Rusty Russell | e5582ca | 2006-09-29 02:01:35 -0700 | [diff] [blame] | 1 | /* Copyright 2005 Rusty Russell rusty@rustcorp.com.au IBM Corporation. |
| 2 | * GPL v2 and any later version. |
| 3 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4 | #include <linux/cpu.h> |
| 5 | #include <linux/err.h> |
Prarit Bhargava | ee527cd | 2007-05-08 00:25:08 -0700 | [diff] [blame] | 6 | #include <linux/kthread.h> |
| 7 | #include <linux/module.h> |
| 8 | #include <linux/sched.h> |
| 9 | #include <linux/stop_machine.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 10 | #include <linux/syscalls.h> |
Benjamin Herrenschmidt | a12bb44 | 2007-05-10 22:22:47 -0700 | [diff] [blame] | 11 | #include <linux/interrupt.h> |
| 12 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | #include <asm/atomic.h> |
| 14 | #include <asm/semaphore.h> |
| 15 | #include <asm/uaccess.h> |
| 16 | |
| 17 | /* Since we effect priority and affinity (both of which are visible |
| 18 | * to, and settable by outside processes) we do indirection via a |
| 19 | * kthread. */ |
| 20 | |
| 21 | /* Thread to stop each CPU in user context. */ |
| 22 | enum stopmachine_state { |
| 23 | STOPMACHINE_WAIT, |
| 24 | STOPMACHINE_PREPARE, |
| 25 | STOPMACHINE_DISABLE_IRQ, |
| 26 | STOPMACHINE_EXIT, |
| 27 | }; |
| 28 | |
| 29 | static enum stopmachine_state stopmachine_state; |
| 30 | static unsigned int stopmachine_num_threads; |
| 31 | static atomic_t stopmachine_thread_ack; |
| 32 | static DECLARE_MUTEX(stopmachine_mutex); |
| 33 | |
Andrew Morton | d8cb7c1 | 2006-07-03 17:32:22 -0700 | [diff] [blame] | 34 | static int stopmachine(void *cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | { |
| 36 | int irqs_disabled = 0; |
| 37 | int prepared = 0; |
| 38 | |
Andrew Morton | d8cb7c1 | 2006-07-03 17:32:22 -0700 | [diff] [blame] | 39 | set_cpus_allowed(current, cpumask_of_cpu((int)(long)cpu)); |
| 40 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | /* Ack: we are alive */ |
akpm@osdl.org | d59dd46 | 2005-05-01 08:58:47 -0700 | [diff] [blame] | 42 | smp_mb(); /* Theoretically the ack = 0 might not be on this CPU yet. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | atomic_inc(&stopmachine_thread_ack); |
| 44 | |
| 45 | /* Simple state machine */ |
| 46 | while (stopmachine_state != STOPMACHINE_EXIT) { |
| 47 | if (stopmachine_state == STOPMACHINE_DISABLE_IRQ |
| 48 | && !irqs_disabled) { |
| 49 | local_irq_disable(); |
Benjamin Herrenschmidt | a12bb44 | 2007-05-10 22:22:47 -0700 | [diff] [blame] | 50 | hard_irq_disable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 51 | irqs_disabled = 1; |
| 52 | /* Ack: irqs disabled. */ |
akpm@osdl.org | d59dd46 | 2005-05-01 08:58:47 -0700 | [diff] [blame] | 53 | smp_mb(); /* Must read state first. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | atomic_inc(&stopmachine_thread_ack); |
| 55 | } else if (stopmachine_state == STOPMACHINE_PREPARE |
| 56 | && !prepared) { |
| 57 | /* Everyone is in place, hold CPU. */ |
| 58 | preempt_disable(); |
| 59 | prepared = 1; |
akpm@osdl.org | d59dd46 | 2005-05-01 08:58:47 -0700 | [diff] [blame] | 60 | smp_mb(); /* Must read state first. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | atomic_inc(&stopmachine_thread_ack); |
| 62 | } |
| 63 | /* Yield in first stage: migration threads need to |
| 64 | * help our sisters onto their CPUs. */ |
| 65 | if (!prepared && !irqs_disabled) |
| 66 | yield(); |
| 67 | else |
| 68 | cpu_relax(); |
| 69 | } |
| 70 | |
| 71 | /* Ack: we are exiting. */ |
akpm@osdl.org | d59dd46 | 2005-05-01 08:58:47 -0700 | [diff] [blame] | 72 | smp_mb(); /* Must read state first. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 73 | atomic_inc(&stopmachine_thread_ack); |
| 74 | |
| 75 | if (irqs_disabled) |
| 76 | local_irq_enable(); |
| 77 | if (prepared) |
| 78 | preempt_enable(); |
| 79 | |
| 80 | return 0; |
| 81 | } |
| 82 | |
| 83 | /* Change the thread state */ |
| 84 | static void stopmachine_set_state(enum stopmachine_state state) |
| 85 | { |
| 86 | atomic_set(&stopmachine_thread_ack, 0); |
akpm@osdl.org | d59dd46 | 2005-05-01 08:58:47 -0700 | [diff] [blame] | 87 | smp_wmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 88 | stopmachine_state = state; |
| 89 | while (atomic_read(&stopmachine_thread_ack) != stopmachine_num_threads) |
| 90 | cpu_relax(); |
| 91 | } |
| 92 | |
| 93 | static int stop_machine(void) |
| 94 | { |
Andrew Morton | d8cb7c1 | 2006-07-03 17:32:22 -0700 | [diff] [blame] | 95 | int i, ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | |
| 97 | atomic_set(&stopmachine_thread_ack, 0); |
| 98 | stopmachine_num_threads = 0; |
| 99 | stopmachine_state = STOPMACHINE_WAIT; |
| 100 | |
| 101 | for_each_online_cpu(i) { |
Ingo Molnar | 39c715b | 2005-06-21 17:14:34 -0700 | [diff] [blame] | 102 | if (i == raw_smp_processor_id()) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 103 | continue; |
Andrew Morton | d8cb7c1 | 2006-07-03 17:32:22 -0700 | [diff] [blame] | 104 | ret = kernel_thread(stopmachine, (void *)(long)i,CLONE_KERNEL); |
| 105 | if (ret < 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | break; |
| 107 | stopmachine_num_threads++; |
| 108 | } |
| 109 | |
| 110 | /* Wait for them all to come to life. */ |
| 111 | while (atomic_read(&stopmachine_thread_ack) != stopmachine_num_threads) |
| 112 | yield(); |
| 113 | |
| 114 | /* If some failed, kill them all. */ |
| 115 | if (ret < 0) { |
| 116 | stopmachine_set_state(STOPMACHINE_EXIT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 117 | return ret; |
| 118 | } |
| 119 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 120 | /* Now they are all started, make them hold the CPUs, ready. */ |
Kirill Korotaev | 4557398 | 2005-11-13 16:07:30 -0800 | [diff] [blame] | 121 | preempt_disable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 122 | stopmachine_set_state(STOPMACHINE_PREPARE); |
| 123 | |
| 124 | /* Make them disable irqs. */ |
Kirill Korotaev | 4557398 | 2005-11-13 16:07:30 -0800 | [diff] [blame] | 125 | local_irq_disable(); |
Benjamin Herrenschmidt | a12bb44 | 2007-05-10 22:22:47 -0700 | [diff] [blame] | 126 | hard_irq_disable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | stopmachine_set_state(STOPMACHINE_DISABLE_IRQ); |
| 128 | |
| 129 | return 0; |
| 130 | } |
| 131 | |
| 132 | static void restart_machine(void) |
| 133 | { |
| 134 | stopmachine_set_state(STOPMACHINE_EXIT); |
| 135 | local_irq_enable(); |
Kirill Korotaev | 4557398 | 2005-11-13 16:07:30 -0800 | [diff] [blame] | 136 | preempt_enable_no_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 137 | } |
| 138 | |
| 139 | struct stop_machine_data |
| 140 | { |
| 141 | int (*fn)(void *); |
| 142 | void *data; |
| 143 | struct completion done; |
| 144 | }; |
| 145 | |
| 146 | static int do_stop(void *_smdata) |
| 147 | { |
| 148 | struct stop_machine_data *smdata = _smdata; |
| 149 | int ret; |
| 150 | |
| 151 | ret = stop_machine(); |
| 152 | if (ret == 0) { |
| 153 | ret = smdata->fn(smdata->data); |
| 154 | restart_machine(); |
| 155 | } |
| 156 | |
| 157 | /* We're done: you can kthread_stop us now */ |
| 158 | complete(&smdata->done); |
| 159 | |
| 160 | /* Wait for kthread_stop */ |
| 161 | set_current_state(TASK_INTERRUPTIBLE); |
| 162 | while (!kthread_should_stop()) { |
| 163 | schedule(); |
| 164 | set_current_state(TASK_INTERRUPTIBLE); |
| 165 | } |
| 166 | __set_current_state(TASK_RUNNING); |
| 167 | return ret; |
| 168 | } |
| 169 | |
| 170 | struct task_struct *__stop_machine_run(int (*fn)(void *), void *data, |
| 171 | unsigned int cpu) |
| 172 | { |
| 173 | struct stop_machine_data smdata; |
| 174 | struct task_struct *p; |
| 175 | |
| 176 | smdata.fn = fn; |
| 177 | smdata.data = data; |
| 178 | init_completion(&smdata.done); |
| 179 | |
| 180 | down(&stopmachine_mutex); |
| 181 | |
| 182 | /* If they don't care which CPU fn runs on, bind to any online one. */ |
| 183 | if (cpu == NR_CPUS) |
Ingo Molnar | 39c715b | 2005-06-21 17:14:34 -0700 | [diff] [blame] | 184 | cpu = raw_smp_processor_id(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | |
| 186 | p = kthread_create(do_stop, &smdata, "kstopmachine"); |
| 187 | if (!IS_ERR(p)) { |
Satoru Takeuchi | 85653af | 2007-07-15 23:39:47 -0700 | [diff] [blame] | 188 | struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 }; |
| 189 | |
| 190 | /* One high-prio thread per cpu. We'll do this one. */ |
| 191 | sched_setscheduler(p, SCHED_FIFO, ¶m); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | kthread_bind(p, cpu); |
| 193 | wake_up_process(p); |
| 194 | wait_for_completion(&smdata.done); |
| 195 | } |
| 196 | up(&stopmachine_mutex); |
| 197 | return p; |
| 198 | } |
| 199 | |
| 200 | int stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu) |
| 201 | { |
| 202 | struct task_struct *p; |
| 203 | int ret; |
| 204 | |
| 205 | /* No CPUs can come up or down during this. */ |
| 206 | lock_cpu_hotplug(); |
| 207 | p = __stop_machine_run(fn, data, cpu); |
| 208 | if (!IS_ERR(p)) |
| 209 | ret = kthread_stop(p); |
| 210 | else |
| 211 | ret = PTR_ERR(p); |
| 212 | unlock_cpu_hotplug(); |
| 213 | |
| 214 | return ret; |
| 215 | } |
Prarit Bhargava | ee527cd | 2007-05-08 00:25:08 -0700 | [diff] [blame] | 216 | EXPORT_SYMBOL_GPL(stop_machine_run); |