blob: cbabb5adccd958d66442f4af17d701225bc81ef8 [file] [log] [blame]
Michael Ellermancc532912005-12-04 18:39:43 +11001/*
2 * Architecture specific (PPC64) functions for kexec based crash dumps.
3 *
4 * Copyright (C) 2005, IBM Corp.
5 *
6 * Created by: Haren Myneni
7 *
8 * This source code is licensed under the GNU General Public License,
9 * Version 2. See the file COPYING for more details.
10 *
11 */
12
Michael Ellermancc532912005-12-04 18:39:43 +110013#include <linux/kernel.h>
14#include <linux/smp.h>
15#include <linux/reboot.h>
16#include <linux/kexec.h>
Paul Gortmaker66b15db2011-05-27 10:46:24 -040017#include <linux/export.h>
Michael Ellermancc532912005-12-04 18:39:43 +110018#include <linux/crash_dump.h>
Michael Ellermancc532912005-12-04 18:39:43 +110019#include <linux/delay.h>
Michael Ellermand6c1a902006-04-04 13:43:01 +020020#include <linux/irq.h>
Michael Ellermancc532912005-12-04 18:39:43 +110021#include <linux/types.h>
22
23#include <asm/processor.h>
24#include <asm/machdep.h>
David Wilderc0ce7d02006-06-23 15:29:34 -070025#include <asm/kexec.h>
Michael Ellermancc532912005-12-04 18:39:43 +110026#include <asm/kdump.h>
David S. Millerd9b2b2a2008-02-13 16:56:49 -080027#include <asm/prom.h>
Haren Mynenif6cc82f2006-01-10 19:25:25 -080028#include <asm/smp.h>
Michael Neuling496b0102008-01-18 15:50:30 +110029#include <asm/setjmp.h>
David Howellsae3a1972012-03-28 18:30:02 +010030#include <asm/debug.h>
Michael Ellermancc532912005-12-04 18:39:43 +110031
Anton Blanchard549e88a2011-11-30 00:23:16 +000032/*
33 * The primary CPU waits a while for all secondary CPUs to enter. This is to
34 * avoid sending an IPI if the secondary CPUs are entering
35 * crash_kexec_secondary on their own (eg via a system reset).
36 *
37 * The secondary timeout has to be longer than the primary. Both timeouts are
38 * in milliseconds.
39 */
40#define PRIMARY_TIMEOUT 500
41#define SECONDARY_TIMEOUT 1000
42
43#define IPI_TIMEOUT 10000
44#define REAL_MODE_TIMEOUT 10000
45
Anton Blanchard2440c012011-11-30 00:23:17 +000046static int time_to_dump;
Michael Ellermancc532912005-12-04 18:39:43 +110047
Anton Blanchard158d5b5e2011-01-21 13:43:59 +110048#define CRASH_HANDLER_MAX 3
Suraj Jitindar Singh1d145162016-05-11 10:57:32 +100049/* List of shutdown handles */
50static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX];
Michael Neuling496b0102008-01-18 15:50:30 +110051static DEFINE_SPINLOCK(crash_handlers_lock);
52
Anton Blanchard07fe0c62011-11-30 00:23:11 +000053static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
54static int crash_shutdown_cpu = -1;
55
56static int handle_fault(struct pt_regs *regs)
57{
58 if (crash_shutdown_cpu == smp_processor_id())
59 longjmp(crash_shutdown_buf, 1);
60 return 0;
61}
62
Michael Ellermancc532912005-12-04 18:39:43 +110063#ifdef CONFIG_SMP
Michael Ellermancc532912005-12-04 18:39:43 +110064
Christian Kujau897e01a2012-01-17 19:13:05 +000065static atomic_t cpus_in_crash;
Daniel Axtens34852ed2016-05-18 11:16:49 +100066static void crash_ipi_callback(struct pt_regs *regs)
Michael Ellermancc532912005-12-04 18:39:43 +110067{
Anton Blanchard2440c012011-11-30 00:23:17 +000068 static cpumask_t cpus_state_saved = CPU_MASK_NONE;
69
Michael Ellermancc532912005-12-04 18:39:43 +110070 int cpu = smp_processor_id();
71
Michael Ellermancc532912005-12-04 18:39:43 +110072 if (!cpu_online(cpu))
73 return;
74
Paul Mackerrasd04c56f2006-10-04 16:47:49 +100075 hard_irq_disable();
Anton Blanchard2440c012011-11-30 00:23:17 +000076 if (!cpumask_test_cpu(cpu, &cpus_state_saved)) {
Magnus Damm85916f82006-12-06 20:40:41 -080077 crash_save_cpu(regs, cpu);
Anton Blanchard2440c012011-11-30 00:23:17 +000078 cpumask_set_cpu(cpu, &cpus_state_saved);
79 }
80
81 atomic_inc(&cpus_in_crash);
Peter Zijlstrac6450732014-03-13 19:00:35 +010082 smp_mb__after_atomic();
David Wilderc0ce7d02006-06-23 15:29:34 -070083
84 /*
David Wilderc0ce7d02006-06-23 15:29:34 -070085 * Starting the kdump boot.
86 * This barrier is needed to make sure that all CPUs are stopped.
David Wilderc0ce7d02006-06-23 15:29:34 -070087 */
Anton Blanchard2440c012011-11-30 00:23:17 +000088 while (!time_to_dump)
David Wilderc0ce7d02006-06-23 15:29:34 -070089 cpu_relax();
90
Michael Ellermancc532912005-12-04 18:39:43 +110091 if (ppc_md.kexec_cpu_down)
92 ppc_md.kexec_cpu_down(1, 1);
Michael Ellermanb6f35b42006-07-05 14:39:43 +100093
94#ifdef CONFIG_PPC64
Michael Ellermancc532912005-12-04 18:39:43 +110095 kexec_smp_wait();
Michael Ellermanb6f35b42006-07-05 14:39:43 +100096#else
97 for (;;); /* FIXME */
98#endif
99
Michael Ellermancc532912005-12-04 18:39:43 +1100100 /* NOTREACHED */
101}
102
David Wilderc0ce7d02006-06-23 15:29:34 -0700103static void crash_kexec_prepare_cpus(int cpu)
Michael Ellermancc532912005-12-04 18:39:43 +1100104{
105 unsigned int msecs;
David Wilderc0ce7d02006-06-23 15:29:34 -0700106 unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
Anton Blanchard07fe0c62011-11-30 00:23:11 +0000107 int tries = 0;
108 int (*old_handler)(struct pt_regs *regs);
Michael Ellermancc532912005-12-04 18:39:43 +1100109
Anton Blanchard9b00ac02011-11-30 00:23:10 +0000110 printk(KERN_EMERG "Sending IPI to other CPUs\n");
111
Michael Ellermancc532912005-12-04 18:39:43 +1100112 crash_send_ipi(crash_ipi_callback);
113 smp_wmb();
114
Anton Blanchard07fe0c62011-11-30 00:23:11 +0000115again:
Michael Ellermancc532912005-12-04 18:39:43 +1100116 /*
Anton Blanchard158d5b5e2011-01-21 13:43:59 +1100117 * FIXME: Until we will have the way to stop other CPUs reliably,
Michael Ellermancc532912005-12-04 18:39:43 +1100118 * the crash CPU will send an IPI and wait for other CPUs to
David Wilderc0ce7d02006-06-23 15:29:34 -0700119 * respond.
Michael Ellermancc532912005-12-04 18:39:43 +1100120 */
Anton Blanchard549e88a2011-11-30 00:23:16 +0000121 msecs = IPI_TIMEOUT;
Anton Blanchard2440c012011-11-30 00:23:17 +0000122 while ((atomic_read(&cpus_in_crash) < ncpus) && (--msecs > 0))
Michael Ellermancc532912005-12-04 18:39:43 +1100123 mdelay(1);
Michael Ellermancc532912005-12-04 18:39:43 +1100124
125 /* Would it be better to replace the trap vector here? */
126
Anton Blanchard2440c012011-11-30 00:23:17 +0000127 if (atomic_read(&cpus_in_crash) >= ncpus) {
Anton Blanchard07fe0c62011-11-30 00:23:11 +0000128 printk(KERN_EMERG "IPI complete\n");
129 return;
David Wilderc0ce7d02006-06-23 15:29:34 -0700130 }
Anton Blanchard9b00ac02011-11-30 00:23:10 +0000131
Anton Blanchard07fe0c62011-11-30 00:23:11 +0000132 printk(KERN_EMERG "ERROR: %d cpu(s) not responding\n",
Anton Blanchard2440c012011-11-30 00:23:17 +0000133 ncpus - atomic_read(&cpus_in_crash));
Anton Blanchard07fe0c62011-11-30 00:23:11 +0000134
135 /*
136 * If we have a panic timeout set then we can't wait indefinitely
137 * for someone to activate system reset. We also give up on the
138 * second time through if system reset fail to work.
139 */
140 if ((panic_timeout > 0) || (tries > 0))
141 return;
142
143 /*
144 * A system reset will cause all CPUs to take an 0x100 exception.
145 * The primary CPU returns here via setjmp, and the secondary
146 * CPUs reexecute the crash_kexec_secondary path.
147 */
148 old_handler = __debugger;
149 __debugger = handle_fault;
150 crash_shutdown_cpu = smp_processor_id();
151
152 if (setjmp(crash_shutdown_buf) == 0) {
153 printk(KERN_EMERG "Activate system reset (dumprestart) "
154 "to stop other cpu(s)\n");
155
156 /*
157 * A system reset will force all CPUs to execute the
158 * crash code again. We need to reset cpus_in_crash so we
159 * wait for everyone to do this.
160 */
Anton Blanchard2440c012011-11-30 00:23:17 +0000161 atomic_set(&cpus_in_crash, 0);
Anton Blanchard07fe0c62011-11-30 00:23:11 +0000162 smp_mb();
163
Anton Blanchard2440c012011-11-30 00:23:17 +0000164 while (atomic_read(&cpus_in_crash) < ncpus)
Anton Blanchard07fe0c62011-11-30 00:23:11 +0000165 cpu_relax();
166 }
167
168 crash_shutdown_cpu = -1;
169 __debugger = old_handler;
170
171 tries++;
172 goto again;
Michael Ellermancc532912005-12-04 18:39:43 +1100173}
David Wilderc0ce7d02006-06-23 15:29:34 -0700174
175/*
Anton Blanchard9b00ac02011-11-30 00:23:10 +0000176 * This function will be called by secondary cpus.
David Wilderc0ce7d02006-06-23 15:29:34 -0700177 */
178void crash_kexec_secondary(struct pt_regs *regs)
179{
David Wilderc0ce7d02006-06-23 15:29:34 -0700180 unsigned long flags;
Anton Blanchard549e88a2011-11-30 00:23:16 +0000181 int msecs = SECONDARY_TIMEOUT;
David Wilderc0ce7d02006-06-23 15:29:34 -0700182
183 local_irq_save(flags);
Anton Blanchard9b00ac02011-11-30 00:23:10 +0000184
Anton Blanchard549e88a2011-11-30 00:23:16 +0000185 /* Wait for the primary crash CPU to signal its progress */
David Wilderc0ce7d02006-06-23 15:29:34 -0700186 while (crashing_cpu < 0) {
187 if (--msecs < 0) {
Anton Blanchard9b00ac02011-11-30 00:23:10 +0000188 /* No response, kdump image may not have been loaded */
David Wilderc0ce7d02006-06-23 15:29:34 -0700189 local_irq_restore(flags);
190 return;
191 }
Anton Blanchard9b00ac02011-11-30 00:23:10 +0000192
David Wilderc0ce7d02006-06-23 15:29:34 -0700193 mdelay(1);
David Wilderc0ce7d02006-06-23 15:29:34 -0700194 }
Anton Blanchard9b00ac02011-11-30 00:23:10 +0000195
David Wilderc0ce7d02006-06-23 15:29:34 -0700196 crash_ipi_callback(regs);
197}
198
Paul Gortmaker7c7a81b2011-04-13 06:30:08 +0000199#else /* ! CONFIG_SMP */
Paul Gortmaker7c7a81b2011-04-13 06:30:08 +0000200
David Wilderc0ce7d02006-06-23 15:29:34 -0700201static void crash_kexec_prepare_cpus(int cpu)
Michael Ellermancc532912005-12-04 18:39:43 +1100202{
203 /*
Anton Blanchard8c274742011-11-30 00:23:12 +0000204 * move the secondaries to us so that we can copy
Michael Ellermancc532912005-12-04 18:39:43 +1100205 * the new kernel 0-0x100 safely
206 *
207 * do this if kexec in setup.c ?
208 */
Michael Ellermanb6f35b42006-07-05 14:39:43 +1000209#ifdef CONFIG_PPC64
Michael Ellermancc532912005-12-04 18:39:43 +1100210 smp_release_cpus();
Michael Ellermanb6f35b42006-07-05 14:39:43 +1000211#else
212 /* FIXME */
213#endif
Michael Ellermancc532912005-12-04 18:39:43 +1100214}
215
David Wilderc0ce7d02006-06-23 15:29:34 -0700216void crash_kexec_secondary(struct pt_regs *regs)
217{
David Wilderc0ce7d02006-06-23 15:29:34 -0700218}
Paul Gortmaker7c7a81b2011-04-13 06:30:08 +0000219#endif /* CONFIG_SMP */
Michael Ellermancc532912005-12-04 18:39:43 +1100220
Ben Hutchings7707e412011-04-24 15:04:31 +0000221/* wait for all the CPUs to hit real mode but timeout if they don't come in */
Scott Woodeeaab662015-10-06 22:48:16 -0500222#if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
223static void __maybe_unused crash_kexec_wait_realmode(int cpu)
Ben Hutchings7707e412011-04-24 15:04:31 +0000224{
225 unsigned int msecs;
226 int i;
227
Anton Blanchard549e88a2011-11-30 00:23:16 +0000228 msecs = REAL_MODE_TIMEOUT;
Milton Millerbd9e5ee2011-05-10 19:28:41 +0000229 for (i=0; i < nr_cpu_ids && msecs > 0; i++) {
Ben Hutchings7707e412011-04-24 15:04:31 +0000230 if (i == cpu)
231 continue;
232
233 while (paca[i].kexec_state < KEXEC_STATE_REAL_MODE) {
234 barrier();
Michael Neuling63f21a52011-07-04 20:40:10 +0000235 if (!cpu_possible(i) || !cpu_online(i) || (msecs <= 0))
Ben Hutchings7707e412011-04-24 15:04:31 +0000236 break;
Ben Hutchings7707e412011-04-24 15:04:31 +0000237 msecs--;
238 mdelay(1);
239 }
240 }
241 mb();
242}
243#else
244static inline void crash_kexec_wait_realmode(int cpu) {}
Scott Woodeeaab662015-10-06 22:48:16 -0500245#endif /* CONFIG_SMP && CONFIG_PPC64 */
Ben Hutchings7707e412011-04-24 15:04:31 +0000246
Michael Neuling496b0102008-01-18 15:50:30 +1100247/*
248 * Register a function to be called on shutdown. Only use this if you
249 * can't reset your device in the second kernel.
250 */
251int crash_shutdown_register(crash_shutdown_t handler)
252{
253 unsigned int i, rc;
254
255 spin_lock(&crash_handlers_lock);
256 for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
257 if (!crash_shutdown_handles[i]) {
258 /* Insert handle at first empty entry */
259 crash_shutdown_handles[i] = handler;
260 rc = 0;
261 break;
262 }
263
264 if (i == CRASH_HANDLER_MAX) {
265 printk(KERN_ERR "Crash shutdown handles full, "
266 "not registered.\n");
267 rc = 1;
268 }
269
270 spin_unlock(&crash_handlers_lock);
271 return rc;
272}
273EXPORT_SYMBOL(crash_shutdown_register);
274
275int crash_shutdown_unregister(crash_shutdown_t handler)
276{
277 unsigned int i, rc;
278
279 spin_lock(&crash_handlers_lock);
280 for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
281 if (crash_shutdown_handles[i] == handler)
282 break;
283
284 if (i == CRASH_HANDLER_MAX) {
285 printk(KERN_ERR "Crash shutdown handle not found\n");
286 rc = 1;
287 } else {
288 /* Shift handles down */
Suraj Jitindar Singh1d145162016-05-11 10:57:32 +1000289 for (; i < (CRASH_HANDLER_MAX - 1); i++)
Michael Neuling496b0102008-01-18 15:50:30 +1100290 crash_shutdown_handles[i] =
291 crash_shutdown_handles[i+1];
Suraj Jitindar Singh1d145162016-05-11 10:57:32 +1000292 /*
293 * Reset last entry to NULL now that it has been shifted down,
294 * this will allow new handles to be added here.
295 */
296 crash_shutdown_handles[i] = NULL;
Michael Neuling496b0102008-01-18 15:50:30 +1100297 rc = 0;
298 }
299
300 spin_unlock(&crash_handlers_lock);
301 return rc;
302}
303EXPORT_SYMBOL(crash_shutdown_unregister);
304
Michael Ellermancc532912005-12-04 18:39:43 +1100305void default_machine_crash_shutdown(struct pt_regs *regs)
306{
Michael Neuling496b0102008-01-18 15:50:30 +1100307 unsigned int i;
308 int (*old_handler)(struct pt_regs *regs);
309
Michael Ellermancc532912005-12-04 18:39:43 +1100310 /*
311 * This function is only called after the system
Lee Revellf18190b2006-06-26 18:30:00 +0200312 * has panicked or is otherwise in a critical state.
Michael Ellermancc532912005-12-04 18:39:43 +1100313 * The minimum amount of code to allow a kexec'd kernel
314 * to run successfully needs to happen here.
315 *
316 * In practice this means stopping other cpus in
317 * an SMP system.
318 * The kernel is broken so disable interrupts.
319 */
Paul Mackerrasd04c56f2006-10-04 16:47:49 +1000320 hard_irq_disable();
Michael Ellermancc532912005-12-04 18:39:43 +1100321
Anton Blanchard249ec222010-08-02 20:39:41 +0000322 /*
323 * Make a note of crashing cpu. Will be used in machine_kexec
324 * such that another IPI will not be sent.
325 */
326 crashing_cpu = smp_processor_id();
Anton Blanchard549e88a2011-11-30 00:23:16 +0000327
328 /*
329 * If we came in via system reset, wait a while for the secondary
330 * CPUs to enter.
331 */
332 if (TRAP(regs) == 0x100)
333 mdelay(PRIMARY_TIMEOUT);
334
Anton Blanchard249ec222010-08-02 20:39:41 +0000335 crash_kexec_prepare_cpus(crashing_cpu);
Anton Blanchard2440c012011-11-30 00:23:17 +0000336
337 crash_save_cpu(regs, crashing_cpu);
338
339 time_to_dump = 1;
340
Anton Blanchard249ec222010-08-02 20:39:41 +0000341 crash_kexec_wait_realmode(crashing_cpu);
Anton Blanchard249ec222010-08-02 20:39:41 +0000342
Matthew McClintockc71635d2010-09-16 17:58:23 -0500343 machine_kexec_mask_interrupts();
Michael Ellermand6c1a902006-04-04 13:43:01 +0200344
Michael Ellermancc532912005-12-04 18:39:43 +1100345 /*
Anton Blanchard8c274742011-11-30 00:23:12 +0000346 * Call registered shutdown routines safely. Swap out
Michael Neuling496b0102008-01-18 15:50:30 +1100347 * __debugger_fault_handler, and replace on exit.
348 */
349 old_handler = __debugger_fault_handler;
350 __debugger_fault_handler = handle_fault;
Anton Blanchard06440792010-05-10 16:25:51 +0000351 crash_shutdown_cpu = smp_processor_id();
Suraj Jitindar Singha7d63922016-07-11 14:17:31 +1000352 for (i = 0; i < CRASH_HANDLER_MAX && crash_shutdown_handles[i]; i++) {
Michael Neuling496b0102008-01-18 15:50:30 +1100353 if (setjmp(crash_shutdown_buf) == 0) {
354 /*
355 * Insert syncs and delay to ensure
356 * instructions in the dangerous region don't
357 * leak away from this protected region.
358 */
359 asm volatile("sync; isync");
360 /* dangerous region */
361 crash_shutdown_handles[i]();
362 asm volatile("sync; isync");
363 }
364 }
Anton Blanchard06440792010-05-10 16:25:51 +0000365 crash_shutdown_cpu = -1;
Michael Neuling496b0102008-01-18 15:50:30 +1100366 __debugger_fault_handler = old_handler;
367
David Wilderc0ce7d02006-06-23 15:29:34 -0700368 if (ppc_md.kexec_cpu_down)
369 ppc_md.kexec_cpu_down(1, 0);
Michael Ellermancc532912005-12-04 18:39:43 +1100370}