]>
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 | ||
13 | #undef DEBUG | |
14 | ||
15 | #include <linux/kernel.h> | |
16 | #include <linux/smp.h> | |
17 | #include <linux/reboot.h> | |
18 | #include <linux/kexec.h> | |
19 | #include <linux/bootmem.h> | |
66b15db6 | 20 | #include <linux/export.h> |
cc532915 | 21 | #include <linux/crash_dump.h> |
cc532915 ME |
22 | #include <linux/delay.h> |
23 | #include <linux/elf.h> | |
24 | #include <linux/elfcore.h> | |
25 | #include <linux/init.h> | |
d6c1a908 | 26 | #include <linux/irq.h> |
cc532915 | 27 | #include <linux/types.h> |
95f72d1e | 28 | #include <linux/memblock.h> |
cc532915 ME |
29 | |
30 | #include <asm/processor.h> | |
31 | #include <asm/machdep.h> | |
c0ce7d08 | 32 | #include <asm/kexec.h> |
cc532915 | 33 | #include <asm/kdump.h> |
d9b2b2a2 | 34 | #include <asm/prom.h> |
cc532915 | 35 | #include <asm/firmware.h> |
f6cc82fc | 36 | #include <asm/smp.h> |
496b010e MN |
37 | #include <asm/system.h> |
38 | #include <asm/setjmp.h> | |
cc532915 ME |
39 | |
40 | #ifdef DEBUG | |
41 | #include <asm/udbg.h> | |
42 | #define DBG(fmt...) udbg_printf(fmt) | |
43 | #else | |
44 | #define DBG(fmt...) | |
45 | #endif | |
46 | ||
47 | /* This keeps a track of which one is crashing cpu. */ | |
48 | int crashing_cpu = -1; | |
c0ce7d08 | 49 | static cpumask_t cpus_in_crash = CPU_MASK_NONE; |
b6f35b49 | 50 | cpumask_t cpus_in_sr = CPU_MASK_NONE; |
cc532915 | 51 | |
158d5b5e | 52 | #define CRASH_HANDLER_MAX 3 |
496b010e MN |
53 | /* NULL terminated list of shutdown handles */ |
54 | static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX+1]; | |
55 | static DEFINE_SPINLOCK(crash_handlers_lock); | |
56 | ||
cc532915 | 57 | #ifdef CONFIG_SMP |
c0ce7d08 | 58 | static atomic_t enter_on_soft_reset = ATOMIC_INIT(0); |
cc532915 ME |
59 | |
60 | void crash_ipi_callback(struct pt_regs *regs) | |
61 | { | |
62 | int cpu = smp_processor_id(); | |
63 | ||
cc532915 ME |
64 | if (!cpu_online(cpu)) |
65 | return; | |
66 | ||
d04c56f7 | 67 | hard_irq_disable(); |
104699c0 | 68 | if (!cpumask_test_cpu(cpu, &cpus_in_crash)) |
85916f81 | 69 | crash_save_cpu(regs, cpu); |
104699c0 | 70 | cpumask_set_cpu(cpu, &cpus_in_crash); |
cc532915 | 71 | |
c0ce7d08 DW |
72 | /* |
73 | * Entered via soft-reset - could be the kdump | |
74 | * process is invoked using soft-reset or user activated | |
75 | * it if some CPU did not respond to an IPI. | |
76 | * For soft-reset, the secondary CPU can enter this func | |
77 | * twice. 1 - using IPI, and 2. soft-reset. | |
78 | * Tell the kexec CPU that entered via soft-reset and ready | |
79 | * to go down. | |
80 | */ | |
104699c0 KM |
81 | if (cpumask_test_cpu(cpu, &cpus_in_sr)) { |
82 | cpumask_clear_cpu(cpu, &cpus_in_sr); | |
c0ce7d08 DW |
83 | atomic_inc(&enter_on_soft_reset); |
84 | } | |
85 | ||
86 | /* | |
87 | * Starting the kdump boot. | |
88 | * This barrier is needed to make sure that all CPUs are stopped. | |
89 | * If not, soft-reset will be invoked to bring other CPUs. | |
90 | */ | |
104699c0 | 91 | while (!cpumask_test_cpu(crashing_cpu, &cpus_in_crash)) |
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 DW |
106 | /* |
107 | * Wait until all CPUs are entered via soft-reset. | |
108 | */ | |
109 | static void crash_soft_reset_check(int cpu) | |
110 | { | |
111 | unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */ | |
112 | ||
104699c0 | 113 | cpumask_clear_cpu(cpu, &cpus_in_sr); |
c0ce7d08 DW |
114 | while (atomic_read(&enter_on_soft_reset) != ncpus) |
115 | cpu_relax(); | |
116 | } | |
117 | ||
118 | ||
119 | static void crash_kexec_prepare_cpus(int cpu) | |
cc532915 ME |
120 | { |
121 | unsigned int msecs; | |
122 | ||
c0ce7d08 | 123 | unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */ |
cc532915 ME |
124 | |
125 | crash_send_ipi(crash_ipi_callback); | |
126 | smp_wmb(); | |
127 | ||
128 | /* | |
158d5b5e | 129 | * FIXME: Until we will have the way to stop other CPUs reliably, |
cc532915 | 130 | * the crash CPU will send an IPI and wait for other CPUs to |
c0ce7d08 | 131 | * respond. |
01aaed9d | 132 | * Delay of at least 10 seconds. |
cc532915 | 133 | */ |
c0ce7d08 | 134 | printk(KERN_EMERG "Sending IPI to other cpus...\n"); |
01aaed9d | 135 | msecs = 10000; |
104699c0 | 136 | while ((cpumask_weight(&cpus_in_crash) < ncpus) && (--msecs > 0)) { |
c0ce7d08 | 137 | cpu_relax(); |
cc532915 ME |
138 | mdelay(1); |
139 | } | |
140 | ||
141 | /* Would it be better to replace the trap vector here? */ | |
142 | ||
143 | /* | |
144 | * FIXME: In case if we do not get all CPUs, one possibility: ask the | |
145 | * user to do soft reset such that we get all. | |
c0ce7d08 DW |
146 | * Soft-reset will be used until better mechanism is implemented. |
147 | */ | |
104699c0 | 148 | if (cpumask_weight(&cpus_in_crash) < ncpus) { |
c0ce7d08 | 149 | printk(KERN_EMERG "done waiting: %d cpu(s) not responding\n", |
104699c0 | 150 | ncpus - cpumask_weight(&cpus_in_crash)); |
c0ce7d08 | 151 | printk(KERN_EMERG "Activate soft-reset to stop other cpu(s)\n"); |
104699c0 | 152 | cpumask_clear(&cpus_in_sr); |
c0ce7d08 | 153 | atomic_set(&enter_on_soft_reset, 0); |
104699c0 | 154 | while (cpumask_weight(&cpus_in_crash) < ncpus) |
c0ce7d08 DW |
155 | cpu_relax(); |
156 | } | |
157 | /* | |
158 | * Make sure all CPUs are entered via soft-reset if the kdump is | |
159 | * invoked using soft-reset. | |
cc532915 | 160 | */ |
104699c0 | 161 | if (cpumask_test_cpu(cpu, &cpus_in_sr)) |
c0ce7d08 | 162 | crash_soft_reset_check(cpu); |
cc532915 ME |
163 | /* Leave the IPI callback set */ |
164 | } | |
c0ce7d08 DW |
165 | |
166 | /* | |
167 | * This function will be called by secondary cpus or by kexec cpu | |
168 | * if soft-reset is activated to stop some CPUs. | |
169 | */ | |
170 | void crash_kexec_secondary(struct pt_regs *regs) | |
171 | { | |
172 | int cpu = smp_processor_id(); | |
173 | unsigned long flags; | |
174 | int msecs = 5; | |
175 | ||
176 | local_irq_save(flags); | |
177 | /* Wait 5ms if the kexec CPU is not entered yet. */ | |
178 | while (crashing_cpu < 0) { | |
179 | if (--msecs < 0) { | |
180 | /* | |
181 | * Either kdump image is not loaded or | |
182 | * kdump process is not started - Probably xmon | |
183 | * exited using 'x'(exit and recover) or | |
184 | * kexec_should_crash() failed for all running tasks. | |
185 | */ | |
104699c0 | 186 | cpumask_clear_cpu(cpu, &cpus_in_sr); |
c0ce7d08 DW |
187 | local_irq_restore(flags); |
188 | return; | |
189 | } | |
190 | mdelay(1); | |
191 | cpu_relax(); | |
192 | } | |
193 | if (cpu == crashing_cpu) { | |
194 | /* | |
195 | * Panic CPU will enter this func only via soft-reset. | |
196 | * Wait until all secondary CPUs entered and | |
197 | * then start kexec boot. | |
198 | */ | |
199 | crash_soft_reset_check(cpu); | |
104699c0 | 200 | cpumask_set_cpu(crashing_cpu, &cpus_in_crash); |
c0ce7d08 DW |
201 | if (ppc_md.kexec_cpu_down) |
202 | ppc_md.kexec_cpu_down(1, 0); | |
203 | machine_kexec(kexec_crash_image); | |
204 | /* NOTREACHED */ | |
205 | } | |
206 | crash_ipi_callback(regs); | |
207 | } | |
208 | ||
7c7a81b5 | 209 | #else /* ! CONFIG_SMP */ |
7c7a81b5 | 210 | |
c0ce7d08 | 211 | static void crash_kexec_prepare_cpus(int cpu) |
cc532915 ME |
212 | { |
213 | /* | |
214 | * move the secondarys to us so that we can copy | |
215 | * the new kernel 0-0x100 safely | |
216 | * | |
217 | * do this if kexec in setup.c ? | |
218 | */ | |
b6f35b49 | 219 | #ifdef CONFIG_PPC64 |
cc532915 | 220 | smp_release_cpus(); |
b6f35b49 ME |
221 | #else |
222 | /* FIXME */ | |
223 | #endif | |
cc532915 ME |
224 | } |
225 | ||
c0ce7d08 DW |
226 | void crash_kexec_secondary(struct pt_regs *regs) |
227 | { | |
104699c0 | 228 | cpumask_clear(&cpus_in_sr); |
c0ce7d08 | 229 | } |
7c7a81b5 | 230 | #endif /* CONFIG_SMP */ |
cc532915 | 231 | |
7707e411 BH |
232 | /* wait for all the CPUs to hit real mode but timeout if they don't come in */ |
233 | #if defined(CONFIG_SMP) && defined(CONFIG_PPC_STD_MMU_64) | |
234 | static void crash_kexec_wait_realmode(int cpu) | |
235 | { | |
236 | unsigned int msecs; | |
237 | int i; | |
238 | ||
239 | msecs = 10000; | |
bd9e5eef | 240 | for (i=0; i < nr_cpu_ids && msecs > 0; i++) { |
7707e411 BH |
241 | if (i == cpu) |
242 | continue; | |
243 | ||
244 | while (paca[i].kexec_state < KEXEC_STATE_REAL_MODE) { | |
245 | barrier(); | |
63f21a56 | 246 | if (!cpu_possible(i) || !cpu_online(i) || (msecs <= 0)) |
7707e411 | 247 | break; |
7707e411 BH |
248 | msecs--; |
249 | mdelay(1); | |
250 | } | |
251 | } | |
252 | mb(); | |
253 | } | |
254 | #else | |
255 | static inline void crash_kexec_wait_realmode(int cpu) {} | |
256 | #endif /* CONFIG_SMP && CONFIG_PPC_STD_MMU_64 */ | |
257 | ||
496b010e MN |
258 | /* |
259 | * Register a function to be called on shutdown. Only use this if you | |
260 | * can't reset your device in the second kernel. | |
261 | */ | |
262 | int crash_shutdown_register(crash_shutdown_t handler) | |
263 | { | |
264 | unsigned int i, rc; | |
265 | ||
266 | spin_lock(&crash_handlers_lock); | |
267 | for (i = 0 ; i < CRASH_HANDLER_MAX; i++) | |
268 | if (!crash_shutdown_handles[i]) { | |
269 | /* Insert handle at first empty entry */ | |
270 | crash_shutdown_handles[i] = handler; | |
271 | rc = 0; | |
272 | break; | |
273 | } | |
274 | ||
275 | if (i == CRASH_HANDLER_MAX) { | |
276 | printk(KERN_ERR "Crash shutdown handles full, " | |
277 | "not registered.\n"); | |
278 | rc = 1; | |
279 | } | |
280 | ||
281 | spin_unlock(&crash_handlers_lock); | |
282 | return rc; | |
283 | } | |
284 | EXPORT_SYMBOL(crash_shutdown_register); | |
285 | ||
286 | int crash_shutdown_unregister(crash_shutdown_t handler) | |
287 | { | |
288 | unsigned int i, rc; | |
289 | ||
290 | spin_lock(&crash_handlers_lock); | |
291 | for (i = 0 ; i < CRASH_HANDLER_MAX; i++) | |
292 | if (crash_shutdown_handles[i] == handler) | |
293 | break; | |
294 | ||
295 | if (i == CRASH_HANDLER_MAX) { | |
296 | printk(KERN_ERR "Crash shutdown handle not found\n"); | |
297 | rc = 1; | |
298 | } else { | |
299 | /* Shift handles down */ | |
300 | for (; crash_shutdown_handles[i]; i++) | |
301 | crash_shutdown_handles[i] = | |
302 | crash_shutdown_handles[i+1]; | |
303 | rc = 0; | |
304 | } | |
305 | ||
306 | spin_unlock(&crash_handlers_lock); | |
307 | return rc; | |
308 | } | |
309 | EXPORT_SYMBOL(crash_shutdown_unregister); | |
310 | ||
311 | static unsigned long crash_shutdown_buf[JMP_BUF_LEN]; | |
06440794 | 312 | static int crash_shutdown_cpu = -1; |
496b010e MN |
313 | |
314 | static int handle_fault(struct pt_regs *regs) | |
315 | { | |
06440794 AB |
316 | if (crash_shutdown_cpu == smp_processor_id()) |
317 | longjmp(crash_shutdown_buf, 1); | |
496b010e MN |
318 | return 0; |
319 | } | |
320 | ||
cc532915 ME |
321 | void default_machine_crash_shutdown(struct pt_regs *regs) |
322 | { | |
496b010e MN |
323 | unsigned int i; |
324 | int (*old_handler)(struct pt_regs *regs); | |
325 | ||
d6c1a908 | 326 | |
cc532915 ME |
327 | /* |
328 | * This function is only called after the system | |
f18190bd | 329 | * has panicked or is otherwise in a critical state. |
cc532915 ME |
330 | * The minimum amount of code to allow a kexec'd kernel |
331 | * to run successfully needs to happen here. | |
332 | * | |
333 | * In practice this means stopping other cpus in | |
334 | * an SMP system. | |
335 | * The kernel is broken so disable interrupts. | |
336 | */ | |
d04c56f7 | 337 | hard_irq_disable(); |
cc532915 | 338 | |
249ec228 AB |
339 | /* |
340 | * Make a note of crashing cpu. Will be used in machine_kexec | |
341 | * such that another IPI will not be sent. | |
342 | */ | |
343 | crashing_cpu = smp_processor_id(); | |
344 | crash_save_cpu(regs, crashing_cpu); | |
345 | crash_kexec_prepare_cpus(crashing_cpu); | |
104699c0 | 346 | cpumask_set_cpu(crashing_cpu, &cpus_in_crash); |
249ec228 | 347 | crash_kexec_wait_realmode(crashing_cpu); |
249ec228 | 348 | |
c71635d2 | 349 | machine_kexec_mask_interrupts(); |
496b010e MN |
350 | |
351 | /* | |
352 | * Call registered shutdown routines savely. Swap out | |
353 | * __debugger_fault_handler, and replace on exit. | |
354 | */ | |
355 | old_handler = __debugger_fault_handler; | |
356 | __debugger_fault_handler = handle_fault; | |
06440794 | 357 | crash_shutdown_cpu = smp_processor_id(); |
496b010e MN |
358 | for (i = 0; crash_shutdown_handles[i]; i++) { |
359 | if (setjmp(crash_shutdown_buf) == 0) { | |
360 | /* | |
361 | * Insert syncs and delay to ensure | |
362 | * instructions in the dangerous region don't | |
363 | * leak away from this protected region. | |
364 | */ | |
365 | asm volatile("sync; isync"); | |
366 | /* dangerous region */ | |
367 | crash_shutdown_handles[i](); | |
368 | asm volatile("sync; isync"); | |
369 | } | |
d6c1a908 | 370 | } |
06440794 | 371 | crash_shutdown_cpu = -1; |
496b010e | 372 | __debugger_fault_handler = old_handler; |
d6c1a908 | 373 | |
c0ce7d08 DW |
374 | if (ppc_md.kexec_cpu_down) |
375 | ppc_md.kexec_cpu_down(1, 0); | |
cc532915 | 376 | } |