]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - arch/powerpc/kernel/crash.c
powerpc/cpuidle: Make it a bool, not a tristate
[mirror_ubuntu-bionic-kernel.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>
cc532915 20#include <linux/init.h>
d6c1a908 21#include <linux/irq.h>
cc532915
ME
22#include <linux/types.h>
23
24#include <asm/processor.h>
25#include <asm/machdep.h>
c0ce7d08 26#include <asm/kexec.h>
cc532915 27#include <asm/kdump.h>
d9b2b2a2 28#include <asm/prom.h>
f6cc82fc 29#include <asm/smp.h>
496b010e
MN
30#include <asm/system.h>
31#include <asm/setjmp.h>
cc532915 32
549e88a1
AB
33/*
34 * The primary CPU waits a while for all secondary CPUs to enter. This is to
35 * avoid sending an IPI if the secondary CPUs are entering
36 * crash_kexec_secondary on their own (eg via a system reset).
37 *
38 * The secondary timeout has to be longer than the primary. Both timeouts are
39 * in milliseconds.
40 */
41#define PRIMARY_TIMEOUT 500
42#define SECONDARY_TIMEOUT 1000
43
44#define IPI_TIMEOUT 10000
45#define REAL_MODE_TIMEOUT 10000
46
8c27474a 47/* This keeps a track of which one is the crashing cpu. */
cc532915 48int crashing_cpu = -1;
2440c01e
AB
49static atomic_t cpus_in_crash;
50static int time_to_dump;
cc532915 51
158d5b5e 52#define CRASH_HANDLER_MAX 3
496b010e
MN
53/* NULL terminated list of shutdown handles */
54static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX+1];
55static DEFINE_SPINLOCK(crash_handlers_lock);
56
07fe0c61
AB
57static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
58static int crash_shutdown_cpu = -1;
59
60static int handle_fault(struct pt_regs *regs)
61{
62 if (crash_shutdown_cpu == smp_processor_id())
63 longjmp(crash_shutdown_buf, 1);
64 return 0;
65}
66
cc532915 67#ifdef CONFIG_SMP
cc532915
ME
68
69void crash_ipi_callback(struct pt_regs *regs)
70{
2440c01e
AB
71 static cpumask_t cpus_state_saved = CPU_MASK_NONE;
72
cc532915
ME
73 int cpu = smp_processor_id();
74
cc532915
ME
75 if (!cpu_online(cpu))
76 return;
77
d04c56f7 78 hard_irq_disable();
2440c01e 79 if (!cpumask_test_cpu(cpu, &cpus_state_saved)) {
85916f81 80 crash_save_cpu(regs, cpu);
2440c01e
AB
81 cpumask_set_cpu(cpu, &cpus_state_saved);
82 }
83
84 atomic_inc(&cpus_in_crash);
85 smp_mb__after_atomic_inc();
cc532915 86
c0ce7d08
DW
87 /*
88 * Starting the kdump boot.
89 * This barrier is needed to make sure that all CPUs are stopped.
c0ce7d08 90 */
2440c01e 91 while (!time_to_dump)
c0ce7d08
DW
92 cpu_relax();
93
94 if (ppc_md.kexec_cpu_down)
95 ppc_md.kexec_cpu_down(1, 1);
b6f35b49
ME
96
97#ifdef CONFIG_PPC64
cc532915 98 kexec_smp_wait();
b6f35b49
ME
99#else
100 for (;;); /* FIXME */
101#endif
102
cc532915
ME
103 /* NOTREACHED */
104}
105
c0ce7d08 106static void crash_kexec_prepare_cpus(int cpu)
cc532915
ME
107{
108 unsigned int msecs;
c0ce7d08 109 unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
07fe0c61
AB
110 int tries = 0;
111 int (*old_handler)(struct pt_regs *regs);
cc532915 112
9b00ac06
AB
113 printk(KERN_EMERG "Sending IPI to other CPUs\n");
114
cc532915
ME
115 crash_send_ipi(crash_ipi_callback);
116 smp_wmb();
117
07fe0c61 118again:
cc532915 119 /*
158d5b5e 120 * FIXME: Until we will have the way to stop other CPUs reliably,
cc532915 121 * the crash CPU will send an IPI and wait for other CPUs to
c0ce7d08 122 * respond.
cc532915 123 */
549e88a1 124 msecs = IPI_TIMEOUT;
2440c01e 125 while ((atomic_read(&cpus_in_crash) < ncpus) && (--msecs > 0))
cc532915 126 mdelay(1);
cc532915
ME
127
128 /* Would it be better to replace the trap vector here? */
129
2440c01e 130 if (atomic_read(&cpus_in_crash) >= ncpus) {
07fe0c61
AB
131 printk(KERN_EMERG "IPI complete\n");
132 return;
c0ce7d08 133 }
9b00ac06 134
07fe0c61 135 printk(KERN_EMERG "ERROR: %d cpu(s) not responding\n",
2440c01e 136 ncpus - atomic_read(&cpus_in_crash));
07fe0c61
AB
137
138 /*
139 * If we have a panic timeout set then we can't wait indefinitely
140 * for someone to activate system reset. We also give up on the
141 * second time through if system reset fail to work.
142 */
143 if ((panic_timeout > 0) || (tries > 0))
144 return;
145
146 /*
147 * A system reset will cause all CPUs to take an 0x100 exception.
148 * The primary CPU returns here via setjmp, and the secondary
149 * CPUs reexecute the crash_kexec_secondary path.
150 */
151 old_handler = __debugger;
152 __debugger = handle_fault;
153 crash_shutdown_cpu = smp_processor_id();
154
155 if (setjmp(crash_shutdown_buf) == 0) {
156 printk(KERN_EMERG "Activate system reset (dumprestart) "
157 "to stop other cpu(s)\n");
158
159 /*
160 * A system reset will force all CPUs to execute the
161 * crash code again. We need to reset cpus_in_crash so we
162 * wait for everyone to do this.
163 */
2440c01e 164 atomic_set(&cpus_in_crash, 0);
07fe0c61
AB
165 smp_mb();
166
2440c01e 167 while (atomic_read(&cpus_in_crash) < ncpus)
07fe0c61
AB
168 cpu_relax();
169 }
170
171 crash_shutdown_cpu = -1;
172 __debugger = old_handler;
173
174 tries++;
175 goto again;
cc532915 176}
c0ce7d08
DW
177
178/*
9b00ac06 179 * This function will be called by secondary cpus.
c0ce7d08
DW
180 */
181void crash_kexec_secondary(struct pt_regs *regs)
182{
c0ce7d08 183 unsigned long flags;
549e88a1 184 int msecs = SECONDARY_TIMEOUT;
c0ce7d08
DW
185
186 local_irq_save(flags);
9b00ac06 187
549e88a1 188 /* Wait for the primary crash CPU to signal its progress */
c0ce7d08
DW
189 while (crashing_cpu < 0) {
190 if (--msecs < 0) {
9b00ac06 191 /* No response, kdump image may not have been loaded */
c0ce7d08
DW
192 local_irq_restore(flags);
193 return;
194 }
9b00ac06 195
c0ce7d08 196 mdelay(1);
c0ce7d08 197 }
9b00ac06 198
c0ce7d08
DW
199 crash_ipi_callback(regs);
200}
201
7c7a81b5 202#else /* ! CONFIG_SMP */
7c7a81b5 203
c0ce7d08 204static void crash_kexec_prepare_cpus(int cpu)
cc532915
ME
205{
206 /*
8c27474a 207 * move the secondaries to us so that we can copy
cc532915
ME
208 * the new kernel 0-0x100 safely
209 *
210 * do this if kexec in setup.c ?
211 */
b6f35b49 212#ifdef CONFIG_PPC64
cc532915 213 smp_release_cpus();
b6f35b49
ME
214#else
215 /* FIXME */
216#endif
cc532915
ME
217}
218
c0ce7d08
DW
219void crash_kexec_secondary(struct pt_regs *regs)
220{
c0ce7d08 221}
7c7a81b5 222#endif /* CONFIG_SMP */
cc532915 223
7707e411
BH
224/* wait for all the CPUs to hit real mode but timeout if they don't come in */
225#if defined(CONFIG_SMP) && defined(CONFIG_PPC_STD_MMU_64)
226static void crash_kexec_wait_realmode(int cpu)
227{
228 unsigned int msecs;
229 int i;
230
549e88a1 231 msecs = REAL_MODE_TIMEOUT;
bd9e5eef 232 for (i=0; i < nr_cpu_ids && msecs > 0; i++) {
7707e411
BH
233 if (i == cpu)
234 continue;
235
236 while (paca[i].kexec_state < KEXEC_STATE_REAL_MODE) {
237 barrier();
63f21a56 238 if (!cpu_possible(i) || !cpu_online(i) || (msecs <= 0))
7707e411 239 break;
7707e411
BH
240 msecs--;
241 mdelay(1);
242 }
243 }
244 mb();
245}
246#else
247static inline void crash_kexec_wait_realmode(int cpu) {}
248#endif /* CONFIG_SMP && CONFIG_PPC_STD_MMU_64 */
249
496b010e
MN
250/*
251 * Register a function to be called on shutdown. Only use this if you
252 * can't reset your device in the second kernel.
253 */
254int crash_shutdown_register(crash_shutdown_t handler)
255{
256 unsigned int i, rc;
257
258 spin_lock(&crash_handlers_lock);
259 for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
260 if (!crash_shutdown_handles[i]) {
261 /* Insert handle at first empty entry */
262 crash_shutdown_handles[i] = handler;
263 rc = 0;
264 break;
265 }
266
267 if (i == CRASH_HANDLER_MAX) {
268 printk(KERN_ERR "Crash shutdown handles full, "
269 "not registered.\n");
270 rc = 1;
271 }
272
273 spin_unlock(&crash_handlers_lock);
274 return rc;
275}
276EXPORT_SYMBOL(crash_shutdown_register);
277
278int crash_shutdown_unregister(crash_shutdown_t handler)
279{
280 unsigned int i, rc;
281
282 spin_lock(&crash_handlers_lock);
283 for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
284 if (crash_shutdown_handles[i] == handler)
285 break;
286
287 if (i == CRASH_HANDLER_MAX) {
288 printk(KERN_ERR "Crash shutdown handle not found\n");
289 rc = 1;
290 } else {
291 /* Shift handles down */
292 for (; crash_shutdown_handles[i]; i++)
293 crash_shutdown_handles[i] =
294 crash_shutdown_handles[i+1];
295 rc = 0;
296 }
297
298 spin_unlock(&crash_handlers_lock);
299 return rc;
300}
301EXPORT_SYMBOL(crash_shutdown_unregister);
302
cc532915
ME
303void default_machine_crash_shutdown(struct pt_regs *regs)
304{
496b010e
MN
305 unsigned int i;
306 int (*old_handler)(struct pt_regs *regs);
307
cc532915
ME
308 /*
309 * This function is only called after the system
f18190bd 310 * has panicked or is otherwise in a critical state.
cc532915
ME
311 * The minimum amount of code to allow a kexec'd kernel
312 * to run successfully needs to happen here.
313 *
314 * In practice this means stopping other cpus in
315 * an SMP system.
316 * The kernel is broken so disable interrupts.
317 */
d04c56f7 318 hard_irq_disable();
cc532915 319
249ec228
AB
320 /*
321 * Make a note of crashing cpu. Will be used in machine_kexec
322 * such that another IPI will not be sent.
323 */
324 crashing_cpu = smp_processor_id();
549e88a1
AB
325
326 /*
327 * If we came in via system reset, wait a while for the secondary
328 * CPUs to enter.
329 */
330 if (TRAP(regs) == 0x100)
331 mdelay(PRIMARY_TIMEOUT);
332
249ec228 333 crash_kexec_prepare_cpus(crashing_cpu);
2440c01e
AB
334
335 crash_save_cpu(regs, crashing_cpu);
336
337 time_to_dump = 1;
338
249ec228 339 crash_kexec_wait_realmode(crashing_cpu);
249ec228 340
c71635d2 341 machine_kexec_mask_interrupts();
496b010e
MN
342
343 /*
8c27474a 344 * Call registered shutdown routines safely. Swap out
496b010e
MN
345 * __debugger_fault_handler, and replace on exit.
346 */
347 old_handler = __debugger_fault_handler;
348 __debugger_fault_handler = handle_fault;
06440794 349 crash_shutdown_cpu = smp_processor_id();
496b010e
MN
350 for (i = 0; crash_shutdown_handles[i]; i++) {
351 if (setjmp(crash_shutdown_buf) == 0) {
352 /*
353 * Insert syncs and delay to ensure
354 * instructions in the dangerous region don't
355 * leak away from this protected region.
356 */
357 asm volatile("sync; isync");
358 /* dangerous region */
359 crash_shutdown_handles[i]();
360 asm volatile("sync; isync");
361 }
d6c1a908 362 }
06440794 363 crash_shutdown_cpu = -1;
496b010e 364 __debugger_fault_handler = old_handler;
d6c1a908 365
c0ce7d08
DW
366 if (ppc_md.kexec_cpu_down)
367 ppc_md.kexec_cpu_down(1, 0);
cc532915 368}