2000 | * |
| 2001 | * - for_each_possible_cpu() is prohibitively expensive on machines with |
| 2002 | * serious number of cpus, therefore we need to take a distributed approach |
| 2003 | * to calculating nr_active. |
| 2004 | * |
| 2005 | * \Sum_i x_i(t) = \Sum_i x_i(t) - x_i(t_0) | x_i(t_0) := 0 |
| 2006 | * = \Sum_i { \Sum_j=1 x_i(t_j) - x_i(t_j-1) } |
| 2007 | * |
| 2008 | * So assuming nr_active := 0 when we start out -- true per definition, we |
| 2009 | * can simply take per-cpu deltas and fold those into a global accumulate |
| 2010 | * to obtain the same result. See calc_load_fold_active(). |
| 2011 | * |
| 2012 | * Furthermore, in order to avoid synchronizing all per-cpu delta folding |
| 2013 | * across the machine, we assume 10 ticks is sufficient time for every |
| 2014 | * cpu to have completed this task. |
| 2015 | * |
| 2016 | * This places an upper-bound on the IRQ-off latency of the machine. Then |
| 2017 | * again, being late doesn't loose the delta, just wrecks the sample. |
| 2018 | * |
| 2019 | * - cpu_rq()->nr_uninterruptible isn't accurately tracked per-cpu because |
| 2020 | * this would add another cross-cpu cacheline miss and atomic operation |
| 2021 | * to the wakeup path. Instead we increment on whatever cpu the task ran |
| 2022 | * when it went into uninterruptible state and decrement on whatever cpu |
| 2023 | * did the wakeup. This means that only the sum of nr_uninterruptible over |
| 2024 | * all cpus yields the correct result. |
| 2025 | * |
| 2026 | * This covers the NO_HZ=n code, for extra head-aches, see the comment below. |
| 2027 | */ |
| 2028 | |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 2029 | /* Variables and functions for calc_load */ |
| 2030 | static atomic_long_t calc_load_tasks; |
| 2031 | static unsigned long calc_load_update; |
| 2032 | unsigned long avenrun[3]; |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2033 | EXPORT_SYMBOL(avenrun); /* should be removed */ |
| 2034 | |
| 2035 | /** |
| 2036 | * get_avenrun - get the load average array |
| 2037 | * @loads: pointer to dest load array |
| 2038 | * @offset: offset to add |
| 2039 | * @shift: shift count to shift the result left |
| 2040 | * |
| 2041 | * These values are estimates at best, so no need for locking. |
| 2042 | */ |
| 2043 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) |
| 2044 | { |
| 2045 | loads[0] = (avenrun[0] + offset) << shift; |
| 2046 | loads[1] = (avenrun[1] + offset) << shift; |
| 2047 | loads[2] = (avenrun[2] + offset) << shift; |
| 2048 | } |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 2049 | |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2050 | static long calc_load_fold_active(struct rq *this_rq) |
| 2051 | { |
| 2052 | long nr_active, delta = 0; |
| 2053 | |
| 2054 | nr_active = this_rq->nr_running; |
| 2055 | nr_active += (long) this_rq->nr_uninterruptible; |
| 2056 | |
| 2057 | if (nr_active != this_rq->calc_load_active) { |
| 2058 | delta = nr_active - this_rq->calc_load_active; |
| 2059 | this_rq->calc_load_active = nr_active; |
| 2060 | } |
| 2061 | |
| 2062 | return delta; |
| 2063 | } |
| 2064 | |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2065 | /* |
| 2066 | * a1 = a0 * e + a * (1 - e) |
| 2067 | */ |
Peter Zijlstra | 0f004f5 | 2010-11-30 19:48:45 +0100 | [diff] [blame] | 2068 | static unsigned long |
| 2069 | calc_load(unsigned long load, unsigned long exp, unsigned long active) |
| 2070 | { |
| 2071 | load *= exp; |
| 2072 | load += active * (FIXED_1 - exp); |
| 2073 | load += 1UL << (FSHIFT - 1); |
| 2074 | return load >> FSHIFT; |
| 2075 | } |
| 2076 | |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2077 | #ifdef CONFIG_NO_HZ |
| 2078 | /* |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2079 | * Handle NO_HZ for the global load-average. |
| 2080 | * |
| 2081 | * Since the above described distributed algorithm to compute the global |
| 2082 | * load-average relies on per-cpu sampling from the tick, it is affected by |
| 2083 | * NO_HZ. |
| 2084 | * |
| 2085 | * The basic idea is to fold the nr_active delta into a global idle-delta upon |
| 2086 | * entering NO_HZ state such that we can include this as an 'extra' cpu delta |
| 2087 | * when we read the global state. |
| 2088 | * |
| 2089 | * Obviously reality has to ruin such a delightfully simple scheme: |
| 2090 | * |
| 2091 | * - When we go NO_HZ idle during the window, we can negate our sample |
| 2092 | * contribution, causing under-accounting. |
| 2093 | * |
| 2094 | * We avoid this by keeping two idle-delta counters and flipping them |
| 2095 | * when the window starts, thus separating old and new NO_HZ load. |
| 2096 | * |
| 2097 | * The only trick is the slight shift in index flip for read vs write. |
| 2098 | * |
| 2099 | * 0s 5s 10s 15s |
| 2100 | * +10 +10 +10 +10 |
| 2101 | * |-|-----------|-|-----------|-|-----------|-| |
| 2102 | * r:0 0 1 1 0 0 1 1 0 |
| 2103 | * w:0 1 1 0 0 1 1 0 0 |
| 2104 | * |
| 2105 | * This ensures we'll fold the old idle contribution in this window while |
| 2106 | * accumlating the new one. |
| 2107 | * |
| 2108 | * - When we wake up from NO_HZ idle during the window, we push up our |
| 2109 | * contribution, since we effectively move our sample point to a known |
| 2110 | * busy state. |
| 2111 | * |
| 2112 | * This is solved by pushing the window forward, and thus skipping the |
| 2113 | * sample, for this cpu (effectively using the idle-delta for this cpu which |
| 2114 | * was in effect at the time the window opened). This also solves the issue |
| 2115 | * of having to deal with a cpu having been in NOHZ idle for multiple |
| 2116 | * LOAD_FREQ intervals. |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2117 | * |
| 2118 | * When making the ILB scale, we should try to pull this in as well. |
| 2119 | */ |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2120 | static atomic_long_t calc_load_idle[2]; |
| 2121 | static int calc_load_idx; |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2122 | |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2123 | static inline int calc_load_write_idx(void) |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2124 | { |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2125 | int idx = calc_load_idx; |
| 2126 | |
| 2127 | /* |
| 2128 | * See calc_global_nohz(), if we observe the new index, we also |
| 2129 | * need to observe the new update time. |
| 2130 | */ |
| 2131 | smp_rmb(); |
| 2132 | |
| 2133 | /* |
| 2134 | * If the folding window started, make sure we start writing in the |
| 2135 | * next idle-delta. |
| 2136 | */ |
| 2137 | if (!time_before(jiffies, calc_load_update)) |
| 2138 | idx++; |
| 2139 | |
| 2140 | return idx & 1; |
| 2141 | } |
| 2142 | |
| 2143 | static inline int calc_load_read_idx(void) |
| 2144 | { |
| 2145 | return calc_load_idx & 1; |
| 2146 | } |
| 2147 | |
| 2148 | void calc_load_enter_idle(void) |
| 2149 | { |
| 2150 | struct rq *this_rq = this_rq(); |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2151 | long delta; |
| 2152 | |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2153 | /* |
| 2154 | * We're going into NOHZ mode, if there's any pending delta, fold it |
| 2155 | * into the pending idle delta. |
| 2156 | */ |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2157 | delta = calc_load_fold_active(this_rq); |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2158 | if (delta) { |
| 2159 | int idx = calc_load_write_idx(); |
| 2160 | atomic_long_add(delta, &calc_load_idle[idx]); |
| 2161 | } |
| 2162 | } |
| 2163 | |
| 2164 | void calc_load_exit_idle(void) |
| 2165 | { |
| 2166 | struct rq *this_rq = this_rq(); |
| 2167 | |
| 2168 | /* |
| 2169 | * If we're still before the sample window, we're done. |
| 2170 | */ |
| 2171 | if (time_before(jiffies, this_rq->calc_load_update)) |
| 2172 | return; |
| 2173 | |
| 2174 | /* |
| 2175 | * We woke inside or after the sample window, this means we're already |
| 2176 | * accounted through the nohz accounting, so skip the entire deal and |
| 2177 | * sync up for the next window. |
| 2178 | */ |
| 2179 | this_rq->calc_load_update = calc_load_update; |
| 2180 | if (time_before(jiffies, this_rq->calc_load_update + 10)) |
| 2181 | this_rq->calc_load_update += LOAD_FREQ; |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2182 | } |
| 2183 | |
| 2184 | static long calc_load_fold_idle(void) |
| 2185 | { |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2186 | int idx = calc_load_read_idx(); |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2187 | long delta = 0; |
| 2188 | |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2189 | if (atomic_long_read(&calc_load_idle[idx])) |
| 2190 | delta = atomic_long_xchg(&calc_load_idle[idx], 0); |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2191 | |
| 2192 | return delta; |
| 2193 | } |
Peter Zijlstra | 0f004f5 | 2010-11-30 19:48:45 +0100 | [diff] [blame] | 2194 | |
| 2195 | /** |
| 2196 | * fixed_power_int - compute: x^n, in O(log n) time |
| 2197 | * |
| 2198 | * @x: base of the power |
| 2199 | * @frac_bits: fractional bits of @x |
| 2200 | * @n: power to raise @x to. |
| 2201 | * |
| 2202 | * By exploiting the relation between the definition of the natural power |
| 2203 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and |
| 2204 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, |
| 2205 | * (where: n_i \elem {0, 1}, the binary vector representing n), |
| 2206 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is |
| 2207 | * of course trivially computable in O(log_2 n), the length of our binary |
| 2208 | * vector. |
| 2209 | */ |
| 2210 | static unsigned long |
| 2211 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) |
| 2212 | { |
| 2213 | unsigned long result = 1UL << frac_bits; |
| 2214 | |
| 2215 | if (n) for (;;) { |
| 2216 | if (n & 1) { |
| 2217 | result *= x; |
| 2218 | result += 1UL << (frac_bits - 1); |
| 2219 | result >>= frac_bits; |
| 2220 | } |
| 2221 | n >>= 1; |
| 2222 | if (!n) |
| 2223 | break; |
| 2224 | x *= x; |
| 2225 | x += 1UL << (frac_bits - 1); |
| 2226 | x >>= frac_bits; |
| 2227 | } |
| 2228 | |
| 2229 | return result; |
| 2230 | } |
| 2231 | |
| 2232 | /* |
| 2233 | * a1 = a0 * e + a * (1 - e) |
| 2234 | * |
| 2235 | * a2 = a1 * e + a * (1 - e) |
| 2236 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) |
| 2237 | * = a0 * e^2 + a * (1 - e) * (1 + e) |
| 2238 | * |
| 2239 | * a3 = a2 * e + a * (1 - e) |
| 2240 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) |
| 2241 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) |
| 2242 | * |
| 2243 | * ... |
| 2244 | * |
| 2245 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] |
| 2246 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) |
| 2247 | * = a0 * e^n + a * (1 - e^n) |
| 2248 | * |
| 2249 | * [1] application of the geometric series: |
| 2250 | * |
| 2251 | * n 1 - x^(n+1) |
| 2252 | * S_n := \Sum x^i = ------------- |
| 2253 | * i=0 1 - x |
| 2254 | */ |
| 2255 | static unsigned long |
| 2256 | calc_load_n(unsigned long load, unsigned long exp, |
| 2257 | unsigned long active, unsigned int n) |
| 2258 | { |
| 2259 | |
| 2260 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); |
| 2261 | } |
| 2262 | |
| 2263 | /* |
| 2264 | * NO_HZ can leave us missing all per-cpu ticks calling |
| 2265 | * calc_load_account_active(), but since an idle CPU folds its delta into |
| 2266 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold |
| 2267 | * in the pending idle delta if our idle period crossed a load cycle boundary. |
| 2268 | * |
| 2269 | * Once we've updated the global active value, we need to apply the exponential |
| 2270 | * weights adjusted to the number of cycles missed. |
| 2271 | */ |
Peter Zijlstra | c308b56 | 2012-03-01 15:04:46 +0100 | [diff] [blame] | 2272 | static void calc_global_nohz(void) |
Peter Zijlstra | 0f004f5 | 2010-11-30 19:48:45 +0100 | [diff] [blame] | 2273 | { |
| 2274 | long delta, active, n; |
| 2275 | |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2276 | if (!time_before(jiffies, calc_load_update + 10)) { |
| 2277 | /* |
| 2278 | * Catch-up, fold however many we are behind still |
| 2279 | */ |
| 2280 | delta = jiffies - calc_load_update - 10; |
| 2281 | n = 1 + (delta / LOAD_FREQ); |
| 2282 | |
| 2283 | active = atomic_long_read(&calc_load_tasks); |
| 2284 | active = active > 0 ? active * FIXED_1 : 0; |
| 2285 | |
| 2286 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); |
| 2287 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); |
| 2288 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); |
| 2289 | |
| 2290 | calc_load_update += n * LOAD_FREQ; |
| 2291 | } |
Peter Zijlstra | 0f004f5 | 2010-11-30 19:48:45 +0100 | [diff] [blame] | 2292 | |
| 2293 | /* |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2294 | * Flip the idle index... |
| 2295 | * |
| 2296 | * Make sure we first write the new time then flip the index, so that |
| 2297 | * calc_load_write_idx() will see the new time when it reads the new |
| 2298 | * index, this avoids a double flip messing things up. |
Peter Zijlstra | 0f004f5 | 2010-11-30 19:48:45 +0100 | [diff] [blame] | 2299 | */ |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2300 | smp_wmb(); |
| 2301 | calc_load_idx++; |
Peter Zijlstra | 0f004f5 | 2010-11-30 19:48:45 +0100 | [diff] [blame] | 2302 | } |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2303 | #else /* !CONFIG_NO_HZ */ |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2304 | |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2305 | static inline long calc_load_fold_idle(void) { return 0; } |
| 2306 | static inline void calc_global_nohz(void) { } |
Peter Zijlstra | 0f004f5 | 2010-11-30 19:48:45 +0100 | [diff] [blame] | 2307 | |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2308 | #endif /* CONFIG_NO_HZ */ |
Thomas Gleixner | 2d02494 | 2009-05-02 20:08:52 +0200 | [diff] [blame] | 2309 | |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 2310 | /* |
| 2311 | * calc_load - update the avenrun load estimates 10 ticks after the |
| 2312 | * CPUs have updated calc_load_tasks. |
| 2313 | */ |
Peter Zijlstra | 0f004f5 | 2010-11-30 19:48:45 +0100 | [diff] [blame] | 2314 | void calc_global_load(unsigned long ticks) |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 2315 | { |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2316 | long active, delta; |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 2317 | |
Peter Zijlstra | 0f004f5 | 2010-11-30 19:48:45 +0100 | [diff] [blame] | 2318 | if (time_before(jiffies, calc_load_update + 10)) |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 2319 | return; |
| 2320 | |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2321 | /* |
| 2322 | * Fold the 'old' idle-delta to include all NO_HZ cpus. |
| 2323 | */ |
| 2324 | delta = calc_load_fold_idle(); |
| 2325 | if (delta) |
| 2326 | atomic_long_add(delta, &calc_load_tasks); |
| 2327 | |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 2328 | active = atomic_long_read(&calc_load_tasks); |
| 2329 | active = active > 0 ? active * FIXED_1 : 0; |
| 2330 | |
| 2331 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
| 2332 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); |
| 2333 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); |
| 2334 | |
| 2335 | calc_load_update += LOAD_FREQ; |
Peter Zijlstra | c308b56 | 2012-03-01 15:04:46 +0100 | [diff] [blame] | 2336 | |
| 2337 | /* |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2338 | * In case we idled for multiple LOAD_FREQ intervals, catch up in bulk. |
Peter Zijlstra | c308b56 | 2012-03-01 15:04:46 +0100 | [diff] [blame] | 2339 | */ |
| 2340 | calc_global_nohz(); |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 2341 | } |
| 2342 | |
| 2343 | /* |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2344 | * Called from update_cpu_load() to periodically update this CPU's |
| 2345 | * active count. |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 2346 | */ |
| 2347 | static void calc_load_account_active(struct rq *this_rq) |
| 2348 | { |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2349 | long delta; |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 2350 | |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2351 | if (time_before(jiffies, this_rq->calc_load_update)) |
| 2352 | return; |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 2353 | |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2354 | delta = calc_load_fold_active(this_rq); |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2355 | if (delta) |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 2356 | atomic_long_add(delta, &calc_load_tasks); |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2357 | |
| 2358 | this_rq->calc_load_update += LOAD_FREQ; |
Jack Steiner | db1b1fe | 2006-03-31 02:31:21 -0800 | [diff] [blame] | 2359 | } |
| 2360 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2361 | /* |
Peter Zijlstra | 5167e8d | 2012-06-22 15:52:09 +0200 | [diff] [blame] | 2362 | * End of global load-average stuff |
| 2363 | */ |
| 2364 | |
| 2365 | /* |
Venkatesh Pallipadi | fdf3e95 | 2010-05-17 18:14:43 -0700 | [diff] [blame] | 2366 | * The exact cpuload at various idx values, calculated at every tick would be |
| 2367 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load |
| 2368 | * |
| 2369 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called |
| 2370 | * on nth tick when cpu may be busy, then we have: |
| 2371 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load |
| 2372 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load |
| 2373 | * |
| 2374 | * decay_load_missed() below does efficient calculation of |
| 2375 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load |
| 2376 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load |
| 2377 | * |
| 2378 | * The calculation is approximated on a 128 point scale. |
| 2379 | * degrade_zero_ticks is the number of ticks after which load at any |
| 2380 | * particular idx is approximated to be zero. |
| 2381 | * degrade_factor is a precomputed table, a row for each load idx. |
| 2382 | * Each column corresponds to degradation factor for a power of two ticks, |
| 2383 | * based on 128 point scale. |
| 2384 | * Example: |
| 2385 | * row 2, col 3 (=12) says that the degradation at load idx 2 after |
| 2386 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). |
| 2387 | * |
| 2388 | * With this power of 2 load factors, we can degrade the load n times |
| 2389 | * by looking at 1 bits in n and doing as many mult/shift instead of |
| 2390 | * n mult/shifts needed by the exact degradation. |
| 2391 | */ |
| 2392 | #define DEGRADE_SHIFT 7 |
| 2393 | static const unsigned char |
| 2394 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; |
| 2395 | static const unsigned char |
| 2396 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { |
| 2397 | {0, 0, 0, 0, 0, 0, 0, 0}, |
| 2398 | {64, 32, 8, 0, 0, 0, 0, 0}, |
| 2399 | {96, 72, 40, 12, 1, 0, 0}, |
| 2400 | {112, 98, 75, 43, 15, 1, 0}, |
| 2401 | {120, 112, 98, 76, 45, 16, 2} }; |
| 2402 | |
| 2403 | /* |
| 2404 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog |
| 2405 | * would be when CPU is idle and so we just decay the old load without |
| 2406 | * adding any new load. |
| 2407 | */ |
| 2408 | static unsigned long |
| 2409 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) |
| 2410 | { |
| 2411 | int j = 0; |
| 2412 | |
| 2413 | if (!missed_updates) |
| 2414 | return load; |
| 2415 | |
| 2416 | if (missed_updates >= degrade_zero_ticks[idx]) |
| 2417 | return 0; |
| 2418 | |
| 2419 | if (idx == 1) |
| 2420 | return load >> missed_updates; |
| 2421 | |
| 2422 | while (missed_updates) { |
| 2423 | if (missed_updates % 2) |
| 2424 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; |
| 2425 | |
| 2426 | missed_updates >>= 1; |
| 2427 | j++; |
| 2428 | } |
| 2429 | return load; |
| 2430 | } |
| 2431 | |
| 2432 | /* |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2433 | * Update rq->cpu_load[] statistics. This function is usually called every |
Venkatesh Pallipadi | fdf3e95 | 2010-05-17 18:14:43 -0700 | [diff] [blame] | 2434 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
| 2435 | * every tick. We fix it up based on jiffies. |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2436 | */ |
Peter Zijlstra | 556061b | 2012-05-11 17:31:26 +0200 | [diff] [blame] | 2437 | static void __update_cpu_load(struct rq *this_rq, unsigned long this_load, |
| 2438 | unsigned long pending_updates) |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2439 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2440 | int i, scale; |
| 2441 | |
| 2442 | this_rq->nr_load_updates++; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2443 | |
| 2444 | /* Update our load: */ |
Venkatesh Pallipadi | fdf3e95 | 2010-05-17 18:14:43 -0700 | [diff] [blame] | 2445 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
| 2446 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2447 | unsigned long old_load, new_load; |
| 2448 | |
| 2449 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
| 2450 | |
| 2451 | old_load = this_rq->cpu_load[i]; |
Venkatesh Pallipadi | fdf3e95 | 2010-05-17 18:14:43 -0700 | [diff] [blame] | 2452 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2453 | new_load = this_load; |
Ingo Molnar | a25707f | 2007-10-15 17:00:03 +0200 | [diff] [blame] | 2454 | /* |
| 2455 | * Round up the averaging division if load is increasing. This |
| 2456 | * prevents us from getting stuck on 9 if the load is 10, for |
| 2457 | * example. |
| 2458 | */ |
| 2459 | if (new_load > old_load) |
Venkatesh Pallipadi | fdf3e95 | 2010-05-17 18:14:43 -0700 | [diff] [blame] | 2460 | new_load += scale - 1; |
| 2461 | |
| 2462 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2463 | } |
Suresh Siddha | da2b71e | 2010-08-23 13:42:51 -0700 | [diff] [blame] | 2464 | |
| 2465 | sched_avg_update(this_rq); |
Venkatesh Pallipadi | fdf3e95 | 2010-05-17 18:14:43 -0700 | [diff] [blame] | 2466 | } |
| 2467 | |
Peter Zijlstra | 5aaa0b7 | 2012-05-17 17:15:29 +0200 | [diff] [blame] | 2468 | #ifdef CONFIG_NO_HZ |
| 2469 | /* |
| 2470 | * There is no sane way to deal with nohz on smp when using jiffies because the |
| 2471 | * cpu doing the jiffies update might drift wrt the cpu doing the jiffy reading |
| 2472 | * causing off-by-one errors in observed deltas; {0,2} instead of {1,1}. |
| 2473 | * |
| 2474 | * Therefore we cannot use the delta approach from the regular tick since that |
| 2475 | * would seriously skew the load calculation. However we'll make do for those |
| 2476 | * updates happening while idle (nohz_idle_balance) or coming out of idle |
| 2477 | * (tick_nohz_idle_exit). |
| 2478 | * |
| 2479 | * This means we might still be one tick off for nohz periods. |
| 2480 | */ |
| 2481 | |
Peter Zijlstra | 556061b | 2012-05-11 17:31:26 +0200 | [diff] [blame] | 2482 | /* |
| 2483 | * Called from nohz_idle_balance() to update the load ratings before doing the |
| 2484 | * idle balance. |
| 2485 | */ |
| 2486 | void update_idle_cpu_load(struct rq *this_rq) |
| 2487 | { |
Peter Zijlstra | 5aaa0b7 | 2012-05-17 17:15:29 +0200 | [diff] [blame] | 2488 | unsigned long curr_jiffies = ACCESS_ONCE(jiffies); |
Peter Zijlstra | 556061b | 2012-05-11 17:31:26 +0200 | [diff] [blame] | 2489 | unsigned long load = this_rq->load.weight; |
| 2490 | unsigned long pending_updates; |
| 2491 | |
| 2492 | /* |
Peter Zijlstra | 5aaa0b7 | 2012-05-17 17:15:29 +0200 | [diff] [blame] | 2493 | * bail if there's load or we're actually up-to-date. |
Peter Zijlstra | 556061b | 2012-05-11 17:31:26 +0200 | [diff] [blame] | 2494 | */ |
| 2495 | if (load || curr_jiffies == this_rq->last_load_update_tick) |
| 2496 | return; |
| 2497 | |
| 2498 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; |
| 2499 | this_rq->last_load_update_tick = curr_jiffies; |
| 2500 | |
| 2501 | __update_cpu_load(this_rq, load, pending_updates); |
| 2502 | } |
| 2503 | |
| 2504 | /* |
Peter Zijlstra | 5aaa0b7 | 2012-05-17 17:15:29 +0200 | [diff] [blame] | 2505 | * Called from tick_nohz_idle_exit() -- try and fix up the ticks we missed. |
| 2506 | */ |
| 2507 | void update_cpu_load_nohz(void) |
| 2508 | { |
| 2509 | struct rq *this_rq = this_rq(); |
| 2510 | unsigned long curr_jiffies = ACCESS_ONCE(jiffies); |
| 2511 | unsigned long pending_updates; |
| 2512 | |
| 2513 | if (curr_jiffies == this_rq->last_load_update_tick) |
| 2514 | return; |
| 2515 | |
| 2516 | raw_spin_lock(&this_rq->lock); |
| 2517 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; |
| 2518 | if (pending_updates) { |
| 2519 | this_rq->last_load_update_tick = curr_jiffies; |
| 2520 | /* |
| 2521 | * We were idle, this means load 0, the current load might be |
| 2522 | * !0 due to remote wakeups and the sort. |
| 2523 | */ |
| 2524 | __update_cpu_load(this_rq, 0, pending_updates); |
| 2525 | } |
| 2526 | raw_spin_unlock(&this_rq->lock); |
| 2527 | } |
| 2528 | #endif /* CONFIG_NO_HZ */ |
| 2529 | |
| 2530 | /* |
Peter Zijlstra | 556061b | 2012-05-11 17:31:26 +0200 | [diff] [blame] | 2531 | * Called from scheduler_tick() |
| 2532 | */ |
Venkatesh Pallipadi | fdf3e95 | 2010-05-17 18:14:43 -0700 | [diff] [blame] | 2533 | static void update_cpu_load_active(struct rq *this_rq) |
| 2534 | { |
Peter Zijlstra | 556061b | 2012-05-11 17:31:26 +0200 | [diff] [blame] | 2535 | /* |
Peter Zijlstra | 5aaa0b7 | 2012-05-17 17:15:29 +0200 | [diff] [blame] | 2536 | * See the mess around update_idle_cpu_load() / update_cpu_load_nohz(). |
Peter Zijlstra | 556061b | 2012-05-11 17:31:26 +0200 | [diff] [blame] | 2537 | */ |
| 2538 | this_rq->last_load_update_tick = jiffies; |
| 2539 | __update_cpu_load(this_rq, this_rq->load.weight, 1); |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 2540 | |
Peter Zijlstra | 74f5187 | 2010-04-22 21:50:19 +0200 | [diff] [blame] | 2541 | calc_load_account_active(this_rq); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2542 | } |
| 2543 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2544 | #ifdef CONFIG_SMP |
| 2545 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2546 | /* |
Peter Zijlstra | 3802290 | 2009-12-16 18:04:37 +0100 | [diff] [blame] | 2547 | * sched_exec - execve() is a valuable balancing opportunity, because at |
| 2548 | * this point the task has the smallest effective memory and cache footprint. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2549 | */ |
Peter Zijlstra | 3802290 | 2009-12-16 18:04:37 +0100 | [diff] [blame] | 2550 | void sched_exec(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2551 | { |
Peter Zijlstra | 3802290 | 2009-12-16 18:04:37 +0100 | [diff] [blame] | 2552 | struct task_struct *p = current; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2553 | unsigned long flags; |
Peter Zijlstra | 0017d73 | 2010-03-24 18:34:10 +0100 | [diff] [blame] | 2554 | int dest_cpu; |
Peter Zijlstra | 3802290 | 2009-12-16 18:04:37 +0100 | [diff] [blame] | 2555 | |
Peter Zijlstra | 8f42ced | 2011-04-05 17:23:53 +0200 | [diff] [blame] | 2556 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
Peter Zijlstra | 7608dec | 2011-04-05 17:23:46 +0200 | [diff] [blame] | 2557 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
Peter Zijlstra | 0017d73 | 2010-03-24 18:34:10 +0100 | [diff] [blame] | 2558 | if (dest_cpu == smp_processor_id()) |
| 2559 | goto unlock; |
Peter Zijlstra | 3802290 | 2009-12-16 18:04:37 +0100 | [diff] [blame] | 2560 | |
Peter Zijlstra | 8f42ced | 2011-04-05 17:23:53 +0200 | [diff] [blame] | 2561 | if (likely(cpu_active(dest_cpu))) { |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 2562 | struct migration_arg arg = { p, dest_cpu }; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 2563 | |
Peter Zijlstra | 8f42ced | 2011-04-05 17:23:53 +0200 | [diff] [blame] | 2564 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
| 2565 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2566 | return; |
| 2567 | } |
Peter Zijlstra | 0017d73 | 2010-03-24 18:34:10 +0100 | [diff] [blame] | 2568 | unlock: |
Peter Zijlstra | 8f42ced | 2011-04-05 17:23:53 +0200 | [diff] [blame] | 2569 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2570 | } |
| 2571 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2572 | #endif |
| 2573 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2574 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
Glauber Costa | 3292beb | 2011-11-28 14:45:17 -0200 | [diff] [blame] | 2575 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2576 | |
| 2577 | EXPORT_PER_CPU_SYMBOL(kstat); |
Glauber Costa | 3292beb | 2011-11-28 14:45:17 -0200 | [diff] [blame] | 2578 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2579 | |
| 2580 | /* |
Hidetoshi Seto | c5f8d99 | 2009-03-31 16:56:03 +0900 | [diff] [blame] | 2581 | * Return any ns on the sched_clock that have not yet been accounted in |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 2582 | * @p in case that task is currently running. |
Hidetoshi Seto | c5f8d99 | 2009-03-31 16:56:03 +0900 | [diff] [blame] | 2583 | * |
| 2584 | * Called with task_rq_lock() held on @rq. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2585 | */ |
Hidetoshi Seto | c5f8d99 | 2009-03-31 16:56:03 +0900 | [diff] [blame] | 2586 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
| 2587 | { |
| 2588 | u64 ns = 0; |
| 2589 | |
| 2590 | if (task_current(rq, p)) { |
| 2591 | update_rq_clock(rq); |
Venkatesh Pallipadi | 305e683 | 2010-10-04 17:03:21 -0700 | [diff] [blame] | 2592 | ns = rq->clock_task - p->se.exec_start; |
Hidetoshi Seto | c5f8d99 | 2009-03-31 16:56:03 +0900 | [diff] [blame] | 2593 | if ((s64)ns < 0) |
| 2594 | ns = 0; |
| 2595 | } |
| 2596 | |
| 2597 | return ns; |
| 2598 | } |
| 2599 | |
Frank Mayhar | bb34d92 | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 2600 | unsigned long long task_delta_exec(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2601 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2602 | unsigned long flags; |
Ingo Molnar | 41b86e9 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 2603 | struct rq *rq; |
Frank Mayhar | bb34d92 | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 2604 | u64 ns = 0; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2605 | |
Ingo Molnar | 41b86e9 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 2606 | rq = task_rq_lock(p, &flags); |
Hidetoshi Seto | c5f8d99 | 2009-03-31 16:56:03 +0900 | [diff] [blame] | 2607 | ns = do_task_delta_exec(p, rq); |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 2608 | task_rq_unlock(rq, p, &flags); |
Ingo Molnar | 1508487 | 2008-09-30 08:28:17 +0200 | [diff] [blame] | 2609 | |
Hidetoshi Seto | c5f8d99 | 2009-03-31 16:56:03 +0900 | [diff] [blame] | 2610 | return ns; |
| 2611 | } |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 2612 | |
Hidetoshi Seto | c5f8d99 | 2009-03-31 16:56:03 +0900 | [diff] [blame] | 2613 | /* |
| 2614 | * Return accounted runtime for the task. |
| 2615 | * In case the task is currently running, return the runtime plus current's |
| 2616 | * pending runtime that have not been accounted yet. |
| 2617 | */ |
| 2618 | unsigned long long task_sched_runtime(struct task_struct *p) |
| 2619 | { |
| 2620 | unsigned long flags; |
| 2621 | struct rq *rq; |
| 2622 | u64 ns = 0; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2623 | |
Hidetoshi Seto | c5f8d99 | 2009-03-31 16:56:03 +0900 | [diff] [blame] | 2624 | rq = task_rq_lock(p, &flags); |
| 2625 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 2626 | task_rq_unlock(rq, p, &flags); |
Hidetoshi Seto | c5f8d99 | 2009-03-31 16:56:03 +0900 | [diff] [blame] | 2627 | |
| 2628 | return ns; |
| 2629 | } |
| 2630 | |
Balbir Singh | 4904862 | 2008-09-05 18:12:23 +0200 | [diff] [blame] | 2631 | /* |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 2632 | * This function gets called by the timer code, with HZ frequency. |
| 2633 | * We call it with interrupts disabled. |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 2634 | */ |
| 2635 | void scheduler_tick(void) |
| 2636 | { |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 2637 | int cpu = smp_processor_id(); |
| 2638 | struct rq *rq = cpu_rq(cpu); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2639 | struct task_struct *curr = rq->curr; |
Peter Zijlstra | 3e51f33 | 2008-05-03 18:29:28 +0200 | [diff] [blame] | 2640 | |
| 2641 | sched_clock_tick(); |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 2642 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 2643 | raw_spin_lock(&rq->lock); |
Peter Zijlstra | 3e51f33 | 2008-05-03 18:29:28 +0200 | [diff] [blame] | 2644 | update_rq_clock(rq); |
Venkatesh Pallipadi | fdf3e95 | 2010-05-17 18:14:43 -0700 | [diff] [blame] | 2645 | update_cpu_load_active(rq); |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 2646 | curr->sched_class->task_tick(rq, curr, 0); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 2647 | raw_spin_unlock(&rq->lock); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2648 | |
Peter Zijlstra | e9d2b06 | 2010-09-17 11:28:50 +0200 | [diff] [blame] | 2649 | perf_event_task_tick(); |
Peter Zijlstra | e220d2d | 2009-05-23 18:28:55 +0200 | [diff] [blame] | 2650 | |
Christoph Lameter | e418e1c | 2006-12-10 02:20:23 -0800 | [diff] [blame] | 2651 | #ifdef CONFIG_SMP |
Suresh Siddha | 6eb57e0 | 2011-10-03 15:09:01 -0700 | [diff] [blame] | 2652 | rq->idle_balance = idle_cpu(cpu); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2653 | trigger_load_balance(rq, cpu); |
Christoph Lameter | e418e1c | 2006-12-10 02:20:23 -0800 | [diff] [blame] | 2654 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2655 | } |
| 2656 | |
Lai Jiangshan | 132380a | 2009-04-02 14:18:25 +0800 | [diff] [blame] | 2657 | notrace unsigned long get_parent_ip(unsigned long addr) |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 2658 | { |
| 2659 | if (in_lock_functions(addr)) { |
| 2660 | addr = CALLER_ADDR2; |
| 2661 | if (in_lock_functions(addr)) |
| 2662 | addr = CALLER_ADDR3; |
| 2663 | } |
| 2664 | return addr; |
| 2665 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2666 | |
Steven Rostedt | 7e49fcc | 2009-01-22 19:01:40 -0500 | [diff] [blame] | 2667 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
| 2668 | defined(CONFIG_PREEMPT_TRACER)) |
| 2669 | |
Srinivasa Ds | 4362758 | 2008-02-23 15:24:04 -0800 | [diff] [blame] | 2670 | void __kprobes add_preempt_count(int val) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2671 | { |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 2672 | #ifdef CONFIG_DEBUG_PREEMPT |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2673 | /* |
| 2674 | * Underflow? |
| 2675 | */ |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 2676 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
| 2677 | return; |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 2678 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2679 | preempt_count() += val; |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 2680 | #ifdef CONFIG_DEBUG_PREEMPT |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2681 | /* |
| 2682 | * Spinlock count overflowing soon? |
| 2683 | */ |
Miguel Ojeda Sandonis | 33859f7 | 2006-12-10 02:20:38 -0800 | [diff] [blame] | 2684 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
| 2685 | PREEMPT_MASK - 10); |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 2686 | #endif |
| 2687 | if (preempt_count() == val) |
| 2688 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2689 | } |
| 2690 | EXPORT_SYMBOL(add_preempt_count); |
| 2691 | |
Srinivasa Ds | 4362758 | 2008-02-23 15:24:04 -0800 | [diff] [blame] | 2692 | void __kprobes sub_preempt_count(int val) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2693 | { |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 2694 | #ifdef CONFIG_DEBUG_PREEMPT |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2695 | /* |
| 2696 | * Underflow? |
| 2697 | */ |
Ingo Molnar | 01e3eb8 | 2009-01-12 13:00:50 +0100 | [diff] [blame] | 2698 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 2699 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2700 | /* |
| 2701 | * Is the spinlock portion underflowing? |
| 2702 | */ |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 2703 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
| 2704 | !(preempt_count() & PREEMPT_MASK))) |
| 2705 | return; |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 2706 | #endif |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 2707 | |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 2708 | if (preempt_count() == val) |
| 2709 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2710 | preempt_count() -= val; |
| 2711 | } |
| 2712 | EXPORT_SYMBOL(sub_preempt_count); |
| 2713 | |
| 2714 | #endif |
| 2715 | |
| 2716 | /* |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2717 | * Print scheduling while atomic bug: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2718 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2719 | static noinline void __schedule_bug(struct task_struct *prev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2720 | { |
Dave Jones | 664dfa6 | 2011-12-22 16:39:30 -0500 | [diff] [blame] | 2721 | if (oops_in_progress) |
| 2722 | return; |
| 2723 | |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 2724 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
| 2725 | prev->comm, prev->pid, preempt_count()); |
Satyam Sharma | 838225b | 2007-10-24 18:23:50 +0200 | [diff] [blame] | 2726 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2727 | debug_show_held_locks(prev); |
Arjan van de Ven | e21f5b1 | 2008-05-23 09:05:58 -0700 | [diff] [blame] | 2728 | print_modules(); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2729 | if (irqs_disabled()) |
| 2730 | print_irqtrace_events(prev); |
Stephen Boyd | 6135fc1 | 2012-03-28 17:10:47 -0700 | [diff] [blame] | 2731 | dump_stack(); |
Konstantin Khlebnikov | 1c2927f | 2012-05-10 16:20:04 +0400 | [diff] [blame] | 2732 | add_taint(TAINT_WARN); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2733 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2734 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2735 | /* |
| 2736 | * Various schedule()-time debugging checks and statistics: |
| 2737 | */ |
| 2738 | static inline void schedule_debug(struct task_struct *prev) |
| 2739 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2740 | /* |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 2741 | * Test if we are atomic. Since do_exit() needs to call into |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2742 | * schedule() atomically, we ignore that path for now. |
| 2743 | * Otherwise, whine if we are scheduling when we should not be. |
| 2744 | */ |
Roel Kluin | 3f33a7c | 2008-05-13 23:44:11 +0200 | [diff] [blame] | 2745 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2746 | __schedule_bug(prev); |
Paul E. McKenney | b3fbab0 | 2011-05-24 08:31:09 -0700 | [diff] [blame] | 2747 | rcu_sleep_check(); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2748 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2749 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
| 2750 | |
Ingo Molnar | 2d72376 | 2007-10-15 17:00:12 +0200 | [diff] [blame] | 2751 | schedstat_inc(this_rq(), sched_count); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2752 | } |
| 2753 | |
Peter Zijlstra | 6cecd08 | 2009-11-30 13:00:37 +0100 | [diff] [blame] | 2754 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
Mike Galbraith | df1c99d | 2009-03-10 19:08:11 +0100 | [diff] [blame] | 2755 | { |
Mike Galbraith | 61eadef | 2011-04-29 08:36:50 +0200 | [diff] [blame] | 2756 | if (prev->on_rq || rq->skip_clock_update < 0) |
Mike Galbraith | a64692a | 2010-03-11 17:16:20 +0100 | [diff] [blame] | 2757 | update_rq_clock(rq); |
Peter Zijlstra | 6cecd08 | 2009-11-30 13:00:37 +0100 | [diff] [blame] | 2758 | prev->sched_class->put_prev_task(rq, prev); |
Mike Galbraith | df1c99d | 2009-03-10 19:08:11 +0100 | [diff] [blame] | 2759 | } |
| 2760 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2761 | /* |
| 2762 | * Pick up the highest-prio task: |
| 2763 | */ |
| 2764 | static inline struct task_struct * |
Wang Chen | b67802e | 2009-03-02 13:55:26 +0800 | [diff] [blame] | 2765 | pick_next_task(struct rq *rq) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2766 | { |
Ingo Molnar | 5522d5d | 2007-10-15 17:00:12 +0200 | [diff] [blame] | 2767 | const struct sched_class *class; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2768 | struct task_struct *p; |
| 2769 | |
| 2770 | /* |
| 2771 | * Optimization: we know that if all tasks are in |
| 2772 | * the fair class we can call that function directly: |
| 2773 | */ |
Paul Turner | 953bfcd | 2011-07-21 09:43:27 -0700 | [diff] [blame] | 2774 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
Ingo Molnar | fb8d472 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 2775 | p = fair_sched_class.pick_next_task(rq); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2776 | if (likely(p)) |
| 2777 | return p; |
| 2778 | } |
| 2779 | |
Peter Zijlstra | 34f971f | 2010-09-22 13:53:15 +0200 | [diff] [blame] | 2780 | for_each_class(class) { |
Ingo Molnar | fb8d472 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 2781 | p = class->pick_next_task(rq); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2782 | if (p) |
| 2783 | return p; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2784 | } |
Peter Zijlstra | 34f971f | 2010-09-22 13:53:15 +0200 | [diff] [blame] | 2785 | |
| 2786 | BUG(); /* the idle class will always have a runnable task */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2787 | } |
| 2788 | |
| 2789 | /* |
Thomas Gleixner | c259e01 | 2011-06-22 19:47:00 +0200 | [diff] [blame] | 2790 | * __schedule() is the main scheduler function. |
Pekka Enberg | edde96e | 2012-08-04 11:49:47 +0300 | [diff] [blame] | 2791 | * |
| 2792 | * The main means of driving the scheduler and thus entering this function are: |
| 2793 | * |
| 2794 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. |
| 2795 | * |
| 2796 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return |
| 2797 | * paths. For example, see arch/x86/entry_64.S. |
| 2798 | * |
| 2799 | * To drive preemption between tasks, the scheduler sets the flag in timer |
| 2800 | * interrupt handler scheduler_tick(). |
| 2801 | * |
| 2802 | * 3. Wakeups don't really cause entry into schedule(). They add a |
| 2803 | * task to the run-queue and that's it. |
| 2804 | * |
| 2805 | * Now, if the new task added to the run-queue preempts the current |
| 2806 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets |
| 2807 | * called on the nearest possible occasion: |
| 2808 | * |
| 2809 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): |
| 2810 | * |
| 2811 | * - in syscall or exception context, at the next outmost |
| 2812 | * preempt_enable(). (this might be as soon as the wake_up()'s |
| 2813 | * spin_unlock()!) |
| 2814 | * |
| 2815 | * - in IRQ context, return from interrupt-handler to |
| 2816 | * preemptible context |
| 2817 | * |
| 2818 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) |
| 2819 | * then at the next: |
| 2820 | * |
| 2821 | * - cond_resched() call |
| 2822 | * - explicit schedule() call |
| 2823 | * - return from syscall or exception to user-space |
| 2824 | * - return from interrupt-handler to user-space |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2825 | */ |
Thomas Gleixner | c259e01 | 2011-06-22 19:47:00 +0200 | [diff] [blame] | 2826 | static void __sched __schedule(void) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2827 | { |
| 2828 | struct task_struct *prev, *next; |
Harvey Harrison | 67ca7bd | 2008-02-15 09:56:36 -0800 | [diff] [blame] | 2829 | unsigned long *switch_count; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2830 | struct rq *rq; |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 2831 | int cpu; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2832 | |
Peter Zijlstra | ff74334 | 2009-03-13 12:21:26 +0100 | [diff] [blame] | 2833 | need_resched: |
| 2834 | preempt_disable(); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2835 | cpu = smp_processor_id(); |
| 2836 | rq = cpu_rq(cpu); |
Paul E. McKenney | 25502a6 | 2010-04-01 17:37:01 -0700 | [diff] [blame] | 2837 | rcu_note_context_switch(cpu); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2838 | prev = rq->curr; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2839 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2840 | schedule_debug(prev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2841 | |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 2842 | if (sched_feat(HRTICK)) |
Mike Galbraith | f333fdc | 2008-05-12 21:20:55 +0200 | [diff] [blame] | 2843 | hrtick_clear(rq); |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 2844 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 2845 | raw_spin_lock_irq(&rq->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2846 | |
Oleg Nesterov | 246d86b | 2010-05-19 14:57:11 +0200 | [diff] [blame] | 2847 | switch_count = &prev->nivcsw; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2848 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2849 | if (unlikely(signal_pending_state(prev->state, prev))) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2850 | prev->state = TASK_RUNNING; |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2851 | } else { |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 2852 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
| 2853 | prev->on_rq = 0; |
| 2854 | |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2855 | /* |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 2856 | * If a worker went to sleep, notify and ask workqueue |
| 2857 | * whether it wants to wake up a task to maintain |
| 2858 | * concurrency. |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2859 | */ |
| 2860 | if (prev->flags & PF_WQ_WORKER) { |
| 2861 | struct task_struct *to_wakeup; |
| 2862 | |
| 2863 | to_wakeup = wq_worker_sleeping(prev, cpu); |
| 2864 | if (to_wakeup) |
| 2865 | try_to_wake_up_local(to_wakeup); |
| 2866 | } |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2867 | } |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2868 | switch_count = &prev->nvcsw; |
| 2869 | } |
| 2870 | |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 2871 | pre_schedule(rq, prev); |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2872 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2873 | if (unlikely(!rq->nr_running)) |
| 2874 | idle_balance(cpu, rq); |
| 2875 | |
Mike Galbraith | df1c99d | 2009-03-10 19:08:11 +0100 | [diff] [blame] | 2876 | put_prev_task(rq, prev); |
Wang Chen | b67802e | 2009-03-02 13:55:26 +0800 | [diff] [blame] | 2877 | next = pick_next_task(rq); |
Mike Galbraith | f26f9af | 2010-12-08 11:05:42 +0100 | [diff] [blame] | 2878 | clear_tsk_need_resched(prev); |
| 2879 | rq->skip_clock_update = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2880 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2881 | if (likely(prev != next)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2882 | rq->nr_switches++; |
| 2883 | rq->curr = next; |
| 2884 | ++*switch_count; |
| 2885 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2886 | context_switch(rq, prev, next); /* unlocks the rq */ |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 2887 | /* |
Oleg Nesterov | 246d86b | 2010-05-19 14:57:11 +0200 | [diff] [blame] | 2888 | * The context switch have flipped the stack from under us |
| 2889 | * and restored the local variables which were saved when |
| 2890 | * this task called schedule() in the past. prev == current |
| 2891 | * is still correct, but it can be moved to another cpu/rq. |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 2892 | */ |
| 2893 | cpu = smp_processor_id(); |
| 2894 | rq = cpu_rq(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2895 | } else |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 2896 | raw_spin_unlock_irq(&rq->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2897 | |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 2898 | post_schedule(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2899 | |
Thomas Gleixner | ba74c14 | 2011-03-21 13:32:17 +0100 | [diff] [blame] | 2900 | sched_preempt_enable_no_resched(); |
Peter Zijlstra | ff74334 | 2009-03-13 12:21:26 +0100 | [diff] [blame] | 2901 | if (need_resched()) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2902 | goto need_resched; |
| 2903 | } |
Thomas Gleixner | c259e01 | 2011-06-22 19:47:00 +0200 | [diff] [blame] | 2904 | |
Thomas Gleixner | 9c40cef | 2011-06-22 19:47:01 +0200 | [diff] [blame] | 2905 | static inline void sched_submit_work(struct task_struct *tsk) |
| 2906 | { |
Thomas Gleixner | 3c7d518 | 2011-07-17 20:46:52 +0200 | [diff] [blame] | 2907 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
Thomas Gleixner | 9c40cef | 2011-06-22 19:47:01 +0200 | [diff] [blame] | 2908 | return; |
| 2909 | /* |
| 2910 | * If we are going to sleep and we have plugged IO queued, |
| 2911 | * make sure to submit it to avoid deadlocks. |
| 2912 | */ |
| 2913 | if (blk_needs_flush_plug(tsk)) |
| 2914 | blk_schedule_flush_plug(tsk); |
| 2915 | } |
| 2916 | |
Simon Kirby | 6ebbe7a | 2011-09-22 17:03:46 -0700 | [diff] [blame] | 2917 | asmlinkage void __sched schedule(void) |
Thomas Gleixner | c259e01 | 2011-06-22 19:47:00 +0200 | [diff] [blame] | 2918 | { |
Thomas Gleixner | 9c40cef | 2011-06-22 19:47:01 +0200 | [diff] [blame] | 2919 | struct task_struct *tsk = current; |
| 2920 | |
| 2921 | sched_submit_work(tsk); |
Thomas Gleixner | c259e01 | 2011-06-22 19:47:00 +0200 | [diff] [blame] | 2922 | __schedule(); |
| 2923 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2924 | EXPORT_SYMBOL(schedule); |
| 2925 | |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 2926 | #ifdef CONFIG_CONTEXT_TRACKING |
Frederic Weisbecker | 20ab65e3 | 2012-07-11 20:26:37 +0200 | [diff] [blame] | 2927 | asmlinkage void __sched schedule_user(void) |
| 2928 | { |
| 2929 | /* |
| 2930 | * If we come here after a random call to set_need_resched(), |
| 2931 | * or we have been woken up remotely but the IPI has not yet arrived, |
| 2932 | * we haven't yet exited the RCU idle mode. Do it here manually until |
| 2933 | * we find a better solution. |
| 2934 | */ |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 2935 | user_exit(); |
Frederic Weisbecker | 20ab65e3 | 2012-07-11 20:26:37 +0200 | [diff] [blame] | 2936 | schedule(); |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 2937 | user_enter(); |
Frederic Weisbecker | 20ab65e3 | 2012-07-11 20:26:37 +0200 | [diff] [blame] | 2938 | } |
| 2939 | #endif |
| 2940 | |
Thomas Gleixner | c5491ea | 2011-03-21 12:09:35 +0100 | [diff] [blame] | 2941 | /** |
| 2942 | * schedule_preempt_disabled - called with preemption disabled |
| 2943 | * |
| 2944 | * Returns with preemption disabled. Note: preempt_count must be 1 |
| 2945 | */ |
| 2946 | void __sched schedule_preempt_disabled(void) |
| 2947 | { |
Thomas Gleixner | ba74c14 | 2011-03-21 13:32:17 +0100 | [diff] [blame] | 2948 | sched_preempt_enable_no_resched(); |
Thomas Gleixner | c5491ea | 2011-03-21 12:09:35 +0100 | [diff] [blame] | 2949 | schedule(); |
| 2950 | preempt_disable(); |
| 2951 | } |
| 2952 | |
Frederic Weisbecker | c08f782 | 2009-12-02 20:49:17 +0100 | [diff] [blame] | 2953 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
Peter Zijlstra | c6eb3dd | 2011-04-05 17:23:41 +0200 | [diff] [blame] | 2954 | |
| 2955 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
| 2956 | { |
Peter Zijlstra | c6eb3dd | 2011-04-05 17:23:41 +0200 | [diff] [blame] | 2957 | if (lock->owner != owner) |
Thomas Gleixner | 307bf98 | 2011-06-10 15:08:55 +0200 | [diff] [blame] | 2958 | return false; |
Peter Zijlstra | c6eb3dd | 2011-04-05 17:23:41 +0200 | [diff] [blame] | 2959 | |
| 2960 | /* |
| 2961 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
| 2962 | * lock->owner still matches owner, if that fails, owner might |
| 2963 | * point to free()d memory, if it still matches, the rcu_read_lock() |
| 2964 | * ensures the memory stays valid. |
| 2965 | */ |
| 2966 | barrier(); |
| 2967 | |
Thomas Gleixner | 307bf98 | 2011-06-10 15:08:55 +0200 | [diff] [blame] | 2968 | return owner->on_cpu; |
Peter Zijlstra | c6eb3dd | 2011-04-05 17:23:41 +0200 | [diff] [blame] | 2969 | } |
| 2970 | |
Peter Zijlstra | 0d66bf6 | 2009-01-12 14:01:47 +0100 | [diff] [blame] | 2971 | /* |
| 2972 | * Look out! "owner" is an entirely speculative pointer |
| 2973 | * access and not reliable. |
| 2974 | */ |
Peter Zijlstra | c6eb3dd | 2011-04-05 17:23:41 +0200 | [diff] [blame] | 2975 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) |
Peter Zijlstra | 0d66bf6 | 2009-01-12 14:01:47 +0100 | [diff] [blame] | 2976 | { |
Peter Zijlstra | 0d66bf6 | 2009-01-12 14:01:47 +0100 | [diff] [blame] | 2977 | if (!sched_feat(OWNER_SPIN)) |
| 2978 | return 0; |
| 2979 | |
Thomas Gleixner | 307bf98 | 2011-06-10 15:08:55 +0200 | [diff] [blame] | 2980 | rcu_read_lock(); |
Peter Zijlstra | c6eb3dd | 2011-04-05 17:23:41 +0200 | [diff] [blame] | 2981 | while (owner_running(lock, owner)) { |
| 2982 | if (need_resched()) |
Thomas Gleixner | 307bf98 | 2011-06-10 15:08:55 +0200 | [diff] [blame] | 2983 | break; |
Peter Zijlstra | 0d66bf6 | 2009-01-12 14:01:47 +0100 | [diff] [blame] | 2984 | |
Gerald Schaefer | 335d7af | 2010-11-22 15:47:36 +0100 | [diff] [blame] | 2985 | arch_mutex_cpu_relax(); |
Peter Zijlstra | 0d66bf6 | 2009-01-12 14:01:47 +0100 | [diff] [blame] | 2986 | } |
Thomas Gleixner | 307bf98 | 2011-06-10 15:08:55 +0200 | [diff] [blame] | 2987 | rcu_read_unlock(); |
Benjamin Herrenschmidt | 4b40221 | 2010-04-16 23:20:00 +0200 | [diff] [blame] | 2988 | |
Peter Zijlstra | c6eb3dd | 2011-04-05 17:23:41 +0200 | [diff] [blame] | 2989 | /* |
Thomas Gleixner | 307bf98 | 2011-06-10 15:08:55 +0200 | [diff] [blame] | 2990 | * We break out the loop above on need_resched() and when the |
| 2991 | * owner changed, which is a sign for heavy contention. Return |
| 2992 | * success only when lock->owner is NULL. |
Peter Zijlstra | c6eb3dd | 2011-04-05 17:23:41 +0200 | [diff] [blame] | 2993 | */ |
Thomas Gleixner | 307bf98 | 2011-06-10 15:08:55 +0200 | [diff] [blame] | 2994 | return lock->owner == NULL; |
Peter Zijlstra | 0d66bf6 | 2009-01-12 14:01:47 +0100 | [diff] [blame] | 2995 | } |
| 2996 | #endif |
| 2997 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2998 | #ifdef CONFIG_PREEMPT |
| 2999 | /* |
Andreas Mohr | 2ed6e34 | 2006-07-10 04:43:52 -0700 | [diff] [blame] | 3000 | * this is the entry point to schedule() from in-kernel preemption |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 3001 | * off of preempt_enable. Kernel preemptions off return from interrupt |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3002 | * occur there and call schedule directly. |
| 3003 | */ |
Steven Rostedt | d1f74e2 | 2010-06-02 21:52:29 -0400 | [diff] [blame] | 3004 | asmlinkage void __sched notrace preempt_schedule(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3005 | { |
| 3006 | struct thread_info *ti = current_thread_info(); |
Ingo Molnar | 6478d88 | 2008-01-25 21:08:33 +0100 | [diff] [blame] | 3007 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3008 | /* |
| 3009 | * If there is a non-zero preempt_count or interrupts are disabled, |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 3010 | * we do not want to preempt the current task. Just return.. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3011 | */ |
Nick Piggin | beed33a | 2006-10-11 01:21:52 -0700 | [diff] [blame] | 3012 | if (likely(ti->preempt_count || irqs_disabled())) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3013 | return; |
| 3014 | |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3015 | do { |
Steven Rostedt | d1f74e2 | 2010-06-02 21:52:29 -0400 | [diff] [blame] | 3016 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
Thomas Gleixner | c259e01 | 2011-06-22 19:47:00 +0200 | [diff] [blame] | 3017 | __schedule(); |
Steven Rostedt | d1f74e2 | 2010-06-02 21:52:29 -0400 | [diff] [blame] | 3018 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3019 | |
| 3020 | /* |
| 3021 | * Check again in case we missed a preemption opportunity |
| 3022 | * between schedule and now. |
| 3023 | */ |
| 3024 | barrier(); |
Lai Jiangshan | 5ed0cec | 2009-03-06 19:40:20 +0800 | [diff] [blame] | 3025 | } while (need_resched()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3026 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3027 | EXPORT_SYMBOL(preempt_schedule); |
| 3028 | |
| 3029 | /* |
Andreas Mohr | 2ed6e34 | 2006-07-10 04:43:52 -0700 | [diff] [blame] | 3030 | * this is the entry point to schedule() from kernel preemption |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3031 | * off of irq context. |
| 3032 | * Note, that this is called and return with irqs disabled. This will |
| 3033 | * protect us against recursive calling from irq. |
| 3034 | */ |
| 3035 | asmlinkage void __sched preempt_schedule_irq(void) |
| 3036 | { |
| 3037 | struct thread_info *ti = current_thread_info(); |
Ingo Molnar | 6478d88 | 2008-01-25 21:08:33 +0100 | [diff] [blame] | 3038 | |
Andreas Mohr | 2ed6e34 | 2006-07-10 04:43:52 -0700 | [diff] [blame] | 3039 | /* Catch callers which need to be fixed */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3040 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
| 3041 | |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 3042 | user_exit(); |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3043 | do { |
| 3044 | add_preempt_count(PREEMPT_ACTIVE); |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3045 | local_irq_enable(); |
Thomas Gleixner | c259e01 | 2011-06-22 19:47:00 +0200 | [diff] [blame] | 3046 | __schedule(); |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3047 | local_irq_disable(); |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3048 | sub_preempt_count(PREEMPT_ACTIVE); |
| 3049 | |
| 3050 | /* |
| 3051 | * Check again in case we missed a preemption opportunity |
| 3052 | * between schedule and now. |
| 3053 | */ |
| 3054 | barrier(); |
Lai Jiangshan | 5ed0cec | 2009-03-06 19:40:20 +0800 | [diff] [blame] | 3055 | } while (need_resched()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3056 | } |
| 3057 | |
| 3058 | #endif /* CONFIG_PREEMPT */ |
| 3059 | |
Peter Zijlstra | 63859d4 | 2009-09-15 19:14:42 +0200 | [diff] [blame] | 3060 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 3061 | void *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3062 | { |
Peter Zijlstra | 63859d4 | 2009-09-15 19:14:42 +0200 | [diff] [blame] | 3063 | return try_to_wake_up(curr->private, mode, wake_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3064 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3065 | EXPORT_SYMBOL(default_wake_function); |
| 3066 | |
| 3067 | /* |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 3068 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
| 3069 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3070 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
| 3071 | * |
| 3072 | * There are circumstances in which we can try to wake a task which has already |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 3073 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3074 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
| 3075 | */ |
Johannes Weiner | 78ddb08 | 2009-04-14 16:53:05 +0200 | [diff] [blame] | 3076 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
Peter Zijlstra | 63859d4 | 2009-09-15 19:14:42 +0200 | [diff] [blame] | 3077 | int nr_exclusive, int wake_flags, void *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3078 | { |
Matthias Kaehlcke | 2e45874 | 2007-10-15 17:00:02 +0200 | [diff] [blame] | 3079 | wait_queue_t *curr, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3080 | |
Matthias Kaehlcke | 2e45874 | 2007-10-15 17:00:02 +0200 | [diff] [blame] | 3081 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 3082 | unsigned flags = curr->flags; |
| 3083 | |
Peter Zijlstra | 63859d4 | 2009-09-15 19:14:42 +0200 | [diff] [blame] | 3084 | if (curr->func(curr, mode, wake_flags, key) && |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 3085 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3086 | break; |
| 3087 | } |
| 3088 | } |
| 3089 | |
| 3090 | /** |
| 3091 | * __wake_up - wake up threads blocked on a waitqueue. |
| 3092 | * @q: the waitqueue |
| 3093 | * @mode: which threads |
| 3094 | * @nr_exclusive: how many wake-one or wake-many threads to wake up |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 3095 | * @key: is directly passed to the wakeup function |
David Howells | 50fa610 | 2009-04-28 15:01:38 +0100 | [diff] [blame] | 3096 | * |
| 3097 | * It may be assumed that this function implies a write memory barrier before |
| 3098 | * changing the task state if and only if any tasks are woken up. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3099 | */ |
Harvey Harrison | 7ad5b3a | 2008-02-08 04:19:53 -0800 | [diff] [blame] | 3100 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 3101 | int nr_exclusive, void *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3102 | { |
| 3103 | unsigned long flags; |
| 3104 | |
| 3105 | spin_lock_irqsave(&q->lock, flags); |
| 3106 | __wake_up_common(q, mode, nr_exclusive, 0, key); |
| 3107 | spin_unlock_irqrestore(&q->lock, flags); |
| 3108 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3109 | EXPORT_SYMBOL(__wake_up); |
| 3110 | |
| 3111 | /* |
| 3112 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. |
| 3113 | */ |
Thomas Gleixner | 63b2001 | 2011-12-01 00:04:00 +0100 | [diff] [blame] | 3114 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3115 | { |
Thomas Gleixner | 63b2001 | 2011-12-01 00:04:00 +0100 | [diff] [blame] | 3116 | __wake_up_common(q, mode, nr, 0, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3117 | } |
Michal Nazarewicz | 22c43c8 | 2010-05-05 12:53:11 +0200 | [diff] [blame] | 3118 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3119 | |
Davide Libenzi | 4ede816 | 2009-03-31 15:24:20 -0700 | [diff] [blame] | 3120 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
| 3121 | { |
| 3122 | __wake_up_common(q, mode, 1, 0, key); |
| 3123 | } |
Trond Myklebust | bf294b4 | 2011-02-21 11:05:41 -0800 | [diff] [blame] | 3124 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
Davide Libenzi | 4ede816 | 2009-03-31 15:24:20 -0700 | [diff] [blame] | 3125 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3126 | /** |
Davide Libenzi | 4ede816 | 2009-03-31 15:24:20 -0700 | [diff] [blame] | 3127 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3128 | * @q: the waitqueue |
| 3129 | * @mode: which threads |
| 3130 | * @nr_exclusive: how many wake-one or wake-many threads to wake up |
Davide Libenzi | 4ede816 | 2009-03-31 15:24:20 -0700 | [diff] [blame] | 3131 | * @key: opaque value to be passed to wakeup targets |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3132 | * |
| 3133 | * The sync wakeup differs that the waker knows that it will schedule |
| 3134 | * away soon, so while the target thread will be woken up, it will not |
| 3135 | * be migrated to another CPU - ie. the two threads are 'synchronized' |
| 3136 | * with each other. This can prevent needless bouncing between CPUs. |
| 3137 | * |
| 3138 | * On UP it can prevent extra preemption. |
David Howells | 50fa610 | 2009-04-28 15:01:38 +0100 | [diff] [blame] | 3139 | * |
| 3140 | * It may be assumed that this function implies a write memory barrier before |
| 3141 | * changing the task state if and only if any tasks are woken up. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3142 | */ |
Davide Libenzi | 4ede816 | 2009-03-31 15:24:20 -0700 | [diff] [blame] | 3143 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
| 3144 | int nr_exclusive, void *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3145 | { |
| 3146 | unsigned long flags; |
Peter Zijlstra | 7d47872 | 2009-09-14 19:55:44 +0200 | [diff] [blame] | 3147 | int wake_flags = WF_SYNC; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3148 | |
| 3149 | if (unlikely(!q)) |
| 3150 | return; |
| 3151 | |
| 3152 | if (unlikely(!nr_exclusive)) |
Peter Zijlstra | 7d47872 | 2009-09-14 19:55:44 +0200 | [diff] [blame] | 3153 | wake_flags = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3154 | |
| 3155 | spin_lock_irqsave(&q->lock, flags); |
Peter Zijlstra | 7d47872 | 2009-09-14 19:55:44 +0200 | [diff] [blame] | 3156 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3157 | spin_unlock_irqrestore(&q->lock, flags); |
| 3158 | } |
Davide Libenzi | 4ede816 | 2009-03-31 15:24:20 -0700 | [diff] [blame] | 3159 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
| 3160 | |
| 3161 | /* |
| 3162 | * __wake_up_sync - see __wake_up_sync_key() |
| 3163 | */ |
| 3164 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
| 3165 | { |
| 3166 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); |
| 3167 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3168 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
| 3169 | |
Kevin Diggs | 65eb3dc | 2008-08-26 10:26:54 +0200 | [diff] [blame] | 3170 | /** |
| 3171 | * complete: - signals a single thread waiting on this completion |
| 3172 | * @x: holds the state of this particular completion |
| 3173 | * |
| 3174 | * This will wake up a single thread waiting on this completion. Threads will be |
| 3175 | * awakened in the same order in which they were queued. |
| 3176 | * |
| 3177 | * See also complete_all(), wait_for_completion() and related routines. |
David Howells | 50fa610 | 2009-04-28 15:01:38 +0100 | [diff] [blame] | 3178 | * |
| 3179 | * It may be assumed that this function implies a write memory barrier before |
| 3180 | * changing the task state if and only if any tasks are woken up. |
Kevin Diggs | 65eb3dc | 2008-08-26 10:26:54 +0200 | [diff] [blame] | 3181 | */ |
Ingo Molnar | b15136e | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 3182 | void complete(struct completion *x) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3183 | { |
| 3184 | unsigned long flags; |
| 3185 | |
| 3186 | spin_lock_irqsave(&x->wait.lock, flags); |
| 3187 | x->done++; |
Matthew Wilcox | d9514f6 | 2007-12-06 11:07:07 -0500 | [diff] [blame] | 3188 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3189 | spin_unlock_irqrestore(&x->wait.lock, flags); |
| 3190 | } |
| 3191 | EXPORT_SYMBOL(complete); |
| 3192 | |
Kevin Diggs | 65eb3dc | 2008-08-26 10:26:54 +0200 | [diff] [blame] | 3193 | /** |
| 3194 | * complete_all: - signals all threads waiting on this completion |
| 3195 | * @x: holds the state of this particular completion |
| 3196 | * |
| 3197 | * This will wake up all threads waiting on this particular completion event. |
David Howells | 50fa610 | 2009-04-28 15:01:38 +0100 | [diff] [blame] | 3198 | * |
| 3199 | * It may be assumed that this function implies a write memory barrier before |
| 3200 | * changing the task state if and only if any tasks are woken up. |
Kevin Diggs | 65eb3dc | 2008-08-26 10:26:54 +0200 | [diff] [blame] | 3201 | */ |
Ingo Molnar | b15136e | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 3202 | void complete_all(struct completion *x) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3203 | { |
| 3204 | unsigned long flags; |
| 3205 | |
| 3206 | spin_lock_irqsave(&x->wait.lock, flags); |
| 3207 | x->done += UINT_MAX/2; |
Matthew Wilcox | d9514f6 | 2007-12-06 11:07:07 -0500 | [diff] [blame] | 3208 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3209 | spin_unlock_irqrestore(&x->wait.lock, flags); |
| 3210 | } |
| 3211 | EXPORT_SYMBOL(complete_all); |
| 3212 | |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3213 | static inline long __sched |
| 3214 | do_wait_for_common(struct completion *x, long timeout, int state) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3215 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3216 | if (!x->done) { |
| 3217 | DECLARE_WAITQUEUE(wait, current); |
| 3218 | |
Changli Gao | a93d2f1 | 2010-05-07 14:33:26 +0800 | [diff] [blame] | 3219 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3220 | do { |
Oleg Nesterov | 94d3d82 | 2008-08-20 16:54:41 -0700 | [diff] [blame] | 3221 | if (signal_pending_state(state, current)) { |
Oleg Nesterov | ea71a54 | 2008-06-20 18:32:20 +0400 | [diff] [blame] | 3222 | timeout = -ERESTARTSYS; |
| 3223 | break; |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3224 | } |
| 3225 | __set_current_state(state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3226 | spin_unlock_irq(&x->wait.lock); |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3227 | timeout = schedule_timeout(timeout); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3228 | spin_lock_irq(&x->wait.lock); |
Oleg Nesterov | ea71a54 | 2008-06-20 18:32:20 +0400 | [diff] [blame] | 3229 | } while (!x->done && timeout); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3230 | __remove_wait_queue(&x->wait, &wait); |
Oleg Nesterov | ea71a54 | 2008-06-20 18:32:20 +0400 | [diff] [blame] | 3231 | if (!x->done) |
| 3232 | return timeout; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3233 | } |
| 3234 | x->done--; |
Oleg Nesterov | ea71a54 | 2008-06-20 18:32:20 +0400 | [diff] [blame] | 3235 | return timeout ?: 1; |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3236 | } |
| 3237 | |
| 3238 | static long __sched |
| 3239 | wait_for_common(struct completion *x, long timeout, int state) |
| 3240 | { |
| 3241 | might_sleep(); |
| 3242 | |
| 3243 | spin_lock_irq(&x->wait.lock); |
| 3244 | timeout = do_wait_for_common(x, timeout, state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3245 | spin_unlock_irq(&x->wait.lock); |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3246 | return timeout; |
| 3247 | } |
| 3248 | |
Kevin Diggs | 65eb3dc | 2008-08-26 10:26:54 +0200 | [diff] [blame] | 3249 | /** |
| 3250 | * wait_for_completion: - waits for completion of a task |
| 3251 | * @x: holds the state of this particular completion |
| 3252 | * |
| 3253 | * This waits to be signaled for completion of a specific task. It is NOT |
| 3254 | * interruptible and there is no timeout. |
| 3255 | * |
| 3256 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout |
| 3257 | * and interrupt capability. Also see complete(). |
| 3258 | */ |
Ingo Molnar | b15136e | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 3259 | void __sched wait_for_completion(struct completion *x) |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3260 | { |
| 3261 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3262 | } |
| 3263 | EXPORT_SYMBOL(wait_for_completion); |
| 3264 | |
Kevin Diggs | 65eb3dc | 2008-08-26 10:26:54 +0200 | [diff] [blame] | 3265 | /** |
| 3266 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) |
| 3267 | * @x: holds the state of this particular completion |
| 3268 | * @timeout: timeout value in jiffies |
| 3269 | * |
| 3270 | * This waits for either a completion of a specific task to be signaled or for a |
| 3271 | * specified timeout to expire. The timeout is in jiffies. It is not |
| 3272 | * interruptible. |
J. Bruce Fields | c6dc7f0 | 2011-10-06 15:22:46 -0400 | [diff] [blame] | 3273 | * |
| 3274 | * The return value is 0 if timed out, and positive (at least 1, or number of |
| 3275 | * jiffies left till timeout) if completed. |
Kevin Diggs | 65eb3dc | 2008-08-26 10:26:54 +0200 | [diff] [blame] | 3276 | */ |
Ingo Molnar | b15136e | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 3277 | unsigned long __sched |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3278 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
| 3279 | { |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3280 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3281 | } |
| 3282 | EXPORT_SYMBOL(wait_for_completion_timeout); |
| 3283 | |
Kevin Diggs | 65eb3dc | 2008-08-26 10:26:54 +0200 | [diff] [blame] | 3284 | /** |
| 3285 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) |
| 3286 | * @x: holds the state of this particular completion |
| 3287 | * |
| 3288 | * This waits for completion of a specific task to be signaled. It is |
| 3289 | * interruptible. |
J. Bruce Fields | c6dc7f0 | 2011-10-06 15:22:46 -0400 | [diff] [blame] | 3290 | * |
| 3291 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. |
Kevin Diggs | 65eb3dc | 2008-08-26 10:26:54 +0200 | [diff] [blame] | 3292 | */ |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3293 | int __sched wait_for_completion_interruptible(struct completion *x) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3294 | { |
Andi Kleen | 51e9799 | 2007-10-18 21:32:55 +0200 | [diff] [blame] | 3295 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
| 3296 | if (t == -ERESTARTSYS) |
| 3297 | return t; |
| 3298 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3299 | } |
| 3300 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
| 3301 | |
Kevin Diggs | 65eb3dc | 2008-08-26 10:26:54 +0200 | [diff] [blame] | 3302 | /** |
| 3303 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) |
| 3304 | * @x: holds the state of this particular completion |
| 3305 | * @timeout: timeout value in jiffies |
| 3306 | * |
| 3307 | * This waits for either a completion of a specific task to be signaled or for a |
| 3308 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. |
J. Bruce Fields | c6dc7f0 | 2011-10-06 15:22:46 -0400 | [diff] [blame] | 3309 | * |
| 3310 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, |
| 3311 | * positive (at least 1, or number of jiffies left till timeout) if completed. |
Kevin Diggs | 65eb3dc | 2008-08-26 10:26:54 +0200 | [diff] [blame] | 3312 | */ |
NeilBrown | 6bf4123 | 2011-01-05 12:50:16 +1100 | [diff] [blame] | 3313 | long __sched |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3314 | wait_for_completion_interruptible_timeout(struct completion *x, |
| 3315 | unsigned long timeout) |
| 3316 | { |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3317 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3318 | } |
| 3319 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
| 3320 | |
Kevin Diggs | 65eb3dc | 2008-08-26 10:26:54 +0200 | [diff] [blame] | 3321 | /** |
| 3322 | * wait_for_completion_killable: - waits for completion of a task (killable) |
| 3323 | * @x: holds the state of this particular completion |
| 3324 | * |
| 3325 | * This waits to be signaled for completion of a specific task. It can be |
| 3326 | * interrupted by a kill signal. |
J. Bruce Fields | c6dc7f0 | 2011-10-06 15:22:46 -0400 | [diff] [blame] | 3327 | * |
| 3328 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. |
Kevin Diggs | 65eb3dc | 2008-08-26 10:26:54 +0200 | [diff] [blame] | 3329 | */ |
Matthew Wilcox | 009e577 | 2007-12-06 12:29:54 -0500 | [diff] [blame] | 3330 | int __sched wait_for_completion_killable(struct completion *x) |
| 3331 | { |
| 3332 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); |
| 3333 | if (t == -ERESTARTSYS) |
| 3334 | return t; |
| 3335 | return 0; |
| 3336 | } |
| 3337 | EXPORT_SYMBOL(wait_for_completion_killable); |
| 3338 | |
Dave Chinner | be4de35 | 2008-08-15 00:40:44 -0700 | [diff] [blame] | 3339 | /** |
Sage Weil | 0aa12fb | 2010-05-29 09:12:30 -0700 | [diff] [blame] | 3340 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) |
| 3341 | * @x: holds the state of this particular completion |
| 3342 | * @timeout: timeout value in jiffies |
| 3343 | * |
| 3344 | * This waits for either a completion of a specific task to be |
| 3345 | * signaled or for a specified timeout to expire. It can be |
| 3346 | * interrupted by a kill signal. The timeout is in jiffies. |
J. Bruce Fields | c6dc7f0 | 2011-10-06 15:22:46 -0400 | [diff] [blame] | 3347 | * |
| 3348 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, |
| 3349 | * positive (at least 1, or number of jiffies left till timeout) if completed. |
Sage Weil | 0aa12fb | 2010-05-29 09:12:30 -0700 | [diff] [blame] | 3350 | */ |
NeilBrown | 6bf4123 | 2011-01-05 12:50:16 +1100 | [diff] [blame] | 3351 | long __sched |
Sage Weil | 0aa12fb | 2010-05-29 09:12:30 -0700 | [diff] [blame] | 3352 | wait_for_completion_killable_timeout(struct completion *x, |
| 3353 | unsigned long timeout) |
| 3354 | { |
| 3355 | return wait_for_common(x, timeout, TASK_KILLABLE); |
| 3356 | } |
| 3357 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); |
| 3358 | |
| 3359 | /** |
Dave Chinner | be4de35 | 2008-08-15 00:40:44 -0700 | [diff] [blame] | 3360 | * try_wait_for_completion - try to decrement a completion without blocking |
| 3361 | * @x: completion structure |
| 3362 | * |
| 3363 | * Returns: 0 if a decrement cannot be done without blocking |
| 3364 | * 1 if a decrement succeeded. |
| 3365 | * |
| 3366 | * If a completion is being used as a counting completion, |
| 3367 | * attempt to decrement the counter without blocking. This |
| 3368 | * enables us to avoid waiting if the resource the completion |
| 3369 | * is protecting is not available. |
| 3370 | */ |
| 3371 | bool try_wait_for_completion(struct completion *x) |
| 3372 | { |
Rafael J. Wysocki | 7539a3b | 2009-12-13 00:07:30 +0100 | [diff] [blame] | 3373 | unsigned long flags; |
Dave Chinner | be4de35 | 2008-08-15 00:40:44 -0700 | [diff] [blame] | 3374 | int ret = 1; |
| 3375 | |
Rafael J. Wysocki | 7539a3b | 2009-12-13 00:07:30 +0100 | [diff] [blame] | 3376 | spin_lock_irqsave(&x->wait.lock, flags); |
Dave Chinner | be4de35 | 2008-08-15 00:40:44 -0700 | [diff] [blame] | 3377 | if (!x->done) |
| 3378 | ret = 0; |
| 3379 | else |
| 3380 | x->done--; |
Rafael J. Wysocki | 7539a3b | 2009-12-13 00:07:30 +0100 | [diff] [blame] | 3381 | spin_unlock_irqrestore(&x->wait.lock, flags); |
Dave Chinner | be4de35 | 2008-08-15 00:40:44 -0700 | [diff] [blame] | 3382 | return ret; |
| 3383 | } |
| 3384 | EXPORT_SYMBOL(try_wait_for_completion); |
| 3385 | |
| 3386 | /** |
| 3387 | * completion_done - Test to see if a completion has any waiters |
| 3388 | * @x: completion structure |
| 3389 | * |
| 3390 | * Returns: 0 if there are waiters (wait_for_completion() in progress) |
| 3391 | * 1 if there are no waiters. |
| 3392 | * |
| 3393 | */ |
| 3394 | bool completion_done(struct completion *x) |
| 3395 | { |
Rafael J. Wysocki | 7539a3b | 2009-12-13 00:07:30 +0100 | [diff] [blame] | 3396 | unsigned long flags; |
Dave Chinner | be4de35 | 2008-08-15 00:40:44 -0700 | [diff] [blame] | 3397 | int ret = 1; |
| 3398 | |
Rafael J. Wysocki | 7539a3b | 2009-12-13 00:07:30 +0100 | [diff] [blame] | 3399 | spin_lock_irqsave(&x->wait.lock, flags); |
Dave Chinner | be4de35 | 2008-08-15 00:40:44 -0700 | [diff] [blame] | 3400 | if (!x->done) |
| 3401 | ret = 0; |
Rafael J. Wysocki | 7539a3b | 2009-12-13 00:07:30 +0100 | [diff] [blame] | 3402 | spin_unlock_irqrestore(&x->wait.lock, flags); |
Dave Chinner | be4de35 | 2008-08-15 00:40:44 -0700 | [diff] [blame] | 3403 | return ret; |
| 3404 | } |
| 3405 | EXPORT_SYMBOL(completion_done); |
| 3406 | |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3407 | static long __sched |
| 3408 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3409 | { |
| 3410 | unsigned long flags; |
| 3411 | wait_queue_t wait; |
| 3412 | |
| 3413 | init_waitqueue_entry(&wait, current); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3414 | |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3415 | __set_current_state(state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3416 | |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3417 | spin_lock_irqsave(&q->lock, flags); |
| 3418 | __add_wait_queue(q, &wait); |
| 3419 | spin_unlock(&q->lock); |
| 3420 | timeout = schedule_timeout(timeout); |
| 3421 | spin_lock_irq(&q->lock); |
| 3422 | __remove_wait_queue(q, &wait); |
| 3423 | spin_unlock_irqrestore(&q->lock, flags); |
| 3424 | |
| 3425 | return timeout; |
| 3426 | } |
| 3427 | |
| 3428 | void __sched interruptible_sleep_on(wait_queue_head_t *q) |
| 3429 | { |
| 3430 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3431 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3432 | EXPORT_SYMBOL(interruptible_sleep_on); |
| 3433 | |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3434 | long __sched |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 3435 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3436 | { |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3437 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3438 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3439 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
| 3440 | |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3441 | void __sched sleep_on(wait_queue_head_t *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3442 | { |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3443 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3444 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3445 | EXPORT_SYMBOL(sleep_on); |
| 3446 | |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3447 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3448 | { |
Andi Kleen | 8cbbe86 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3449 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3450 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3451 | EXPORT_SYMBOL(sleep_on_timeout); |
| 3452 | |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3453 | #ifdef CONFIG_RT_MUTEXES |
| 3454 | |
| 3455 | /* |
| 3456 | * rt_mutex_setprio - set the current priority of a task |
| 3457 | * @p: task |
| 3458 | * @prio: prio value (kernel-internal form) |
| 3459 | * |
| 3460 | * This function changes the 'effective' priority of a task. It does |
| 3461 | * not touch ->normal_prio like __setscheduler(). |
| 3462 | * |
| 3463 | * Used by the rt_mutex code to implement priority inheritance logic. |
| 3464 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3465 | void rt_mutex_setprio(struct task_struct *p, int prio) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3466 | { |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 3467 | int oldprio, on_rq, running; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 3468 | struct rq *rq; |
Thomas Gleixner | 83ab0aa | 2010-02-17 09:05:48 +0100 | [diff] [blame] | 3469 | const struct sched_class *prev_class; |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3470 | |
| 3471 | BUG_ON(prio < 0 || prio > MAX_PRIO); |
| 3472 | |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 3473 | rq = __task_rq_lock(p); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3474 | |
Thomas Gleixner | 1c4dd99 | 2011-06-06 20:07:38 +0200 | [diff] [blame] | 3475 | /* |
| 3476 | * Idle task boosting is a nono in general. There is one |
| 3477 | * exception, when PREEMPT_RT and NOHZ is active: |
| 3478 | * |
| 3479 | * The idle task calls get_next_timer_interrupt() and holds |
| 3480 | * the timer wheel base->lock on the CPU and another CPU wants |
| 3481 | * to access the timer (probably to cancel it). We can safely |
| 3482 | * ignore the boosting request, as the idle CPU runs this code |
| 3483 | * with interrupts disabled and will complete the lock |
| 3484 | * protected section without being interrupted. So there is no |
| 3485 | * real need to boost. |
| 3486 | */ |
| 3487 | if (unlikely(p == rq->idle)) { |
| 3488 | WARN_ON(p != rq->curr); |
| 3489 | WARN_ON(p->pi_blocked_on); |
| 3490 | goto out_unlock; |
| 3491 | } |
| 3492 | |
Steven Rostedt | a802707 | 2010-09-20 15:13:34 -0400 | [diff] [blame] | 3493 | trace_sched_pi_setprio(p, prio); |
Andrew Morton | d5f9f94 | 2007-05-08 20:27:06 -0700 | [diff] [blame] | 3494 | oldprio = p->prio; |
Thomas Gleixner | 83ab0aa | 2010-02-17 09:05:48 +0100 | [diff] [blame] | 3495 | prev_class = p->sched_class; |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 3496 | on_rq = p->on_rq; |
Dmitry Adamushko | 051a1d1 | 2007-12-18 15:21:13 +0100 | [diff] [blame] | 3497 | running = task_current(rq, p); |
Hiroshi Shimamoto | 0e1f348 | 2008-03-10 11:01:20 -0700 | [diff] [blame] | 3498 | if (on_rq) |
Ingo Molnar | 69be72c | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 3499 | dequeue_task(rq, p, 0); |
Hiroshi Shimamoto | 0e1f348 | 2008-03-10 11:01:20 -0700 | [diff] [blame] | 3500 | if (running) |
| 3501 | p->sched_class->put_prev_task(rq, p); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3502 | |
| 3503 | if (rt_prio(prio)) |
| 3504 | p->sched_class = &rt_sched_class; |
| 3505 | else |
| 3506 | p->sched_class = &fair_sched_class; |
| 3507 | |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3508 | p->prio = prio; |
| 3509 | |
Hiroshi Shimamoto | 0e1f348 | 2008-03-10 11:01:20 -0700 | [diff] [blame] | 3510 | if (running) |
| 3511 | p->sched_class->set_curr_task(rq); |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 3512 | if (on_rq) |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 3513 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 3514 | |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 3515 | check_class_changed(rq, p, prev_class, oldprio); |
Thomas Gleixner | 1c4dd99 | 2011-06-06 20:07:38 +0200 | [diff] [blame] | 3516 | out_unlock: |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 3517 | __task_rq_unlock(rq); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3518 | } |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3519 | #endif |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3520 | void set_user_nice(struct task_struct *p, long nice) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3521 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3522 | int old_prio, delta, on_rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3523 | unsigned long flags; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 3524 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3525 | |
| 3526 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) |
| 3527 | return; |
| 3528 | /* |
| 3529 | * We have to be careful, if called from sys_setpriority(), |
| 3530 | * the task might be in the middle of scheduling on another CPU. |
| 3531 | */ |
| 3532 | rq = task_rq_lock(p, &flags); |
| 3533 | /* |
| 3534 | * The RT priorities are set via sched_setscheduler(), but we still |
| 3535 | * allow the 'normal' nice value to be set - but as expected |
| 3536 | * it wont have any effect on scheduling until the task is |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3537 | * SCHED_FIFO/SCHED_RR: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3538 | */ |
Ingo Molnar | e05606d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3539 | if (task_has_rt_policy(p)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3540 | p->static_prio = NICE_TO_PRIO(nice); |
| 3541 | goto out_unlock; |
| 3542 | } |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 3543 | on_rq = p->on_rq; |
Peter Zijlstra | c09595f | 2008-06-27 13:41:14 +0200 | [diff] [blame] | 3544 | if (on_rq) |
Ingo Molnar | 69be72c | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 3545 | dequeue_task(rq, p, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3546 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3547 | p->static_prio = NICE_TO_PRIO(nice); |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 3548 | set_load_weight(p); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3549 | old_prio = p->prio; |
| 3550 | p->prio = effective_prio(p); |
| 3551 | delta = p->prio - old_prio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3552 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3553 | if (on_rq) { |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 3554 | enqueue_task(rq, p, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3555 | /* |
Andrew Morton | d5f9f94 | 2007-05-08 20:27:06 -0700 | [diff] [blame] | 3556 | * If the task increased its priority or is running and |
| 3557 | * lowered its priority, then reschedule its CPU: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3558 | */ |
Andrew Morton | d5f9f94 | 2007-05-08 20:27:06 -0700 | [diff] [blame] | 3559 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3560 | resched_task(rq->curr); |
| 3561 | } |
| 3562 | out_unlock: |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 3563 | task_rq_unlock(rq, p, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3564 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3565 | EXPORT_SYMBOL(set_user_nice); |
| 3566 | |
Matt Mackall | e43379f | 2005-05-01 08:59:00 -0700 | [diff] [blame] | 3567 | /* |
| 3568 | * can_nice - check if a task can reduce its nice value |
| 3569 | * @p: task |
| 3570 | * @nice: nice value |
| 3571 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3572 | int can_nice(const struct task_struct *p, const int nice) |
Matt Mackall | e43379f | 2005-05-01 08:59:00 -0700 | [diff] [blame] | 3573 | { |
Matt Mackall | 024f474 | 2005-08-18 11:24:19 -0700 | [diff] [blame] | 3574 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
| 3575 | int nice_rlim = 20 - nice; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 3576 | |
Jiri Slaby | 78d7d40 | 2010-03-05 13:42:54 -0800 | [diff] [blame] | 3577 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
Matt Mackall | e43379f | 2005-05-01 08:59:00 -0700 | [diff] [blame] | 3578 | capable(CAP_SYS_NICE)); |
| 3579 | } |
| 3580 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3581 | #ifdef __ARCH_WANT_SYS_NICE |
| 3582 | |
| 3583 | /* |
| 3584 | * sys_nice - change the priority of the current process. |
| 3585 | * @increment: priority increment |
| 3586 | * |
| 3587 | * sys_setpriority is a more generic, but much slower function that |
| 3588 | * does similar things. |
| 3589 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 3590 | SYSCALL_DEFINE1(nice, int, increment) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3591 | { |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 3592 | long nice, retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3593 | |
| 3594 | /* |
| 3595 | * Setpriority might change our priority at the same moment. |
| 3596 | * We don't have to worry. Conceptually one call occurs first |
| 3597 | * and we have a single winner. |
| 3598 | */ |
Matt Mackall | e43379f | 2005-05-01 08:59:00 -0700 | [diff] [blame] | 3599 | if (increment < -40) |
| 3600 | increment = -40; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3601 | if (increment > 40) |
| 3602 | increment = 40; |
| 3603 | |
Américo Wang | 2b8f836 | 2009-02-16 18:54:21 +0800 | [diff] [blame] | 3604 | nice = TASK_NICE(current) + increment; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3605 | if (nice < -20) |
| 3606 | nice = -20; |
| 3607 | if (nice > 19) |
| 3608 | nice = 19; |
| 3609 | |
Matt Mackall | e43379f | 2005-05-01 08:59:00 -0700 | [diff] [blame] | 3610 | if (increment < 0 && !can_nice(current, nice)) |
| 3611 | return -EPERM; |
| 3612 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3613 | retval = security_task_setnice(current, nice); |
| 3614 | if (retval) |
| 3615 | return retval; |
| 3616 | |
| 3617 | set_user_nice(current, nice); |
| 3618 | return 0; |
| 3619 | } |
| 3620 | |
| 3621 | #endif |
| 3622 | |
| 3623 | /** |
| 3624 | * task_prio - return the priority value of a given task. |
| 3625 | * @p: the task in question. |
| 3626 | * |
| 3627 | * This is the priority value as seen by users in /proc. |
| 3628 | * RT tasks are offset by -200. Normal tasks are centered |
| 3629 | * around 0, value goes from -16 to +15. |
| 3630 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3631 | int task_prio(const struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3632 | { |
| 3633 | return p->prio - MAX_RT_PRIO; |
| 3634 | } |
| 3635 | |
| 3636 | /** |
| 3637 | * task_nice - return the nice value of a given task. |
| 3638 | * @p: the task in question. |
| 3639 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3640 | int task_nice(const struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3641 | { |
| 3642 | return TASK_NICE(p); |
| 3643 | } |
Pavel Roskin | 150d8be | 2008-03-05 16:56:37 -0500 | [diff] [blame] | 3644 | EXPORT_SYMBOL(task_nice); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3645 | |
| 3646 | /** |
| 3647 | * idle_cpu - is a given cpu idle currently? |
| 3648 | * @cpu: the processor in question. |
| 3649 | */ |
| 3650 | int idle_cpu(int cpu) |
| 3651 | { |
Thomas Gleixner | 908a328 | 2011-09-15 15:32:06 +0200 | [diff] [blame] | 3652 | struct rq *rq = cpu_rq(cpu); |
| 3653 | |
| 3654 | if (rq->curr != rq->idle) |
| 3655 | return 0; |
| 3656 | |
| 3657 | if (rq->nr_running) |
| 3658 | return 0; |
| 3659 | |
| 3660 | #ifdef CONFIG_SMP |
| 3661 | if (!llist_empty(&rq->wake_list)) |
| 3662 | return 0; |
| 3663 | #endif |
| 3664 | |
| 3665 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3666 | } |
| 3667 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3668 | /** |
| 3669 | * idle_task - return the idle task for a given cpu. |
| 3670 | * @cpu: the processor in question. |
| 3671 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3672 | struct task_struct *idle_task(int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3673 | { |
| 3674 | return cpu_rq(cpu)->idle; |
| 3675 | } |
| 3676 | |
| 3677 | /** |
| 3678 | * find_process_by_pid - find a process with a matching PID value. |
| 3679 | * @pid: the pid in question. |
| 3680 | */ |
Alexey Dobriyan | a995744 | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 3681 | static struct task_struct *find_process_by_pid(pid_t pid) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3682 | { |
Pavel Emelyanov | 228ebcb | 2007-10-18 23:40:16 -0700 | [diff] [blame] | 3683 | return pid ? find_task_by_vpid(pid) : current; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3684 | } |
| 3685 | |
| 3686 | /* Actually do priority change: must hold rq lock. */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3687 | static void |
| 3688 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3689 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3690 | p->policy = policy; |
| 3691 | p->rt_priority = prio; |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3692 | p->normal_prio = normal_prio(p); |
| 3693 | /* we are holding p->pi_lock already */ |
| 3694 | p->prio = rt_mutex_getprio(p); |
Peter Zijlstra | ffd44db | 2009-11-10 20:12:01 +0100 | [diff] [blame] | 3695 | if (rt_prio(p->prio)) |
| 3696 | p->sched_class = &rt_sched_class; |
| 3697 | else |
| 3698 | p->sched_class = &fair_sched_class; |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 3699 | set_load_weight(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3700 | } |
| 3701 | |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 3702 | /* |
| 3703 | * check the target process has a UID that matches the current process's |
| 3704 | */ |
| 3705 | static bool check_same_owner(struct task_struct *p) |
| 3706 | { |
| 3707 | const struct cred *cred = current_cred(), *pcred; |
| 3708 | bool match; |
| 3709 | |
| 3710 | rcu_read_lock(); |
| 3711 | pcred = __task_cred(p); |
Eric W. Biederman | 9c806aa | 2012-02-02 18:54:02 -0800 | [diff] [blame] | 3712 | match = (uid_eq(cred->euid, pcred->euid) || |
| 3713 | uid_eq(cred->euid, pcred->uid)); |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 3714 | rcu_read_unlock(); |
| 3715 | return match; |
| 3716 | } |
| 3717 | |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 3718 | static int __sched_setscheduler(struct task_struct *p, int policy, |
KOSAKI Motohiro | fe7de49 | 2010-10-20 16:01:12 -0700 | [diff] [blame] | 3719 | const struct sched_param *param, bool user) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3720 | { |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 3721 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3722 | unsigned long flags; |
Thomas Gleixner | 83ab0aa | 2010-02-17 09:05:48 +0100 | [diff] [blame] | 3723 | const struct sched_class *prev_class; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 3724 | struct rq *rq; |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 3725 | int reset_on_fork; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3726 | |
Steven Rostedt | 66e5393 | 2006-06-27 02:54:44 -0700 | [diff] [blame] | 3727 | /* may grab non-irq protected spin_locks */ |
| 3728 | BUG_ON(in_interrupt()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3729 | recheck: |
| 3730 | /* double check policy once rq lock held */ |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 3731 | if (policy < 0) { |
| 3732 | reset_on_fork = p->sched_reset_on_fork; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3733 | policy = oldpolicy = p->policy; |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 3734 | } else { |
| 3735 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); |
| 3736 | policy &= ~SCHED_RESET_ON_FORK; |
| 3737 | |
| 3738 | if (policy != SCHED_FIFO && policy != SCHED_RR && |
| 3739 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
| 3740 | policy != SCHED_IDLE) |
| 3741 | return -EINVAL; |
| 3742 | } |
| 3743 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3744 | /* |
| 3745 | * Valid priorities for SCHED_FIFO and SCHED_RR are |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3746 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
| 3747 | * SCHED_BATCH and SCHED_IDLE is 0. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3748 | */ |
| 3749 | if (param->sched_priority < 0 || |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 3750 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
Steven Rostedt | d46523e | 2005-07-25 16:28:39 -0400 | [diff] [blame] | 3751 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3752 | return -EINVAL; |
Ingo Molnar | e05606d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3753 | if (rt_policy(policy) != (param->sched_priority != 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3754 | return -EINVAL; |
| 3755 | |
Olivier Croquette | 37e4ab3 | 2005-06-25 14:57:32 -0700 | [diff] [blame] | 3756 | /* |
| 3757 | * Allow unprivileged RT tasks to decrease priority: |
| 3758 | */ |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 3759 | if (user && !capable(CAP_SYS_NICE)) { |
Ingo Molnar | e05606d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3760 | if (rt_policy(policy)) { |
Oleg Nesterov | a44702e | 2010-06-11 01:09:44 +0200 | [diff] [blame] | 3761 | unsigned long rlim_rtprio = |
| 3762 | task_rlimit(p, RLIMIT_RTPRIO); |
Oleg Nesterov | 5fe1d75 | 2006-09-29 02:00:48 -0700 | [diff] [blame] | 3763 | |
Oleg Nesterov | 8dc3e90 | 2006-09-29 02:00:50 -0700 | [diff] [blame] | 3764 | /* can't set/change the rt policy */ |
| 3765 | if (policy != p->policy && !rlim_rtprio) |
| 3766 | return -EPERM; |
| 3767 | |
| 3768 | /* can't increase priority */ |
| 3769 | if (param->sched_priority > p->rt_priority && |
| 3770 | param->sched_priority > rlim_rtprio) |
| 3771 | return -EPERM; |
| 3772 | } |
Darren Hart | c02aa73 | 2011-02-17 15:37:07 -0800 | [diff] [blame] | 3773 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3774 | /* |
Darren Hart | c02aa73 | 2011-02-17 15:37:07 -0800 | [diff] [blame] | 3775 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
| 3776 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3777 | */ |
Darren Hart | c02aa73 | 2011-02-17 15:37:07 -0800 | [diff] [blame] | 3778 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
| 3779 | if (!can_nice(p, TASK_NICE(p))) |
| 3780 | return -EPERM; |
| 3781 | } |
Oleg Nesterov | 8dc3e90 | 2006-09-29 02:00:50 -0700 | [diff] [blame] | 3782 | |
Olivier Croquette | 37e4ab3 | 2005-06-25 14:57:32 -0700 | [diff] [blame] | 3783 | /* can't change other user's priorities */ |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 3784 | if (!check_same_owner(p)) |
Olivier Croquette | 37e4ab3 | 2005-06-25 14:57:32 -0700 | [diff] [blame] | 3785 | return -EPERM; |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 3786 | |
| 3787 | /* Normal users shall not reset the sched_reset_on_fork flag */ |
| 3788 | if (p->sched_reset_on_fork && !reset_on_fork) |
| 3789 | return -EPERM; |
Olivier Croquette | 37e4ab3 | 2005-06-25 14:57:32 -0700 | [diff] [blame] | 3790 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3791 | |
Jeremy Fitzhardinge | 725aad2 | 2008-08-03 09:33:03 -0700 | [diff] [blame] | 3792 | if (user) { |
KOSAKI Motohiro | b0ae198 | 2010-10-15 04:21:18 +0900 | [diff] [blame] | 3793 | retval = security_task_setscheduler(p); |
Jeremy Fitzhardinge | 725aad2 | 2008-08-03 09:33:03 -0700 | [diff] [blame] | 3794 | if (retval) |
| 3795 | return retval; |
| 3796 | } |
| 3797 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3798 | /* |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3799 | * make sure no PI-waiters arrive (or leave) while we are |
| 3800 | * changing the priority of the task: |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 3801 | * |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 3802 | * To be able to change p->policy safely, the appropriate |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3803 | * runqueue lock must be held. |
| 3804 | */ |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 3805 | rq = task_rq_lock(p, &flags); |
Peter Zijlstra | dc61b1d | 2010-06-08 11:40:42 +0200 | [diff] [blame] | 3806 | |
Peter Zijlstra | 34f971f | 2010-09-22 13:53:15 +0200 | [diff] [blame] | 3807 | /* |
| 3808 | * Changing the policy of the stop threads its a very bad idea |
| 3809 | */ |
| 3810 | if (p == rq->stop) { |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 3811 | task_rq_unlock(rq, p, &flags); |
Peter Zijlstra | 34f971f | 2010-09-22 13:53:15 +0200 | [diff] [blame] | 3812 | return -EINVAL; |
| 3813 | } |
| 3814 | |
Dario Faggioli | a51e919 | 2011-03-24 14:00:18 +0100 | [diff] [blame] | 3815 | /* |
| 3816 | * If not changing anything there's no need to proceed further: |
| 3817 | */ |
| 3818 | if (unlikely(policy == p->policy && (!rt_policy(policy) || |
| 3819 | param->sched_priority == p->rt_priority))) { |
Namhyung Kim | 45afb17 | 2012-07-07 16:49:02 +0900 | [diff] [blame] | 3820 | task_rq_unlock(rq, p, &flags); |
Dario Faggioli | a51e919 | 2011-03-24 14:00:18 +0100 | [diff] [blame] | 3821 | return 0; |
| 3822 | } |
| 3823 | |
Peter Zijlstra | dc61b1d | 2010-06-08 11:40:42 +0200 | [diff] [blame] | 3824 | #ifdef CONFIG_RT_GROUP_SCHED |
| 3825 | if (user) { |
| 3826 | /* |
| 3827 | * Do not allow realtime tasks into groups that have no runtime |
| 3828 | * assigned. |
| 3829 | */ |
| 3830 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
Mike Galbraith | f449377 | 2011-01-13 04:54:50 +0100 | [diff] [blame] | 3831 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
| 3832 | !task_group_is_autogroup(task_group(p))) { |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 3833 | task_rq_unlock(rq, p, &flags); |
Peter Zijlstra | dc61b1d | 2010-06-08 11:40:42 +0200 | [diff] [blame] | 3834 | return -EPERM; |
| 3835 | } |
| 3836 | } |
| 3837 | #endif |
| 3838 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3839 | /* recheck policy now with rq lock held */ |
| 3840 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
| 3841 | policy = oldpolicy = -1; |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 3842 | task_rq_unlock(rq, p, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3843 | goto recheck; |
| 3844 | } |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 3845 | on_rq = p->on_rq; |
Dmitry Adamushko | 051a1d1 | 2007-12-18 15:21:13 +0100 | [diff] [blame] | 3846 | running = task_current(rq, p); |
Hiroshi Shimamoto | 0e1f348 | 2008-03-10 11:01:20 -0700 | [diff] [blame] | 3847 | if (on_rq) |
Peter Zijlstra | 4ca9b72 | 2012-01-25 11:50:51 +0100 | [diff] [blame] | 3848 | dequeue_task(rq, p, 0); |
Hiroshi Shimamoto | 0e1f348 | 2008-03-10 11:01:20 -0700 | [diff] [blame] | 3849 | if (running) |
| 3850 | p->sched_class->put_prev_task(rq, p); |
Dmitry Adamushko | f6b5320 | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 3851 | |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 3852 | p->sched_reset_on_fork = reset_on_fork; |
| 3853 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3854 | oldprio = p->prio; |
Thomas Gleixner | 83ab0aa | 2010-02-17 09:05:48 +0100 | [diff] [blame] | 3855 | prev_class = p->sched_class; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3856 | __setscheduler(rq, p, policy, param->sched_priority); |
Dmitry Adamushko | f6b5320 | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 3857 | |
Hiroshi Shimamoto | 0e1f348 | 2008-03-10 11:01:20 -0700 | [diff] [blame] | 3858 | if (running) |
| 3859 | p->sched_class->set_curr_task(rq); |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 3860 | if (on_rq) |
Peter Zijlstra | 4ca9b72 | 2012-01-25 11:50:51 +0100 | [diff] [blame] | 3861 | enqueue_task(rq, p, 0); |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 3862 | |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 3863 | check_class_changed(rq, p, prev_class, oldprio); |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 3864 | task_rq_unlock(rq, p, &flags); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3865 | |
Thomas Gleixner | 95e02ca | 2006-06-27 02:55:02 -0700 | [diff] [blame] | 3866 | rt_mutex_adjust_pi(p); |
| 3867 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3868 | return 0; |
| 3869 | } |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 3870 | |
| 3871 | /** |
| 3872 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
| 3873 | * @p: the task in question. |
| 3874 | * @policy: new policy. |
| 3875 | * @param: structure containing the new RT priority. |
| 3876 | * |
| 3877 | * NOTE that the task may be already dead. |
| 3878 | */ |
| 3879 | int sched_setscheduler(struct task_struct *p, int policy, |
KOSAKI Motohiro | fe7de49 | 2010-10-20 16:01:12 -0700 | [diff] [blame] | 3880 | const struct sched_param *param) |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 3881 | { |
| 3882 | return __sched_setscheduler(p, policy, param, true); |
| 3883 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3884 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
| 3885 | |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 3886 | /** |
| 3887 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. |
| 3888 | * @p: the task in question. |
| 3889 | * @policy: new policy. |
| 3890 | * @param: structure containing the new RT priority. |
| 3891 | * |
| 3892 | * Just like sched_setscheduler, only don't bother checking if the |
| 3893 | * current context has permission. For example, this is needed in |
| 3894 | * stop_machine(): we create temporary high priority worker threads, |
| 3895 | * but our caller might not have that capability. |
| 3896 | */ |
| 3897 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, |
KOSAKI Motohiro | fe7de49 | 2010-10-20 16:01:12 -0700 | [diff] [blame] | 3898 | const struct sched_param *param) |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 3899 | { |
| 3900 | return __sched_setscheduler(p, policy, param, false); |
| 3901 | } |
| 3902 | |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 3903 | static int |
| 3904 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3905 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3906 | struct sched_param lparam; |
| 3907 | struct task_struct *p; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3908 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3909 | |
| 3910 | if (!param || pid < 0) |
| 3911 | return -EINVAL; |
| 3912 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) |
| 3913 | return -EFAULT; |
Oleg Nesterov | 5fe1d75 | 2006-09-29 02:00:48 -0700 | [diff] [blame] | 3914 | |
| 3915 | rcu_read_lock(); |
| 3916 | retval = -ESRCH; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3917 | p = find_process_by_pid(pid); |
Oleg Nesterov | 5fe1d75 | 2006-09-29 02:00:48 -0700 | [diff] [blame] | 3918 | if (p != NULL) |
| 3919 | retval = sched_setscheduler(p, policy, &lparam); |
| 3920 | rcu_read_unlock(); |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3921 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3922 | return retval; |
| 3923 | } |
| 3924 | |
| 3925 | /** |
| 3926 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority |
| 3927 | * @pid: the pid in question. |
| 3928 | * @policy: new policy. |
| 3929 | * @param: structure containing the new RT priority. |
| 3930 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 3931 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
| 3932 | struct sched_param __user *, param) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3933 | { |
Jason Baron | c21761f | 2006-01-18 17:43:03 -0800 | [diff] [blame] | 3934 | /* negative values for policy are not valid */ |
| 3935 | if (policy < 0) |
| 3936 | return -EINVAL; |
| 3937 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3938 | return do_sched_setscheduler(pid, policy, param); |
| 3939 | } |
| 3940 | |
| 3941 | /** |
| 3942 | * sys_sched_setparam - set/change the RT priority of a thread |
| 3943 | * @pid: the pid in question. |
| 3944 | * @param: structure containing the new RT priority. |
| 3945 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 3946 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3947 | { |
| 3948 | return do_sched_setscheduler(pid, -1, param); |
| 3949 | } |
| 3950 | |
| 3951 | /** |
| 3952 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread |
| 3953 | * @pid: the pid in question. |
| 3954 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 3955 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3956 | { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3957 | struct task_struct *p; |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3958 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3959 | |
| 3960 | if (pid < 0) |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3961 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3962 | |
| 3963 | retval = -ESRCH; |
Thomas Gleixner | 5fe85be | 2009-12-09 10:14:58 +0000 | [diff] [blame] | 3964 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3965 | p = find_process_by_pid(pid); |
| 3966 | if (p) { |
| 3967 | retval = security_task_getscheduler(p); |
| 3968 | if (!retval) |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 3969 | retval = p->policy |
| 3970 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3971 | } |
Thomas Gleixner | 5fe85be | 2009-12-09 10:14:58 +0000 | [diff] [blame] | 3972 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3973 | return retval; |
| 3974 | } |
| 3975 | |
| 3976 | /** |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 3977 | * sys_sched_getparam - get the RT priority of a thread |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3978 | * @pid: the pid in question. |
| 3979 | * @param: structure containing the RT priority. |
| 3980 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 3981 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3982 | { |
| 3983 | struct sched_param lp; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3984 | struct task_struct *p; |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3985 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3986 | |
| 3987 | if (!param || pid < 0) |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3988 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3989 | |
Thomas Gleixner | 5fe85be | 2009-12-09 10:14:58 +0000 | [diff] [blame] | 3990 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3991 | p = find_process_by_pid(pid); |
| 3992 | retval = -ESRCH; |
| 3993 | if (!p) |
| 3994 | goto out_unlock; |
| 3995 | |
| 3996 | retval = security_task_getscheduler(p); |
| 3997 | if (retval) |
| 3998 | goto out_unlock; |
| 3999 | |
| 4000 | lp.sched_priority = p->rt_priority; |
Thomas Gleixner | 5fe85be | 2009-12-09 10:14:58 +0000 | [diff] [blame] | 4001 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4002 | |
| 4003 | /* |
| 4004 | * This one might sleep, we cannot do it with a spinlock held ... |
| 4005 | */ |
| 4006 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; |
| 4007 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4008 | return retval; |
| 4009 | |
| 4010 | out_unlock: |
Thomas Gleixner | 5fe85be | 2009-12-09 10:14:58 +0000 | [diff] [blame] | 4011 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4012 | return retval; |
| 4013 | } |
| 4014 | |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 4015 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4016 | { |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4017 | cpumask_var_t cpus_allowed, new_mask; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4018 | struct task_struct *p; |
| 4019 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4020 | |
Gautham R Shenoy | 95402b3 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 4021 | get_online_cpus(); |
Thomas Gleixner | 23f5d14 | 2009-12-09 10:15:01 +0000 | [diff] [blame] | 4022 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4023 | |
| 4024 | p = find_process_by_pid(pid); |
| 4025 | if (!p) { |
Thomas Gleixner | 23f5d14 | 2009-12-09 10:15:01 +0000 | [diff] [blame] | 4026 | rcu_read_unlock(); |
Gautham R Shenoy | 95402b3 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 4027 | put_online_cpus(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4028 | return -ESRCH; |
| 4029 | } |
| 4030 | |
Thomas Gleixner | 23f5d14 | 2009-12-09 10:15:01 +0000 | [diff] [blame] | 4031 | /* Prevent p going away */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4032 | get_task_struct(p); |
Thomas Gleixner | 23f5d14 | 2009-12-09 10:15:01 +0000 | [diff] [blame] | 4033 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4034 | |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4035 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
| 4036 | retval = -ENOMEM; |
| 4037 | goto out_put_task; |
| 4038 | } |
| 4039 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { |
| 4040 | retval = -ENOMEM; |
| 4041 | goto out_free_cpus_allowed; |
| 4042 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4043 | retval = -EPERM; |
Linus Torvalds | c49c41a | 2012-01-14 18:36:33 -0800 | [diff] [blame] | 4044 | if (!check_same_owner(p) && !ns_capable(task_user_ns(p), CAP_SYS_NICE)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4045 | goto out_unlock; |
| 4046 | |
KOSAKI Motohiro | b0ae198 | 2010-10-15 04:21:18 +0900 | [diff] [blame] | 4047 | retval = security_task_setscheduler(p); |
David Quigley | e7834f8 | 2006-06-23 02:03:59 -0700 | [diff] [blame] | 4048 | if (retval) |
| 4049 | goto out_unlock; |
| 4050 | |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4051 | cpuset_cpus_allowed(p, cpus_allowed); |
| 4052 | cpumask_and(new_mask, in_mask, cpus_allowed); |
Peter Zijlstra | 4924627 | 2010-10-17 21:46:10 +0200 | [diff] [blame] | 4053 | again: |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4054 | retval = set_cpus_allowed_ptr(p, new_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4055 | |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 4056 | if (!retval) { |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4057 | cpuset_cpus_allowed(p, cpus_allowed); |
| 4058 | if (!cpumask_subset(new_mask, cpus_allowed)) { |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 4059 | /* |
| 4060 | * We must have raced with a concurrent cpuset |
| 4061 | * update. Just reset the cpus_allowed to the |
| 4062 | * cpuset's cpus_allowed |
| 4063 | */ |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4064 | cpumask_copy(new_mask, cpus_allowed); |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 4065 | goto again; |
| 4066 | } |
| 4067 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4068 | out_unlock: |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4069 | free_cpumask_var(new_mask); |
| 4070 | out_free_cpus_allowed: |
| 4071 | free_cpumask_var(cpus_allowed); |
| 4072 | out_put_task: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4073 | put_task_struct(p); |
Gautham R Shenoy | 95402b3 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 4074 | put_online_cpus(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4075 | return retval; |
| 4076 | } |
| 4077 | |
| 4078 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 4079 | struct cpumask *new_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4080 | { |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 4081 | if (len < cpumask_size()) |
| 4082 | cpumask_clear(new_mask); |
| 4083 | else if (len > cpumask_size()) |
| 4084 | len = cpumask_size(); |
| 4085 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4086 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
| 4087 | } |
| 4088 | |
| 4089 | /** |
| 4090 | * sys_sched_setaffinity - set the cpu affinity of a process |
| 4091 | * @pid: pid of the process |
| 4092 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr |
| 4093 | * @user_mask_ptr: user-space pointer to the new cpu mask |
| 4094 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 4095 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
| 4096 | unsigned long __user *, user_mask_ptr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4097 | { |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4098 | cpumask_var_t new_mask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4099 | int retval; |
| 4100 | |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4101 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
| 4102 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4103 | |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4104 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
| 4105 | if (retval == 0) |
| 4106 | retval = sched_setaffinity(pid, new_mask); |
| 4107 | free_cpumask_var(new_mask); |
| 4108 | return retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4109 | } |
| 4110 | |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 4111 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4112 | { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4113 | struct task_struct *p; |
Thomas Gleixner | 3160568 | 2009-12-08 20:24:16 +0000 | [diff] [blame] | 4114 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4115 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4116 | |
Gautham R Shenoy | 95402b3 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 4117 | get_online_cpus(); |
Thomas Gleixner | 23f5d14 | 2009-12-09 10:15:01 +0000 | [diff] [blame] | 4118 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4119 | |
| 4120 | retval = -ESRCH; |
| 4121 | p = find_process_by_pid(pid); |
| 4122 | if (!p) |
| 4123 | goto out_unlock; |
| 4124 | |
David Quigley | e7834f8 | 2006-06-23 02:03:59 -0700 | [diff] [blame] | 4125 | retval = security_task_getscheduler(p); |
| 4126 | if (retval) |
| 4127 | goto out_unlock; |
| 4128 | |
Peter Zijlstra | 013fdb8 | 2011-04-05 17:23:45 +0200 | [diff] [blame] | 4129 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 4130 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
Peter Zijlstra | 013fdb8 | 2011-04-05 17:23:45 +0200 | [diff] [blame] | 4131 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4132 | |
| 4133 | out_unlock: |
Thomas Gleixner | 23f5d14 | 2009-12-09 10:15:01 +0000 | [diff] [blame] | 4134 | rcu_read_unlock(); |
Gautham R Shenoy | 95402b3 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 4135 | put_online_cpus(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4136 | |
Ulrich Drepper | 9531b62 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4137 | return retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4138 | } |
| 4139 | |
| 4140 | /** |
| 4141 | * sys_sched_getaffinity - get the cpu affinity of a process |
| 4142 | * @pid: pid of the process |
| 4143 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr |
| 4144 | * @user_mask_ptr: user-space pointer to hold the current cpu mask |
| 4145 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 4146 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
| 4147 | unsigned long __user *, user_mask_ptr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4148 | { |
| 4149 | int ret; |
Rusty Russell | f17c860 | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4150 | cpumask_var_t mask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4151 | |
Anton Blanchard | 84fba5e | 2010-04-06 17:02:19 +1000 | [diff] [blame] | 4152 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
KOSAKI Motohiro | cd3d803 | 2010-03-12 16:15:36 +0900 | [diff] [blame] | 4153 | return -EINVAL; |
| 4154 | if (len & (sizeof(unsigned long)-1)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4155 | return -EINVAL; |
| 4156 | |
Rusty Russell | f17c860 | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4157 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
| 4158 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4159 | |
Rusty Russell | f17c860 | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4160 | ret = sched_getaffinity(pid, mask); |
| 4161 | if (ret == 0) { |
KOSAKI Motohiro | 8bc037f | 2010-03-17 09:36:58 +0900 | [diff] [blame] | 4162 | size_t retlen = min_t(size_t, len, cpumask_size()); |
KOSAKI Motohiro | cd3d803 | 2010-03-12 16:15:36 +0900 | [diff] [blame] | 4163 | |
| 4164 | if (copy_to_user(user_mask_ptr, mask, retlen)) |
Rusty Russell | f17c860 | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4165 | ret = -EFAULT; |
| 4166 | else |
KOSAKI Motohiro | cd3d803 | 2010-03-12 16:15:36 +0900 | [diff] [blame] | 4167 | ret = retlen; |
Rusty Russell | f17c860 | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4168 | } |
| 4169 | free_cpumask_var(mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4170 | |
Rusty Russell | f17c860 | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4171 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4172 | } |
| 4173 | |
| 4174 | /** |
| 4175 | * sys_sched_yield - yield the current processor to other threads. |
| 4176 | * |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4177 | * This function yields the current CPU to other tasks. If there are no |
| 4178 | * other threads running on this CPU then this function will return. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4179 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 4180 | SYSCALL_DEFINE0(sched_yield) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4181 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 4182 | struct rq *rq = this_rq_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4183 | |
Ingo Molnar | 2d72376 | 2007-10-15 17:00:12 +0200 | [diff] [blame] | 4184 | schedstat_inc(rq, yld_count); |
Dmitry Adamushko | 4530d7a | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 4185 | current->sched_class->yield_task(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4186 | |
| 4187 | /* |
| 4188 | * Since we are going to call schedule() anyway, there's |
| 4189 | * no need to preempt or enable interrupts: |
| 4190 | */ |
| 4191 | __release(rq->lock); |
Ingo Molnar | 8a25d5d | 2006-07-03 00:24:54 -0700 | [diff] [blame] | 4192 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
Thomas Gleixner | 9828ea9 | 2009-12-03 20:55:53 +0100 | [diff] [blame] | 4193 | do_raw_spin_unlock(&rq->lock); |
Thomas Gleixner | ba74c14 | 2011-03-21 13:32:17 +0100 | [diff] [blame] | 4194 | sched_preempt_enable_no_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4195 | |
| 4196 | schedule(); |
| 4197 | |
| 4198 | return 0; |
| 4199 | } |
| 4200 | |
Peter Zijlstra | d86ee48 | 2009-07-10 14:57:57 +0200 | [diff] [blame] | 4201 | static inline int should_resched(void) |
| 4202 | { |
| 4203 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); |
| 4204 | } |
| 4205 | |
Andrew Morton | e7b3840 | 2006-06-30 01:56:00 -0700 | [diff] [blame] | 4206 | static void __cond_resched(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4207 | { |
Frederic Weisbecker | e7aaaa6 | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 4208 | add_preempt_count(PREEMPT_ACTIVE); |
Thomas Gleixner | c259e01 | 2011-06-22 19:47:00 +0200 | [diff] [blame] | 4209 | __schedule(); |
Frederic Weisbecker | e7aaaa6 | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 4210 | sub_preempt_count(PREEMPT_ACTIVE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4211 | } |
| 4212 | |
Herbert Xu | 02b67cc3 | 2008-01-25 21:08:28 +0100 | [diff] [blame] | 4213 | int __sched _cond_resched(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4214 | { |
Peter Zijlstra | d86ee48 | 2009-07-10 14:57:57 +0200 | [diff] [blame] | 4215 | if (should_resched()) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4216 | __cond_resched(); |
| 4217 | return 1; |
| 4218 | } |
| 4219 | return 0; |
| 4220 | } |
Herbert Xu | 02b67cc3 | 2008-01-25 21:08:28 +0100 | [diff] [blame] | 4221 | EXPORT_SYMBOL(_cond_resched); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4222 | |
| 4223 | /* |
Frederic Weisbecker | 613afbf | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 4224 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4225 | * call schedule, and on return reacquire the lock. |
| 4226 | * |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 4227 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4228 | * operations here to prevent schedule() from being called twice (once via |
| 4229 | * spin_unlock(), once by hand). |
| 4230 | */ |
Frederic Weisbecker | 613afbf | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 4231 | int __cond_resched_lock(spinlock_t *lock) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4232 | { |
Peter Zijlstra | d86ee48 | 2009-07-10 14:57:57 +0200 | [diff] [blame] | 4233 | int resched = should_resched(); |
Jan Kara | 6df3cec | 2005-06-13 15:52:32 -0700 | [diff] [blame] | 4234 | int ret = 0; |
| 4235 | |
Peter Zijlstra | f607c66 | 2009-07-20 19:16:29 +0200 | [diff] [blame] | 4236 | lockdep_assert_held(lock); |
| 4237 | |
Nick Piggin | 95c354f | 2008-01-30 13:31:20 +0100 | [diff] [blame] | 4238 | if (spin_needbreak(lock) || resched) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4239 | spin_unlock(lock); |
Peter Zijlstra | d86ee48 | 2009-07-10 14:57:57 +0200 | [diff] [blame] | 4240 | if (resched) |
Nick Piggin | 95c354f | 2008-01-30 13:31:20 +0100 | [diff] [blame] | 4241 | __cond_resched(); |
| 4242 | else |
| 4243 | cpu_relax(); |
Jan Kara | 6df3cec | 2005-06-13 15:52:32 -0700 | [diff] [blame] | 4244 | ret = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4245 | spin_lock(lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4246 | } |
Jan Kara | 6df3cec | 2005-06-13 15:52:32 -0700 | [diff] [blame] | 4247 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4248 | } |
Frederic Weisbecker | 613afbf | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 4249 | EXPORT_SYMBOL(__cond_resched_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4250 | |
Frederic Weisbecker | 613afbf | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 4251 | int __sched __cond_resched_softirq(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4252 | { |
| 4253 | BUG_ON(!in_softirq()); |
| 4254 | |
Peter Zijlstra | d86ee48 | 2009-07-10 14:57:57 +0200 | [diff] [blame] | 4255 | if (should_resched()) { |
Thomas Gleixner | 98d82567 | 2007-05-23 13:58:18 -0700 | [diff] [blame] | 4256 | local_bh_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4257 | __cond_resched(); |
| 4258 | local_bh_disable(); |
| 4259 | return 1; |
| 4260 | } |
| 4261 | return 0; |
| 4262 | } |
Frederic Weisbecker | 613afbf | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 4263 | EXPORT_SYMBOL(__cond_resched_softirq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4264 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4265 | /** |
| 4266 | * yield - yield the current processor to other threads. |
| 4267 | * |
Peter Zijlstra | 8e3fabf | 2012-03-06 18:54:26 +0100 | [diff] [blame] | 4268 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
| 4269 | * |
| 4270 | * The scheduler is at all times free to pick the calling task as the most |
| 4271 | * eligible task to run, if removing the yield() call from your code breaks |
| 4272 | * it, its already broken. |
| 4273 | * |
| 4274 | * Typical broken usage is: |
| 4275 | * |
| 4276 | * while (!event) |
| 4277 | * yield(); |
| 4278 | * |
| 4279 | * where one assumes that yield() will let 'the other' process run that will |
| 4280 | * make event true. If the current task is a SCHED_FIFO task that will never |
| 4281 | * happen. Never use yield() as a progress guarantee!! |
| 4282 | * |
| 4283 | * If you want to use yield() to wait for something, use wait_event(). |
| 4284 | * If you want to use yield() to be 'nice' for others, use cond_resched(). |
| 4285 | * If you still want to use yield(), do not! |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4286 | */ |
| 4287 | void __sched yield(void) |
| 4288 | { |
| 4289 | set_current_state(TASK_RUNNING); |
| 4290 | sys_sched_yield(); |
| 4291 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4292 | EXPORT_SYMBOL(yield); |
| 4293 | |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 4294 | /** |
| 4295 | * yield_to - yield the current processor to another thread in |
| 4296 | * your thread group, or accelerate that thread toward the |
| 4297 | * processor it's on. |
Randy Dunlap | 16addf9 | 2011-03-18 09:34:53 -0700 | [diff] [blame] | 4298 | * @p: target task |
| 4299 | * @preempt: whether task preemption is allowed or not |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 4300 | * |
| 4301 | * It's the caller's job to ensure that the target task struct |
| 4302 | * can't go away on us before we can do any checks. |
| 4303 | * |
| 4304 | * Returns true if we indeed boosted the target task. |
| 4305 | */ |
| 4306 | bool __sched yield_to(struct task_struct *p, bool preempt) |
| 4307 | { |
| 4308 | struct task_struct *curr = current; |
| 4309 | struct rq *rq, *p_rq; |
| 4310 | unsigned long flags; |
| 4311 | bool yielded = 0; |
| 4312 | |
| 4313 | local_irq_save(flags); |
| 4314 | rq = this_rq(); |
| 4315 | |
| 4316 | again: |
| 4317 | p_rq = task_rq(p); |
| 4318 | double_rq_lock(rq, p_rq); |
| 4319 | while (task_rq(p) != p_rq) { |
| 4320 | double_rq_unlock(rq, p_rq); |
| 4321 | goto again; |
| 4322 | } |
| 4323 | |
| 4324 | if (!curr->sched_class->yield_to_task) |
| 4325 | goto out; |
| 4326 | |
| 4327 | if (curr->sched_class != p->sched_class) |
| 4328 | goto out; |
| 4329 | |
| 4330 | if (task_running(p_rq, p) || p->state) |
| 4331 | goto out; |
| 4332 | |
| 4333 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); |
Venkatesh Pallipadi | 6d1cafd | 2011-03-01 16:28:21 -0800 | [diff] [blame] | 4334 | if (yielded) { |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 4335 | schedstat_inc(rq, yld_count); |
Venkatesh Pallipadi | 6d1cafd | 2011-03-01 16:28:21 -0800 | [diff] [blame] | 4336 | /* |
| 4337 | * Make p's CPU reschedule; pick_next_entity takes care of |
| 4338 | * fairness. |
| 4339 | */ |
| 4340 | if (preempt && rq != p_rq) |
| 4341 | resched_task(p_rq->curr); |
| 4342 | } |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 4343 | |
| 4344 | out: |
| 4345 | double_rq_unlock(rq, p_rq); |
| 4346 | local_irq_restore(flags); |
| 4347 | |
| 4348 | if (yielded) |
| 4349 | schedule(); |
| 4350 | |
| 4351 | return yielded; |
| 4352 | } |
| 4353 | EXPORT_SYMBOL_GPL(yield_to); |
| 4354 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4355 | /* |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 4356 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4357 | * that process accounting knows that this is a task in IO wait state. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4358 | */ |
| 4359 | void __sched io_schedule(void) |
| 4360 | { |
Hitoshi Mitake | 54d35f2 | 2009-06-29 14:44:57 +0900 | [diff] [blame] | 4361 | struct rq *rq = raw_rq(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4362 | |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 4363 | delayacct_blkio_start(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4364 | atomic_inc(&rq->nr_iowait); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 4365 | blk_flush_plug(current); |
Arjan van de Ven | 8f0dfc3 | 2009-07-20 11:26:58 -0700 | [diff] [blame] | 4366 | current->in_iowait = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4367 | schedule(); |
Arjan van de Ven | 8f0dfc3 | 2009-07-20 11:26:58 -0700 | [diff] [blame] | 4368 | current->in_iowait = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4369 | atomic_dec(&rq->nr_iowait); |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 4370 | delayacct_blkio_end(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4371 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4372 | EXPORT_SYMBOL(io_schedule); |
| 4373 | |
| 4374 | long __sched io_schedule_timeout(long timeout) |
| 4375 | { |
Hitoshi Mitake | 54d35f2 | 2009-06-29 14:44:57 +0900 | [diff] [blame] | 4376 | struct rq *rq = raw_rq(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4377 | long ret; |
| 4378 | |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 4379 | delayacct_blkio_start(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4380 | atomic_inc(&rq->nr_iowait); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 4381 | blk_flush_plug(current); |
Arjan van de Ven | 8f0dfc3 | 2009-07-20 11:26:58 -0700 | [diff] [blame] | 4382 | current->in_iowait = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4383 | ret = schedule_timeout(timeout); |
Arjan van de Ven | 8f0dfc3 | 2009-07-20 11:26:58 -0700 | [diff] [blame] | 4384 | current->in_iowait = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4385 | atomic_dec(&rq->nr_iowait); |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 4386 | delayacct_blkio_end(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4387 | return ret; |
| 4388 | } |
| 4389 | |
| 4390 | /** |
| 4391 | * sys_sched_get_priority_max - return maximum RT priority. |
| 4392 | * @policy: scheduling class. |
| 4393 | * |
| 4394 | * this syscall returns the maximum rt_priority that can be used |
| 4395 | * by a given scheduling class. |
| 4396 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 4397 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4398 | { |
| 4399 | int ret = -EINVAL; |
| 4400 | |
| 4401 | switch (policy) { |
| 4402 | case SCHED_FIFO: |
| 4403 | case SCHED_RR: |
| 4404 | ret = MAX_USER_RT_PRIO-1; |
| 4405 | break; |
| 4406 | case SCHED_NORMAL: |
Ingo Molnar | b0a9499 | 2006-01-14 13:20:41 -0800 | [diff] [blame] | 4407 | case SCHED_BATCH: |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4408 | case SCHED_IDLE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4409 | ret = 0; |
| 4410 | break; |
| 4411 | } |
| 4412 | return ret; |
| 4413 | } |
| 4414 | |
| 4415 | /** |
| 4416 | * sys_sched_get_priority_min - return minimum RT priority. |
| 4417 | * @policy: scheduling class. |
| 4418 | * |
| 4419 | * this syscall returns the minimum rt_priority that can be used |
| 4420 | * by a given scheduling class. |
| 4421 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 4422 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4423 | { |
| 4424 | int ret = -EINVAL; |
| 4425 | |
| 4426 | switch (policy) { |
| 4427 | case SCHED_FIFO: |
| 4428 | case SCHED_RR: |
| 4429 | ret = 1; |
| 4430 | break; |
| 4431 | case SCHED_NORMAL: |
Ingo Molnar | b0a9499 | 2006-01-14 13:20:41 -0800 | [diff] [blame] | 4432 | case SCHED_BATCH: |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4433 | case SCHED_IDLE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4434 | ret = 0; |
| 4435 | } |
| 4436 | return ret; |
| 4437 | } |
| 4438 | |
| 4439 | /** |
| 4440 | * sys_sched_rr_get_interval - return the default timeslice of a process. |
| 4441 | * @pid: pid of the process. |
| 4442 | * @interval: userspace pointer to the timeslice value. |
| 4443 | * |
| 4444 | * this syscall writes the default timeslice value of a given process |
| 4445 | * into the user-space timespec buffer. A value of '0' means infinity. |
| 4446 | */ |
Heiko Carstens | 17da2bd | 2009-01-14 14:14:10 +0100 | [diff] [blame] | 4447 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
Heiko Carstens | 754fe8d | 2009-01-14 14:14:09 +0100 | [diff] [blame] | 4448 | struct timespec __user *, interval) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4449 | { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4450 | struct task_struct *p; |
Dmitry Adamushko | a4ec24b | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 4451 | unsigned int time_slice; |
Thomas Gleixner | dba091b | 2009-12-09 09:32:03 +0100 | [diff] [blame] | 4452 | unsigned long flags; |
| 4453 | struct rq *rq; |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 4454 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4455 | struct timespec t; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4456 | |
| 4457 | if (pid < 0) |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 4458 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4459 | |
| 4460 | retval = -ESRCH; |
Thomas Gleixner | 1a551ae | 2009-12-09 10:15:11 +0000 | [diff] [blame] | 4461 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4462 | p = find_process_by_pid(pid); |
| 4463 | if (!p) |
| 4464 | goto out_unlock; |
| 4465 | |
| 4466 | retval = security_task_getscheduler(p); |
| 4467 | if (retval) |
| 4468 | goto out_unlock; |
| 4469 | |
Thomas Gleixner | dba091b | 2009-12-09 09:32:03 +0100 | [diff] [blame] | 4470 | rq = task_rq_lock(p, &flags); |
| 4471 | time_slice = p->sched_class->get_rr_interval(rq, p); |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 4472 | task_rq_unlock(rq, p, &flags); |
Dmitry Adamushko | a4ec24b | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 4473 | |
Thomas Gleixner | 1a551ae | 2009-12-09 10:15:11 +0000 | [diff] [blame] | 4474 | rcu_read_unlock(); |
Dmitry Adamushko | a4ec24b | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 4475 | jiffies_to_timespec(time_slice, &t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4476 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4477 | return retval; |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 4478 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4479 | out_unlock: |
Thomas Gleixner | 1a551ae | 2009-12-09 10:15:11 +0000 | [diff] [blame] | 4480 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4481 | return retval; |
| 4482 | } |
| 4483 | |
Steven Rostedt | 7c731e0 | 2008-05-12 21:20:41 +0200 | [diff] [blame] | 4484 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4485 | |
Ingo Molnar | 82a1fcb | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 4486 | void sched_show_task(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4487 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4488 | unsigned long free = 0; |
Paul E. McKenney | 4e79752 | 2012-11-07 13:35:32 -0800 | [diff] [blame] | 4489 | int ppid; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4490 | unsigned state; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4491 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4492 | state = p->state ? __ffs(p->state) + 1 : 0; |
Erik Gilling | 28d0686 | 2010-11-19 18:08:51 -0800 | [diff] [blame] | 4493 | printk(KERN_INFO "%-15.15s %c", p->comm, |
Andreas Mohr | 2ed6e34 | 2006-07-10 04:43:52 -0700 | [diff] [blame] | 4494 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
Ingo Molnar | 4bd7732 | 2007-07-11 21:21:47 +0200 | [diff] [blame] | 4495 | #if BITS_PER_LONG == 32 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4496 | if (state == TASK_RUNNING) |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 4497 | printk(KERN_CONT " running "); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4498 | else |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 4499 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4500 | #else |
| 4501 | if (state == TASK_RUNNING) |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 4502 | printk(KERN_CONT " running task "); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4503 | else |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 4504 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4505 | #endif |
| 4506 | #ifdef CONFIG_DEBUG_STACK_USAGE |
Eric Sandeen | 7c9f886 | 2008-04-22 16:38:23 -0500 | [diff] [blame] | 4507 | free = stack_not_used(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4508 | #endif |
Paul E. McKenney | 4e79752 | 2012-11-07 13:35:32 -0800 | [diff] [blame] | 4509 | rcu_read_lock(); |
| 4510 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); |
| 4511 | rcu_read_unlock(); |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 4512 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
Paul E. McKenney | 4e79752 | 2012-11-07 13:35:32 -0800 | [diff] [blame] | 4513 | task_pid_nr(p), ppid, |
David Rientjes | aa47b7e | 2009-05-04 01:38:05 -0700 | [diff] [blame] | 4514 | (unsigned long)task_thread_info(p)->flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4515 | |
Nick Piggin | 5fb5e6d | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 4516 | show_stack(p, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4517 | } |
| 4518 | |
Ingo Molnar | e59e2ae | 2006-12-06 20:35:59 -0800 | [diff] [blame] | 4519 | void show_state_filter(unsigned long state_filter) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4520 | { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4521 | struct task_struct *g, *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4522 | |
Ingo Molnar | 4bd7732 | 2007-07-11 21:21:47 +0200 | [diff] [blame] | 4523 | #if BITS_PER_LONG == 32 |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 4524 | printk(KERN_INFO |
| 4525 | " task PC stack pid father\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4526 | #else |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 4527 | printk(KERN_INFO |
| 4528 | " task PC stack pid father\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4529 | #endif |
Thomas Gleixner | 510f5ac | 2011-07-17 20:47:54 +0200 | [diff] [blame] | 4530 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4531 | do_each_thread(g, p) { |
| 4532 | /* |
| 4533 | * reset the NMI-timeout, listing all files on a slow |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 4534 | * console might take a lot of time: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4535 | */ |
| 4536 | touch_nmi_watchdog(); |
Ingo Molnar | 39bc89f | 2007-04-25 20:50:03 -0700 | [diff] [blame] | 4537 | if (!state_filter || (p->state & state_filter)) |
Ingo Molnar | 82a1fcb | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 4538 | sched_show_task(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4539 | } while_each_thread(g, p); |
| 4540 | |
Jeremy Fitzhardinge | 04c9167 | 2007-05-08 00:28:05 -0700 | [diff] [blame] | 4541 | touch_all_softlockup_watchdogs(); |
| 4542 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4543 | #ifdef CONFIG_SCHED_DEBUG |
| 4544 | sysrq_sched_debug_show(); |
| 4545 | #endif |
Thomas Gleixner | 510f5ac | 2011-07-17 20:47:54 +0200 | [diff] [blame] | 4546 | rcu_read_unlock(); |
Ingo Molnar | e59e2ae | 2006-12-06 20:35:59 -0800 | [diff] [blame] | 4547 | /* |
| 4548 | * Only show locks if all tasks are dumped: |
| 4549 | */ |
Shmulik Ladkani | 93335a2 | 2009-11-25 15:23:41 +0200 | [diff] [blame] | 4550 | if (!state_filter) |
Ingo Molnar | e59e2ae | 2006-12-06 20:35:59 -0800 | [diff] [blame] | 4551 | debug_show_all_locks(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4552 | } |
| 4553 | |
Ingo Molnar | 1df2105 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 4554 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
| 4555 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4556 | idle->sched_class = &idle_sched_class; |
Ingo Molnar | 1df2105 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 4557 | } |
| 4558 | |
Ingo Molnar | f340c0d | 2005-06-28 16:40:42 +0200 | [diff] [blame] | 4559 | /** |
| 4560 | * init_idle - set up an idle thread for a given CPU |
| 4561 | * @idle: task in question |
| 4562 | * @cpu: cpu the idle task belongs to |
| 4563 | * |
| 4564 | * NOTE: this function does not set the idle thread's NEED_RESCHED |
| 4565 | * flag, to make booting more robust. |
| 4566 | */ |
Nick Piggin | 5c1e176 | 2006-10-03 01:14:04 -0700 | [diff] [blame] | 4567 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4568 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 4569 | struct rq *rq = cpu_rq(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4570 | unsigned long flags; |
| 4571 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 4572 | raw_spin_lock_irqsave(&rq->lock, flags); |
Ingo Molnar | 5cbd54e | 2008-11-12 20:05:50 +0100 | [diff] [blame] | 4573 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4574 | __sched_fork(idle); |
Peter Zijlstra | 06b83b5 | 2009-12-16 18:04:35 +0100 | [diff] [blame] | 4575 | idle->state = TASK_RUNNING; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4576 | idle->se.exec_start = sched_clock(); |
| 4577 | |
KOSAKI Motohiro | 1e1b6c5 | 2011-05-19 15:08:58 +0900 | [diff] [blame] | 4578 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
Peter Zijlstra | 6506cf6c | 2010-09-16 17:50:31 +0200 | [diff] [blame] | 4579 | /* |
| 4580 | * We're having a chicken and egg problem, even though we are |
| 4581 | * holding rq->lock, the cpu isn't yet set to this cpu so the |
| 4582 | * lockdep check in task_group() will fail. |
| 4583 | * |
| 4584 | * Similar case to sched_fork(). / Alternatively we could |
| 4585 | * use task_rq_lock() here and obtain the other rq->lock. |
| 4586 | * |
| 4587 | * Silence PROVE_RCU |
| 4588 | */ |
| 4589 | rcu_read_lock(); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4590 | __set_task_cpu(idle, cpu); |
Peter Zijlstra | 6506cf6c | 2010-09-16 17:50:31 +0200 | [diff] [blame] | 4591 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4592 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4593 | rq->curr = rq->idle = idle; |
Peter Zijlstra | 3ca7a44 | 2011-04-05 17:23:40 +0200 | [diff] [blame] | 4594 | #if defined(CONFIG_SMP) |
| 4595 | idle->on_cpu = 1; |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 4596 | #endif |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 4597 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4598 | |
| 4599 | /* Set the preempt count _outside_ the spinlocks! */ |
Al Viro | a1261f5 | 2005-11-13 16:06:55 -0800 | [diff] [blame] | 4600 | task_thread_info(idle)->preempt_count = 0; |
Jonathan Corbet | 625f2a3 | 2011-04-22 11:19:10 -0600 | [diff] [blame] | 4601 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4602 | /* |
| 4603 | * The idle tasks have their own, simple scheduling class: |
| 4604 | */ |
| 4605 | idle->sched_class = &idle_sched_class; |
Steven Rostedt | 868baf0 | 2011-02-10 21:26:13 -0500 | [diff] [blame] | 4606 | ftrace_graph_init_idle_task(idle, cpu); |
Carsten Emde | f1c6f1a | 2011-10-26 23:14:16 +0200 | [diff] [blame] | 4607 | #if defined(CONFIG_SMP) |
| 4608 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
| 4609 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4610 | } |
| 4611 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4612 | #ifdef CONFIG_SMP |
KOSAKI Motohiro | 1e1b6c5 | 2011-05-19 15:08:58 +0900 | [diff] [blame] | 4613 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
| 4614 | { |
| 4615 | if (p->sched_class && p->sched_class->set_cpus_allowed) |
| 4616 | p->sched_class->set_cpus_allowed(p, new_mask); |
Peter Zijlstra | 4939602 | 2011-06-25 15:45:46 +0200 | [diff] [blame] | 4617 | |
| 4618 | cpumask_copy(&p->cpus_allowed, new_mask); |
Peter Zijlstra | 29baa74 | 2012-04-23 12:11:21 +0200 | [diff] [blame] | 4619 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
KOSAKI Motohiro | 1e1b6c5 | 2011-05-19 15:08:58 +0900 | [diff] [blame] | 4620 | } |
| 4621 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4622 | /* |
| 4623 | * This is how migration works: |
| 4624 | * |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 4625 | * 1) we invoke migration_cpu_stop() on the target CPU using |
| 4626 | * stop_one_cpu(). |
| 4627 | * 2) stopper starts to run (implicitly forcing the migrated thread |
| 4628 | * off the CPU) |
| 4629 | * 3) it checks whether the migrated task is still in the wrong runqueue. |
| 4630 | * 4) if it's in the wrong runqueue then the migration thread removes |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4631 | * it and puts it into the right queue. |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 4632 | * 5) stopper completes and stop_one_cpu() returns and the migration |
| 4633 | * is done. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4634 | */ |
| 4635 | |
| 4636 | /* |
| 4637 | * Change a given task's CPU affinity. Migrate the thread to a |
| 4638 | * proper CPU and schedule it away if the CPU it's executing on |
| 4639 | * is removed from the allowed bitmask. |
| 4640 | * |
| 4641 | * NOTE: the caller must have a valid reference to the task, the |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 4642 | * task must not exit() & deallocate itself prematurely. The |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4643 | * call is not atomic; no spinlocks may be held. |
| 4644 | */ |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 4645 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4646 | { |
| 4647 | unsigned long flags; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 4648 | struct rq *rq; |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 4649 | unsigned int dest_cpu; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 4650 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4651 | |
| 4652 | rq = task_rq_lock(p, &flags); |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 4653 | |
Yong Zhang | db44fc0 | 2011-05-09 22:07:05 +0800 | [diff] [blame] | 4654 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
| 4655 | goto out; |
| 4656 | |
Peter Zijlstra | 6ad4c18 | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 4657 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4658 | ret = -EINVAL; |
| 4659 | goto out; |
| 4660 | } |
| 4661 | |
Yong Zhang | db44fc0 | 2011-05-09 22:07:05 +0800 | [diff] [blame] | 4662 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { |
David Rientjes | 9985b0b | 2008-06-05 12:57:11 -0700 | [diff] [blame] | 4663 | ret = -EINVAL; |
| 4664 | goto out; |
| 4665 | } |
| 4666 | |
KOSAKI Motohiro | 1e1b6c5 | 2011-05-19 15:08:58 +0900 | [diff] [blame] | 4667 | do_set_cpus_allowed(p, new_mask); |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 4668 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4669 | /* Can the task run on the task's current CPU? If so, we're done */ |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 4670 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4671 | goto out; |
| 4672 | |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 4673 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
Peter Zijlstra | bd8e7dd | 2011-04-05 17:23:59 +0200 | [diff] [blame] | 4674 | if (p->on_rq) { |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 4675 | struct migration_arg arg = { p, dest_cpu }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4676 | /* Need help from migration thread: drop lock and wait. */ |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 4677 | task_rq_unlock(rq, p, &flags); |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 4678 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4679 | tlb_migrate_finish(p->mm); |
| 4680 | return 0; |
| 4681 | } |
| 4682 | out: |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 4683 | task_rq_unlock(rq, p, &flags); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 4684 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4685 | return ret; |
| 4686 | } |
Mike Travis | cd8ba7c | 2008-03-26 14:23:49 -0700 | [diff] [blame] | 4687 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4688 | |
| 4689 | /* |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 4690 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4691 | * this because either it can't run here any more (set_cpus_allowed() |
| 4692 | * away from this CPU, or CPU going down), or because we're |
| 4693 | * attempting to rebalance this task on exec (sched_exec). |
| 4694 | * |
| 4695 | * So we race with normal scheduler movements, but that's OK, as long |
| 4696 | * as the task is no longer on this CPU. |
Kirill Korotaev | efc3081 | 2006-06-27 02:54:32 -0700 | [diff] [blame] | 4697 | * |
| 4698 | * Returns non-zero if task was successfully migrated. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4699 | */ |
Kirill Korotaev | efc3081 | 2006-06-27 02:54:32 -0700 | [diff] [blame] | 4700 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4701 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 4702 | struct rq *rq_dest, *rq_src; |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 4703 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4704 | |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 4705 | if (unlikely(!cpu_active(dest_cpu))) |
Kirill Korotaev | efc3081 | 2006-06-27 02:54:32 -0700 | [diff] [blame] | 4706 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4707 | |
| 4708 | rq_src = cpu_rq(src_cpu); |
| 4709 | rq_dest = cpu_rq(dest_cpu); |
| 4710 | |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 4711 | raw_spin_lock(&p->pi_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4712 | double_rq_lock(rq_src, rq_dest); |
| 4713 | /* Already moved. */ |
| 4714 | if (task_cpu(p) != src_cpu) |
Linus Torvalds | b1e3873 | 2008-07-10 11:25:03 -0700 | [diff] [blame] | 4715 | goto done; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4716 | /* Affinity changed (again). */ |
Peter Zijlstra | fa17b50 | 2011-06-16 12:23:22 +0200 | [diff] [blame] | 4717 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
Linus Torvalds | b1e3873 | 2008-07-10 11:25:03 -0700 | [diff] [blame] | 4718 | goto fail; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4719 | |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 4720 | /* |
| 4721 | * If we're not on a rq, the next wake-up will ensure we're |
| 4722 | * placed properly. |
| 4723 | */ |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 4724 | if (p->on_rq) { |
Peter Zijlstra | 4ca9b72 | 2012-01-25 11:50:51 +0100 | [diff] [blame] | 4725 | dequeue_task(rq_src, p, 0); |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 4726 | set_task_cpu(p, dest_cpu); |
Peter Zijlstra | 4ca9b72 | 2012-01-25 11:50:51 +0100 | [diff] [blame] | 4727 | enqueue_task(rq_dest, p, 0); |
Peter Zijlstra | 15afe09 | 2008-09-20 23:38:02 +0200 | [diff] [blame] | 4728 | check_preempt_curr(rq_dest, p, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4729 | } |
Linus Torvalds | b1e3873 | 2008-07-10 11:25:03 -0700 | [diff] [blame] | 4730 | done: |
Kirill Korotaev | efc3081 | 2006-06-27 02:54:32 -0700 | [diff] [blame] | 4731 | ret = 1; |
Linus Torvalds | b1e3873 | 2008-07-10 11:25:03 -0700 | [diff] [blame] | 4732 | fail: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4733 | double_rq_unlock(rq_src, rq_dest); |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 4734 | raw_spin_unlock(&p->pi_lock); |
Kirill Korotaev | efc3081 | 2006-06-27 02:54:32 -0700 | [diff] [blame] | 4735 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4736 | } |
| 4737 | |
| 4738 | /* |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 4739 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
| 4740 | * and performs thread migration by bumping thread off CPU then |
| 4741 | * 'pushing' onto another runqueue. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4742 | */ |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 4743 | static int migration_cpu_stop(void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4744 | { |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 4745 | struct migration_arg *arg = data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4746 | |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 4747 | /* |
| 4748 | * The original target cpu might have gone down and we might |
| 4749 | * be on another cpu but it doesn't matter. |
| 4750 | */ |
| 4751 | local_irq_disable(); |
| 4752 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
| 4753 | local_irq_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4754 | return 0; |
| 4755 | } |
| 4756 | |
| 4757 | #ifdef CONFIG_HOTPLUG_CPU |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4758 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 4759 | /* |
| 4760 | * Ensures that the idle task is using init_mm right before its cpu goes |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4761 | * offline. |
| 4762 | */ |
| 4763 | void idle_task_exit(void) |
| 4764 | { |
| 4765 | struct mm_struct *mm = current->active_mm; |
| 4766 | |
| 4767 | BUG_ON(cpu_online(smp_processor_id())); |
| 4768 | |
| 4769 | if (mm != &init_mm) |
| 4770 | switch_mm(mm, &init_mm, current); |
| 4771 | mmdrop(mm); |
| 4772 | } |
| 4773 | |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 4774 | /* |
Peter Zijlstra | 5d18023 | 2012-08-20 11:26:57 +0200 | [diff] [blame] | 4775 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
| 4776 | * we might have. Assumes we're called after migrate_tasks() so that the |
| 4777 | * nr_active count is stable. |
| 4778 | * |
| 4779 | * Also see the comment "Global load-average calculations". |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 4780 | */ |
Peter Zijlstra | 5d18023 | 2012-08-20 11:26:57 +0200 | [diff] [blame] | 4781 | static void calc_load_migrate(struct rq *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4782 | { |
Peter Zijlstra | 5d18023 | 2012-08-20 11:26:57 +0200 | [diff] [blame] | 4783 | long delta = calc_load_fold_active(rq); |
| 4784 | if (delta) |
| 4785 | atomic_long_add(delta, &calc_load_tasks); |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 4786 | } |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 4787 | |
| 4788 | /* |
| 4789 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
| 4790 | * try_to_wake_up()->select_task_rq(). |
| 4791 | * |
| 4792 | * Called with rq->lock held even though we'er in stop_machine() and |
| 4793 | * there's no concurrency possible, we hold the required locks anyway |
| 4794 | * because of lock validation efforts. |
| 4795 | */ |
| 4796 | static void migrate_tasks(unsigned int dead_cpu) |
| 4797 | { |
| 4798 | struct rq *rq = cpu_rq(dead_cpu); |
| 4799 | struct task_struct *next, *stop = rq->stop; |
| 4800 | int dest_cpu; |
| 4801 | |
| 4802 | /* |
| 4803 | * Fudge the rq selection such that the below task selection loop |
| 4804 | * doesn't get stuck on the currently eligible stop task. |
| 4805 | * |
| 4806 | * We're currently inside stop_machine() and the rq is either stuck |
| 4807 | * in the stop_machine_cpu_stop() loop, or we're executing this code, |
| 4808 | * either way we should never end up calling schedule() until we're |
| 4809 | * done here. |
| 4810 | */ |
| 4811 | rq->stop = NULL; |
| 4812 | |
| 4813 | for ( ; ; ) { |
| 4814 | /* |
| 4815 | * There's this thread running, bail when that's the only |
| 4816 | * remaining thread. |
| 4817 | */ |
| 4818 | if (rq->nr_running == 1) |
| 4819 | break; |
| 4820 | |
| 4821 | next = pick_next_task(rq); |
| 4822 | BUG_ON(!next); |
| 4823 | next->sched_class->put_prev_task(rq, next); |
| 4824 | |
| 4825 | /* Find suitable destination for @next, with force if needed. */ |
| 4826 | dest_cpu = select_fallback_rq(dead_cpu, next); |
| 4827 | raw_spin_unlock(&rq->lock); |
| 4828 | |
| 4829 | __migrate_task(next, dead_cpu, dest_cpu); |
| 4830 | |
| 4831 | raw_spin_lock(&rq->lock); |
| 4832 | } |
| 4833 | |
| 4834 | rq->stop = stop; |
| 4835 | } |
| 4836 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4837 | #endif /* CONFIG_HOTPLUG_CPU */ |
| 4838 | |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4839 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
| 4840 | |
| 4841 | static struct ctl_table sd_ctl_dir[] = { |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4842 | { |
| 4843 | .procname = "sched_domain", |
Eric W. Biederman | c57baf1 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 4844 | .mode = 0555, |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4845 | }, |
Eric W. Biederman | 5699230 | 2009-11-05 15:38:40 -0800 | [diff] [blame] | 4846 | {} |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4847 | }; |
| 4848 | |
| 4849 | static struct ctl_table sd_ctl_root[] = { |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4850 | { |
| 4851 | .procname = "kernel", |
Eric W. Biederman | c57baf1 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 4852 | .mode = 0555, |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4853 | .child = sd_ctl_dir, |
| 4854 | }, |
Eric W. Biederman | 5699230 | 2009-11-05 15:38:40 -0800 | [diff] [blame] | 4855 | {} |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4856 | }; |
| 4857 | |
| 4858 | static struct ctl_table *sd_alloc_ctl_entry(int n) |
| 4859 | { |
| 4860 | struct ctl_table *entry = |
Milton Miller | 5cf9f06 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 4861 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4862 | |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4863 | return entry; |
| 4864 | } |
| 4865 | |
Milton Miller | 6382bc9 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 4866 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
| 4867 | { |
Milton Miller | cd790076 | 2007-10-17 16:55:11 +0200 | [diff] [blame] | 4868 | struct ctl_table *entry; |
Milton Miller | 6382bc9 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 4869 | |
Milton Miller | cd790076 | 2007-10-17 16:55:11 +0200 | [diff] [blame] | 4870 | /* |
| 4871 | * In the intermediate directories, both the child directory and |
| 4872 | * procname are dynamically allocated and could fail but the mode |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 4873 | * will always be set. In the lowest directory the names are |
Milton Miller | cd790076 | 2007-10-17 16:55:11 +0200 | [diff] [blame] | 4874 | * static strings and all have proc handlers. |
| 4875 | */ |
| 4876 | for (entry = *tablep; entry->mode; entry++) { |
Milton Miller | 6382bc9 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 4877 | if (entry->child) |
| 4878 | sd_free_ctl_entry(&entry->child); |
Milton Miller | cd790076 | 2007-10-17 16:55:11 +0200 | [diff] [blame] | 4879 | if (entry->proc_handler == NULL) |
| 4880 | kfree(entry->procname); |
| 4881 | } |
Milton Miller | 6382bc9 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 4882 | |
| 4883 | kfree(*tablep); |
| 4884 | *tablep = NULL; |
| 4885 | } |
| 4886 | |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4887 | static int min_load_idx = 0; |
| 4888 | static int max_load_idx = CPU_LOAD_IDX_MAX; |
| 4889 | |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4890 | static void |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4891 | set_table_entry(struct ctl_table *entry, |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4892 | const char *procname, void *data, int maxlen, |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4893 | umode_t mode, proc_handler *proc_handler, |
| 4894 | bool load_idx) |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4895 | { |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4896 | entry->procname = procname; |
| 4897 | entry->data = data; |
| 4898 | entry->maxlen = maxlen; |
| 4899 | entry->mode = mode; |
| 4900 | entry->proc_handler = proc_handler; |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4901 | |
| 4902 | if (load_idx) { |
| 4903 | entry->extra1 = &min_load_idx; |
| 4904 | entry->extra2 = &max_load_idx; |
| 4905 | } |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4906 | } |
| 4907 | |
| 4908 | static struct ctl_table * |
| 4909 | sd_alloc_ctl_domain_table(struct sched_domain *sd) |
| 4910 | { |
Ingo Molnar | a5d8c34 | 2008-10-09 11:35:51 +0200 | [diff] [blame] | 4911 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4912 | |
Milton Miller | ad1cdc1 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 4913 | if (table == NULL) |
| 4914 | return NULL; |
| 4915 | |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4916 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4917 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4918 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4919 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4920 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4921 | sizeof(int), 0644, proc_dointvec_minmax, true); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4922 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4923 | sizeof(int), 0644, proc_dointvec_minmax, true); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4924 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4925 | sizeof(int), 0644, proc_dointvec_minmax, true); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4926 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4927 | sizeof(int), 0644, proc_dointvec_minmax, true); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4928 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4929 | sizeof(int), 0644, proc_dointvec_minmax, true); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4930 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4931 | sizeof(int), 0644, proc_dointvec_minmax, false); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4932 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4933 | sizeof(int), 0644, proc_dointvec_minmax, false); |
Zou Nan hai | ace8b3d | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 4934 | set_table_entry(&table[9], "cache_nice_tries", |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4935 | &sd->cache_nice_tries, |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4936 | sizeof(int), 0644, proc_dointvec_minmax, false); |
Zou Nan hai | ace8b3d | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 4937 | set_table_entry(&table[10], "flags", &sd->flags, |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4938 | sizeof(int), 0644, proc_dointvec_minmax, false); |
Ingo Molnar | a5d8c34 | 2008-10-09 11:35:51 +0200 | [diff] [blame] | 4939 | set_table_entry(&table[11], "name", sd->name, |
Namhyung Kim | 201c373 | 2012-08-16 17:03:24 +0900 | [diff] [blame] | 4940 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
Ingo Molnar | a5d8c34 | 2008-10-09 11:35:51 +0200 | [diff] [blame] | 4941 | /* &table[12] is terminator */ |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4942 | |
| 4943 | return table; |
| 4944 | } |
| 4945 | |
Ingo Molnar | 9a4e715 | 2007-11-28 15:52:56 +0100 | [diff] [blame] | 4946 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4947 | { |
| 4948 | struct ctl_table *entry, *table; |
| 4949 | struct sched_domain *sd; |
| 4950 | int domain_num = 0, i; |
| 4951 | char buf[32]; |
| 4952 | |
| 4953 | for_each_domain(cpu, sd) |
| 4954 | domain_num++; |
| 4955 | entry = table = sd_alloc_ctl_entry(domain_num + 1); |
Milton Miller | ad1cdc1 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 4956 | if (table == NULL) |
| 4957 | return NULL; |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4958 | |
| 4959 | i = 0; |
| 4960 | for_each_domain(cpu, sd) { |
| 4961 | snprintf(buf, 32, "domain%d", i); |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4962 | entry->procname = kstrdup(buf, GFP_KERNEL); |
Eric W. Biederman | c57baf1 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 4963 | entry->mode = 0555; |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4964 | entry->child = sd_alloc_ctl_domain_table(sd); |
| 4965 | entry++; |
| 4966 | i++; |
| 4967 | } |
| 4968 | return table; |
| 4969 | } |
| 4970 | |
| 4971 | static struct ctl_table_header *sd_sysctl_header; |
Milton Miller | 6382bc9 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 4972 | static void register_sched_domain_sysctl(void) |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4973 | { |
Peter Zijlstra | 6ad4c18 | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 4974 | int i, cpu_num = num_possible_cpus(); |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4975 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
| 4976 | char buf[32]; |
| 4977 | |
Milton Miller | 7378547 | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 4978 | WARN_ON(sd_ctl_dir[0].child); |
| 4979 | sd_ctl_dir[0].child = entry; |
| 4980 | |
Milton Miller | ad1cdc1 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 4981 | if (entry == NULL) |
| 4982 | return; |
| 4983 | |
Peter Zijlstra | 6ad4c18 | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 4984 | for_each_possible_cpu(i) { |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4985 | snprintf(buf, 32, "cpu%d", i); |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4986 | entry->procname = kstrdup(buf, GFP_KERNEL); |
Eric W. Biederman | c57baf1 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 4987 | entry->mode = 0555; |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4988 | entry->child = sd_alloc_ctl_cpu_table(i); |
Milton Miller | 97b6ea7 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 4989 | entry++; |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4990 | } |
Milton Miller | 7378547 | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 4991 | |
| 4992 | WARN_ON(sd_sysctl_header); |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 4993 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
| 4994 | } |
Milton Miller | 6382bc9 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 4995 | |
Milton Miller | 7378547 | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 4996 | /* may be called multiple times per register */ |
Milton Miller | 6382bc9 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 4997 | static void unregister_sched_domain_sysctl(void) |
| 4998 | { |
Milton Miller | 7378547 | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 4999 | if (sd_sysctl_header) |
| 5000 | unregister_sysctl_table(sd_sysctl_header); |
Milton Miller | 6382bc9 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 5001 | sd_sysctl_header = NULL; |
Milton Miller | 7378547 | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5002 | if (sd_ctl_dir[0].child) |
| 5003 | sd_free_ctl_entry(&sd_ctl_dir[0].child); |
Milton Miller | 6382bc9 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 5004 | } |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5005 | #else |
Milton Miller | 6382bc9 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 5006 | static void register_sched_domain_sysctl(void) |
| 5007 | { |
| 5008 | } |
| 5009 | static void unregister_sched_domain_sysctl(void) |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5010 | { |
| 5011 | } |
| 5012 | #endif |
| 5013 | |
Gregory Haskins | 1f11eb6 | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 5014 | static void set_rq_online(struct rq *rq) |
| 5015 | { |
| 5016 | if (!rq->online) { |
| 5017 | const struct sched_class *class; |
| 5018 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5019 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
Gregory Haskins | 1f11eb6 | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 5020 | rq->online = 1; |
| 5021 | |
| 5022 | for_each_class(class) { |
| 5023 | if (class->rq_online) |
| 5024 | class->rq_online(rq); |
| 5025 | } |
| 5026 | } |
| 5027 | } |
| 5028 | |
| 5029 | static void set_rq_offline(struct rq *rq) |
| 5030 | { |
| 5031 | if (rq->online) { |
| 5032 | const struct sched_class *class; |
| 5033 | |
| 5034 | for_each_class(class) { |
| 5035 | if (class->rq_offline) |
| 5036 | class->rq_offline(rq); |
| 5037 | } |
| 5038 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5039 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
Gregory Haskins | 1f11eb6 | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 5040 | rq->online = 0; |
| 5041 | } |
| 5042 | } |
| 5043 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5044 | /* |
| 5045 | * migration_call - callback that gets triggered when a CPU is added. |
| 5046 | * Here we can start up the necessary migration thread for the new CPU. |
| 5047 | */ |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5048 | static int __cpuinit |
| 5049 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5050 | { |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5051 | int cpu = (long)hcpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5052 | unsigned long flags; |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 5053 | struct rq *rq = cpu_rq(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5054 | |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5055 | switch (action & ~CPU_TASKS_FROZEN) { |
Gautham R Shenoy | 5be9361 | 2007-05-09 02:34:04 -0700 | [diff] [blame] | 5056 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5057 | case CPU_UP_PREPARE: |
Thomas Gleixner | a468d38 | 2009-07-17 14:15:46 +0200 | [diff] [blame] | 5058 | rq->calc_load_update = calc_load_update; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5059 | break; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5060 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5061 | case CPU_ONLINE: |
Gregory Haskins | 1f94ef5 | 2008-03-10 16:52:41 -0400 | [diff] [blame] | 5062 | /* Update our root-domain */ |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 5063 | raw_spin_lock_irqsave(&rq->lock, flags); |
Gregory Haskins | 1f94ef5 | 2008-03-10 16:52:41 -0400 | [diff] [blame] | 5064 | if (rq->rd) { |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5065 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
Gregory Haskins | 1f11eb6 | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 5066 | |
| 5067 | set_rq_online(rq); |
Gregory Haskins | 1f94ef5 | 2008-03-10 16:52:41 -0400 | [diff] [blame] | 5068 | } |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 5069 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5070 | break; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5071 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5072 | #ifdef CONFIG_HOTPLUG_CPU |
Gregory Haskins | 08f503b | 2008-03-10 17:59:11 -0400 | [diff] [blame] | 5073 | case CPU_DYING: |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 5074 | sched_ttwu_pending(); |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5075 | /* Update our root-domain */ |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 5076 | raw_spin_lock_irqsave(&rq->lock, flags); |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5077 | if (rq->rd) { |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5078 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
Gregory Haskins | 1f11eb6 | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 5079 | set_rq_offline(rq); |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5080 | } |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5081 | migrate_tasks(cpu); |
| 5082 | BUG_ON(rq->nr_running != 1); /* the migration thread */ |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 5083 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
Peter Zijlstra | 5d18023 | 2012-08-20 11:26:57 +0200 | [diff] [blame] | 5084 | break; |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5085 | |
Peter Zijlstra | 5d18023 | 2012-08-20 11:26:57 +0200 | [diff] [blame] | 5086 | case CPU_DEAD: |
Peter Zijlstra | f319da0 | 2012-08-20 11:26:57 +0200 | [diff] [blame] | 5087 | calc_load_migrate(rq); |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5088 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5089 | #endif |
| 5090 | } |
Peter Zijlstra | 49c022e | 2011-04-05 10:14:25 +0200 | [diff] [blame] | 5091 | |
| 5092 | update_max_interval(); |
| 5093 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5094 | return NOTIFY_OK; |
| 5095 | } |
| 5096 | |
Paul Mackerras | f38b082 | 2009-06-02 21:05:16 +1000 | [diff] [blame] | 5097 | /* |
| 5098 | * Register at high priority so that task migration (migrate_all_tasks) |
| 5099 | * happens before everything else. This has to be lower priority than |
Ingo Molnar | cdd6c48 | 2009-09-21 12:02:48 +0200 | [diff] [blame] | 5100 | * the notifier in the perf_event subsystem, though. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5101 | */ |
Chandra Seetharaman | 26c2143 | 2006-06-27 02:54:10 -0700 | [diff] [blame] | 5102 | static struct notifier_block __cpuinitdata migration_notifier = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5103 | .notifier_call = migration_call, |
Tejun Heo | 50a323b | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 5104 | .priority = CPU_PRI_MIGRATION, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5105 | }; |
| 5106 | |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 5107 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
| 5108 | unsigned long action, void *hcpu) |
| 5109 | { |
| 5110 | switch (action & ~CPU_TASKS_FROZEN) { |
Peter Zijlstra | 5fbd036 | 2011-12-15 17:09:22 +0100 | [diff] [blame] | 5111 | case CPU_STARTING: |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 5112 | case CPU_DOWN_FAILED: |
| 5113 | set_cpu_active((long)hcpu, true); |
| 5114 | return NOTIFY_OK; |
| 5115 | default: |
| 5116 | return NOTIFY_DONE; |
| 5117 | } |
| 5118 | } |
| 5119 | |
| 5120 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, |
| 5121 | unsigned long action, void *hcpu) |
| 5122 | { |
| 5123 | switch (action & ~CPU_TASKS_FROZEN) { |
| 5124 | case CPU_DOWN_PREPARE: |
| 5125 | set_cpu_active((long)hcpu, false); |
| 5126 | return NOTIFY_OK; |
| 5127 | default: |
| 5128 | return NOTIFY_DONE; |
| 5129 | } |
| 5130 | } |
| 5131 | |
Eduard - Gabriel Munteanu | 7babe8d | 2008-07-25 19:45:11 -0700 | [diff] [blame] | 5132 | static int __init migration_init(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5133 | { |
| 5134 | void *cpu = (void *)(long)smp_processor_id(); |
Akinobu Mita | 07dccf3 | 2006-09-29 02:00:22 -0700 | [diff] [blame] | 5135 | int err; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5136 | |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 5137 | /* Initialize migration for the boot CPU */ |
Akinobu Mita | 07dccf3 | 2006-09-29 02:00:22 -0700 | [diff] [blame] | 5138 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
| 5139 | BUG_ON(err == NOTIFY_BAD); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5140 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
| 5141 | register_cpu_notifier(&migration_notifier); |
Eduard - Gabriel Munteanu | 7babe8d | 2008-07-25 19:45:11 -0700 | [diff] [blame] | 5142 | |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 5143 | /* Register cpu active notifiers */ |
| 5144 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); |
| 5145 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); |
| 5146 | |
Thomas Gleixner | a004cd4 | 2009-07-21 09:54:05 +0200 | [diff] [blame] | 5147 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5148 | } |
Eduard - Gabriel Munteanu | 7babe8d | 2008-07-25 19:45:11 -0700 | [diff] [blame] | 5149 | early_initcall(migration_init); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5150 | #endif |
| 5151 | |
| 5152 | #ifdef CONFIG_SMP |
Christoph Lameter | 476f353 | 2007-05-06 14:48:58 -0700 | [diff] [blame] | 5153 | |
Peter Zijlstra | 4cb9883 | 2011-04-07 14:09:58 +0200 | [diff] [blame] | 5154 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
| 5155 | |
Ingo Molnar | 3e9830d | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 5156 | #ifdef CONFIG_SCHED_DEBUG |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5157 | |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 5158 | static __read_mostly int sched_debug_enabled; |
Mike Travis | f663011 | 2009-11-17 18:22:15 -0600 | [diff] [blame] | 5159 | |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 5160 | static int __init sched_debug_setup(char *str) |
Mike Travis | f663011 | 2009-11-17 18:22:15 -0600 | [diff] [blame] | 5161 | { |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 5162 | sched_debug_enabled = 1; |
Mike Travis | f663011 | 2009-11-17 18:22:15 -0600 | [diff] [blame] | 5163 | |
| 5164 | return 0; |
| 5165 | } |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 5166 | early_param("sched_debug", sched_debug_setup); |
| 5167 | |
| 5168 | static inline bool sched_debug(void) |
| 5169 | { |
| 5170 | return sched_debug_enabled; |
| 5171 | } |
Mike Travis | f663011 | 2009-11-17 18:22:15 -0600 | [diff] [blame] | 5172 | |
Mike Travis | 7c16ec5 | 2008-04-04 18:11:11 -0700 | [diff] [blame] | 5173 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 5174 | struct cpumask *groupmask) |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5175 | { |
| 5176 | struct sched_group *group = sd->groups; |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 5177 | char str[256]; |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5178 | |
Rusty Russell | 968ea6d | 2008-12-13 21:55:51 +1030 | [diff] [blame] | 5179 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 5180 | cpumask_clear(groupmask); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5181 | |
| 5182 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); |
| 5183 | |
| 5184 | if (!(sd->flags & SD_LOAD_BALANCE)) { |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5185 | printk("does not load-balance\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5186 | if (sd->parent) |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5187 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
| 5188 | " has parent"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5189 | return -1; |
| 5190 | } |
| 5191 | |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5192 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5193 | |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5194 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5195 | printk(KERN_ERR "ERROR: domain->span does not contain " |
| 5196 | "CPU%d\n", cpu); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5197 | } |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5198 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5199 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
| 5200 | " CPU%d\n", cpu); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5201 | } |
| 5202 | |
| 5203 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
| 5204 | do { |
| 5205 | if (!group) { |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5206 | printk("\n"); |
| 5207 | printk(KERN_ERR "ERROR: group is NULL\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5208 | break; |
| 5209 | } |
| 5210 | |
Peter Zijlstra | c3decf0 | 2012-05-31 12:05:32 +0200 | [diff] [blame] | 5211 | /* |
| 5212 | * Even though we initialize ->power to something semi-sane, |
| 5213 | * we leave power_orig unset. This allows us to detect if |
| 5214 | * domain iteration is still funny without causing /0 traps. |
| 5215 | */ |
| 5216 | if (!group->sgp->power_orig) { |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5217 | printk(KERN_CONT "\n"); |
| 5218 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
| 5219 | "set\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5220 | break; |
| 5221 | } |
| 5222 | |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5223 | if (!cpumask_weight(sched_group_cpus(group))) { |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5224 | printk(KERN_CONT "\n"); |
| 5225 | printk(KERN_ERR "ERROR: empty group\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5226 | break; |
| 5227 | } |
| 5228 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 5229 | if (!(sd->flags & SD_OVERLAP) && |
| 5230 | cpumask_intersects(groupmask, sched_group_cpus(group))) { |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5231 | printk(KERN_CONT "\n"); |
| 5232 | printk(KERN_ERR "ERROR: repeated CPUs\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5233 | break; |
| 5234 | } |
| 5235 | |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5236 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5237 | |
Rusty Russell | 968ea6d | 2008-12-13 21:55:51 +1030 | [diff] [blame] | 5238 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
Gautham R Shenoy | 381512c | 2009-04-14 09:09:36 +0530 | [diff] [blame] | 5239 | |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5240 | printk(KERN_CONT " %s", str); |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 5241 | if (group->sgp->power != SCHED_POWER_SCALE) { |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5242 | printk(KERN_CONT " (cpu_power = %d)", |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 5243 | group->sgp->power); |
Gautham R Shenoy | 381512c | 2009-04-14 09:09:36 +0530 | [diff] [blame] | 5244 | } |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5245 | |
| 5246 | group = group->next; |
| 5247 | } while (group != sd->groups); |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5248 | printk(KERN_CONT "\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5249 | |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5250 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5251 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5252 | |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5253 | if (sd->parent && |
| 5254 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5255 | printk(KERN_ERR "ERROR: parent span is not a superset " |
| 5256 | "of domain->span\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5257 | return 0; |
| 5258 | } |
| 5259 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5260 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
| 5261 | { |
| 5262 | int level = 0; |
| 5263 | |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 5264 | if (!sched_debug_enabled) |
Mike Travis | f663011 | 2009-11-17 18:22:15 -0600 | [diff] [blame] | 5265 | return; |
| 5266 | |
Nick Piggin | 41c7ce9 | 2005-06-25 14:57:24 -0700 | [diff] [blame] | 5267 | if (!sd) { |
| 5268 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); |
| 5269 | return; |
| 5270 | } |
| 5271 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5272 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
| 5273 | |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5274 | for (;;) { |
Peter Zijlstra | 4cb9883 | 2011-04-07 14:09:58 +0200 | [diff] [blame] | 5275 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5276 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5277 | level++; |
| 5278 | sd = sd->parent; |
Miguel Ojeda Sandonis | 33859f7 | 2006-12-10 02:20:38 -0800 | [diff] [blame] | 5279 | if (!sd) |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5280 | break; |
| 5281 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5282 | } |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 5283 | #else /* !CONFIG_SCHED_DEBUG */ |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5284 | # define sched_domain_debug(sd, cpu) do { } while (0) |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 5285 | static inline bool sched_debug(void) |
| 5286 | { |
| 5287 | return false; |
| 5288 | } |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 5289 | #endif /* CONFIG_SCHED_DEBUG */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5290 | |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 5291 | static int sd_degenerate(struct sched_domain *sd) |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5292 | { |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5293 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5294 | return 1; |
| 5295 | |
| 5296 | /* Following flags need at least 2 groups */ |
| 5297 | if (sd->flags & (SD_LOAD_BALANCE | |
| 5298 | SD_BALANCE_NEWIDLE | |
| 5299 | SD_BALANCE_FORK | |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 5300 | SD_BALANCE_EXEC | |
| 5301 | SD_SHARE_CPUPOWER | |
| 5302 | SD_SHARE_PKG_RESOURCES)) { |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5303 | if (sd->groups != sd->groups->next) |
| 5304 | return 0; |
| 5305 | } |
| 5306 | |
| 5307 | /* Following flags don't use groups */ |
Peter Zijlstra | c88d591 | 2009-09-10 13:50:02 +0200 | [diff] [blame] | 5308 | if (sd->flags & (SD_WAKE_AFFINE)) |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5309 | return 0; |
| 5310 | |
| 5311 | return 1; |
| 5312 | } |
| 5313 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5314 | static int |
| 5315 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5316 | { |
| 5317 | unsigned long cflags = sd->flags, pflags = parent->flags; |
| 5318 | |
| 5319 | if (sd_degenerate(parent)) |
| 5320 | return 1; |
| 5321 | |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5322 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5323 | return 0; |
| 5324 | |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5325 | /* Flags needing groups don't count if only 1 group in parent */ |
| 5326 | if (parent->groups == parent->groups->next) { |
| 5327 | pflags &= ~(SD_LOAD_BALANCE | |
| 5328 | SD_BALANCE_NEWIDLE | |
| 5329 | SD_BALANCE_FORK | |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 5330 | SD_BALANCE_EXEC | |
| 5331 | SD_SHARE_CPUPOWER | |
| 5332 | SD_SHARE_PKG_RESOURCES); |
Ken Chen | 5436499 | 2008-12-07 18:47:37 -0800 | [diff] [blame] | 5333 | if (nr_node_ids == 1) |
| 5334 | pflags &= ~SD_SERIALIZE; |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5335 | } |
| 5336 | if (~cflags & pflags) |
| 5337 | return 0; |
| 5338 | |
| 5339 | return 1; |
| 5340 | } |
| 5341 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5342 | static void free_rootdomain(struct rcu_head *rcu) |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5343 | { |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5344 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
Peter Zijlstra | 047106a | 2009-11-16 10:28:09 +0100 | [diff] [blame] | 5345 | |
Rusty Russell | 68e7456 | 2008-11-25 02:35:13 +1030 | [diff] [blame] | 5346 | cpupri_cleanup(&rd->cpupri); |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5347 | free_cpumask_var(rd->rto_mask); |
| 5348 | free_cpumask_var(rd->online); |
| 5349 | free_cpumask_var(rd->span); |
| 5350 | kfree(rd); |
| 5351 | } |
| 5352 | |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5353 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
| 5354 | { |
Ingo Molnar | a0490fa | 2009-02-12 11:35:40 +0100 | [diff] [blame] | 5355 | struct root_domain *old_rd = NULL; |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5356 | unsigned long flags; |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5357 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 5358 | raw_spin_lock_irqsave(&rq->lock, flags); |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5359 | |
| 5360 | if (rq->rd) { |
Ingo Molnar | a0490fa | 2009-02-12 11:35:40 +0100 | [diff] [blame] | 5361 | old_rd = rq->rd; |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5362 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5363 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
Gregory Haskins | 1f11eb6 | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 5364 | set_rq_offline(rq); |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5365 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5366 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
Gregory Haskins | dc93852 | 2008-01-25 21:08:26 +0100 | [diff] [blame] | 5367 | |
Ingo Molnar | a0490fa | 2009-02-12 11:35:40 +0100 | [diff] [blame] | 5368 | /* |
| 5369 | * If we dont want to free the old_rt yet then |
| 5370 | * set old_rd to NULL to skip the freeing later |
| 5371 | * in this function: |
| 5372 | */ |
| 5373 | if (!atomic_dec_and_test(&old_rd->refcount)) |
| 5374 | old_rd = NULL; |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5375 | } |
| 5376 | |
| 5377 | atomic_inc(&rd->refcount); |
| 5378 | rq->rd = rd; |
| 5379 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5380 | cpumask_set_cpu(rq->cpu, rd->span); |
Gregory Haskins | 00aec93 | 2009-07-30 10:57:23 -0400 | [diff] [blame] | 5381 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
Gregory Haskins | 1f11eb6 | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 5382 | set_rq_online(rq); |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5383 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 5384 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
Ingo Molnar | a0490fa | 2009-02-12 11:35:40 +0100 | [diff] [blame] | 5385 | |
| 5386 | if (old_rd) |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5387 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5388 | } |
| 5389 | |
Pekka Enberg | 68c38fc | 2010-07-15 23:18:22 +0300 | [diff] [blame] | 5390 | static int init_rootdomain(struct root_domain *rd) |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5391 | { |
| 5392 | memset(rd, 0, sizeof(*rd)); |
| 5393 | |
Pekka Enberg | 68c38fc | 2010-07-15 23:18:22 +0300 | [diff] [blame] | 5394 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
Li Zefan | 0c910d2 | 2009-01-06 17:39:06 +0800 | [diff] [blame] | 5395 | goto out; |
Pekka Enberg | 68c38fc | 2010-07-15 23:18:22 +0300 | [diff] [blame] | 5396 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5397 | goto free_span; |
Pekka Enberg | 68c38fc | 2010-07-15 23:18:22 +0300 | [diff] [blame] | 5398 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5399 | goto free_online; |
Gregory Haskins | 6e0534f | 2008-05-12 21:21:01 +0200 | [diff] [blame] | 5400 | |
Pekka Enberg | 68c38fc | 2010-07-15 23:18:22 +0300 | [diff] [blame] | 5401 | if (cpupri_init(&rd->cpupri) != 0) |
Rusty Russell | 68e7456 | 2008-11-25 02:35:13 +1030 | [diff] [blame] | 5402 | goto free_rto_mask; |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5403 | return 0; |
| 5404 | |
Rusty Russell | 68e7456 | 2008-11-25 02:35:13 +1030 | [diff] [blame] | 5405 | free_rto_mask: |
| 5406 | free_cpumask_var(rd->rto_mask); |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5407 | free_online: |
| 5408 | free_cpumask_var(rd->online); |
| 5409 | free_span: |
| 5410 | free_cpumask_var(rd->span); |
Li Zefan | 0c910d2 | 2009-01-06 17:39:06 +0800 | [diff] [blame] | 5411 | out: |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5412 | return -ENOMEM; |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5413 | } |
| 5414 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 5415 | /* |
| 5416 | * By default the system creates a single root-domain with all cpus as |
| 5417 | * members (mimicking the global state we have today). |
| 5418 | */ |
| 5419 | struct root_domain def_root_domain; |
| 5420 | |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5421 | static void init_defrootdomain(void) |
| 5422 | { |
Pekka Enberg | 68c38fc | 2010-07-15 23:18:22 +0300 | [diff] [blame] | 5423 | init_rootdomain(&def_root_domain); |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5424 | |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5425 | atomic_set(&def_root_domain.refcount, 1); |
| 5426 | } |
| 5427 | |
Gregory Haskins | dc93852 | 2008-01-25 21:08:26 +0100 | [diff] [blame] | 5428 | static struct root_domain *alloc_rootdomain(void) |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5429 | { |
| 5430 | struct root_domain *rd; |
| 5431 | |
| 5432 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); |
| 5433 | if (!rd) |
| 5434 | return NULL; |
| 5435 | |
Pekka Enberg | 68c38fc | 2010-07-15 23:18:22 +0300 | [diff] [blame] | 5436 | if (init_rootdomain(rd) != 0) { |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5437 | kfree(rd); |
| 5438 | return NULL; |
| 5439 | } |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5440 | |
| 5441 | return rd; |
| 5442 | } |
| 5443 | |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5444 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
| 5445 | { |
| 5446 | struct sched_group *tmp, *first; |
| 5447 | |
| 5448 | if (!sg) |
| 5449 | return; |
| 5450 | |
| 5451 | first = sg; |
| 5452 | do { |
| 5453 | tmp = sg->next; |
| 5454 | |
| 5455 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) |
| 5456 | kfree(sg->sgp); |
| 5457 | |
| 5458 | kfree(sg); |
| 5459 | sg = tmp; |
| 5460 | } while (sg != first); |
| 5461 | } |
| 5462 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5463 | static void free_sched_domain(struct rcu_head *rcu) |
| 5464 | { |
| 5465 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5466 | |
| 5467 | /* |
| 5468 | * If its an overlapping domain it has private groups, iterate and |
| 5469 | * nuke them all. |
| 5470 | */ |
| 5471 | if (sd->flags & SD_OVERLAP) { |
| 5472 | free_sched_groups(sd->groups, 1); |
| 5473 | } else if (atomic_dec_and_test(&sd->groups->ref)) { |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 5474 | kfree(sd->groups->sgp); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5475 | kfree(sd->groups); |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 5476 | } |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5477 | kfree(sd); |
| 5478 | } |
| 5479 | |
| 5480 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) |
| 5481 | { |
| 5482 | call_rcu(&sd->rcu, free_sched_domain); |
| 5483 | } |
| 5484 | |
| 5485 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) |
| 5486 | { |
| 5487 | for (; sd; sd = sd->parent) |
| 5488 | destroy_sched_domain(sd, cpu); |
| 5489 | } |
| 5490 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5491 | /* |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 5492 | * Keep a special pointer to the highest sched_domain that has |
| 5493 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this |
| 5494 | * allows us to avoid some pointer chasing select_idle_sibling(). |
| 5495 | * |
| 5496 | * Also keep a unique ID per domain (we use the first cpu number in |
| 5497 | * the cpumask of the domain), this allows us to quickly tell if |
Peter Zijlstra | 39be350 | 2012-01-26 12:44:34 +0100 | [diff] [blame] | 5498 | * two cpus are in the same cache domain, see cpus_share_cache(). |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 5499 | */ |
| 5500 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); |
| 5501 | DEFINE_PER_CPU(int, sd_llc_id); |
| 5502 | |
| 5503 | static void update_top_cache_domain(int cpu) |
| 5504 | { |
| 5505 | struct sched_domain *sd; |
| 5506 | int id = cpu; |
| 5507 | |
| 5508 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); |
Linus Torvalds | 37407ea | 2012-09-16 12:29:43 -0700 | [diff] [blame] | 5509 | if (sd) |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 5510 | id = cpumask_first(sched_domain_span(sd)); |
| 5511 | |
| 5512 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); |
| 5513 | per_cpu(sd_llc_id, cpu) = id; |
| 5514 | } |
| 5515 | |
| 5516 | /* |
Ingo Molnar | 0eab914 | 2008-01-25 21:08:19 +0100 | [diff] [blame] | 5517 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5518 | * hold the hotplug lock. |
| 5519 | */ |
Ingo Molnar | 0eab914 | 2008-01-25 21:08:19 +0100 | [diff] [blame] | 5520 | static void |
| 5521 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5522 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5523 | struct rq *rq = cpu_rq(cpu); |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5524 | struct sched_domain *tmp; |
| 5525 | |
| 5526 | /* Remove the sched domains which do not contribute to scheduling. */ |
Li Zefan | f29c9b1 | 2008-11-06 09:45:16 +0800 | [diff] [blame] | 5527 | for (tmp = sd; tmp; ) { |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5528 | struct sched_domain *parent = tmp->parent; |
| 5529 | if (!parent) |
| 5530 | break; |
Li Zefan | f29c9b1 | 2008-11-06 09:45:16 +0800 | [diff] [blame] | 5531 | |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 5532 | if (sd_parent_degenerate(tmp, parent)) { |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5533 | tmp->parent = parent->parent; |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 5534 | if (parent->parent) |
| 5535 | parent->parent->child = tmp; |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5536 | destroy_sched_domain(parent, cpu); |
Li Zefan | f29c9b1 | 2008-11-06 09:45:16 +0800 | [diff] [blame] | 5537 | } else |
| 5538 | tmp = tmp->parent; |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5539 | } |
| 5540 | |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 5541 | if (sd && sd_degenerate(sd)) { |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5542 | tmp = sd; |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5543 | sd = sd->parent; |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5544 | destroy_sched_domain(tmp, cpu); |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 5545 | if (sd) |
| 5546 | sd->child = NULL; |
| 5547 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5548 | |
Peter Zijlstra | 4cb9883 | 2011-04-07 14:09:58 +0200 | [diff] [blame] | 5549 | sched_domain_debug(sd, cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5550 | |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5551 | rq_attach_root(rq, rd); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5552 | tmp = rq->sd; |
Nick Piggin | 674311d | 2005-06-25 14:57:27 -0700 | [diff] [blame] | 5553 | rcu_assign_pointer(rq->sd, sd); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5554 | destroy_sched_domains(tmp, cpu); |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 5555 | |
| 5556 | update_top_cache_domain(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5557 | } |
| 5558 | |
| 5559 | /* cpus with isolated domains */ |
Rusty Russell | dcc30a3 | 2008-11-25 02:35:12 +1030 | [diff] [blame] | 5560 | static cpumask_var_t cpu_isolated_map; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5561 | |
| 5562 | /* Setup the mask of cpus configured for isolated domains */ |
| 5563 | static int __init isolated_cpu_setup(char *str) |
| 5564 | { |
Rusty Russell | bdddd29 | 2009-12-02 14:09:16 +1030 | [diff] [blame] | 5565 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
Rusty Russell | 968ea6d | 2008-12-13 21:55:51 +1030 | [diff] [blame] | 5566 | cpulist_parse(str, cpu_isolated_map); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5567 | return 1; |
| 5568 | } |
| 5569 | |
Ingo Molnar | 8927f49 | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 5570 | __setup("isolcpus=", isolated_cpu_setup); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5571 | |
Peter Zijlstra | d3081f5 | 2011-04-07 14:09:59 +0200 | [diff] [blame] | 5572 | static const struct cpumask *cpu_cpu_mask(int cpu) |
| 5573 | { |
| 5574 | return cpumask_of_node(cpu_to_node(cpu)); |
| 5575 | } |
| 5576 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5577 | struct sd_data { |
| 5578 | struct sched_domain **__percpu sd; |
| 5579 | struct sched_group **__percpu sg; |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 5580 | struct sched_group_power **__percpu sgp; |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5581 | }; |
| 5582 | |
Andreas Herrmann | 49a02c5 | 2009-08-18 12:51:52 +0200 | [diff] [blame] | 5583 | struct s_data { |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 5584 | struct sched_domain ** __percpu sd; |
Andreas Herrmann | 49a02c5 | 2009-08-18 12:51:52 +0200 | [diff] [blame] | 5585 | struct root_domain *rd; |
| 5586 | }; |
| 5587 | |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 5588 | enum s_alloc { |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 5589 | sa_rootdomain, |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 5590 | sa_sd, |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5591 | sa_sd_storage, |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 5592 | sa_none, |
| 5593 | }; |
| 5594 | |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 5595 | struct sched_domain_topology_level; |
| 5596 | |
| 5597 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); |
Peter Zijlstra | eb7a74e6 | 2011-04-07 14:10:00 +0200 | [diff] [blame] | 5598 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
| 5599 | |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5600 | #define SDTL_OVERLAP 0x01 |
| 5601 | |
Peter Zijlstra | eb7a74e6 | 2011-04-07 14:10:00 +0200 | [diff] [blame] | 5602 | struct sched_domain_topology_level { |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 5603 | sched_domain_init_f init; |
| 5604 | sched_domain_mask_f mask; |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5605 | int flags; |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 5606 | int numa_level; |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 5607 | struct sd_data data; |
Peter Zijlstra | eb7a74e6 | 2011-04-07 14:10:00 +0200 | [diff] [blame] | 5608 | }; |
| 5609 | |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 5610 | /* |
| 5611 | * Build an iteration mask that can exclude certain CPUs from the upwards |
| 5612 | * domain traversal. |
| 5613 | * |
| 5614 | * Asymmetric node setups can result in situations where the domain tree is of |
| 5615 | * unequal depth, make sure to skip domains that already cover the entire |
| 5616 | * range. |
| 5617 | * |
| 5618 | * In that case build_sched_domains() will have terminated the iteration early |
| 5619 | * and our sibling sd spans will be empty. Domains should always include the |
| 5620 | * cpu they're built on, so check that. |
| 5621 | * |
| 5622 | */ |
| 5623 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) |
| 5624 | { |
| 5625 | const struct cpumask *span = sched_domain_span(sd); |
| 5626 | struct sd_data *sdd = sd->private; |
| 5627 | struct sched_domain *sibling; |
| 5628 | int i; |
| 5629 | |
| 5630 | for_each_cpu(i, span) { |
| 5631 | sibling = *per_cpu_ptr(sdd->sd, i); |
| 5632 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) |
| 5633 | continue; |
| 5634 | |
| 5635 | cpumask_set_cpu(i, sched_group_mask(sg)); |
| 5636 | } |
| 5637 | } |
| 5638 | |
| 5639 | /* |
| 5640 | * Return the canonical balance cpu for this group, this is the first cpu |
| 5641 | * of this group that's also in the iteration mask. |
| 5642 | */ |
| 5643 | int group_balance_cpu(struct sched_group *sg) |
| 5644 | { |
| 5645 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); |
| 5646 | } |
| 5647 | |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5648 | static int |
| 5649 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) |
| 5650 | { |
| 5651 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; |
| 5652 | const struct cpumask *span = sched_domain_span(sd); |
| 5653 | struct cpumask *covered = sched_domains_tmpmask; |
| 5654 | struct sd_data *sdd = sd->private; |
| 5655 | struct sched_domain *child; |
| 5656 | int i; |
| 5657 | |
| 5658 | cpumask_clear(covered); |
| 5659 | |
| 5660 | for_each_cpu(i, span) { |
| 5661 | struct cpumask *sg_span; |
| 5662 | |
| 5663 | if (cpumask_test_cpu(i, covered)) |
| 5664 | continue; |
| 5665 | |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 5666 | child = *per_cpu_ptr(sdd->sd, i); |
| 5667 | |
| 5668 | /* See the comment near build_group_mask(). */ |
| 5669 | if (!cpumask_test_cpu(i, sched_domain_span(child))) |
| 5670 | continue; |
| 5671 | |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5672 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
Suresh Siddha | 4d78a22 | 2011-11-18 15:03:29 -0800 | [diff] [blame] | 5673 | GFP_KERNEL, cpu_to_node(cpu)); |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5674 | |
| 5675 | if (!sg) |
| 5676 | goto fail; |
| 5677 | |
| 5678 | sg_span = sched_group_cpus(sg); |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5679 | if (child->child) { |
| 5680 | child = child->child; |
| 5681 | cpumask_copy(sg_span, sched_domain_span(child)); |
| 5682 | } else |
| 5683 | cpumask_set_cpu(i, sg_span); |
| 5684 | |
| 5685 | cpumask_or(covered, covered, sg_span); |
| 5686 | |
Peter Zijlstra | 74a5ce2 | 2012-05-23 18:00:43 +0200 | [diff] [blame] | 5687 | sg->sgp = *per_cpu_ptr(sdd->sgp, i); |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 5688 | if (atomic_inc_return(&sg->sgp->ref) == 1) |
| 5689 | build_group_mask(sd, sg); |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5690 | |
Peter Zijlstra | c3decf0 | 2012-05-31 12:05:32 +0200 | [diff] [blame] | 5691 | /* |
| 5692 | * Initialize sgp->power such that even if we mess up the |
| 5693 | * domains and no possible iteration will get us here, we won't |
| 5694 | * die on a /0 trap. |
| 5695 | */ |
| 5696 | sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span); |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 5697 | |
| 5698 | /* |
| 5699 | * Make sure the first group of this domain contains the |
| 5700 | * canonical balance cpu. Otherwise the sched_domain iteration |
| 5701 | * breaks. See update_sg_lb_stats(). |
| 5702 | */ |
Peter Zijlstra | 74a5ce2 | 2012-05-23 18:00:43 +0200 | [diff] [blame] | 5703 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 5704 | group_balance_cpu(sg) == cpu) |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5705 | groups = sg; |
| 5706 | |
| 5707 | if (!first) |
| 5708 | first = sg; |
| 5709 | if (last) |
| 5710 | last->next = sg; |
| 5711 | last = sg; |
| 5712 | last->next = first; |
| 5713 | } |
| 5714 | sd->groups = groups; |
| 5715 | |
| 5716 | return 0; |
| 5717 | |
| 5718 | fail: |
| 5719 | free_sched_groups(first, 0); |
| 5720 | |
| 5721 | return -ENOMEM; |
| 5722 | } |
| 5723 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5724 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5725 | { |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5726 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
| 5727 | struct sched_domain *child = sd->child; |
| 5728 | |
| 5729 | if (child) |
| 5730 | cpu = cpumask_first(sched_domain_span(child)); |
| 5731 | |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 5732 | if (sg) { |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5733 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 5734 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5735 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 5736 | } |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5737 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5738 | return cpu; |
| 5739 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5740 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5741 | /* |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5742 | * build_sched_groups will build a circular linked list of the groups |
| 5743 | * covered by the given span, and will set each group's ->cpumask correctly, |
| 5744 | * and ->cpu_power to 0. |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5745 | * |
| 5746 | * Assumes the sched_domain tree is fully constructed |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5747 | */ |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5748 | static int |
| 5749 | build_sched_groups(struct sched_domain *sd, int cpu) |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 5750 | { |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5751 | struct sched_group *first = NULL, *last = NULL; |
| 5752 | struct sd_data *sdd = sd->private; |
| 5753 | const struct cpumask *span = sched_domain_span(sd); |
Peter Zijlstra | f96225f | 2011-04-07 14:09:57 +0200 | [diff] [blame] | 5754 | struct cpumask *covered; |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5755 | int i; |
| 5756 | |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5757 | get_group(cpu, sdd, &sd->groups); |
| 5758 | atomic_inc(&sd->groups->ref); |
| 5759 | |
| 5760 | if (cpu != cpumask_first(sched_domain_span(sd))) |
| 5761 | return 0; |
| 5762 | |
Peter Zijlstra | f96225f | 2011-04-07 14:09:57 +0200 | [diff] [blame] | 5763 | lockdep_assert_held(&sched_domains_mutex); |
| 5764 | covered = sched_domains_tmpmask; |
| 5765 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5766 | cpumask_clear(covered); |
| 5767 | |
| 5768 | for_each_cpu(i, span) { |
| 5769 | struct sched_group *sg; |
| 5770 | int group = get_group(i, sdd, &sg); |
| 5771 | int j; |
| 5772 | |
| 5773 | if (cpumask_test_cpu(i, covered)) |
| 5774 | continue; |
| 5775 | |
| 5776 | cpumask_clear(sched_group_cpus(sg)); |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 5777 | sg->sgp->power = 0; |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 5778 | cpumask_setall(sched_group_mask(sg)); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5779 | |
| 5780 | for_each_cpu(j, span) { |
| 5781 | if (get_group(j, sdd, NULL) != group) |
| 5782 | continue; |
| 5783 | |
| 5784 | cpumask_set_cpu(j, covered); |
| 5785 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
| 5786 | } |
| 5787 | |
| 5788 | if (!first) |
| 5789 | first = sg; |
| 5790 | if (last) |
| 5791 | last->next = sg; |
| 5792 | last = sg; |
| 5793 | } |
| 5794 | last->next = first; |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5795 | |
| 5796 | return 0; |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 5797 | } |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 5798 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5799 | /* |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 5800 | * Initialize sched groups cpu_power. |
| 5801 | * |
| 5802 | * cpu_power indicates the capacity of sched group, which is used while |
| 5803 | * distributing the load between different sched groups in a sched domain. |
| 5804 | * Typically cpu_power for all the groups in a sched domain will be same unless |
| 5805 | * there are asymmetries in the topology. If there are asymmetries, group |
| 5806 | * having more cpu_power will pickup more load compared to the group having |
| 5807 | * less cpu_power. |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 5808 | */ |
| 5809 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) |
| 5810 | { |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5811 | struct sched_group *sg = sd->groups; |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 5812 | |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5813 | WARN_ON(!sd || !sg); |
| 5814 | |
| 5815 | do { |
| 5816 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); |
| 5817 | sg = sg->next; |
| 5818 | } while (sg != sd->groups); |
| 5819 | |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 5820 | if (cpu != group_balance_cpu(sg)) |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 5821 | return; |
| 5822 | |
Peter Zijlstra | d274cb3 | 2011-04-07 14:09:43 +0200 | [diff] [blame] | 5823 | update_group_power(sd, cpu); |
Suresh Siddha | 69e1e81 | 2011-12-01 17:07:33 -0800 | [diff] [blame] | 5824 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 5825 | } |
| 5826 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 5827 | int __weak arch_sd_sibling_asym_packing(void) |
| 5828 | { |
| 5829 | return 0*SD_ASYM_PACKING; |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 5830 | } |
| 5831 | |
| 5832 | /* |
Mike Travis | 7c16ec5 | 2008-04-04 18:11:11 -0700 | [diff] [blame] | 5833 | * Initializers for schedule domains |
| 5834 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() |
| 5835 | */ |
| 5836 | |
Ingo Molnar | a5d8c34 | 2008-10-09 11:35:51 +0200 | [diff] [blame] | 5837 | #ifdef CONFIG_SCHED_DEBUG |
| 5838 | # define SD_INIT_NAME(sd, type) sd->name = #type |
| 5839 | #else |
| 5840 | # define SD_INIT_NAME(sd, type) do { } while (0) |
| 5841 | #endif |
| 5842 | |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 5843 | #define SD_INIT_FUNC(type) \ |
| 5844 | static noinline struct sched_domain * \ |
| 5845 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ |
| 5846 | { \ |
| 5847 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ |
| 5848 | *sd = SD_##type##_INIT; \ |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 5849 | SD_INIT_NAME(sd, type); \ |
| 5850 | sd->private = &tl->data; \ |
| 5851 | return sd; \ |
Mike Travis | 7c16ec5 | 2008-04-04 18:11:11 -0700 | [diff] [blame] | 5852 | } |
| 5853 | |
| 5854 | SD_INIT_FUNC(CPU) |
Mike Travis | 7c16ec5 | 2008-04-04 18:11:11 -0700 | [diff] [blame] | 5855 | #ifdef CONFIG_SCHED_SMT |
| 5856 | SD_INIT_FUNC(SIBLING) |
| 5857 | #endif |
| 5858 | #ifdef CONFIG_SCHED_MC |
| 5859 | SD_INIT_FUNC(MC) |
| 5860 | #endif |
Heiko Carstens | 01a0854 | 2010-08-31 10:28:16 +0200 | [diff] [blame] | 5861 | #ifdef CONFIG_SCHED_BOOK |
| 5862 | SD_INIT_FUNC(BOOK) |
| 5863 | #endif |
Mike Travis | 7c16ec5 | 2008-04-04 18:11:11 -0700 | [diff] [blame] | 5864 | |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 5865 | static int default_relax_domain_level = -1; |
Peter Zijlstra | 60495e7 | 2011-04-07 14:10:04 +0200 | [diff] [blame] | 5866 | int sched_domain_level_max; |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 5867 | |
| 5868 | static int __init setup_relax_domain_level(char *str) |
| 5869 | { |
Dimitri Sivanich | a841f8c | 2012-06-05 13:44:36 -0500 | [diff] [blame] | 5870 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
| 5871 | pr_warn("Unable to set relax_domain_level\n"); |
Li Zefan | 30e0e17 | 2008-05-13 10:27:17 +0800 | [diff] [blame] | 5872 | |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 5873 | return 1; |
| 5874 | } |
| 5875 | __setup("relax_domain_level=", setup_relax_domain_level); |
| 5876 | |
| 5877 | static void set_domain_attribute(struct sched_domain *sd, |
| 5878 | struct sched_domain_attr *attr) |
| 5879 | { |
| 5880 | int request; |
| 5881 | |
| 5882 | if (!attr || attr->relax_domain_level < 0) { |
| 5883 | if (default_relax_domain_level < 0) |
| 5884 | return; |
| 5885 | else |
| 5886 | request = default_relax_domain_level; |
| 5887 | } else |
| 5888 | request = attr->relax_domain_level; |
| 5889 | if (request < sd->level) { |
| 5890 | /* turn off idle balance on this domain */ |
Peter Zijlstra | c88d591 | 2009-09-10 13:50:02 +0200 | [diff] [blame] | 5891 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 5892 | } else { |
| 5893 | /* turn on idle balance on this domain */ |
Peter Zijlstra | c88d591 | 2009-09-10 13:50:02 +0200 | [diff] [blame] | 5894 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 5895 | } |
| 5896 | } |
| 5897 | |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 5898 | static void __sdt_free(const struct cpumask *cpu_map); |
| 5899 | static int __sdt_alloc(const struct cpumask *cpu_map); |
| 5900 | |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 5901 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
| 5902 | const struct cpumask *cpu_map) |
| 5903 | { |
| 5904 | switch (what) { |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 5905 | case sa_rootdomain: |
Peter Zijlstra | 822ff79 | 2011-04-07 14:09:51 +0200 | [diff] [blame] | 5906 | if (!atomic_read(&d->rd->refcount)) |
| 5907 | free_rootdomain(&d->rd->rcu); /* fall through */ |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 5908 | case sa_sd: |
| 5909 | free_percpu(d->sd); /* fall through */ |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5910 | case sa_sd_storage: |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 5911 | __sdt_free(cpu_map); /* fall through */ |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 5912 | case sa_none: |
| 5913 | break; |
| 5914 | } |
| 5915 | } |
| 5916 | |
| 5917 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
| 5918 | const struct cpumask *cpu_map) |
| 5919 | { |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5920 | memset(d, 0, sizeof(*d)); |
| 5921 | |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 5922 | if (__sdt_alloc(cpu_map)) |
| 5923 | return sa_sd_storage; |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 5924 | d->sd = alloc_percpu(struct sched_domain *); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5925 | if (!d->sd) |
| 5926 | return sa_sd_storage; |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 5927 | d->rd = alloc_rootdomain(); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5928 | if (!d->rd) |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 5929 | return sa_sd; |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 5930 | return sa_rootdomain; |
| 5931 | } |
| 5932 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5933 | /* |
| 5934 | * NULL the sd_data elements we've used to build the sched_domain and |
| 5935 | * sched_group structure so that the subsequent __free_domain_allocs() |
| 5936 | * will not free the data we're using. |
| 5937 | */ |
| 5938 | static void claim_allocations(int cpu, struct sched_domain *sd) |
| 5939 | { |
| 5940 | struct sd_data *sdd = sd->private; |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5941 | |
| 5942 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); |
| 5943 | *per_cpu_ptr(sdd->sd, cpu) = NULL; |
| 5944 | |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5945 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5946 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5947 | |
| 5948 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 5949 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5950 | } |
| 5951 | |
Andreas Herrmann | d817353 | 2009-08-18 12:57:03 +0200 | [diff] [blame] | 5952 | #ifdef CONFIG_SCHED_SMT |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 5953 | static const struct cpumask *cpu_smt_mask(int cpu) |
| 5954 | { |
| 5955 | return topology_thread_cpumask(cpu); |
Andreas Herrmann | d817353 | 2009-08-18 12:57:03 +0200 | [diff] [blame] | 5956 | } |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 5957 | #endif |
Andreas Herrmann | d817353 | 2009-08-18 12:57:03 +0200 | [diff] [blame] | 5958 | |
Peter Zijlstra | d069b91 | 2011-04-07 14:10:02 +0200 | [diff] [blame] | 5959 | /* |
| 5960 | * Topology list, bottom-up. |
| 5961 | */ |
Peter Zijlstra | eb7a74e6 | 2011-04-07 14:10:00 +0200 | [diff] [blame] | 5962 | static struct sched_domain_topology_level default_topology[] = { |
Peter Zijlstra | d069b91 | 2011-04-07 14:10:02 +0200 | [diff] [blame] | 5963 | #ifdef CONFIG_SCHED_SMT |
| 5964 | { sd_init_SIBLING, cpu_smt_mask, }, |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 5965 | #endif |
| 5966 | #ifdef CONFIG_SCHED_MC |
| 5967 | { sd_init_MC, cpu_coregroup_mask, }, |
| 5968 | #endif |
Peter Zijlstra | d069b91 | 2011-04-07 14:10:02 +0200 | [diff] [blame] | 5969 | #ifdef CONFIG_SCHED_BOOK |
| 5970 | { sd_init_BOOK, cpu_book_mask, }, |
| 5971 | #endif |
| 5972 | { sd_init_CPU, cpu_cpu_mask, }, |
Peter Zijlstra | eb7a74e6 | 2011-04-07 14:10:00 +0200 | [diff] [blame] | 5973 | { NULL, }, |
| 5974 | }; |
| 5975 | |
| 5976 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; |
| 5977 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 5978 | #ifdef CONFIG_NUMA |
| 5979 | |
| 5980 | static int sched_domains_numa_levels; |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 5981 | static int *sched_domains_numa_distance; |
| 5982 | static struct cpumask ***sched_domains_numa_masks; |
| 5983 | static int sched_domains_curr_level; |
| 5984 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 5985 | static inline int sd_local_flags(int level) |
| 5986 | { |
Alex Shi | 10717dc | 2012-06-06 14:52:51 +0800 | [diff] [blame] | 5987 | if (sched_domains_numa_distance[level] > RECLAIM_DISTANCE) |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 5988 | return 0; |
| 5989 | |
| 5990 | return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE; |
| 5991 | } |
| 5992 | |
| 5993 | static struct sched_domain * |
| 5994 | sd_numa_init(struct sched_domain_topology_level *tl, int cpu) |
| 5995 | { |
| 5996 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); |
| 5997 | int level = tl->numa_level; |
| 5998 | int sd_weight = cpumask_weight( |
| 5999 | sched_domains_numa_masks[level][cpu_to_node(cpu)]); |
| 6000 | |
| 6001 | *sd = (struct sched_domain){ |
| 6002 | .min_interval = sd_weight, |
| 6003 | .max_interval = 2*sd_weight, |
| 6004 | .busy_factor = 32, |
Peter Zijlstra | 870a0bb | 2012-05-11 00:26:27 +0200 | [diff] [blame] | 6005 | .imbalance_pct = 125, |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6006 | .cache_nice_tries = 2, |
| 6007 | .busy_idx = 3, |
| 6008 | .idle_idx = 2, |
| 6009 | .newidle_idx = 0, |
| 6010 | .wake_idx = 0, |
| 6011 | .forkexec_idx = 0, |
| 6012 | |
| 6013 | .flags = 1*SD_LOAD_BALANCE |
| 6014 | | 1*SD_BALANCE_NEWIDLE |
| 6015 | | 0*SD_BALANCE_EXEC |
| 6016 | | 0*SD_BALANCE_FORK |
| 6017 | | 0*SD_BALANCE_WAKE |
| 6018 | | 0*SD_WAKE_AFFINE |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6019 | | 0*SD_SHARE_CPUPOWER |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6020 | | 0*SD_SHARE_PKG_RESOURCES |
| 6021 | | 1*SD_SERIALIZE |
| 6022 | | 0*SD_PREFER_SIBLING |
| 6023 | | sd_local_flags(level) |
| 6024 | , |
| 6025 | .last_balance = jiffies, |
| 6026 | .balance_interval = sd_weight, |
| 6027 | }; |
| 6028 | SD_INIT_NAME(sd, NUMA); |
| 6029 | sd->private = &tl->data; |
| 6030 | |
| 6031 | /* |
| 6032 | * Ugly hack to pass state to sd_numa_mask()... |
| 6033 | */ |
| 6034 | sched_domains_curr_level = tl->numa_level; |
| 6035 | |
| 6036 | return sd; |
| 6037 | } |
| 6038 | |
| 6039 | static const struct cpumask *sd_numa_mask(int cpu) |
| 6040 | { |
| 6041 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; |
| 6042 | } |
| 6043 | |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 6044 | static void sched_numa_warn(const char *str) |
| 6045 | { |
| 6046 | static int done = false; |
| 6047 | int i,j; |
| 6048 | |
| 6049 | if (done) |
| 6050 | return; |
| 6051 | |
| 6052 | done = true; |
| 6053 | |
| 6054 | printk(KERN_WARNING "ERROR: %s\n\n", str); |
| 6055 | |
| 6056 | for (i = 0; i < nr_node_ids; i++) { |
| 6057 | printk(KERN_WARNING " "); |
| 6058 | for (j = 0; j < nr_node_ids; j++) |
| 6059 | printk(KERN_CONT "%02d ", node_distance(i,j)); |
| 6060 | printk(KERN_CONT "\n"); |
| 6061 | } |
| 6062 | printk(KERN_WARNING "\n"); |
| 6063 | } |
| 6064 | |
| 6065 | static bool find_numa_distance(int distance) |
| 6066 | { |
| 6067 | int i; |
| 6068 | |
| 6069 | if (distance == node_distance(0, 0)) |
| 6070 | return true; |
| 6071 | |
| 6072 | for (i = 0; i < sched_domains_numa_levels; i++) { |
| 6073 | if (sched_domains_numa_distance[i] == distance) |
| 6074 | return true; |
| 6075 | } |
| 6076 | |
| 6077 | return false; |
| 6078 | } |
| 6079 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6080 | static void sched_init_numa(void) |
| 6081 | { |
| 6082 | int next_distance, curr_distance = node_distance(0, 0); |
| 6083 | struct sched_domain_topology_level *tl; |
| 6084 | int level = 0; |
| 6085 | int i, j, k; |
| 6086 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6087 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
| 6088 | if (!sched_domains_numa_distance) |
| 6089 | return; |
| 6090 | |
| 6091 | /* |
| 6092 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the |
| 6093 | * unique distances in the node_distance() table. |
| 6094 | * |
| 6095 | * Assumes node_distance(0,j) includes all distances in |
| 6096 | * node_distance(i,j) in order to avoid cubic time. |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6097 | */ |
| 6098 | next_distance = curr_distance; |
| 6099 | for (i = 0; i < nr_node_ids; i++) { |
| 6100 | for (j = 0; j < nr_node_ids; j++) { |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 6101 | for (k = 0; k < nr_node_ids; k++) { |
| 6102 | int distance = node_distance(i, k); |
| 6103 | |
| 6104 | if (distance > curr_distance && |
| 6105 | (distance < next_distance || |
| 6106 | next_distance == curr_distance)) |
| 6107 | next_distance = distance; |
| 6108 | |
| 6109 | /* |
| 6110 | * While not a strong assumption it would be nice to know |
| 6111 | * about cases where if node A is connected to B, B is not |
| 6112 | * equally connected to A. |
| 6113 | */ |
| 6114 | if (sched_debug() && node_distance(k, i) != distance) |
| 6115 | sched_numa_warn("Node-distance not symmetric"); |
| 6116 | |
| 6117 | if (sched_debug() && i && !find_numa_distance(distance)) |
| 6118 | sched_numa_warn("Node-0 not representative"); |
| 6119 | } |
| 6120 | if (next_distance != curr_distance) { |
| 6121 | sched_domains_numa_distance[level++] = next_distance; |
| 6122 | sched_domains_numa_levels = level; |
| 6123 | curr_distance = next_distance; |
| 6124 | } else break; |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6125 | } |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 6126 | |
| 6127 | /* |
| 6128 | * In case of sched_debug() we verify the above assumption. |
| 6129 | */ |
| 6130 | if (!sched_debug()) |
| 6131 | break; |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6132 | } |
| 6133 | /* |
| 6134 | * 'level' contains the number of unique distances, excluding the |
| 6135 | * identity distance node_distance(i,i). |
| 6136 | * |
| 6137 | * The sched_domains_nume_distance[] array includes the actual distance |
| 6138 | * numbers. |
| 6139 | */ |
| 6140 | |
Tang Chen | 5f7865f | 2012-09-25 21:12:30 +0800 | [diff] [blame] | 6141 | /* |
| 6142 | * Here, we should temporarily reset sched_domains_numa_levels to 0. |
| 6143 | * If it fails to allocate memory for array sched_domains_numa_masks[][], |
| 6144 | * the array will contain less then 'level' members. This could be |
| 6145 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] |
| 6146 | * in other functions. |
| 6147 | * |
| 6148 | * We reset it to 'level' at the end of this function. |
| 6149 | */ |
| 6150 | sched_domains_numa_levels = 0; |
| 6151 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6152 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
| 6153 | if (!sched_domains_numa_masks) |
| 6154 | return; |
| 6155 | |
| 6156 | /* |
| 6157 | * Now for each level, construct a mask per node which contains all |
| 6158 | * cpus of nodes that are that many hops away from us. |
| 6159 | */ |
| 6160 | for (i = 0; i < level; i++) { |
| 6161 | sched_domains_numa_masks[i] = |
| 6162 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); |
| 6163 | if (!sched_domains_numa_masks[i]) |
| 6164 | return; |
| 6165 | |
| 6166 | for (j = 0; j < nr_node_ids; j++) { |
Peter Zijlstra | 2ea4580 | 2012-05-25 09:26:43 +0200 | [diff] [blame] | 6167 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6168 | if (!mask) |
| 6169 | return; |
| 6170 | |
| 6171 | sched_domains_numa_masks[i][j] = mask; |
| 6172 | |
| 6173 | for (k = 0; k < nr_node_ids; k++) { |
Peter Zijlstra | dd7d863 | 2012-05-11 00:56:20 +0200 | [diff] [blame] | 6174 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6175 | continue; |
| 6176 | |
| 6177 | cpumask_or(mask, mask, cpumask_of_node(k)); |
| 6178 | } |
| 6179 | } |
| 6180 | } |
| 6181 | |
| 6182 | tl = kzalloc((ARRAY_SIZE(default_topology) + level) * |
| 6183 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
| 6184 | if (!tl) |
| 6185 | return; |
| 6186 | |
| 6187 | /* |
| 6188 | * Copy the default topology bits.. |
| 6189 | */ |
| 6190 | for (i = 0; default_topology[i].init; i++) |
| 6191 | tl[i] = default_topology[i]; |
| 6192 | |
| 6193 | /* |
| 6194 | * .. and append 'j' levels of NUMA goodness. |
| 6195 | */ |
| 6196 | for (j = 0; j < level; i++, j++) { |
| 6197 | tl[i] = (struct sched_domain_topology_level){ |
| 6198 | .init = sd_numa_init, |
| 6199 | .mask = sd_numa_mask, |
| 6200 | .flags = SDTL_OVERLAP, |
| 6201 | .numa_level = j, |
| 6202 | }; |
| 6203 | } |
| 6204 | |
| 6205 | sched_domain_topology = tl; |
Tang Chen | 5f7865f | 2012-09-25 21:12:30 +0800 | [diff] [blame] | 6206 | |
| 6207 | sched_domains_numa_levels = level; |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6208 | } |
Tang Chen | 301a5cb | 2012-09-25 21:12:31 +0800 | [diff] [blame] | 6209 | |
| 6210 | static void sched_domains_numa_masks_set(int cpu) |
| 6211 | { |
| 6212 | int i, j; |
| 6213 | int node = cpu_to_node(cpu); |
| 6214 | |
| 6215 | for (i = 0; i < sched_domains_numa_levels; i++) { |
| 6216 | for (j = 0; j < nr_node_ids; j++) { |
| 6217 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) |
| 6218 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); |
| 6219 | } |
| 6220 | } |
| 6221 | } |
| 6222 | |
| 6223 | static void sched_domains_numa_masks_clear(int cpu) |
| 6224 | { |
| 6225 | int i, j; |
| 6226 | for (i = 0; i < sched_domains_numa_levels; i++) { |
| 6227 | for (j = 0; j < nr_node_ids; j++) |
| 6228 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); |
| 6229 | } |
| 6230 | } |
| 6231 | |
| 6232 | /* |
| 6233 | * Update sched_domains_numa_masks[level][node] array when new cpus |
| 6234 | * are onlined. |
| 6235 | */ |
| 6236 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, |
| 6237 | unsigned long action, |
| 6238 | void *hcpu) |
| 6239 | { |
| 6240 | int cpu = (long)hcpu; |
| 6241 | |
| 6242 | switch (action & ~CPU_TASKS_FROZEN) { |
| 6243 | case CPU_ONLINE: |
| 6244 | sched_domains_numa_masks_set(cpu); |
| 6245 | break; |
| 6246 | |
| 6247 | case CPU_DEAD: |
| 6248 | sched_domains_numa_masks_clear(cpu); |
| 6249 | break; |
| 6250 | |
| 6251 | default: |
| 6252 | return NOTIFY_DONE; |
| 6253 | } |
| 6254 | |
| 6255 | return NOTIFY_OK; |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6256 | } |
| 6257 | #else |
| 6258 | static inline void sched_init_numa(void) |
| 6259 | { |
| 6260 | } |
Tang Chen | 301a5cb | 2012-09-25 21:12:31 +0800 | [diff] [blame] | 6261 | |
| 6262 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, |
| 6263 | unsigned long action, |
| 6264 | void *hcpu) |
| 6265 | { |
| 6266 | return 0; |
| 6267 | } |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6268 | #endif /* CONFIG_NUMA */ |
| 6269 | |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6270 | static int __sdt_alloc(const struct cpumask *cpu_map) |
| 6271 | { |
| 6272 | struct sched_domain_topology_level *tl; |
| 6273 | int j; |
| 6274 | |
| 6275 | for (tl = sched_domain_topology; tl->init; tl++) { |
| 6276 | struct sd_data *sdd = &tl->data; |
| 6277 | |
| 6278 | sdd->sd = alloc_percpu(struct sched_domain *); |
| 6279 | if (!sdd->sd) |
| 6280 | return -ENOMEM; |
| 6281 | |
| 6282 | sdd->sg = alloc_percpu(struct sched_group *); |
| 6283 | if (!sdd->sg) |
| 6284 | return -ENOMEM; |
| 6285 | |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 6286 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
| 6287 | if (!sdd->sgp) |
| 6288 | return -ENOMEM; |
| 6289 | |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6290 | for_each_cpu(j, cpu_map) { |
| 6291 | struct sched_domain *sd; |
| 6292 | struct sched_group *sg; |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 6293 | struct sched_group_power *sgp; |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6294 | |
| 6295 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), |
| 6296 | GFP_KERNEL, cpu_to_node(j)); |
| 6297 | if (!sd) |
| 6298 | return -ENOMEM; |
| 6299 | |
| 6300 | *per_cpu_ptr(sdd->sd, j) = sd; |
| 6301 | |
| 6302 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
| 6303 | GFP_KERNEL, cpu_to_node(j)); |
| 6304 | if (!sg) |
| 6305 | return -ENOMEM; |
| 6306 | |
Igor Mammedov | 30b4e9e | 2012-05-09 12:38:28 +0200 | [diff] [blame] | 6307 | sg->next = sg; |
| 6308 | |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6309 | *per_cpu_ptr(sdd->sg, j) = sg; |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 6310 | |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 6311 | sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(), |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 6312 | GFP_KERNEL, cpu_to_node(j)); |
| 6313 | if (!sgp) |
| 6314 | return -ENOMEM; |
| 6315 | |
| 6316 | *per_cpu_ptr(sdd->sgp, j) = sgp; |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6317 | } |
| 6318 | } |
| 6319 | |
| 6320 | return 0; |
| 6321 | } |
| 6322 | |
| 6323 | static void __sdt_free(const struct cpumask *cpu_map) |
| 6324 | { |
| 6325 | struct sched_domain_topology_level *tl; |
| 6326 | int j; |
| 6327 | |
| 6328 | for (tl = sched_domain_topology; tl->init; tl++) { |
| 6329 | struct sd_data *sdd = &tl->data; |
| 6330 | |
| 6331 | for_each_cpu(j, cpu_map) { |
he, bo | fb2cf2c | 2012-04-25 19:59:21 +0800 | [diff] [blame] | 6332 | struct sched_domain *sd; |
| 6333 | |
| 6334 | if (sdd->sd) { |
| 6335 | sd = *per_cpu_ptr(sdd->sd, j); |
| 6336 | if (sd && (sd->flags & SD_OVERLAP)) |
| 6337 | free_sched_groups(sd->groups, 0); |
| 6338 | kfree(*per_cpu_ptr(sdd->sd, j)); |
| 6339 | } |
| 6340 | |
| 6341 | if (sdd->sg) |
| 6342 | kfree(*per_cpu_ptr(sdd->sg, j)); |
| 6343 | if (sdd->sgp) |
| 6344 | kfree(*per_cpu_ptr(sdd->sgp, j)); |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6345 | } |
| 6346 | free_percpu(sdd->sd); |
he, bo | fb2cf2c | 2012-04-25 19:59:21 +0800 | [diff] [blame] | 6347 | sdd->sd = NULL; |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6348 | free_percpu(sdd->sg); |
he, bo | fb2cf2c | 2012-04-25 19:59:21 +0800 | [diff] [blame] | 6349 | sdd->sg = NULL; |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 6350 | free_percpu(sdd->sgp); |
he, bo | fb2cf2c | 2012-04-25 19:59:21 +0800 | [diff] [blame] | 6351 | sdd->sgp = NULL; |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6352 | } |
| 6353 | } |
| 6354 | |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 6355 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
| 6356 | struct s_data *d, const struct cpumask *cpu_map, |
Peter Zijlstra | d069b91 | 2011-04-07 14:10:02 +0200 | [diff] [blame] | 6357 | struct sched_domain_attr *attr, struct sched_domain *child, |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 6358 | int cpu) |
| 6359 | { |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6360 | struct sched_domain *sd = tl->init(tl, cpu); |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 6361 | if (!sd) |
Peter Zijlstra | d069b91 | 2011-04-07 14:10:02 +0200 | [diff] [blame] | 6362 | return child; |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 6363 | |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 6364 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
Peter Zijlstra | 60495e7 | 2011-04-07 14:10:04 +0200 | [diff] [blame] | 6365 | if (child) { |
| 6366 | sd->level = child->level + 1; |
| 6367 | sched_domain_level_max = max(sched_domain_level_max, sd->level); |
Peter Zijlstra | d069b91 | 2011-04-07 14:10:02 +0200 | [diff] [blame] | 6368 | child->parent = sd; |
Peter Zijlstra | 60495e7 | 2011-04-07 14:10:04 +0200 | [diff] [blame] | 6369 | } |
Peter Zijlstra | d069b91 | 2011-04-07 14:10:02 +0200 | [diff] [blame] | 6370 | sd->child = child; |
Dimitri Sivanich | a841f8c | 2012-06-05 13:44:36 -0500 | [diff] [blame] | 6371 | set_domain_attribute(sd, attr); |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 6372 | |
| 6373 | return sd; |
| 6374 | } |
| 6375 | |
Mike Travis | 7c16ec5 | 2008-04-04 18:11:11 -0700 | [diff] [blame] | 6376 | /* |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6377 | * Build sched domains for a given set of cpus and attach the sched domains |
| 6378 | * to the individual cpus |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6379 | */ |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6380 | static int build_sched_domains(const struct cpumask *cpu_map, |
| 6381 | struct sched_domain_attr *attr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6382 | { |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 6383 | enum s_alloc alloc_state = sa_none; |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6384 | struct sched_domain *sd; |
Andreas Herrmann | 49a02c5 | 2009-08-18 12:51:52 +0200 | [diff] [blame] | 6385 | struct s_data d; |
Peter Zijlstra | 822ff79 | 2011-04-07 14:09:51 +0200 | [diff] [blame] | 6386 | int i, ret = -ENOMEM; |
Rusty Russell | 3404c8d | 2008-11-25 02:35:03 +1030 | [diff] [blame] | 6387 | |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 6388 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
| 6389 | if (alloc_state != sa_rootdomain) |
| 6390 | goto error; |
Mike Travis | 7c16ec5 | 2008-04-04 18:11:11 -0700 | [diff] [blame] | 6391 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6392 | /* Set up domains for cpus specified by the cpu_map. */ |
Rusty Russell | abcd083 | 2008-11-25 02:35:02 +1030 | [diff] [blame] | 6393 | for_each_cpu(i, cpu_map) { |
Peter Zijlstra | eb7a74e6 | 2011-04-07 14:10:00 +0200 | [diff] [blame] | 6394 | struct sched_domain_topology_level *tl; |
| 6395 | |
Peter Zijlstra | 3bd65a8 | 2011-04-07 14:09:54 +0200 | [diff] [blame] | 6396 | sd = NULL; |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6397 | for (tl = sched_domain_topology; tl->init; tl++) { |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 6398 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6399 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
| 6400 | sd->flags |= SD_OVERLAP; |
Peter Zijlstra | d110235 | 2011-07-20 18:42:57 +0200 | [diff] [blame] | 6401 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
| 6402 | break; |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6403 | } |
Peter Zijlstra | d274cb3 | 2011-04-07 14:09:43 +0200 | [diff] [blame] | 6404 | |
Peter Zijlstra | d069b91 | 2011-04-07 14:10:02 +0200 | [diff] [blame] | 6405 | while (sd->child) |
| 6406 | sd = sd->child; |
| 6407 | |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 6408 | *per_cpu_ptr(d.sd, i) = sd; |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6409 | } |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 6410 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6411 | /* Build the groups for the domains */ |
| 6412 | for_each_cpu(i, cpu_map) { |
| 6413 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
| 6414 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6415 | if (sd->flags & SD_OVERLAP) { |
| 6416 | if (build_overlap_sched_groups(sd, i)) |
| 6417 | goto error; |
| 6418 | } else { |
| 6419 | if (build_sched_groups(sd, i)) |
| 6420 | goto error; |
| 6421 | } |
Peter Zijlstra | 1cf51902 | 2011-04-07 14:09:47 +0200 | [diff] [blame] | 6422 | } |
Peter Zijlstra | a06dadb | 2011-04-07 14:09:44 +0200 | [diff] [blame] | 6423 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6424 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6425 | /* Calculate CPU power for physical packages and nodes */ |
Peter Zijlstra | a9c9a9b | 2011-04-07 14:09:49 +0200 | [diff] [blame] | 6426 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
| 6427 | if (!cpumask_test_cpu(i, cpu_map)) |
| 6428 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6429 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6430 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
| 6431 | claim_allocations(i, sd); |
Peter Zijlstra | cd4ea6a | 2011-04-07 14:09:45 +0200 | [diff] [blame] | 6432 | init_sched_groups_power(i, sd); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6433 | } |
Siddha, Suresh B | f712c0c | 2006-07-30 03:02:59 -0700 | [diff] [blame] | 6434 | } |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6435 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6436 | /* Attach the domains */ |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6437 | rcu_read_lock(); |
Rusty Russell | abcd083 | 2008-11-25 02:35:02 +1030 | [diff] [blame] | 6438 | for_each_cpu(i, cpu_map) { |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 6439 | sd = *per_cpu_ptr(d.sd, i); |
Andreas Herrmann | 49a02c5 | 2009-08-18 12:51:52 +0200 | [diff] [blame] | 6440 | cpu_attach_domain(sd, d.rd, i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6441 | } |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6442 | rcu_read_unlock(); |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6443 | |
Peter Zijlstra | 822ff79 | 2011-04-07 14:09:51 +0200 | [diff] [blame] | 6444 | ret = 0; |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6445 | error: |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 6446 | __free_domain_allocs(&d, alloc_state, cpu_map); |
Peter Zijlstra | 822ff79 | 2011-04-07 14:09:51 +0200 | [diff] [blame] | 6447 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6448 | } |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6449 | |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 6450 | static cpumask_var_t *doms_cur; /* current sched domains */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6451 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
Ingo Molnar | 4285f594 | 2008-05-16 17:47:14 +0200 | [diff] [blame] | 6452 | static struct sched_domain_attr *dattr_cur; |
| 6453 | /* attribues of custom domains in 'doms_cur' */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6454 | |
| 6455 | /* |
| 6456 | * Special case: If a kmalloc of a doms_cur partition (array of |
Rusty Russell | 4212823 | 2008-11-25 02:35:12 +1030 | [diff] [blame] | 6457 | * cpumask) fails, then fallback to a single sched domain, |
| 6458 | * as determined by the single cpumask fallback_doms. |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6459 | */ |
Rusty Russell | 4212823 | 2008-11-25 02:35:12 +1030 | [diff] [blame] | 6460 | static cpumask_var_t fallback_doms; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6461 | |
Heiko Carstens | ee79d1b | 2008-12-09 18:49:50 +0100 | [diff] [blame] | 6462 | /* |
| 6463 | * arch_update_cpu_topology lets virtualized architectures update the |
| 6464 | * cpu core maps. It is supposed to return 1 if the topology changed |
| 6465 | * or 0 if it stayed the same. |
| 6466 | */ |
| 6467 | int __attribute__((weak)) arch_update_cpu_topology(void) |
Heiko Carstens | 22e52b0 | 2008-03-12 18:31:59 +0100 | [diff] [blame] | 6468 | { |
Heiko Carstens | ee79d1b | 2008-12-09 18:49:50 +0100 | [diff] [blame] | 6469 | return 0; |
Heiko Carstens | 22e52b0 | 2008-03-12 18:31:59 +0100 | [diff] [blame] | 6470 | } |
| 6471 | |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 6472 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
| 6473 | { |
| 6474 | int i; |
| 6475 | cpumask_var_t *doms; |
| 6476 | |
| 6477 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); |
| 6478 | if (!doms) |
| 6479 | return NULL; |
| 6480 | for (i = 0; i < ndoms; i++) { |
| 6481 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { |
| 6482 | free_sched_domains(doms, i); |
| 6483 | return NULL; |
| 6484 | } |
| 6485 | } |
| 6486 | return doms; |
| 6487 | } |
| 6488 | |
| 6489 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) |
| 6490 | { |
| 6491 | unsigned int i; |
| 6492 | for (i = 0; i < ndoms; i++) |
| 6493 | free_cpumask_var(doms[i]); |
| 6494 | kfree(doms); |
| 6495 | } |
| 6496 | |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6497 | /* |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 6498 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6499 | * For now this just excludes isolated cpus, but could be used to |
| 6500 | * exclude other special cases in the future. |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6501 | */ |
Peter Zijlstra | c4a8849 | 2011-04-07 14:09:42 +0200 | [diff] [blame] | 6502 | static int init_sched_domains(const struct cpumask *cpu_map) |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6503 | { |
Milton Miller | 7378547 | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 6504 | int err; |
| 6505 | |
Heiko Carstens | 22e52b0 | 2008-03-12 18:31:59 +0100 | [diff] [blame] | 6506 | arch_update_cpu_topology(); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6507 | ndoms_cur = 1; |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 6508 | doms_cur = alloc_sched_domains(ndoms_cur); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6509 | if (!doms_cur) |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 6510 | doms_cur = &fallback_doms; |
| 6511 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6512 | err = build_sched_domains(doms_cur[0], NULL); |
Milton Miller | 6382bc9 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 6513 | register_sched_domain_sysctl(); |
Milton Miller | 7378547 | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 6514 | |
| 6515 | return err; |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6516 | } |
| 6517 | |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6518 | /* |
| 6519 | * Detach sched domains from a group of cpus specified in cpu_map |
| 6520 | * These cpus will now be attached to the NULL domain |
| 6521 | */ |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 6522 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6523 | { |
| 6524 | int i; |
| 6525 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6526 | rcu_read_lock(); |
Rusty Russell | abcd083 | 2008-11-25 02:35:02 +1030 | [diff] [blame] | 6527 | for_each_cpu(i, cpu_map) |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 6528 | cpu_attach_domain(NULL, &def_root_domain, i); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6529 | rcu_read_unlock(); |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6530 | } |
| 6531 | |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 6532 | /* handle null as "default" */ |
| 6533 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, |
| 6534 | struct sched_domain_attr *new, int idx_new) |
| 6535 | { |
| 6536 | struct sched_domain_attr tmp; |
| 6537 | |
| 6538 | /* fast path */ |
| 6539 | if (!new && !cur) |
| 6540 | return 1; |
| 6541 | |
| 6542 | tmp = SD_ATTR_INIT; |
| 6543 | return !memcmp(cur ? (cur + idx_cur) : &tmp, |
| 6544 | new ? (new + idx_new) : &tmp, |
| 6545 | sizeof(struct sched_domain_attr)); |
| 6546 | } |
| 6547 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6548 | /* |
| 6549 | * Partition sched domains as specified by the 'ndoms_new' |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 6550 | * cpumasks in the array doms_new[] of cpumasks. This compares |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6551 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
| 6552 | * It destroys each deleted domain and builds each new domain. |
| 6553 | * |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 6554 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 6555 | * The masks don't intersect (don't overlap.) We should setup one |
| 6556 | * sched domain for each mask. CPUs not in any of the cpumasks will |
| 6557 | * not be load balanced. If the same cpumask appears both in the |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6558 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
| 6559 | * it as it is. |
| 6560 | * |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 6561 | * The passed in 'doms_new' should be allocated using |
| 6562 | * alloc_sched_domains. This routine takes ownership of it and will |
| 6563 | * free_sched_domains it when done with it. If the caller failed the |
| 6564 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, |
| 6565 | * and partition_sched_domains() will fallback to the single partition |
| 6566 | * 'fallback_doms', it also forces the domains to be rebuilt. |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6567 | * |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 6568 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
Li Zefan | 700018e | 2008-11-18 14:02:03 +0800 | [diff] [blame] | 6569 | * ndoms_new == 0 is a special case for destroying existing domains, |
| 6570 | * and it will not create the default domain. |
Max Krasnyansky | dfb512e | 2008-08-29 13:11:41 -0700 | [diff] [blame] | 6571 | * |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6572 | * Call with hotplug lock held |
| 6573 | */ |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 6574 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 6575 | struct sched_domain_attr *dattr_new) |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6576 | { |
Max Krasnyansky | dfb512e | 2008-08-29 13:11:41 -0700 | [diff] [blame] | 6577 | int i, j, n; |
Heiko Carstens | d65bd5e | 2008-12-09 18:49:51 +0100 | [diff] [blame] | 6578 | int new_topology; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6579 | |
Heiko Carstens | 712555e | 2008-04-28 11:33:07 +0200 | [diff] [blame] | 6580 | mutex_lock(&sched_domains_mutex); |
Srivatsa Vaddagiri | a183561 | 2008-01-25 21:08:00 +0100 | [diff] [blame] | 6581 | |
Milton Miller | 7378547 | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 6582 | /* always unregister in case we don't destroy any domains */ |
| 6583 | unregister_sched_domain_sysctl(); |
| 6584 | |
Heiko Carstens | d65bd5e | 2008-12-09 18:49:51 +0100 | [diff] [blame] | 6585 | /* Let architecture update cpu core mappings. */ |
| 6586 | new_topology = arch_update_cpu_topology(); |
| 6587 | |
Max Krasnyansky | dfb512e | 2008-08-29 13:11:41 -0700 | [diff] [blame] | 6588 | n = doms_new ? ndoms_new : 0; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6589 | |
| 6590 | /* Destroy deleted domains */ |
| 6591 | for (i = 0; i < ndoms_cur; i++) { |
Heiko Carstens | d65bd5e | 2008-12-09 18:49:51 +0100 | [diff] [blame] | 6592 | for (j = 0; j < n && !new_topology; j++) { |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 6593 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 6594 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6595 | goto match1; |
| 6596 | } |
| 6597 | /* no match - a current sched domain not in new doms_new[] */ |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 6598 | detach_destroy_domains(doms_cur[i]); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6599 | match1: |
| 6600 | ; |
| 6601 | } |
| 6602 | |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 6603 | if (doms_new == NULL) { |
| 6604 | ndoms_cur = 0; |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 6605 | doms_new = &fallback_doms; |
Peter Zijlstra | 6ad4c18 | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 6606 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
Li Zefan | faa2f98 | 2008-11-04 16:20:23 +0800 | [diff] [blame] | 6607 | WARN_ON_ONCE(dattr_new); |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 6608 | } |
| 6609 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6610 | /* Build new domains */ |
| 6611 | for (i = 0; i < ndoms_new; i++) { |
Heiko Carstens | d65bd5e | 2008-12-09 18:49:51 +0100 | [diff] [blame] | 6612 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 6613 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 6614 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6615 | goto match2; |
| 6616 | } |
| 6617 | /* no match - add a new doms_new */ |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6618 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6619 | match2: |
| 6620 | ; |
| 6621 | } |
| 6622 | |
| 6623 | /* Remember the new sched domains */ |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 6624 | if (doms_cur != &fallback_doms) |
| 6625 | free_sched_domains(doms_cur, ndoms_cur); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 6626 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6627 | doms_cur = doms_new; |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 6628 | dattr_cur = dattr_new; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6629 | ndoms_cur = ndoms_new; |
Milton Miller | 7378547 | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 6630 | |
| 6631 | register_sched_domain_sysctl(); |
Srivatsa Vaddagiri | a183561 | 2008-01-25 21:08:00 +0100 | [diff] [blame] | 6632 | |
Heiko Carstens | 712555e | 2008-04-28 11:33:07 +0200 | [diff] [blame] | 6633 | mutex_unlock(&sched_domains_mutex); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 6634 | } |
| 6635 | |
Srivatsa S. Bhat | d35be8b | 2012-05-24 19:46:26 +0530 | [diff] [blame] | 6636 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
| 6637 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6638 | /* |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 6639 | * Update cpusets according to cpu_active mask. If cpusets are |
| 6640 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper |
| 6641 | * around partition_sched_domains(). |
Srivatsa S. Bhat | d35be8b | 2012-05-24 19:46:26 +0530 | [diff] [blame] | 6642 | * |
| 6643 | * If we come here as part of a suspend/resume, don't touch cpusets because we |
| 6644 | * want to restore it back to its original state upon resume anyway. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6645 | */ |
Tejun Heo | 0b2e918 | 2010-06-21 23:53:31 +0200 | [diff] [blame] | 6646 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
| 6647 | void *hcpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6648 | { |
Srivatsa S. Bhat | d35be8b | 2012-05-24 19:46:26 +0530 | [diff] [blame] | 6649 | switch (action) { |
| 6650 | case CPU_ONLINE_FROZEN: |
| 6651 | case CPU_DOWN_FAILED_FROZEN: |
| 6652 | |
| 6653 | /* |
| 6654 | * num_cpus_frozen tracks how many CPUs are involved in suspend |
| 6655 | * resume sequence. As long as this is not the last online |
| 6656 | * operation in the resume sequence, just build a single sched |
| 6657 | * domain, ignoring cpusets. |
| 6658 | */ |
| 6659 | num_cpus_frozen--; |
| 6660 | if (likely(num_cpus_frozen)) { |
| 6661 | partition_sched_domains(1, NULL, NULL); |
| 6662 | break; |
| 6663 | } |
| 6664 | |
| 6665 | /* |
| 6666 | * This is the last CPU online operation. So fall through and |
| 6667 | * restore the original sched domains by considering the |
| 6668 | * cpuset configurations. |
| 6669 | */ |
| 6670 | |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 6671 | case CPU_ONLINE: |
Peter Zijlstra | 6ad4c18 | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 6672 | case CPU_DOWN_FAILED: |
Srivatsa S. Bhat | 7ddf96b | 2012-05-24 19:46:55 +0530 | [diff] [blame] | 6673 | cpuset_update_active_cpus(true); |
Srivatsa S. Bhat | d35be8b | 2012-05-24 19:46:26 +0530 | [diff] [blame] | 6674 | break; |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 6675 | default: |
| 6676 | return NOTIFY_DONE; |
| 6677 | } |
Srivatsa S. Bhat | d35be8b | 2012-05-24 19:46:26 +0530 | [diff] [blame] | 6678 | return NOTIFY_OK; |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 6679 | } |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 6680 | |
Tejun Heo | 0b2e918 | 2010-06-21 23:53:31 +0200 | [diff] [blame] | 6681 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
| 6682 | void *hcpu) |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 6683 | { |
Srivatsa S. Bhat | d35be8b | 2012-05-24 19:46:26 +0530 | [diff] [blame] | 6684 | switch (action) { |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 6685 | case CPU_DOWN_PREPARE: |
Srivatsa S. Bhat | 7ddf96b | 2012-05-24 19:46:55 +0530 | [diff] [blame] | 6686 | cpuset_update_active_cpus(false); |
Srivatsa S. Bhat | d35be8b | 2012-05-24 19:46:26 +0530 | [diff] [blame] | 6687 | break; |
| 6688 | case CPU_DOWN_PREPARE_FROZEN: |
| 6689 | num_cpus_frozen++; |
| 6690 | partition_sched_domains(1, NULL, NULL); |
| 6691 | break; |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 6692 | default: |
| 6693 | return NOTIFY_DONE; |
| 6694 | } |
Srivatsa S. Bhat | d35be8b | 2012-05-24 19:46:26 +0530 | [diff] [blame] | 6695 | return NOTIFY_OK; |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 6696 | } |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 6697 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6698 | void __init sched_init_smp(void) |
| 6699 | { |
Rusty Russell | dcc30a3 | 2008-11-25 02:35:12 +1030 | [diff] [blame] | 6700 | cpumask_var_t non_isolated_cpus; |
| 6701 | |
| 6702 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); |
Yong Zhang | cb5fd13 | 2009-09-14 20:20:16 +0800 | [diff] [blame] | 6703 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
Nick Piggin | 5c1e176 | 2006-10-03 01:14:04 -0700 | [diff] [blame] | 6704 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6705 | sched_init_numa(); |
| 6706 | |
Gautham R Shenoy | 95402b3 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 6707 | get_online_cpus(); |
Heiko Carstens | 712555e | 2008-04-28 11:33:07 +0200 | [diff] [blame] | 6708 | mutex_lock(&sched_domains_mutex); |
Peter Zijlstra | c4a8849 | 2011-04-07 14:09:42 +0200 | [diff] [blame] | 6709 | init_sched_domains(cpu_active_mask); |
Rusty Russell | dcc30a3 | 2008-11-25 02:35:12 +1030 | [diff] [blame] | 6710 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
| 6711 | if (cpumask_empty(non_isolated_cpus)) |
| 6712 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); |
Heiko Carstens | 712555e | 2008-04-28 11:33:07 +0200 | [diff] [blame] | 6713 | mutex_unlock(&sched_domains_mutex); |
Gautham R Shenoy | 95402b3 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 6714 | put_online_cpus(); |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 6715 | |
Tang Chen | 301a5cb | 2012-09-25 21:12:31 +0800 | [diff] [blame] | 6716 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 6717 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
| 6718 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 6719 | |
| 6720 | /* RT runtime code needs to handle some hotplug events */ |
| 6721 | hotcpu_notifier(update_runtime, 0); |
| 6722 | |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 6723 | init_hrtick(); |
Nick Piggin | 5c1e176 | 2006-10-03 01:14:04 -0700 | [diff] [blame] | 6724 | |
| 6725 | /* Move init over to a non-isolated CPU */ |
Rusty Russell | dcc30a3 | 2008-11-25 02:35:12 +1030 | [diff] [blame] | 6726 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
Nick Piggin | 5c1e176 | 2006-10-03 01:14:04 -0700 | [diff] [blame] | 6727 | BUG(); |
Ingo Molnar | 19978ca | 2007-11-09 22:39:38 +0100 | [diff] [blame] | 6728 | sched_init_granularity(); |
Rusty Russell | dcc30a3 | 2008-11-25 02:35:12 +1030 | [diff] [blame] | 6729 | free_cpumask_var(non_isolated_cpus); |
Rusty Russell | 4212823 | 2008-11-25 02:35:12 +1030 | [diff] [blame] | 6730 | |
Rusty Russell | 0e3900e | 2008-11-25 02:35:13 +1030 | [diff] [blame] | 6731 | init_sched_rt_class(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6732 | } |
| 6733 | #else |
| 6734 | void __init sched_init_smp(void) |
| 6735 | { |
Ingo Molnar | 19978ca | 2007-11-09 22:39:38 +0100 | [diff] [blame] | 6736 | sched_init_granularity(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6737 | } |
| 6738 | #endif /* CONFIG_SMP */ |
| 6739 | |
Arun R Bharadwaj | cd1bb94 | 2009-04-16 12:15:34 +0530 | [diff] [blame] | 6740 | const_debug unsigned int sysctl_timer_migration = 1; |
| 6741 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6742 | int in_sched_functions(unsigned long addr) |
| 6743 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6744 | return in_lock_functions(addr) || |
| 6745 | (addr >= (unsigned long)__sched_text_start |
| 6746 | && addr < (unsigned long)__sched_text_end); |
| 6747 | } |
| 6748 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 6749 | #ifdef CONFIG_CGROUP_SCHED |
| 6750 | struct task_group root_task_group; |
Mike Galbraith | 35cf4e5 | 2012-08-07 05:00:13 +0200 | [diff] [blame] | 6751 | LIST_HEAD(task_groups); |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 6752 | #endif |
| 6753 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 6754 | DECLARE_PER_CPU(cpumask_var_t, load_balance_tmpmask); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 6755 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6756 | void __init sched_init(void) |
| 6757 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6758 | int i, j; |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 6759 | unsigned long alloc_size = 0, ptr; |
| 6760 | |
| 6761 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 6762 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); |
| 6763 | #endif |
| 6764 | #ifdef CONFIG_RT_GROUP_SCHED |
| 6765 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); |
| 6766 | #endif |
Rusty Russell | df7c8e8 | 2009-03-19 15:22:20 +1030 | [diff] [blame] | 6767 | #ifdef CONFIG_CPUMASK_OFFSTACK |
Rusty Russell | 8c083f0 | 2009-03-19 15:22:20 +1030 | [diff] [blame] | 6768 | alloc_size += num_possible_cpus() * cpumask_size(); |
Rusty Russell | df7c8e8 | 2009-03-19 15:22:20 +1030 | [diff] [blame] | 6769 | #endif |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 6770 | if (alloc_size) { |
Pekka Enberg | 36b7b6d | 2009-06-10 23:42:36 +0300 | [diff] [blame] | 6771 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 6772 | |
| 6773 | #ifdef CONFIG_FAIR_GROUP_SCHED |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 6774 | root_task_group.se = (struct sched_entity **)ptr; |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 6775 | ptr += nr_cpu_ids * sizeof(void **); |
| 6776 | |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 6777 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 6778 | ptr += nr_cpu_ids * sizeof(void **); |
Peter Zijlstra | eff766a | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 6779 | |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 6780 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 6781 | #ifdef CONFIG_RT_GROUP_SCHED |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 6782 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 6783 | ptr += nr_cpu_ids * sizeof(void **); |
| 6784 | |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 6785 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
Peter Zijlstra | eff766a | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 6786 | ptr += nr_cpu_ids * sizeof(void **); |
| 6787 | |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 6788 | #endif /* CONFIG_RT_GROUP_SCHED */ |
Rusty Russell | df7c8e8 | 2009-03-19 15:22:20 +1030 | [diff] [blame] | 6789 | #ifdef CONFIG_CPUMASK_OFFSTACK |
| 6790 | for_each_possible_cpu(i) { |
| 6791 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; |
| 6792 | ptr += cpumask_size(); |
| 6793 | } |
| 6794 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 6795 | } |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6796 | |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 6797 | #ifdef CONFIG_SMP |
| 6798 | init_defrootdomain(); |
| 6799 | #endif |
| 6800 | |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 6801 | init_rt_bandwidth(&def_rt_bandwidth, |
| 6802 | global_rt_period(), global_rt_runtime()); |
| 6803 | |
| 6804 | #ifdef CONFIG_RT_GROUP_SCHED |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 6805 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 6806 | global_rt_period(), global_rt_runtime()); |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 6807 | #endif /* CONFIG_RT_GROUP_SCHED */ |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 6808 | |
Dhaval Giani | 7c94143 | 2010-01-20 13:26:18 +0100 | [diff] [blame] | 6809 | #ifdef CONFIG_CGROUP_SCHED |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 6810 | list_add(&root_task_group.list, &task_groups); |
| 6811 | INIT_LIST_HEAD(&root_task_group.children); |
Glauber Costa | f4d6f6c | 2011-11-01 19:19:07 -0200 | [diff] [blame] | 6812 | INIT_LIST_HEAD(&root_task_group.siblings); |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 6813 | autogroup_init(&init_task); |
Glauber Costa | 54c707e | 2011-11-28 14:45:19 -0200 | [diff] [blame] | 6814 | |
Dhaval Giani | 7c94143 | 2010-01-20 13:26:18 +0100 | [diff] [blame] | 6815 | #endif /* CONFIG_CGROUP_SCHED */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 6816 | |
Glauber Costa | 54c707e | 2011-11-28 14:45:19 -0200 | [diff] [blame] | 6817 | #ifdef CONFIG_CGROUP_CPUACCT |
| 6818 | root_cpuacct.cpustat = &kernel_cpustat; |
| 6819 | root_cpuacct.cpuusage = alloc_percpu(u64); |
| 6820 | /* Too early, not expected to fail */ |
| 6821 | BUG_ON(!root_cpuacct.cpuusage); |
| 6822 | #endif |
KAMEZAWA Hiroyuki | 0a94502 | 2006-03-28 01:56:37 -0800 | [diff] [blame] | 6823 | for_each_possible_cpu(i) { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 6824 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6825 | |
| 6826 | rq = cpu_rq(i); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 6827 | raw_spin_lock_init(&rq->lock); |
Nick Piggin | 7897986 | 2005-06-25 14:57:13 -0700 | [diff] [blame] | 6828 | rq->nr_running = 0; |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 6829 | rq->calc_load_active = 0; |
| 6830 | rq->calc_load_update = jiffies + LOAD_FREQ; |
Jan H. Schönherr | acb5a9b | 2011-07-14 18:32:43 +0200 | [diff] [blame] | 6831 | init_cfs_rq(&rq->cfs); |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 6832 | init_rt_rq(&rq->rt, rq); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 6833 | #ifdef CONFIG_FAIR_GROUP_SCHED |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 6834 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 6835 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 6836 | /* |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 6837 | * How much cpu bandwidth does root_task_group get? |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 6838 | * |
| 6839 | * In case of task-groups formed thr' the cgroup filesystem, it |
| 6840 | * gets 100% of the cpu resources in the system. This overall |
| 6841 | * system cpu resource is divided among the tasks of |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 6842 | * root_task_group and its child task-groups in a fair manner, |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 6843 | * based on each entity's (task or task-group's) weight |
| 6844 | * (se->load.weight). |
| 6845 | * |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 6846 | * In other words, if root_task_group has 10 tasks of weight |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 6847 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
| 6848 | * then A0's share of the cpu resource is: |
| 6849 | * |
Ingo Molnar | 0d905bc | 2009-05-04 19:13:30 +0200 | [diff] [blame] | 6850 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 6851 | * |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 6852 | * We achieve this by letting root_task_group's tasks sit |
| 6853 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 6854 | */ |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 6855 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 6856 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 6857 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
| 6858 | |
| 6859 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 6860 | #ifdef CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 6861 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 6862 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 6863 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6864 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6865 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
| 6866 | rq->cpu_load[j] = 0; |
Venkatesh Pallipadi | fdf3e95 | 2010-05-17 18:14:43 -0700 | [diff] [blame] | 6867 | |
| 6868 | rq->last_load_update_tick = jiffies; |
| 6869 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6870 | #ifdef CONFIG_SMP |
Nick Piggin | 41c7ce9 | 2005-06-25 14:57:24 -0700 | [diff] [blame] | 6871 | rq->sd = NULL; |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 6872 | rq->rd = NULL; |
Nikhil Rao | 1399fa7 | 2011-05-18 10:09:39 -0700 | [diff] [blame] | 6873 | rq->cpu_power = SCHED_POWER_SCALE; |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 6874 | rq->post_schedule = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6875 | rq->active_balance = 0; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6876 | rq->next_balance = jiffies; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6877 | rq->push_cpu = 0; |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 6878 | rq->cpu = i; |
Gregory Haskins | 1f11eb6 | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 6879 | rq->online = 0; |
Mike Galbraith | eae0c9d | 2009-11-10 03:50:02 +0100 | [diff] [blame] | 6880 | rq->idle_stamp = 0; |
| 6881 | rq->avg_idle = 2*sysctl_sched_migration_cost; |
Peter Zijlstra | 367456c | 2012-02-20 21:49:09 +0100 | [diff] [blame] | 6882 | |
| 6883 | INIT_LIST_HEAD(&rq->cfs_tasks); |
| 6884 | |
Gregory Haskins | dc93852 | 2008-01-25 21:08:26 +0100 | [diff] [blame] | 6885 | rq_attach_root(rq, &def_root_domain); |
Venkatesh Pallipadi | 83cd4fe | 2010-05-21 17:09:41 -0700 | [diff] [blame] | 6886 | #ifdef CONFIG_NO_HZ |
Suresh Siddha | 1c792db | 2011-12-01 17:07:32 -0800 | [diff] [blame] | 6887 | rq->nohz_flags = 0; |
Venkatesh Pallipadi | 83cd4fe | 2010-05-21 17:09:41 -0700 | [diff] [blame] | 6888 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6889 | #endif |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 6890 | init_rq_hrtick(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6891 | atomic_set(&rq->nr_iowait, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6892 | } |
| 6893 | |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 6894 | set_load_weight(&init_task); |
Heiko Carstens | b50f60c | 2006-07-30 03:03:52 -0700 | [diff] [blame] | 6895 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 6896 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
| 6897 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); |
| 6898 | #endif |
| 6899 | |
Heiko Carstens | b50f60c | 2006-07-30 03:03:52 -0700 | [diff] [blame] | 6900 | #ifdef CONFIG_RT_MUTEXES |
Dima Zavin | 732375c | 2011-07-07 17:27:59 -0700 | [diff] [blame] | 6901 | plist_head_init(&init_task.pi_waiters); |
Heiko Carstens | b50f60c | 2006-07-30 03:03:52 -0700 | [diff] [blame] | 6902 | #endif |
| 6903 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6904 | /* |
| 6905 | * The boot idle thread does lazy MMU switching as well: |
| 6906 | */ |
| 6907 | atomic_inc(&init_mm.mm_count); |
| 6908 | enter_lazy_tlb(&init_mm, current); |
| 6909 | |
| 6910 | /* |
| 6911 | * Make us the idle thread. Technically, schedule() should not be |
| 6912 | * called from this thread, however somewhere below it might be, |
| 6913 | * but because we are the idle thread, we just pick up running again |
| 6914 | * when this runqueue becomes "idle". |
| 6915 | */ |
| 6916 | init_idle(current, smp_processor_id()); |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 6917 | |
| 6918 | calc_load_update = jiffies + LOAD_FREQ; |
| 6919 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6920 | /* |
| 6921 | * During early bootup we pretend to be a normal task: |
| 6922 | */ |
| 6923 | current->sched_class = &fair_sched_class; |
Ingo Molnar | 6892b75 | 2008-02-13 14:02:36 +0100 | [diff] [blame] | 6924 | |
Rusty Russell | bf4d83f | 2008-11-25 09:57:51 +1030 | [diff] [blame] | 6925 | #ifdef CONFIG_SMP |
Peter Zijlstra | 4cb9883 | 2011-04-07 14:09:58 +0200 | [diff] [blame] | 6926 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
Rusty Russell | bdddd29 | 2009-12-02 14:09:16 +1030 | [diff] [blame] | 6927 | /* May be allocated at isolcpus cmdline parse time */ |
| 6928 | if (cpu_isolated_map == NULL) |
| 6929 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); |
Thomas Gleixner | 29d5e04 | 2012-04-20 13:05:45 +0000 | [diff] [blame] | 6930 | idle_thread_set_boot_cpu(); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 6931 | #endif |
| 6932 | init_sched_fair_class(); |
Rusty Russell | 6a7b3dc | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 6933 | |
Ingo Molnar | 6892b75 | 2008-02-13 14:02:36 +0100 | [diff] [blame] | 6934 | scheduler_running = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6935 | } |
| 6936 | |
Frederic Weisbecker | d902db1 | 2011-06-08 19:31:56 +0200 | [diff] [blame] | 6937 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
Frederic Weisbecker | e4aafea | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 6938 | static inline int preempt_count_equals(int preempt_offset) |
| 6939 | { |
Frederic Weisbecker | 234da7b | 2009-12-16 20:21:05 +0100 | [diff] [blame] | 6940 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
Frederic Weisbecker | e4aafea | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 6941 | |
Arnd Bergmann | 4ba8216 | 2011-01-25 22:52:22 +0100 | [diff] [blame] | 6942 | return (nested == preempt_offset); |
Frederic Weisbecker | e4aafea | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 6943 | } |
| 6944 | |
Simon Kagstrom | d894837 | 2009-12-23 11:08:18 +0100 | [diff] [blame] | 6945 | void __might_sleep(const char *file, int line, int preempt_offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6946 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6947 | static unsigned long prev_jiffy; /* ratelimiting */ |
| 6948 | |
Paul E. McKenney | b3fbab0 | 2011-05-24 08:31:09 -0700 | [diff] [blame] | 6949 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
Frederic Weisbecker | e4aafea | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 6950 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
| 6951 | system_state != SYSTEM_RUNNING || oops_in_progress) |
Ingo Molnar | aef745f | 2008-08-28 11:34:43 +0200 | [diff] [blame] | 6952 | return; |
| 6953 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
| 6954 | return; |
| 6955 | prev_jiffy = jiffies; |
| 6956 | |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 6957 | printk(KERN_ERR |
| 6958 | "BUG: sleeping function called from invalid context at %s:%d\n", |
| 6959 | file, line); |
| 6960 | printk(KERN_ERR |
| 6961 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", |
| 6962 | in_atomic(), irqs_disabled(), |
| 6963 | current->pid, current->comm); |
Ingo Molnar | aef745f | 2008-08-28 11:34:43 +0200 | [diff] [blame] | 6964 | |
| 6965 | debug_show_held_locks(current); |
| 6966 | if (irqs_disabled()) |
| 6967 | print_irqtrace_events(current); |
| 6968 | dump_stack(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6969 | } |
| 6970 | EXPORT_SYMBOL(__might_sleep); |
| 6971 | #endif |
| 6972 | |
| 6973 | #ifdef CONFIG_MAGIC_SYSRQ |
Andi Kleen | 3a5e4dc | 2007-10-15 17:00:15 +0200 | [diff] [blame] | 6974 | static void normalize_task(struct rq *rq, struct task_struct *p) |
| 6975 | { |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 6976 | const struct sched_class *prev_class = p->sched_class; |
| 6977 | int old_prio = p->prio; |
Andi Kleen | 3a5e4dc | 2007-10-15 17:00:15 +0200 | [diff] [blame] | 6978 | int on_rq; |
Peter Zijlstra | 3e51f33 | 2008-05-03 18:29:28 +0200 | [diff] [blame] | 6979 | |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 6980 | on_rq = p->on_rq; |
Andi Kleen | 3a5e4dc | 2007-10-15 17:00:15 +0200 | [diff] [blame] | 6981 | if (on_rq) |
Peter Zijlstra | 4ca9b72 | 2012-01-25 11:50:51 +0100 | [diff] [blame] | 6982 | dequeue_task(rq, p, 0); |
Andi Kleen | 3a5e4dc | 2007-10-15 17:00:15 +0200 | [diff] [blame] | 6983 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
| 6984 | if (on_rq) { |
Peter Zijlstra | 4ca9b72 | 2012-01-25 11:50:51 +0100 | [diff] [blame] | 6985 | enqueue_task(rq, p, 0); |
Andi Kleen | 3a5e4dc | 2007-10-15 17:00:15 +0200 | [diff] [blame] | 6986 | resched_task(rq->curr); |
| 6987 | } |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 6988 | |
| 6989 | check_class_changed(rq, p, prev_class, old_prio); |
Andi Kleen | 3a5e4dc | 2007-10-15 17:00:15 +0200 | [diff] [blame] | 6990 | } |
| 6991 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6992 | void normalize_rt_tasks(void) |
| 6993 | { |
Ingo Molnar | a0f98a1 | 2007-06-17 18:37:45 +0200 | [diff] [blame] | 6994 | struct task_struct *g, *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6995 | unsigned long flags; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 6996 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6997 | |
Peter Zijlstra | 4cf5d77 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 6998 | read_lock_irqsave(&tasklist_lock, flags); |
Ingo Molnar | a0f98a1 | 2007-06-17 18:37:45 +0200 | [diff] [blame] | 6999 | do_each_thread(g, p) { |
Ingo Molnar | 178be79 | 2007-10-15 17:00:18 +0200 | [diff] [blame] | 7000 | /* |
| 7001 | * Only normalize user tasks: |
| 7002 | */ |
| 7003 | if (!p->mm) |
| 7004 | continue; |
| 7005 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 7006 | p->se.exec_start = 0; |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 7007 | #ifdef CONFIG_SCHEDSTATS |
Lucas De Marchi | 41acab8 | 2010-03-10 23:37:45 -0300 | [diff] [blame] | 7008 | p->se.statistics.wait_start = 0; |
| 7009 | p->se.statistics.sleep_start = 0; |
| 7010 | p->se.statistics.block_start = 0; |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 7011 | #endif |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 7012 | |
| 7013 | if (!rt_task(p)) { |
| 7014 | /* |
| 7015 | * Renice negative nice level userspace |
| 7016 | * tasks back to 0: |
| 7017 | */ |
| 7018 | if (TASK_NICE(p) < 0 && p->mm) |
| 7019 | set_user_nice(p, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7020 | continue; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 7021 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7022 | |
Thomas Gleixner | 1d61548 | 2009-11-17 14:54:03 +0100 | [diff] [blame] | 7023 | raw_spin_lock(&p->pi_lock); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 7024 | rq = __task_rq_lock(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7025 | |
Ingo Molnar | 178be79 | 2007-10-15 17:00:18 +0200 | [diff] [blame] | 7026 | normalize_task(rq, p); |
Andi Kleen | 3a5e4dc | 2007-10-15 17:00:15 +0200 | [diff] [blame] | 7027 | |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 7028 | __task_rq_unlock(rq); |
Thomas Gleixner | 1d61548 | 2009-11-17 14:54:03 +0100 | [diff] [blame] | 7029 | raw_spin_unlock(&p->pi_lock); |
Ingo Molnar | a0f98a1 | 2007-06-17 18:37:45 +0200 | [diff] [blame] | 7030 | } while_each_thread(g, p); |
| 7031 | |
Peter Zijlstra | 4cf5d77 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7032 | read_unlock_irqrestore(&tasklist_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7033 | } |
| 7034 | |
| 7035 | #endif /* CONFIG_MAGIC_SYSRQ */ |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7036 | |
Jason Wessel | 67fc4e0 | 2010-05-20 21:04:21 -0500 | [diff] [blame] | 7037 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7038 | /* |
Jason Wessel | 67fc4e0 | 2010-05-20 21:04:21 -0500 | [diff] [blame] | 7039 | * These functions are only useful for the IA64 MCA handling, or kdb. |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7040 | * |
| 7041 | * They can only be called when the whole system has been |
| 7042 | * stopped - every CPU needs to be quiescent, and no scheduling |
| 7043 | * activity can take place. Using them for anything else would |
| 7044 | * be a serious bug, and as a result, they aren't even visible |
| 7045 | * under any other configuration. |
| 7046 | */ |
| 7047 | |
| 7048 | /** |
| 7049 | * curr_task - return the current task for a given cpu. |
| 7050 | * @cpu: the processor in question. |
| 7051 | * |
| 7052 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! |
| 7053 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 7054 | struct task_struct *curr_task(int cpu) |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7055 | { |
| 7056 | return cpu_curr(cpu); |
| 7057 | } |
| 7058 | |
Jason Wessel | 67fc4e0 | 2010-05-20 21:04:21 -0500 | [diff] [blame] | 7059 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
| 7060 | |
| 7061 | #ifdef CONFIG_IA64 |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7062 | /** |
| 7063 | * set_curr_task - set the current task for a given cpu. |
| 7064 | * @cpu: the processor in question. |
| 7065 | * @p: the task pointer to set. |
| 7066 | * |
| 7067 | * Description: This function must only be used when non-maskable interrupts |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 7068 | * are serviced on a separate stack. It allows the architecture to switch the |
| 7069 | * notion of the current task on a cpu in a non-blocking manner. This function |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7070 | * must be called with all CPU's synchronized, and interrupts disabled, the |
| 7071 | * and caller must save the original value of the current task (see |
| 7072 | * curr_task() above) and restore that value before reenabling interrupts and |
| 7073 | * re-starting the system. |
| 7074 | * |
| 7075 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! |
| 7076 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 7077 | void set_curr_task(int cpu, struct task_struct *p) |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7078 | { |
| 7079 | cpu_curr(cpu) = p; |
| 7080 | } |
| 7081 | |
| 7082 | #endif |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7083 | |
Dhaval Giani | 7c94143 | 2010-01-20 13:26:18 +0100 | [diff] [blame] | 7084 | #ifdef CONFIG_CGROUP_SCHED |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7085 | /* task_group_lock serializes the addition/removal of task groups */ |
| 7086 | static DEFINE_SPINLOCK(task_group_lock); |
| 7087 | |
Peter Zijlstra | bccbe08 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7088 | static void free_sched_group(struct task_group *tg) |
| 7089 | { |
| 7090 | free_fair_sched_group(tg); |
| 7091 | free_rt_sched_group(tg); |
Mike Galbraith | e9aa1dd | 2011-01-05 11:11:25 +0100 | [diff] [blame] | 7092 | autogroup_free(tg); |
Peter Zijlstra | bccbe08 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7093 | kfree(tg); |
| 7094 | } |
| 7095 | |
| 7096 | /* allocate runqueue etc for a new task group */ |
Dhaval Giani | ec7dc8a | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7097 | struct task_group *sched_create_group(struct task_group *parent) |
Peter Zijlstra | bccbe08 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7098 | { |
| 7099 | struct task_group *tg; |
| 7100 | unsigned long flags; |
Peter Zijlstra | bccbe08 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7101 | |
| 7102 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); |
| 7103 | if (!tg) |
| 7104 | return ERR_PTR(-ENOMEM); |
| 7105 | |
Dhaval Giani | ec7dc8a | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7106 | if (!alloc_fair_sched_group(tg, parent)) |
Peter Zijlstra | bccbe08 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7107 | goto err; |
| 7108 | |
Dhaval Giani | ec7dc8a | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7109 | if (!alloc_rt_sched_group(tg, parent)) |
Peter Zijlstra | bccbe08 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7110 | goto err; |
| 7111 | |
Peter Zijlstra | 8ed3699 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7112 | spin_lock_irqsave(&task_group_lock, flags); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7113 | list_add_rcu(&tg->list, &task_groups); |
Peter Zijlstra | f473aa5 | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 7114 | |
| 7115 | WARN_ON(!parent); /* root should already exist */ |
| 7116 | |
| 7117 | tg->parent = parent; |
Peter Zijlstra | f473aa5 | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 7118 | INIT_LIST_HEAD(&tg->children); |
Zhang, Yanmin | 09f2724 | 2030-08-14 15:56:40 +0800 | [diff] [blame] | 7119 | list_add_rcu(&tg->siblings, &parent->children); |
Peter Zijlstra | 8ed3699 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7120 | spin_unlock_irqrestore(&task_group_lock, flags); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7121 | |
Srivatsa Vaddagiri | 9b5b775 | 2007-10-15 17:00:09 +0200 | [diff] [blame] | 7122 | return tg; |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7123 | |
| 7124 | err: |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7125 | free_sched_group(tg); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7126 | return ERR_PTR(-ENOMEM); |
| 7127 | } |
| 7128 | |
Srivatsa Vaddagiri | 9b5b775 | 2007-10-15 17:00:09 +0200 | [diff] [blame] | 7129 | /* rcu callback to free various structures associated with a task group */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7130 | static void free_sched_group_rcu(struct rcu_head *rhp) |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7131 | { |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7132 | /* now it should be safe to free those cfs_rqs */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7133 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7134 | } |
| 7135 | |
Srivatsa Vaddagiri | 9b5b775 | 2007-10-15 17:00:09 +0200 | [diff] [blame] | 7136 | /* Destroy runqueue etc associated with a task group */ |
Ingo Molnar | 4cf86d7 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 7137 | void sched_destroy_group(struct task_group *tg) |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7138 | { |
Peter Zijlstra | 8ed3699 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7139 | unsigned long flags; |
Srivatsa Vaddagiri | 9b5b775 | 2007-10-15 17:00:09 +0200 | [diff] [blame] | 7140 | int i; |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7141 | |
Peter Zijlstra | 3d4b47b | 2010-11-15 15:47:01 -0800 | [diff] [blame] | 7142 | /* end participation in shares distribution */ |
| 7143 | for_each_possible_cpu(i) |
Peter Zijlstra | bccbe08 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7144 | unregister_fair_sched_group(tg, i); |
Peter Zijlstra | 3d4b47b | 2010-11-15 15:47:01 -0800 | [diff] [blame] | 7145 | |
| 7146 | spin_lock_irqsave(&task_group_lock, flags); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7147 | list_del_rcu(&tg->list); |
Peter Zijlstra | f473aa5 | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 7148 | list_del_rcu(&tg->siblings); |
Peter Zijlstra | 8ed3699 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7149 | spin_unlock_irqrestore(&task_group_lock, flags); |
Srivatsa Vaddagiri | 9b5b775 | 2007-10-15 17:00:09 +0200 | [diff] [blame] | 7150 | |
Srivatsa Vaddagiri | 9b5b775 | 2007-10-15 17:00:09 +0200 | [diff] [blame] | 7151 | /* wait for possible concurrent references to cfs_rqs complete */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7152 | call_rcu(&tg->rcu, free_sched_group_rcu); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7153 | } |
| 7154 | |
Srivatsa Vaddagiri | 9b5b775 | 2007-10-15 17:00:09 +0200 | [diff] [blame] | 7155 | /* change task's runqueue when it moves between groups. |
Ingo Molnar | 3a25201 | 2007-10-15 17:00:12 +0200 | [diff] [blame] | 7156 | * The caller of this function should have put the task in its new group |
| 7157 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to |
| 7158 | * reflect its new group. |
Srivatsa Vaddagiri | 9b5b775 | 2007-10-15 17:00:09 +0200 | [diff] [blame] | 7159 | */ |
| 7160 | void sched_move_task(struct task_struct *tsk) |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7161 | { |
Peter Zijlstra | 8323f26 | 2012-06-22 13:36:05 +0200 | [diff] [blame] | 7162 | struct task_group *tg; |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7163 | int on_rq, running; |
| 7164 | unsigned long flags; |
| 7165 | struct rq *rq; |
| 7166 | |
| 7167 | rq = task_rq_lock(tsk, &flags); |
| 7168 | |
Dmitry Adamushko | 051a1d1 | 2007-12-18 15:21:13 +0100 | [diff] [blame] | 7169 | running = task_current(rq, tsk); |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 7170 | on_rq = tsk->on_rq; |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7171 | |
Hiroshi Shimamoto | 0e1f348 | 2008-03-10 11:01:20 -0700 | [diff] [blame] | 7172 | if (on_rq) |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7173 | dequeue_task(rq, tsk, 0); |
Hiroshi Shimamoto | 0e1f348 | 2008-03-10 11:01:20 -0700 | [diff] [blame] | 7174 | if (unlikely(running)) |
| 7175 | tsk->sched_class->put_prev_task(rq, tsk); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7176 | |
Peter Zijlstra | 8323f26 | 2012-06-22 13:36:05 +0200 | [diff] [blame] | 7177 | tg = container_of(task_subsys_state_check(tsk, cpu_cgroup_subsys_id, |
| 7178 | lockdep_is_held(&tsk->sighand->siglock)), |
| 7179 | struct task_group, css); |
| 7180 | tg = autogroup_task_group(tsk, tg); |
| 7181 | tsk->sched_task_group = tg; |
| 7182 | |
Peter Zijlstra | 810b381 | 2008-02-29 15:21:01 -0500 | [diff] [blame] | 7183 | #ifdef CONFIG_FAIR_GROUP_SCHED |
Peter Zijlstra | b2b5ce0 | 2010-10-15 15:24:15 +0200 | [diff] [blame] | 7184 | if (tsk->sched_class->task_move_group) |
| 7185 | tsk->sched_class->task_move_group(tsk, on_rq); |
| 7186 | else |
Peter Zijlstra | 810b381 | 2008-02-29 15:21:01 -0500 | [diff] [blame] | 7187 | #endif |
Peter Zijlstra | b2b5ce0 | 2010-10-15 15:24:15 +0200 | [diff] [blame] | 7188 | set_task_rq(tsk, task_cpu(tsk)); |
Peter Zijlstra | 810b381 | 2008-02-29 15:21:01 -0500 | [diff] [blame] | 7189 | |
Hiroshi Shimamoto | 0e1f348 | 2008-03-10 11:01:20 -0700 | [diff] [blame] | 7190 | if (unlikely(running)) |
| 7191 | tsk->sched_class->set_curr_task(rq); |
| 7192 | if (on_rq) |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 7193 | enqueue_task(rq, tsk, 0); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7194 | |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 7195 | task_rq_unlock(rq, tsk, &flags); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7196 | } |
Dhaval Giani | 7c94143 | 2010-01-20 13:26:18 +0100 | [diff] [blame] | 7197 | #endif /* CONFIG_CGROUP_SCHED */ |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7198 | |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 7199 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7200 | static unsigned long to_ratio(u64 period, u64 runtime) |
| 7201 | { |
| 7202 | if (runtime == RUNTIME_INF) |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 7203 | return 1ULL << 20; |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7204 | |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 7205 | return div64_u64(runtime << 20, period); |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7206 | } |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 7207 | #endif |
| 7208 | |
| 7209 | #ifdef CONFIG_RT_GROUP_SCHED |
| 7210 | /* |
| 7211 | * Ensure that the real time constraints are schedulable. |
| 7212 | */ |
| 7213 | static DEFINE_MUTEX(rt_constraints_mutex); |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7214 | |
Dhaval Giani | 521f1a24 | 2008-02-28 15:21:56 +0530 | [diff] [blame] | 7215 | /* Must be called with tasklist_lock held */ |
| 7216 | static inline int tg_has_rt_tasks(struct task_group *tg) |
| 7217 | { |
| 7218 | struct task_struct *g, *p; |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 7219 | |
Dhaval Giani | 521f1a24 | 2008-02-28 15:21:56 +0530 | [diff] [blame] | 7220 | do_each_thread(g, p) { |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7221 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
Dhaval Giani | 521f1a24 | 2008-02-28 15:21:56 +0530 | [diff] [blame] | 7222 | return 1; |
| 7223 | } while_each_thread(g, p); |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 7224 | |
Dhaval Giani | 521f1a24 | 2008-02-28 15:21:56 +0530 | [diff] [blame] | 7225 | return 0; |
| 7226 | } |
| 7227 | |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 7228 | struct rt_schedulable_data { |
| 7229 | struct task_group *tg; |
| 7230 | u64 rt_period; |
| 7231 | u64 rt_runtime; |
| 7232 | }; |
| 7233 | |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 7234 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 7235 | { |
| 7236 | struct rt_schedulable_data *d = data; |
| 7237 | struct task_group *child; |
| 7238 | unsigned long total, sum = 0; |
| 7239 | u64 period, runtime; |
| 7240 | |
| 7241 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
| 7242 | runtime = tg->rt_bandwidth.rt_runtime; |
| 7243 | |
| 7244 | if (tg == d->tg) { |
| 7245 | period = d->rt_period; |
| 7246 | runtime = d->rt_runtime; |
| 7247 | } |
| 7248 | |
Peter Zijlstra | 4653f80 | 2008-09-23 15:33:44 +0200 | [diff] [blame] | 7249 | /* |
| 7250 | * Cannot have more runtime than the period. |
| 7251 | */ |
| 7252 | if (runtime > period && runtime != RUNTIME_INF) |
| 7253 | return -EINVAL; |
| 7254 | |
| 7255 | /* |
| 7256 | * Ensure we don't starve existing RT tasks. |
| 7257 | */ |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 7258 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
| 7259 | return -EBUSY; |
| 7260 | |
| 7261 | total = to_ratio(period, runtime); |
| 7262 | |
Peter Zijlstra | 4653f80 | 2008-09-23 15:33:44 +0200 | [diff] [blame] | 7263 | /* |
| 7264 | * Nobody can have more than the global setting allows. |
| 7265 | */ |
| 7266 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) |
| 7267 | return -EINVAL; |
| 7268 | |
| 7269 | /* |
| 7270 | * The sum of our children's runtime should not exceed our own. |
| 7271 | */ |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 7272 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
| 7273 | period = ktime_to_ns(child->rt_bandwidth.rt_period); |
| 7274 | runtime = child->rt_bandwidth.rt_runtime; |
| 7275 | |
| 7276 | if (child == d->tg) { |
| 7277 | period = d->rt_period; |
| 7278 | runtime = d->rt_runtime; |
| 7279 | } |
| 7280 | |
| 7281 | sum += to_ratio(period, runtime); |
| 7282 | } |
| 7283 | |
| 7284 | if (sum > total) |
| 7285 | return -EINVAL; |
| 7286 | |
| 7287 | return 0; |
| 7288 | } |
| 7289 | |
| 7290 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
| 7291 | { |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 7292 | int ret; |
| 7293 | |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 7294 | struct rt_schedulable_data data = { |
| 7295 | .tg = tg, |
| 7296 | .rt_period = period, |
| 7297 | .rt_runtime = runtime, |
| 7298 | }; |
| 7299 | |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 7300 | rcu_read_lock(); |
| 7301 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); |
| 7302 | rcu_read_unlock(); |
| 7303 | |
| 7304 | return ret; |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 7305 | } |
| 7306 | |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7307 | static int tg_set_rt_bandwidth(struct task_group *tg, |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7308 | u64 rt_period, u64 rt_runtime) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7309 | { |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 7310 | int i, err = 0; |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7311 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7312 | mutex_lock(&rt_constraints_mutex); |
Dhaval Giani | 521f1a24 | 2008-02-28 15:21:56 +0530 | [diff] [blame] | 7313 | read_lock(&tasklist_lock); |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 7314 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
| 7315 | if (err) |
Dhaval Giani | 521f1a24 | 2008-02-28 15:21:56 +0530 | [diff] [blame] | 7316 | goto unlock; |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 7317 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 7318 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7319 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
| 7320 | tg->rt_bandwidth.rt_runtime = rt_runtime; |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 7321 | |
| 7322 | for_each_possible_cpu(i) { |
| 7323 | struct rt_rq *rt_rq = tg->rt_rq[i]; |
| 7324 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 7325 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 7326 | rt_rq->rt_runtime = rt_runtime; |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 7327 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 7328 | } |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 7329 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
Peter Zijlstra | 4924627 | 2010-10-17 21:46:10 +0200 | [diff] [blame] | 7330 | unlock: |
Dhaval Giani | 521f1a24 | 2008-02-28 15:21:56 +0530 | [diff] [blame] | 7331 | read_unlock(&tasklist_lock); |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7332 | mutex_unlock(&rt_constraints_mutex); |
| 7333 | |
| 7334 | return err; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7335 | } |
| 7336 | |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7337 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
| 7338 | { |
| 7339 | u64 rt_runtime, rt_period; |
| 7340 | |
| 7341 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
| 7342 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; |
| 7343 | if (rt_runtime_us < 0) |
| 7344 | rt_runtime = RUNTIME_INF; |
| 7345 | |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7346 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7347 | } |
| 7348 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7349 | long sched_group_rt_runtime(struct task_group *tg) |
| 7350 | { |
| 7351 | u64 rt_runtime_us; |
| 7352 | |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7353 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7354 | return -1; |
| 7355 | |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7356 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7357 | do_div(rt_runtime_us, NSEC_PER_USEC); |
| 7358 | return rt_runtime_us; |
| 7359 | } |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7360 | |
| 7361 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
| 7362 | { |
| 7363 | u64 rt_runtime, rt_period; |
| 7364 | |
| 7365 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; |
| 7366 | rt_runtime = tg->rt_bandwidth.rt_runtime; |
| 7367 | |
Raistlin | 619b048 | 2008-06-26 18:54:09 +0200 | [diff] [blame] | 7368 | if (rt_period == 0) |
| 7369 | return -EINVAL; |
| 7370 | |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7371 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7372 | } |
| 7373 | |
| 7374 | long sched_group_rt_period(struct task_group *tg) |
| 7375 | { |
| 7376 | u64 rt_period_us; |
| 7377 | |
| 7378 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); |
| 7379 | do_div(rt_period_us, NSEC_PER_USEC); |
| 7380 | return rt_period_us; |
| 7381 | } |
| 7382 | |
| 7383 | static int sched_rt_global_constraints(void) |
| 7384 | { |
Peter Zijlstra | 4653f80 | 2008-09-23 15:33:44 +0200 | [diff] [blame] | 7385 | u64 runtime, period; |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7386 | int ret = 0; |
| 7387 | |
Hiroshi Shimamoto | ec5d498 | 2008-09-10 17:00:19 -0700 | [diff] [blame] | 7388 | if (sysctl_sched_rt_period <= 0) |
| 7389 | return -EINVAL; |
| 7390 | |
Peter Zijlstra | 4653f80 | 2008-09-23 15:33:44 +0200 | [diff] [blame] | 7391 | runtime = global_rt_runtime(); |
| 7392 | period = global_rt_period(); |
| 7393 | |
| 7394 | /* |
| 7395 | * Sanity check on the sysctl variables. |
| 7396 | */ |
| 7397 | if (runtime > period && runtime != RUNTIME_INF) |
| 7398 | return -EINVAL; |
Peter Zijlstra | 10b612f | 2008-06-19 14:22:27 +0200 | [diff] [blame] | 7399 | |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7400 | mutex_lock(&rt_constraints_mutex); |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 7401 | read_lock(&tasklist_lock); |
Peter Zijlstra | 4653f80 | 2008-09-23 15:33:44 +0200 | [diff] [blame] | 7402 | ret = __rt_schedulable(NULL, 0, 0); |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 7403 | read_unlock(&tasklist_lock); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7404 | mutex_unlock(&rt_constraints_mutex); |
| 7405 | |
| 7406 | return ret; |
| 7407 | } |
Dhaval Giani | 54e9912 | 2009-02-27 15:13:54 +0530 | [diff] [blame] | 7408 | |
| 7409 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
| 7410 | { |
| 7411 | /* Don't accept realtime tasks when there is no way for them to run */ |
| 7412 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) |
| 7413 | return 0; |
| 7414 | |
| 7415 | return 1; |
| 7416 | } |
| 7417 | |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 7418 | #else /* !CONFIG_RT_GROUP_SCHED */ |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7419 | static int sched_rt_global_constraints(void) |
| 7420 | { |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 7421 | unsigned long flags; |
| 7422 | int i; |
| 7423 | |
Hiroshi Shimamoto | ec5d498 | 2008-09-10 17:00:19 -0700 | [diff] [blame] | 7424 | if (sysctl_sched_rt_period <= 0) |
| 7425 | return -EINVAL; |
| 7426 | |
Peter Zijlstra | 60aa605 | 2009-05-05 17:50:21 +0200 | [diff] [blame] | 7427 | /* |
| 7428 | * There's always some RT tasks in the root group |
| 7429 | * -- migration, kstopmachine etc.. |
| 7430 | */ |
| 7431 | if (sysctl_sched_rt_runtime == 0) |
| 7432 | return -EBUSY; |
| 7433 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 7434 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 7435 | for_each_possible_cpu(i) { |
| 7436 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; |
| 7437 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 7438 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 7439 | rt_rq->rt_runtime = global_rt_runtime(); |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 7440 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 7441 | } |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 7442 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 7443 | |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7444 | return 0; |
| 7445 | } |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 7446 | #endif /* CONFIG_RT_GROUP_SCHED */ |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7447 | |
| 7448 | int sched_rt_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 7449 | void __user *buffer, size_t *lenp, |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7450 | loff_t *ppos) |
| 7451 | { |
| 7452 | int ret; |
| 7453 | int old_period, old_runtime; |
| 7454 | static DEFINE_MUTEX(mutex); |
| 7455 | |
| 7456 | mutex_lock(&mutex); |
| 7457 | old_period = sysctl_sched_rt_period; |
| 7458 | old_runtime = sysctl_sched_rt_runtime; |
| 7459 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 7460 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7461 | |
| 7462 | if (!ret && write) { |
| 7463 | ret = sched_rt_global_constraints(); |
| 7464 | if (ret) { |
| 7465 | sysctl_sched_rt_period = old_period; |
| 7466 | sysctl_sched_rt_runtime = old_runtime; |
| 7467 | } else { |
| 7468 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); |
| 7469 | def_rt_bandwidth.rt_period = |
| 7470 | ns_to_ktime(global_rt_period()); |
| 7471 | } |
| 7472 | } |
| 7473 | mutex_unlock(&mutex); |
| 7474 | |
| 7475 | return ret; |
| 7476 | } |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7477 | |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7478 | #ifdef CONFIG_CGROUP_SCHED |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7479 | |
| 7480 | /* return corresponding task_group object of a cgroup */ |
Paul Menage | 2b01dfe | 2007-10-24 18:23:50 +0200 | [diff] [blame] | 7481 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7482 | { |
Paul Menage | 2b01dfe | 2007-10-24 18:23:50 +0200 | [diff] [blame] | 7483 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
| 7484 | struct task_group, css); |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7485 | } |
| 7486 | |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 7487 | static struct cgroup_subsys_state *cpu_cgroup_css_alloc(struct cgroup *cgrp) |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7488 | { |
Dhaval Giani | ec7dc8a | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7489 | struct task_group *tg, *parent; |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7490 | |
Paul Menage | 2b01dfe | 2007-10-24 18:23:50 +0200 | [diff] [blame] | 7491 | if (!cgrp->parent) { |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7492 | /* This is early initialization for the top cgroup */ |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 7493 | return &root_task_group.css; |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7494 | } |
| 7495 | |
Dhaval Giani | ec7dc8a | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7496 | parent = cgroup_tg(cgrp->parent); |
| 7497 | tg = sched_create_group(parent); |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7498 | if (IS_ERR(tg)) |
| 7499 | return ERR_PTR(-ENOMEM); |
| 7500 | |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7501 | return &tg->css; |
| 7502 | } |
| 7503 | |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 7504 | static void cpu_cgroup_css_free(struct cgroup *cgrp) |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7505 | { |
Paul Menage | 2b01dfe | 2007-10-24 18:23:50 +0200 | [diff] [blame] | 7506 | struct task_group *tg = cgroup_tg(cgrp); |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7507 | |
| 7508 | sched_destroy_group(tg); |
| 7509 | } |
| 7510 | |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 7511 | static int cpu_cgroup_can_attach(struct cgroup *cgrp, |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 7512 | struct cgroup_taskset *tset) |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7513 | { |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 7514 | struct task_struct *task; |
| 7515 | |
| 7516 | cgroup_taskset_for_each(task, cgrp, tset) { |
Peter Zijlstra | b68aa23 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7517 | #ifdef CONFIG_RT_GROUP_SCHED |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 7518 | if (!sched_rt_can_attach(cgroup_tg(cgrp), task)) |
| 7519 | return -EINVAL; |
Peter Zijlstra | b68aa23 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7520 | #else |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 7521 | /* We don't support RT-tasks being in separate groups */ |
| 7522 | if (task->sched_class != &fair_sched_class) |
| 7523 | return -EINVAL; |
Peter Zijlstra | b68aa23 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7524 | #endif |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 7525 | } |
Ben Blum | be367d0 | 2009-09-23 15:56:31 -0700 | [diff] [blame] | 7526 | return 0; |
| 7527 | } |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7528 | |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 7529 | static void cpu_cgroup_attach(struct cgroup *cgrp, |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 7530 | struct cgroup_taskset *tset) |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7531 | { |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 7532 | struct task_struct *task; |
| 7533 | |
| 7534 | cgroup_taskset_for_each(task, cgrp, tset) |
| 7535 | sched_move_task(task); |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7536 | } |
| 7537 | |
Peter Zijlstra | 068c5cc | 2011-01-19 12:26:11 +0100 | [diff] [blame] | 7538 | static void |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 7539 | cpu_cgroup_exit(struct cgroup *cgrp, struct cgroup *old_cgrp, |
| 7540 | struct task_struct *task) |
Peter Zijlstra | 068c5cc | 2011-01-19 12:26:11 +0100 | [diff] [blame] | 7541 | { |
| 7542 | /* |
| 7543 | * cgroup_exit() is called in the copy_process() failure path. |
| 7544 | * Ignore this case since the task hasn't ran yet, this avoids |
| 7545 | * trying to poke a half freed task state from generic code. |
| 7546 | */ |
| 7547 | if (!(task->flags & PF_EXITING)) |
| 7548 | return; |
| 7549 | |
| 7550 | sched_move_task(task); |
| 7551 | } |
| 7552 | |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7553 | #ifdef CONFIG_FAIR_GROUP_SCHED |
Paul Menage | f4c753b | 2008-04-29 00:59:56 -0700 | [diff] [blame] | 7554 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
Paul Menage | 2b01dfe | 2007-10-24 18:23:50 +0200 | [diff] [blame] | 7555 | u64 shareval) |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7556 | { |
Nikhil Rao | c8b2811 | 2011-05-18 14:37:48 -0700 | [diff] [blame] | 7557 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7558 | } |
| 7559 | |
Paul Menage | f4c753b | 2008-04-29 00:59:56 -0700 | [diff] [blame] | 7560 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7561 | { |
Paul Menage | 2b01dfe | 2007-10-24 18:23:50 +0200 | [diff] [blame] | 7562 | struct task_group *tg = cgroup_tg(cgrp); |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7563 | |
Nikhil Rao | c8b2811 | 2011-05-18 14:37:48 -0700 | [diff] [blame] | 7564 | return (u64) scale_load_down(tg->shares); |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7565 | } |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7566 | |
| 7567 | #ifdef CONFIG_CFS_BANDWIDTH |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 7568 | static DEFINE_MUTEX(cfs_constraints_mutex); |
| 7569 | |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7570 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
| 7571 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ |
| 7572 | |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 7573 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
| 7574 | |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7575 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
| 7576 | { |
Paul Turner | 56f570e | 2011-11-07 20:26:33 -0800 | [diff] [blame] | 7577 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7578 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7579 | |
| 7580 | if (tg == &root_task_group) |
| 7581 | return -EINVAL; |
| 7582 | |
| 7583 | /* |
| 7584 | * Ensure we have at some amount of bandwidth every period. This is |
| 7585 | * to prevent reaching a state of large arrears when throttled via |
| 7586 | * entity_tick() resulting in prolonged exit starvation. |
| 7587 | */ |
| 7588 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) |
| 7589 | return -EINVAL; |
| 7590 | |
| 7591 | /* |
| 7592 | * Likewise, bound things on the otherside by preventing insane quota |
| 7593 | * periods. This also allows us to normalize in computing quota |
| 7594 | * feasibility. |
| 7595 | */ |
| 7596 | if (period > max_cfs_quota_period) |
| 7597 | return -EINVAL; |
| 7598 | |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 7599 | mutex_lock(&cfs_constraints_mutex); |
| 7600 | ret = __cfs_schedulable(tg, period, quota); |
| 7601 | if (ret) |
| 7602 | goto out_unlock; |
| 7603 | |
Paul Turner | 58088ad | 2011-07-21 09:43:31 -0700 | [diff] [blame] | 7604 | runtime_enabled = quota != RUNTIME_INF; |
Paul Turner | 56f570e | 2011-11-07 20:26:33 -0800 | [diff] [blame] | 7605 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
| 7606 | account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7607 | raw_spin_lock_irq(&cfs_b->lock); |
| 7608 | cfs_b->period = ns_to_ktime(period); |
| 7609 | cfs_b->quota = quota; |
Paul Turner | 58088ad | 2011-07-21 09:43:31 -0700 | [diff] [blame] | 7610 | |
Paul Turner | a9cf55b | 2011-07-21 09:43:32 -0700 | [diff] [blame] | 7611 | __refill_cfs_bandwidth_runtime(cfs_b); |
Paul Turner | 58088ad | 2011-07-21 09:43:31 -0700 | [diff] [blame] | 7612 | /* restart the period timer (if active) to handle new period expiry */ |
| 7613 | if (runtime_enabled && cfs_b->timer_active) { |
| 7614 | /* force a reprogram */ |
| 7615 | cfs_b->timer_active = 0; |
| 7616 | __start_cfs_bandwidth(cfs_b); |
| 7617 | } |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7618 | raw_spin_unlock_irq(&cfs_b->lock); |
| 7619 | |
| 7620 | for_each_possible_cpu(i) { |
| 7621 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7622 | struct rq *rq = cfs_rq->rq; |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7623 | |
| 7624 | raw_spin_lock_irq(&rq->lock); |
Paul Turner | 58088ad | 2011-07-21 09:43:31 -0700 | [diff] [blame] | 7625 | cfs_rq->runtime_enabled = runtime_enabled; |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7626 | cfs_rq->runtime_remaining = 0; |
Paul Turner | 671fd9d | 2011-07-21 09:43:34 -0700 | [diff] [blame] | 7627 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7628 | if (cfs_rq->throttled) |
Paul Turner | 671fd9d | 2011-07-21 09:43:34 -0700 | [diff] [blame] | 7629 | unthrottle_cfs_rq(cfs_rq); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7630 | raw_spin_unlock_irq(&rq->lock); |
| 7631 | } |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 7632 | out_unlock: |
| 7633 | mutex_unlock(&cfs_constraints_mutex); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7634 | |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 7635 | return ret; |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7636 | } |
| 7637 | |
| 7638 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) |
| 7639 | { |
| 7640 | u64 quota, period; |
| 7641 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7642 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7643 | if (cfs_quota_us < 0) |
| 7644 | quota = RUNTIME_INF; |
| 7645 | else |
| 7646 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; |
| 7647 | |
| 7648 | return tg_set_cfs_bandwidth(tg, period, quota); |
| 7649 | } |
| 7650 | |
| 7651 | long tg_get_cfs_quota(struct task_group *tg) |
| 7652 | { |
| 7653 | u64 quota_us; |
| 7654 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7655 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7656 | return -1; |
| 7657 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7658 | quota_us = tg->cfs_bandwidth.quota; |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7659 | do_div(quota_us, NSEC_PER_USEC); |
| 7660 | |
| 7661 | return quota_us; |
| 7662 | } |
| 7663 | |
| 7664 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) |
| 7665 | { |
| 7666 | u64 quota, period; |
| 7667 | |
| 7668 | period = (u64)cfs_period_us * NSEC_PER_USEC; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7669 | quota = tg->cfs_bandwidth.quota; |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7670 | |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7671 | return tg_set_cfs_bandwidth(tg, period, quota); |
| 7672 | } |
| 7673 | |
| 7674 | long tg_get_cfs_period(struct task_group *tg) |
| 7675 | { |
| 7676 | u64 cfs_period_us; |
| 7677 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7678 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7679 | do_div(cfs_period_us, NSEC_PER_USEC); |
| 7680 | |
| 7681 | return cfs_period_us; |
| 7682 | } |
| 7683 | |
| 7684 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) |
| 7685 | { |
| 7686 | return tg_get_cfs_quota(cgroup_tg(cgrp)); |
| 7687 | } |
| 7688 | |
| 7689 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, |
| 7690 | s64 cfs_quota_us) |
| 7691 | { |
| 7692 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); |
| 7693 | } |
| 7694 | |
| 7695 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) |
| 7696 | { |
| 7697 | return tg_get_cfs_period(cgroup_tg(cgrp)); |
| 7698 | } |
| 7699 | |
| 7700 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
| 7701 | u64 cfs_period_us) |
| 7702 | { |
| 7703 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); |
| 7704 | } |
| 7705 | |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 7706 | struct cfs_schedulable_data { |
| 7707 | struct task_group *tg; |
| 7708 | u64 period, quota; |
| 7709 | }; |
| 7710 | |
| 7711 | /* |
| 7712 | * normalize group quota/period to be quota/max_period |
| 7713 | * note: units are usecs |
| 7714 | */ |
| 7715 | static u64 normalize_cfs_quota(struct task_group *tg, |
| 7716 | struct cfs_schedulable_data *d) |
| 7717 | { |
| 7718 | u64 quota, period; |
| 7719 | |
| 7720 | if (tg == d->tg) { |
| 7721 | period = d->period; |
| 7722 | quota = d->quota; |
| 7723 | } else { |
| 7724 | period = tg_get_cfs_period(tg); |
| 7725 | quota = tg_get_cfs_quota(tg); |
| 7726 | } |
| 7727 | |
| 7728 | /* note: these should typically be equivalent */ |
| 7729 | if (quota == RUNTIME_INF || quota == -1) |
| 7730 | return RUNTIME_INF; |
| 7731 | |
| 7732 | return to_ratio(period, quota); |
| 7733 | } |
| 7734 | |
| 7735 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) |
| 7736 | { |
| 7737 | struct cfs_schedulable_data *d = data; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7738 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 7739 | s64 quota = 0, parent_quota = -1; |
| 7740 | |
| 7741 | if (!tg->parent) { |
| 7742 | quota = RUNTIME_INF; |
| 7743 | } else { |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7744 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 7745 | |
| 7746 | quota = normalize_cfs_quota(tg, d); |
| 7747 | parent_quota = parent_b->hierarchal_quota; |
| 7748 | |
| 7749 | /* |
| 7750 | * ensure max(child_quota) <= parent_quota, inherit when no |
| 7751 | * limit is set |
| 7752 | */ |
| 7753 | if (quota == RUNTIME_INF) |
| 7754 | quota = parent_quota; |
| 7755 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) |
| 7756 | return -EINVAL; |
| 7757 | } |
| 7758 | cfs_b->hierarchal_quota = quota; |
| 7759 | |
| 7760 | return 0; |
| 7761 | } |
| 7762 | |
| 7763 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) |
| 7764 | { |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 7765 | int ret; |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 7766 | struct cfs_schedulable_data data = { |
| 7767 | .tg = tg, |
| 7768 | .period = period, |
| 7769 | .quota = quota, |
| 7770 | }; |
| 7771 | |
| 7772 | if (quota != RUNTIME_INF) { |
| 7773 | do_div(data.period, NSEC_PER_USEC); |
| 7774 | do_div(data.quota, NSEC_PER_USEC); |
| 7775 | } |
| 7776 | |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 7777 | rcu_read_lock(); |
| 7778 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); |
| 7779 | rcu_read_unlock(); |
| 7780 | |
| 7781 | return ret; |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 7782 | } |
Nikhil Rao | e8da1b1 | 2011-07-21 09:43:40 -0700 | [diff] [blame] | 7783 | |
| 7784 | static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft, |
| 7785 | struct cgroup_map_cb *cb) |
| 7786 | { |
| 7787 | struct task_group *tg = cgroup_tg(cgrp); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7788 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
Nikhil Rao | e8da1b1 | 2011-07-21 09:43:40 -0700 | [diff] [blame] | 7789 | |
| 7790 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); |
| 7791 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); |
| 7792 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); |
| 7793 | |
| 7794 | return 0; |
| 7795 | } |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7796 | #endif /* CONFIG_CFS_BANDWIDTH */ |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 7797 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7798 | |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7799 | #ifdef CONFIG_RT_GROUP_SCHED |
Mirco Tischler | 0c70814 | 2008-05-14 16:05:46 -0700 | [diff] [blame] | 7800 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
Paul Menage | 06ecb27 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 7801 | s64 val) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7802 | { |
Paul Menage | 06ecb27 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 7803 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7804 | } |
| 7805 | |
Paul Menage | 06ecb27 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 7806 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7807 | { |
Paul Menage | 06ecb27 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 7808 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7809 | } |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7810 | |
| 7811 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, |
| 7812 | u64 rt_period_us) |
| 7813 | { |
| 7814 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); |
| 7815 | } |
| 7816 | |
| 7817 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) |
| 7818 | { |
| 7819 | return sched_group_rt_period(cgroup_tg(cgrp)); |
| 7820 | } |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 7821 | #endif /* CONFIG_RT_GROUP_SCHED */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7822 | |
Paul Menage | fe5c7cc | 2007-10-29 21:18:11 +0100 | [diff] [blame] | 7823 | static struct cftype cpu_files[] = { |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7824 | #ifdef CONFIG_FAIR_GROUP_SCHED |
Paul Menage | fe5c7cc | 2007-10-29 21:18:11 +0100 | [diff] [blame] | 7825 | { |
| 7826 | .name = "shares", |
Paul Menage | f4c753b | 2008-04-29 00:59:56 -0700 | [diff] [blame] | 7827 | .read_u64 = cpu_shares_read_u64, |
| 7828 | .write_u64 = cpu_shares_write_u64, |
Paul Menage | fe5c7cc | 2007-10-29 21:18:11 +0100 | [diff] [blame] | 7829 | }, |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7830 | #endif |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7831 | #ifdef CONFIG_CFS_BANDWIDTH |
| 7832 | { |
| 7833 | .name = "cfs_quota_us", |
| 7834 | .read_s64 = cpu_cfs_quota_read_s64, |
| 7835 | .write_s64 = cpu_cfs_quota_write_s64, |
| 7836 | }, |
| 7837 | { |
| 7838 | .name = "cfs_period_us", |
| 7839 | .read_u64 = cpu_cfs_period_read_u64, |
| 7840 | .write_u64 = cpu_cfs_period_write_u64, |
| 7841 | }, |
Nikhil Rao | e8da1b1 | 2011-07-21 09:43:40 -0700 | [diff] [blame] | 7842 | { |
| 7843 | .name = "stat", |
| 7844 | .read_map = cpu_stats_show, |
| 7845 | }, |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7846 | #endif |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7847 | #ifdef CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7848 | { |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7849 | .name = "rt_runtime_us", |
Paul Menage | 06ecb27 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 7850 | .read_s64 = cpu_rt_runtime_read, |
| 7851 | .write_s64 = cpu_rt_runtime_write, |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7852 | }, |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7853 | { |
| 7854 | .name = "rt_period_us", |
Paul Menage | f4c753b | 2008-04-29 00:59:56 -0700 | [diff] [blame] | 7855 | .read_u64 = cpu_rt_period_read_uint, |
| 7856 | .write_u64 = cpu_rt_period_write_uint, |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7857 | }, |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7858 | #endif |
Tejun Heo | 4baf6e3 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 7859 | { } /* terminate */ |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7860 | }; |
| 7861 | |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7862 | struct cgroup_subsys cpu_cgroup_subsys = { |
Ingo Molnar | 38605ca | 2007-10-29 21:18:11 +0100 | [diff] [blame] | 7863 | .name = "cpu", |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 7864 | .css_alloc = cpu_cgroup_css_alloc, |
| 7865 | .css_free = cpu_cgroup_css_free, |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 7866 | .can_attach = cpu_cgroup_can_attach, |
| 7867 | .attach = cpu_cgroup_attach, |
Peter Zijlstra | 068c5cc | 2011-01-19 12:26:11 +0100 | [diff] [blame] | 7868 | .exit = cpu_cgroup_exit, |
Ingo Molnar | 38605ca | 2007-10-29 21:18:11 +0100 | [diff] [blame] | 7869 | .subsys_id = cpu_cgroup_subsys_id, |
Tejun Heo | 4baf6e3 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 7870 | .base_cftypes = cpu_files, |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 7871 | .early_init = 1, |
| 7872 | }; |
| 7873 | |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7874 | #endif /* CONFIG_CGROUP_SCHED */ |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7875 | |
| 7876 | #ifdef CONFIG_CGROUP_CPUACCT |
| 7877 | |
| 7878 | /* |
| 7879 | * CPU accounting code for task groups. |
| 7880 | * |
| 7881 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh |
| 7882 | * (balbir@in.ibm.com). |
| 7883 | */ |
| 7884 | |
Frederic Weisbecker | 73fbec6 | 2012-06-16 15:57:37 +0200 | [diff] [blame] | 7885 | struct cpuacct root_cpuacct; |
| 7886 | |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7887 | /* create a new cpu accounting group */ |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 7888 | static struct cgroup_subsys_state *cpuacct_css_alloc(struct cgroup *cgrp) |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7889 | { |
Glauber Costa | 54c707e | 2011-11-28 14:45:19 -0200 | [diff] [blame] | 7890 | struct cpuacct *ca; |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7891 | |
Glauber Costa | 54c707e | 2011-11-28 14:45:19 -0200 | [diff] [blame] | 7892 | if (!cgrp->parent) |
| 7893 | return &root_cpuacct.css; |
| 7894 | |
| 7895 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7896 | if (!ca) |
Bharata B Rao | ef12fef | 2009-03-31 10:02:22 +0530 | [diff] [blame] | 7897 | goto out; |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7898 | |
| 7899 | ca->cpuusage = alloc_percpu(u64); |
Bharata B Rao | ef12fef | 2009-03-31 10:02:22 +0530 | [diff] [blame] | 7900 | if (!ca->cpuusage) |
| 7901 | goto out_free_ca; |
| 7902 | |
Glauber Costa | 54c707e | 2011-11-28 14:45:19 -0200 | [diff] [blame] | 7903 | ca->cpustat = alloc_percpu(struct kernel_cpustat); |
| 7904 | if (!ca->cpustat) |
| 7905 | goto out_free_cpuusage; |
Bharata B Rao | 934352f | 2008-11-10 20:41:13 +0530 | [diff] [blame] | 7906 | |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7907 | return &ca->css; |
Bharata B Rao | ef12fef | 2009-03-31 10:02:22 +0530 | [diff] [blame] | 7908 | |
Glauber Costa | 54c707e | 2011-11-28 14:45:19 -0200 | [diff] [blame] | 7909 | out_free_cpuusage: |
Bharata B Rao | ef12fef | 2009-03-31 10:02:22 +0530 | [diff] [blame] | 7910 | free_percpu(ca->cpuusage); |
| 7911 | out_free_ca: |
| 7912 | kfree(ca); |
| 7913 | out: |
| 7914 | return ERR_PTR(-ENOMEM); |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7915 | } |
| 7916 | |
| 7917 | /* destroy an existing cpu accounting group */ |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 7918 | static void cpuacct_css_free(struct cgroup *cgrp) |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7919 | { |
Dhaval Giani | 32cd756 | 2008-02-29 10:02:43 +0530 | [diff] [blame] | 7920 | struct cpuacct *ca = cgroup_ca(cgrp); |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7921 | |
Glauber Costa | 54c707e | 2011-11-28 14:45:19 -0200 | [diff] [blame] | 7922 | free_percpu(ca->cpustat); |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7923 | free_percpu(ca->cpuusage); |
| 7924 | kfree(ca); |
| 7925 | } |
| 7926 | |
Ken Chen | 720f549 | 2008-12-15 22:02:01 -0800 | [diff] [blame] | 7927 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
| 7928 | { |
Rusty Russell | b36128c | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 7929 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
Ken Chen | 720f549 | 2008-12-15 22:02:01 -0800 | [diff] [blame] | 7930 | u64 data; |
| 7931 | |
| 7932 | #ifndef CONFIG_64BIT |
| 7933 | /* |
| 7934 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. |
| 7935 | */ |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 7936 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
Ken Chen | 720f549 | 2008-12-15 22:02:01 -0800 | [diff] [blame] | 7937 | data = *cpuusage; |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 7938 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
Ken Chen | 720f549 | 2008-12-15 22:02:01 -0800 | [diff] [blame] | 7939 | #else |
| 7940 | data = *cpuusage; |
| 7941 | #endif |
| 7942 | |
| 7943 | return data; |
| 7944 | } |
| 7945 | |
| 7946 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) |
| 7947 | { |
Rusty Russell | b36128c | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 7948 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
Ken Chen | 720f549 | 2008-12-15 22:02:01 -0800 | [diff] [blame] | 7949 | |
| 7950 | #ifndef CONFIG_64BIT |
| 7951 | /* |
| 7952 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. |
| 7953 | */ |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 7954 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
Ken Chen | 720f549 | 2008-12-15 22:02:01 -0800 | [diff] [blame] | 7955 | *cpuusage = val; |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 7956 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
Ken Chen | 720f549 | 2008-12-15 22:02:01 -0800 | [diff] [blame] | 7957 | #else |
| 7958 | *cpuusage = val; |
| 7959 | #endif |
| 7960 | } |
| 7961 | |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7962 | /* return total cpu usage (in nanoseconds) of a group */ |
Dhaval Giani | 32cd756 | 2008-02-29 10:02:43 +0530 | [diff] [blame] | 7963 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7964 | { |
Dhaval Giani | 32cd756 | 2008-02-29 10:02:43 +0530 | [diff] [blame] | 7965 | struct cpuacct *ca = cgroup_ca(cgrp); |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7966 | u64 totalcpuusage = 0; |
| 7967 | int i; |
| 7968 | |
Ken Chen | 720f549 | 2008-12-15 22:02:01 -0800 | [diff] [blame] | 7969 | for_each_present_cpu(i) |
| 7970 | totalcpuusage += cpuacct_cpuusage_read(ca, i); |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 7971 | |
| 7972 | return totalcpuusage; |
| 7973 | } |
| 7974 | |
Dhaval Giani | 0297b80 | 2008-02-29 10:02:44 +0530 | [diff] [blame] | 7975 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
| 7976 | u64 reset) |
| 7977 | { |
| 7978 | struct cpuacct *ca = cgroup_ca(cgrp); |
| 7979 | int err = 0; |
| 7980 | int i; |
| 7981 | |
| 7982 | if (reset) { |
| 7983 | err = -EINVAL; |
| 7984 | goto out; |
| 7985 | } |
| 7986 | |
Ken Chen | 720f549 | 2008-12-15 22:02:01 -0800 | [diff] [blame] | 7987 | for_each_present_cpu(i) |
| 7988 | cpuacct_cpuusage_write(ca, i, 0); |
Dhaval Giani | 0297b80 | 2008-02-29 10:02:44 +0530 | [diff] [blame] | 7989 | |
Dhaval Giani | 0297b80 | 2008-02-29 10:02:44 +0530 | [diff] [blame] | 7990 | out: |
| 7991 | return err; |
| 7992 | } |
| 7993 | |
Ken Chen | e9515c3 | 2008-12-15 22:04:15 -0800 | [diff] [blame] | 7994 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
| 7995 | struct seq_file *m) |
| 7996 | { |
| 7997 | struct cpuacct *ca = cgroup_ca(cgroup); |
| 7998 | u64 percpu; |
| 7999 | int i; |
| 8000 | |
| 8001 | for_each_present_cpu(i) { |
| 8002 | percpu = cpuacct_cpuusage_read(ca, i); |
| 8003 | seq_printf(m, "%llu ", (unsigned long long) percpu); |
| 8004 | } |
| 8005 | seq_printf(m, "\n"); |
| 8006 | return 0; |
| 8007 | } |
| 8008 | |
Bharata B Rao | ef12fef | 2009-03-31 10:02:22 +0530 | [diff] [blame] | 8009 | static const char *cpuacct_stat_desc[] = { |
| 8010 | [CPUACCT_STAT_USER] = "user", |
| 8011 | [CPUACCT_STAT_SYSTEM] = "system", |
| 8012 | }; |
| 8013 | |
| 8014 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, |
Glauber Costa | 54c707e | 2011-11-28 14:45:19 -0200 | [diff] [blame] | 8015 | struct cgroup_map_cb *cb) |
Bharata B Rao | ef12fef | 2009-03-31 10:02:22 +0530 | [diff] [blame] | 8016 | { |
| 8017 | struct cpuacct *ca = cgroup_ca(cgrp); |
Glauber Costa | 54c707e | 2011-11-28 14:45:19 -0200 | [diff] [blame] | 8018 | int cpu; |
| 8019 | s64 val = 0; |
Bharata B Rao | ef12fef | 2009-03-31 10:02:22 +0530 | [diff] [blame] | 8020 | |
Glauber Costa | 54c707e | 2011-11-28 14:45:19 -0200 | [diff] [blame] | 8021 | for_each_online_cpu(cpu) { |
| 8022 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); |
| 8023 | val += kcpustat->cpustat[CPUTIME_USER]; |
| 8024 | val += kcpustat->cpustat[CPUTIME_NICE]; |
Bharata B Rao | ef12fef | 2009-03-31 10:02:22 +0530 | [diff] [blame] | 8025 | } |
Glauber Costa | 54c707e | 2011-11-28 14:45:19 -0200 | [diff] [blame] | 8026 | val = cputime64_to_clock_t(val); |
| 8027 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_USER], val); |
| 8028 | |
| 8029 | val = 0; |
| 8030 | for_each_online_cpu(cpu) { |
| 8031 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); |
| 8032 | val += kcpustat->cpustat[CPUTIME_SYSTEM]; |
| 8033 | val += kcpustat->cpustat[CPUTIME_IRQ]; |
| 8034 | val += kcpustat->cpustat[CPUTIME_SOFTIRQ]; |
| 8035 | } |
| 8036 | |
| 8037 | val = cputime64_to_clock_t(val); |
| 8038 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_SYSTEM], val); |
| 8039 | |
Bharata B Rao | ef12fef | 2009-03-31 10:02:22 +0530 | [diff] [blame] | 8040 | return 0; |
| 8041 | } |
| 8042 | |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 8043 | static struct cftype files[] = { |
| 8044 | { |
| 8045 | .name = "usage", |
Paul Menage | f4c753b | 2008-04-29 00:59:56 -0700 | [diff] [blame] | 8046 | .read_u64 = cpuusage_read, |
| 8047 | .write_u64 = cpuusage_write, |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 8048 | }, |
Ken Chen | e9515c3 | 2008-12-15 22:04:15 -0800 | [diff] [blame] | 8049 | { |
| 8050 | .name = "usage_percpu", |
| 8051 | .read_seq_string = cpuacct_percpu_seq_read, |
| 8052 | }, |
Bharata B Rao | ef12fef | 2009-03-31 10:02:22 +0530 | [diff] [blame] | 8053 | { |
| 8054 | .name = "stat", |
| 8055 | .read_map = cpuacct_stats_show, |
| 8056 | }, |
Tejun Heo | 4baf6e3 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 8057 | { } /* terminate */ |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 8058 | }; |
| 8059 | |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 8060 | /* |
| 8061 | * charge this task's execution time to its accounting group. |
| 8062 | * |
| 8063 | * called with rq->lock held. |
| 8064 | */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 8065 | void cpuacct_charge(struct task_struct *tsk, u64 cputime) |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 8066 | { |
| 8067 | struct cpuacct *ca; |
Bharata B Rao | 934352f | 2008-11-10 20:41:13 +0530 | [diff] [blame] | 8068 | int cpu; |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 8069 | |
Li Zefan | c40c6f8 | 2009-02-26 15:40:15 +0800 | [diff] [blame] | 8070 | if (unlikely(!cpuacct_subsys.active)) |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 8071 | return; |
| 8072 | |
Bharata B Rao | 934352f | 2008-11-10 20:41:13 +0530 | [diff] [blame] | 8073 | cpu = task_cpu(tsk); |
Bharata B Rao | a18b83b | 2009-03-23 10:02:53 +0530 | [diff] [blame] | 8074 | |
| 8075 | rcu_read_lock(); |
| 8076 | |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 8077 | ca = task_ca(tsk); |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 8078 | |
Glauber Costa | 44252e4 | 2011-11-28 14:45:18 -0200 | [diff] [blame] | 8079 | for (; ca; ca = parent_ca(ca)) { |
Rusty Russell | b36128c | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 8080 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 8081 | *cpuusage += cputime; |
| 8082 | } |
Bharata B Rao | a18b83b | 2009-03-23 10:02:53 +0530 | [diff] [blame] | 8083 | |
| 8084 | rcu_read_unlock(); |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 8085 | } |
| 8086 | |
| 8087 | struct cgroup_subsys cpuacct_subsys = { |
| 8088 | .name = "cpuacct", |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 8089 | .css_alloc = cpuacct_css_alloc, |
| 8090 | .css_free = cpuacct_css_free, |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 8091 | .subsys_id = cpuacct_subsys_id, |
Tejun Heo | 4baf6e3 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 8092 | .base_cftypes = files, |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 8093 | }; |
| 8094 | #endif /* CONFIG_CGROUP_CPUACCT */ |
Paul E. McKenney | b637a32 | 2012-09-19 16:58:38 -0700 | [diff] [blame] | 8095 | |
| 8096 | void dump_cpu_task(int cpu) |
| 8097 | { |
| 8098 | pr_info("Task dump for CPU %d:\n", cpu); |
| 8099 | sched_show_task(cpu_curr(cpu)); |
| 8100 | } |