]>
Commit | Line | Data |
---|---|---|
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> |
cc532915 | 26 | #include <asm/kdump.h> |
d9b2b2a2 | 27 | #include <asm/prom.h> |
f6cc82fc | 28 | #include <asm/smp.h> |
496b010e | 29 | #include <asm/setjmp.h> |
ae3a197e | 30 | #include <asm/debug.h> |
cc532915 | 31 | |
549e88a1 AB |
32 | /* |
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 | ||
2440c01e | 46 | static int time_to_dump; |
cc532915 | 47 | |
158d5b5e | 48 | #define CRASH_HANDLER_MAX 3 |
1d145165 SJS |
49 | /* List of shutdown handles */ |
50 | static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX]; | |
496b010e MN |
51 | static DEFINE_SPINLOCK(crash_handlers_lock); |
52 | ||
07fe0c61 AB |
53 | static unsigned long crash_shutdown_buf[JMP_BUF_LEN]; |
54 | static int crash_shutdown_cpu = -1; | |
55 | ||
56 | static 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 | ||
cc532915 | 63 | #ifdef CONFIG_SMP |
cc532915 | 64 | |
897e01a0 | 65 | static atomic_t cpus_in_crash; |
34852ed5 | 66 | static void crash_ipi_callback(struct pt_regs *regs) |
cc532915 | 67 | { |
2440c01e AB |
68 | static cpumask_t cpus_state_saved = CPU_MASK_NONE; |
69 | ||
cc532915 ME |
70 | int cpu = smp_processor_id(); |
71 | ||
cc532915 ME |
72 | if (!cpu_online(cpu)) |
73 | return; | |
74 | ||
d04c56f7 | 75 | hard_irq_disable(); |
2440c01e | 76 | if (!cpumask_test_cpu(cpu, &cpus_state_saved)) { |
85916f81 | 77 | crash_save_cpu(regs, cpu); |
2440c01e AB |
78 | cpumask_set_cpu(cpu, &cpus_state_saved); |
79 | } | |
80 | ||
81 | atomic_inc(&cpus_in_crash); | |
c645073f | 82 | smp_mb__after_atomic(); |
cc532915 | 83 | |
c0ce7d08 DW |
84 | /* |
85 | * Starting the kdump boot. | |
86 | * This barrier is needed to make sure that all CPUs are stopped. | |
c0ce7d08 | 87 | */ |
2440c01e | 88 | while (!time_to_dump) |
c0ce7d08 DW |
89 | cpu_relax(); |
90 | ||
91 | if (ppc_md.kexec_cpu_down) | |
92 | ppc_md.kexec_cpu_down(1, 1); | |
b6f35b49 ME |
93 | |
94 | #ifdef CONFIG_PPC64 | |
cc532915 | 95 | kexec_smp_wait(); |
b6f35b49 ME |
96 | #else |
97 | for (;;); /* FIXME */ | |
98 | #endif | |
99 | ||
cc532915 ME |
100 | /* NOTREACHED */ |
101 | } | |
102 | ||
c0ce7d08 | 103 | static void crash_kexec_prepare_cpus(int cpu) |
cc532915 ME |
104 | { |
105 | unsigned int msecs; | |
c0ce7d08 | 106 | unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */ |
07fe0c61 AB |
107 | int tries = 0; |
108 | int (*old_handler)(struct pt_regs *regs); | |
cc532915 | 109 | |
9b00ac06 AB |
110 | printk(KERN_EMERG "Sending IPI to other CPUs\n"); |
111 | ||
cc532915 ME |
112 | crash_send_ipi(crash_ipi_callback); |
113 | smp_wmb(); | |
114 | ||
07fe0c61 | 115 | again: |
cc532915 | 116 | /* |
158d5b5e | 117 | * FIXME: Until we will have the way to stop other CPUs reliably, |
cc532915 | 118 | * the crash CPU will send an IPI and wait for other CPUs to |
c0ce7d08 | 119 | * respond. |
cc532915 | 120 | */ |
549e88a1 | 121 | msecs = IPI_TIMEOUT; |
2440c01e | 122 | while ((atomic_read(&cpus_in_crash) < ncpus) && (--msecs > 0)) |
cc532915 | 123 | mdelay(1); |
cc532915 ME |
124 | |
125 | /* Would it be better to replace the trap vector here? */ | |
126 | ||
2440c01e | 127 | if (atomic_read(&cpus_in_crash) >= ncpus) { |
07fe0c61 AB |
128 | printk(KERN_EMERG "IPI complete\n"); |
129 | return; | |
c0ce7d08 | 130 | } |
9b00ac06 | 131 | |
07fe0c61 | 132 | printk(KERN_EMERG "ERROR: %d cpu(s) not responding\n", |
2440c01e | 133 | ncpus - atomic_read(&cpus_in_crash)); |
07fe0c61 AB |
134 | |
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 | */ | |
2440c01e | 161 | atomic_set(&cpus_in_crash, 0); |
07fe0c61 AB |
162 | smp_mb(); |
163 | ||
2440c01e | 164 | while (atomic_read(&cpus_in_crash) < ncpus) |
07fe0c61 AB |
165 | cpu_relax(); |
166 | } | |
167 | ||
168 | crash_shutdown_cpu = -1; | |
169 | __debugger = old_handler; | |
170 | ||
171 | tries++; | |
172 | goto again; | |
cc532915 | 173 | } |
c0ce7d08 DW |
174 | |
175 | /* | |
9b00ac06 | 176 | * This function will be called by secondary cpus. |
c0ce7d08 DW |
177 | */ |
178 | void crash_kexec_secondary(struct pt_regs *regs) | |
179 | { | |
c0ce7d08 | 180 | unsigned long flags; |
549e88a1 | 181 | int msecs = SECONDARY_TIMEOUT; |
c0ce7d08 DW |
182 | |
183 | local_irq_save(flags); | |
9b00ac06 | 184 | |
549e88a1 | 185 | /* Wait for the primary crash CPU to signal its progress */ |
c0ce7d08 DW |
186 | while (crashing_cpu < 0) { |
187 | if (--msecs < 0) { | |
9b00ac06 | 188 | /* No response, kdump image may not have been loaded */ |
c0ce7d08 DW |
189 | local_irq_restore(flags); |
190 | return; | |
191 | } | |
9b00ac06 | 192 | |
c0ce7d08 | 193 | mdelay(1); |
c0ce7d08 | 194 | } |
9b00ac06 | 195 | |
c0ce7d08 DW |
196 | crash_ipi_callback(regs); |
197 | } | |
198 | ||
7c7a81b5 | 199 | #else /* ! CONFIG_SMP */ |
7c7a81b5 | 200 | |
c0ce7d08 | 201 | static void crash_kexec_prepare_cpus(int cpu) |
cc532915 ME |
202 | { |
203 | /* | |
8c27474a | 204 | * move the secondaries to us so that we can copy |
cc532915 ME |
205 | * the new kernel 0-0x100 safely |
206 | * | |
207 | * do this if kexec in setup.c ? | |
208 | */ | |
b6f35b49 | 209 | #ifdef CONFIG_PPC64 |
cc532915 | 210 | smp_release_cpus(); |
b6f35b49 ME |
211 | #else |
212 | /* FIXME */ | |
213 | #endif | |
cc532915 ME |
214 | } |
215 | ||
c0ce7d08 DW |
216 | void crash_kexec_secondary(struct pt_regs *regs) |
217 | { | |
c0ce7d08 | 218 | } |
7c7a81b5 | 219 | #endif /* CONFIG_SMP */ |
cc532915 | 220 | |
7707e411 | 221 | /* wait for all the CPUs to hit real mode but timeout if they don't come in */ |
eeaab663 SW |
222 | #if defined(CONFIG_SMP) && defined(CONFIG_PPC64) |
223 | static void __maybe_unused crash_kexec_wait_realmode(int cpu) | |
7707e411 BH |
224 | { |
225 | unsigned int msecs; | |
226 | int i; | |
227 | ||
549e88a1 | 228 | msecs = REAL_MODE_TIMEOUT; |
bd9e5eef | 229 | for (i=0; i < nr_cpu_ids && msecs > 0; i++) { |
7707e411 BH |
230 | if (i == cpu) |
231 | continue; | |
232 | ||
233 | while (paca[i].kexec_state < KEXEC_STATE_REAL_MODE) { | |
234 | barrier(); | |
63f21a56 | 235 | if (!cpu_possible(i) || !cpu_online(i) || (msecs <= 0)) |
7707e411 | 236 | break; |
7707e411 BH |
237 | msecs--; |
238 | mdelay(1); | |
239 | } | |
240 | } | |
241 | mb(); | |
242 | } | |
243 | #else | |
244 | static inline void crash_kexec_wait_realmode(int cpu) {} | |
eeaab663 | 245 | #endif /* CONFIG_SMP && CONFIG_PPC64 */ |
7707e411 | 246 | |
496b010e MN |
247 | /* |
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 | */ | |
251 | int 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 | } | |
273 | EXPORT_SYMBOL(crash_shutdown_register); | |
274 | ||
275 | int 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 */ | |
1d145165 | 289 | for (; i < (CRASH_HANDLER_MAX - 1); i++) |
496b010e MN |
290 | crash_shutdown_handles[i] = |
291 | crash_shutdown_handles[i+1]; | |
1d145165 SJS |
292 | /* |
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; | |
496b010e MN |
297 | rc = 0; |
298 | } | |
299 | ||
300 | spin_unlock(&crash_handlers_lock); | |
301 | return rc; | |
302 | } | |
303 | EXPORT_SYMBOL(crash_shutdown_unregister); | |
304 | ||
cc532915 ME |
305 | void default_machine_crash_shutdown(struct pt_regs *regs) |
306 | { | |
496b010e MN |
307 | unsigned int i; |
308 | int (*old_handler)(struct pt_regs *regs); | |
309 | ||
cc532915 ME |
310 | /* |
311 | * This function is only called after the system | |
f18190bd | 312 | * has panicked or is otherwise in a critical state. |
cc532915 ME |
313 | * 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 | */ | |
d04c56f7 | 320 | hard_irq_disable(); |
cc532915 | 321 | |
249ec228 AB |
322 | /* |
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(); | |
549e88a1 AB |
327 | |
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 | ||
249ec228 | 335 | crash_kexec_prepare_cpus(crashing_cpu); |
2440c01e AB |
336 | |
337 | crash_save_cpu(regs, crashing_cpu); | |
338 | ||
339 | time_to_dump = 1; | |
340 | ||
249ec228 | 341 | crash_kexec_wait_realmode(crashing_cpu); |
249ec228 | 342 | |
c71635d2 | 343 | machine_kexec_mask_interrupts(); |
496b010e MN |
344 | |
345 | /* | |
8c27474a | 346 | * Call registered shutdown routines safely. Swap out |
496b010e MN |
347 | * __debugger_fault_handler, and replace on exit. |
348 | */ | |
349 | old_handler = __debugger_fault_handler; | |
350 | __debugger_fault_handler = handle_fault; | |
06440794 | 351 | crash_shutdown_cpu = smp_processor_id(); |
a7d63928 | 352 | for (i = 0; i < CRASH_HANDLER_MAX && crash_shutdown_handles[i]; i++) { |
496b010e MN |
353 | 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 | } | |
d6c1a908 | 364 | } |
06440794 | 365 | crash_shutdown_cpu = -1; |
496b010e | 366 | __debugger_fault_handler = old_handler; |
d6c1a908 | 367 | |
c0ce7d08 DW |
368 | if (ppc_md.kexec_cpu_down) |
369 | ppc_md.kexec_cpu_down(1, 0); | |
cc532915 | 370 | } |