Merge tag 'iio-fixes-4.19a' of git://git.kernel.org/pub/scm/linux/kernel/git/jic23...
[linux-2.6-block.git] / arch / powerpc / kernel / crash.c
CommitLineData
cc532915
ME
1/*
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
cc532915
ME
13#include <linux/kernel.h>
14#include <linux/smp.h>
15#include <linux/reboot.h>
16#include <linux/kexec.h>
66b15db6 17#include <linux/export.h>
cc532915 18#include <linux/crash_dump.h>
cc532915 19#include <linux/delay.h>
d6c1a908 20#include <linux/irq.h>
cc532915
ME
21#include <linux/types.h>
22
23#include <asm/processor.h>
24#include <asm/machdep.h>
c0ce7d08 25#include <asm/kexec.h>
d9b2b2a2 26#include <asm/prom.h>
f6cc82fc 27#include <asm/smp.h>
496b010e 28#include <asm/setjmp.h>
ae3a197e 29#include <asm/debug.h>
cc532915 30
549e88a1
AB
31/*
32 * The primary CPU waits a while for all secondary CPUs to enter. This is to
33 * avoid sending an IPI if the secondary CPUs are entering
34 * crash_kexec_secondary on their own (eg via a system reset).
35 *
36 * The secondary timeout has to be longer than the primary. Both timeouts are
37 * in milliseconds.
38 */
39#define PRIMARY_TIMEOUT 500
40#define SECONDARY_TIMEOUT 1000
41
42#define IPI_TIMEOUT 10000
43#define REAL_MODE_TIMEOUT 10000
44
2440c01e 45static int time_to_dump;
4145f358
BS
46/*
47 * crash_wake_offline should be set to 1 by platforms that intend to wake
48 * up offline cpus prior to jumping to a kdump kernel. Currently powernv
49 * sets it to 1, since we want to avoid things from happening when an
50 * offline CPU wakes up due to something like an HMI (malfunction error),
51 * which propagates to all threads.
52 */
53int crash_wake_offline;
cc532915 54
158d5b5e 55#define CRASH_HANDLER_MAX 3
1d145165
SJS
56/* List of shutdown handles */
57static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX];
496b010e
MN
58static DEFINE_SPINLOCK(crash_handlers_lock);
59
07fe0c61
AB
60static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
61static int crash_shutdown_cpu = -1;
62
63static int handle_fault(struct pt_regs *regs)
64{
65 if (crash_shutdown_cpu == smp_processor_id())
66 longjmp(crash_shutdown_buf, 1);
67 return 0;
68}
69
cc532915 70#ifdef CONFIG_SMP
cc532915 71
897e01a0 72static atomic_t cpus_in_crash;
4145f358 73void crash_ipi_callback(struct pt_regs *regs)
cc532915 74{
2440c01e
AB
75 static cpumask_t cpus_state_saved = CPU_MASK_NONE;
76
cc532915
ME
77 int cpu = smp_processor_id();
78
d04c56f7 79 hard_irq_disable();
2440c01e 80 if (!cpumask_test_cpu(cpu, &cpus_state_saved)) {
85916f81 81 crash_save_cpu(regs, cpu);
2440c01e
AB
82 cpumask_set_cpu(cpu, &cpus_state_saved);
83 }
84
85 atomic_inc(&cpus_in_crash);
c645073f 86 smp_mb__after_atomic();
cc532915 87
c0ce7d08
DW
88 /*
89 * Starting the kdump boot.
90 * This barrier is needed to make sure that all CPUs are stopped.
c0ce7d08 91 */
2440c01e 92 while (!time_to_dump)
c0ce7d08
DW
93 cpu_relax();
94
95 if (ppc_md.kexec_cpu_down)
96 ppc_md.kexec_cpu_down(1, 1);
b6f35b49
ME
97
98#ifdef CONFIG_PPC64
cc532915 99 kexec_smp_wait();
b6f35b49
ME
100#else
101 for (;;); /* FIXME */
102#endif
103
cc532915
ME
104 /* NOTREACHED */
105}
106
c0ce7d08 107static void crash_kexec_prepare_cpus(int cpu)
cc532915
ME
108{
109 unsigned int msecs;
c0ce7d08 110 unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
07fe0c61
AB
111 int tries = 0;
112 int (*old_handler)(struct pt_regs *regs);
cc532915 113
9b00ac06
AB
114 printk(KERN_EMERG "Sending IPI to other CPUs\n");
115
4145f358
BS
116 if (crash_wake_offline)
117 ncpus = num_present_cpus() - 1;
118
cc532915
ME
119 crash_send_ipi(crash_ipi_callback);
120 smp_wmb();
121
07fe0c61 122again:
cc532915 123 /*
158d5b5e 124 * FIXME: Until we will have the way to stop other CPUs reliably,
cc532915 125 * the crash CPU will send an IPI and wait for other CPUs to
c0ce7d08 126 * respond.
cc532915 127 */
549e88a1 128 msecs = IPI_TIMEOUT;
2440c01e 129 while ((atomic_read(&cpus_in_crash) < ncpus) && (--msecs > 0))
cc532915 130 mdelay(1);
cc532915
ME
131
132 /* Would it be better to replace the trap vector here? */
133
2440c01e 134 if (atomic_read(&cpus_in_crash) >= ncpus) {
07fe0c61
AB
135 printk(KERN_EMERG "IPI complete\n");
136 return;
c0ce7d08 137 }
9b00ac06 138
07fe0c61 139 printk(KERN_EMERG "ERROR: %d cpu(s) not responding\n",
2440c01e 140 ncpus - atomic_read(&cpus_in_crash));
07fe0c61
AB
141
142 /*
143 * If we have a panic timeout set then we can't wait indefinitely
144 * for someone to activate system reset. We also give up on the
145 * second time through if system reset fail to work.
146 */
147 if ((panic_timeout > 0) || (tries > 0))
148 return;
149
150 /*
151 * A system reset will cause all CPUs to take an 0x100 exception.
152 * The primary CPU returns here via setjmp, and the secondary
153 * CPUs reexecute the crash_kexec_secondary path.
154 */
155 old_handler = __debugger;
156 __debugger = handle_fault;
157 crash_shutdown_cpu = smp_processor_id();
158
159 if (setjmp(crash_shutdown_buf) == 0) {
160 printk(KERN_EMERG "Activate system reset (dumprestart) "
161 "to stop other cpu(s)\n");
162
163 /*
164 * A system reset will force all CPUs to execute the
165 * crash code again. We need to reset cpus_in_crash so we
166 * wait for everyone to do this.
167 */
2440c01e 168 atomic_set(&cpus_in_crash, 0);
07fe0c61
AB
169 smp_mb();
170
2440c01e 171 while (atomic_read(&cpus_in_crash) < ncpus)
07fe0c61
AB
172 cpu_relax();
173 }
174
175 crash_shutdown_cpu = -1;
176 __debugger = old_handler;
177
178 tries++;
179 goto again;
cc532915 180}
c0ce7d08
DW
181
182/*
9b00ac06 183 * This function will be called by secondary cpus.
c0ce7d08
DW
184 */
185void crash_kexec_secondary(struct pt_regs *regs)
186{
c0ce7d08 187 unsigned long flags;
549e88a1 188 int msecs = SECONDARY_TIMEOUT;
c0ce7d08
DW
189
190 local_irq_save(flags);
9b00ac06 191
549e88a1 192 /* Wait for the primary crash CPU to signal its progress */
c0ce7d08
DW
193 while (crashing_cpu < 0) {
194 if (--msecs < 0) {
9b00ac06 195 /* No response, kdump image may not have been loaded */
c0ce7d08
DW
196 local_irq_restore(flags);
197 return;
198 }
9b00ac06 199
c0ce7d08 200 mdelay(1);
c0ce7d08 201 }
9b00ac06 202
c0ce7d08
DW
203 crash_ipi_callback(regs);
204}
205
7c7a81b5 206#else /* ! CONFIG_SMP */
7c7a81b5 207
c0ce7d08 208static void crash_kexec_prepare_cpus(int cpu)
cc532915
ME
209{
210 /*
8c27474a 211 * move the secondaries to us so that we can copy
cc532915
ME
212 * the new kernel 0-0x100 safely
213 *
214 * do this if kexec in setup.c ?
215 */
b6f35b49 216#ifdef CONFIG_PPC64
cc532915 217 smp_release_cpus();
b6f35b49
ME
218#else
219 /* FIXME */
220#endif
cc532915
ME
221}
222
c0ce7d08
DW
223void crash_kexec_secondary(struct pt_regs *regs)
224{
c0ce7d08 225}
7c7a81b5 226#endif /* CONFIG_SMP */
cc532915 227
7707e411 228/* wait for all the CPUs to hit real mode but timeout if they don't come in */
eeaab663
SW
229#if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
230static void __maybe_unused crash_kexec_wait_realmode(int cpu)
7707e411
BH
231{
232 unsigned int msecs;
233 int i;
234
549e88a1 235 msecs = REAL_MODE_TIMEOUT;
bd9e5eef 236 for (i=0; i < nr_cpu_ids && msecs > 0; i++) {
7707e411
BH
237 if (i == cpu)
238 continue;
239
d2e60075 240 while (paca_ptrs[i]->kexec_state < KEXEC_STATE_REAL_MODE) {
7707e411 241 barrier();
63f21a56 242 if (!cpu_possible(i) || !cpu_online(i) || (msecs <= 0))
7707e411 243 break;
7707e411
BH
244 msecs--;
245 mdelay(1);
246 }
247 }
248 mb();
249}
250#else
251static inline void crash_kexec_wait_realmode(int cpu) {}
eeaab663 252#endif /* CONFIG_SMP && CONFIG_PPC64 */
7707e411 253
496b010e
MN
254/*
255 * Register a function to be called on shutdown. Only use this if you
256 * can't reset your device in the second kernel.
257 */
258int crash_shutdown_register(crash_shutdown_t handler)
259{
260 unsigned int i, rc;
261
262 spin_lock(&crash_handlers_lock);
263 for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
264 if (!crash_shutdown_handles[i]) {
265 /* Insert handle at first empty entry */
266 crash_shutdown_handles[i] = handler;
267 rc = 0;
268 break;
269 }
270
271 if (i == CRASH_HANDLER_MAX) {
272 printk(KERN_ERR "Crash shutdown handles full, "
273 "not registered.\n");
274 rc = 1;
275 }
276
277 spin_unlock(&crash_handlers_lock);
278 return rc;
279}
280EXPORT_SYMBOL(crash_shutdown_register);
281
282int crash_shutdown_unregister(crash_shutdown_t handler)
283{
284 unsigned int i, rc;
285
286 spin_lock(&crash_handlers_lock);
287 for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
288 if (crash_shutdown_handles[i] == handler)
289 break;
290
291 if (i == CRASH_HANDLER_MAX) {
292 printk(KERN_ERR "Crash shutdown handle not found\n");
293 rc = 1;
294 } else {
295 /* Shift handles down */
1d145165 296 for (; i < (CRASH_HANDLER_MAX - 1); i++)
496b010e
MN
297 crash_shutdown_handles[i] =
298 crash_shutdown_handles[i+1];
1d145165
SJS
299 /*
300 * Reset last entry to NULL now that it has been shifted down,
301 * this will allow new handles to be added here.
302 */
303 crash_shutdown_handles[i] = NULL;
496b010e
MN
304 rc = 0;
305 }
306
307 spin_unlock(&crash_handlers_lock);
308 return rc;
309}
310EXPORT_SYMBOL(crash_shutdown_unregister);
311
cc532915
ME
312void default_machine_crash_shutdown(struct pt_regs *regs)
313{
496b010e
MN
314 unsigned int i;
315 int (*old_handler)(struct pt_regs *regs);
316
cc532915
ME
317 /*
318 * This function is only called after the system
f18190bd 319 * has panicked or is otherwise in a critical state.
cc532915
ME
320 * The minimum amount of code to allow a kexec'd kernel
321 * to run successfully needs to happen here.
322 *
323 * In practice this means stopping other cpus in
324 * an SMP system.
325 * The kernel is broken so disable interrupts.
326 */
d04c56f7 327 hard_irq_disable();
cc532915 328
249ec228
AB
329 /*
330 * Make a note of crashing cpu. Will be used in machine_kexec
331 * such that another IPI will not be sent.
332 */
333 crashing_cpu = smp_processor_id();
549e88a1
AB
334
335 /*
336 * If we came in via system reset, wait a while for the secondary
337 * CPUs to enter.
338 */
339 if (TRAP(regs) == 0x100)
340 mdelay(PRIMARY_TIMEOUT);
341
249ec228 342 crash_kexec_prepare_cpus(crashing_cpu);
2440c01e
AB
343
344 crash_save_cpu(regs, crashing_cpu);
345
346 time_to_dump = 1;
347
249ec228 348 crash_kexec_wait_realmode(crashing_cpu);
249ec228 349
c71635d2 350 machine_kexec_mask_interrupts();
496b010e
MN
351
352 /*
8c27474a 353 * Call registered shutdown routines safely. Swap out
496b010e
MN
354 * __debugger_fault_handler, and replace on exit.
355 */
356 old_handler = __debugger_fault_handler;
357 __debugger_fault_handler = handle_fault;
06440794 358 crash_shutdown_cpu = smp_processor_id();
a7d63928 359 for (i = 0; i < CRASH_HANDLER_MAX && crash_shutdown_handles[i]; i++) {
496b010e
MN
360 if (setjmp(crash_shutdown_buf) == 0) {
361 /*
362 * Insert syncs and delay to ensure
363 * instructions in the dangerous region don't
364 * leak away from this protected region.
365 */
366 asm volatile("sync; isync");
367 /* dangerous region */
368 crash_shutdown_handles[i]();
369 asm volatile("sync; isync");
370 }
d6c1a908 371 }
06440794 372 crash_shutdown_cpu = -1;
496b010e 373 __debugger_fault_handler = old_handler;
d6c1a908 374
c0ce7d08
DW
375 if (ppc_md.kexec_cpu_down)
376 ppc_md.kexec_cpu_down(1, 0);
cc532915 377}