]>
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> | |
20 | #include <linux/crash_dump.h> | |
cc532915 ME |
21 | #include <linux/delay.h> |
22 | #include <linux/elf.h> | |
23 | #include <linux/elfcore.h> | |
24 | #include <linux/init.h> | |
d6c1a908 | 25 | #include <linux/irq.h> |
cc532915 | 26 | #include <linux/types.h> |
95f72d1e | 27 | #include <linux/memblock.h> |
cc532915 ME |
28 | |
29 | #include <asm/processor.h> | |
30 | #include <asm/machdep.h> | |
c0ce7d08 | 31 | #include <asm/kexec.h> |
cc532915 | 32 | #include <asm/kdump.h> |
d9b2b2a2 | 33 | #include <asm/prom.h> |
cc532915 | 34 | #include <asm/firmware.h> |
f6cc82fc | 35 | #include <asm/smp.h> |
496b010e MN |
36 | #include <asm/system.h> |
37 | #include <asm/setjmp.h> | |
cc532915 ME |
38 | |
39 | #ifdef DEBUG | |
40 | #include <asm/udbg.h> | |
41 | #define DBG(fmt...) udbg_printf(fmt) | |
42 | #else | |
43 | #define DBG(fmt...) | |
44 | #endif | |
45 | ||
46 | /* This keeps a track of which one is crashing cpu. */ | |
47 | int crashing_cpu = -1; | |
c0ce7d08 | 48 | static cpumask_t cpus_in_crash = CPU_MASK_NONE; |
b6f35b49 | 49 | cpumask_t cpus_in_sr = CPU_MASK_NONE; |
cc532915 | 50 | |
158d5b5e | 51 | #define CRASH_HANDLER_MAX 3 |
496b010e MN |
52 | /* NULL terminated list of shutdown handles */ |
53 | static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX+1]; | |
54 | static DEFINE_SPINLOCK(crash_handlers_lock); | |
55 | ||
cc532915 | 56 | #ifdef CONFIG_SMP |
c0ce7d08 | 57 | static atomic_t enter_on_soft_reset = ATOMIC_INIT(0); |
cc532915 ME |
58 | |
59 | void crash_ipi_callback(struct pt_regs *regs) | |
60 | { | |
61 | int cpu = smp_processor_id(); | |
62 | ||
cc532915 ME |
63 | if (!cpu_online(cpu)) |
64 | return; | |
65 | ||
d04c56f7 | 66 | hard_irq_disable(); |
104699c0 | 67 | if (!cpumask_test_cpu(cpu, &cpus_in_crash)) |
85916f81 | 68 | crash_save_cpu(regs, cpu); |
104699c0 | 69 | cpumask_set_cpu(cpu, &cpus_in_crash); |
cc532915 | 70 | |
c0ce7d08 DW |
71 | /* |
72 | * Entered via soft-reset - could be the kdump | |
73 | * process is invoked using soft-reset or user activated | |
74 | * it if some CPU did not respond to an IPI. | |
75 | * For soft-reset, the secondary CPU can enter this func | |
76 | * twice. 1 - using IPI, and 2. soft-reset. | |
77 | * Tell the kexec CPU that entered via soft-reset and ready | |
78 | * to go down. | |
79 | */ | |
104699c0 KM |
80 | if (cpumask_test_cpu(cpu, &cpus_in_sr)) { |
81 | cpumask_clear_cpu(cpu, &cpus_in_sr); | |
c0ce7d08 DW |
82 | atomic_inc(&enter_on_soft_reset); |
83 | } | |
84 | ||
85 | /* | |
86 | * Starting the kdump boot. | |
87 | * This barrier is needed to make sure that all CPUs are stopped. | |
88 | * If not, soft-reset will be invoked to bring other CPUs. | |
89 | */ | |
104699c0 | 90 | while (!cpumask_test_cpu(crashing_cpu, &cpus_in_crash)) |
c0ce7d08 DW |
91 | cpu_relax(); |
92 | ||
93 | if (ppc_md.kexec_cpu_down) | |
94 | ppc_md.kexec_cpu_down(1, 1); | |
b6f35b49 ME |
95 | |
96 | #ifdef CONFIG_PPC64 | |
cc532915 | 97 | kexec_smp_wait(); |
b6f35b49 ME |
98 | #else |
99 | for (;;); /* FIXME */ | |
100 | #endif | |
101 | ||
cc532915 ME |
102 | /* NOTREACHED */ |
103 | } | |
104 | ||
c0ce7d08 DW |
105 | /* |
106 | * Wait until all CPUs are entered via soft-reset. | |
107 | */ | |
108 | static void crash_soft_reset_check(int cpu) | |
109 | { | |
110 | unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */ | |
111 | ||
104699c0 | 112 | cpumask_clear_cpu(cpu, &cpus_in_sr); |
c0ce7d08 DW |
113 | while (atomic_read(&enter_on_soft_reset) != ncpus) |
114 | cpu_relax(); | |
115 | } | |
116 | ||
117 | ||
118 | static void crash_kexec_prepare_cpus(int cpu) | |
cc532915 ME |
119 | { |
120 | unsigned int msecs; | |
121 | ||
c0ce7d08 | 122 | unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */ |
cc532915 ME |
123 | |
124 | crash_send_ipi(crash_ipi_callback); | |
125 | smp_wmb(); | |
126 | ||
127 | /* | |
158d5b5e | 128 | * FIXME: Until we will have the way to stop other CPUs reliably, |
cc532915 | 129 | * the crash CPU will send an IPI and wait for other CPUs to |
c0ce7d08 | 130 | * respond. |
01aaed9d | 131 | * Delay of at least 10 seconds. |
cc532915 | 132 | */ |
c0ce7d08 | 133 | printk(KERN_EMERG "Sending IPI to other cpus...\n"); |
01aaed9d | 134 | msecs = 10000; |
104699c0 | 135 | while ((cpumask_weight(&cpus_in_crash) < ncpus) && (--msecs > 0)) { |
c0ce7d08 | 136 | cpu_relax(); |
cc532915 ME |
137 | mdelay(1); |
138 | } | |
139 | ||
140 | /* Would it be better to replace the trap vector here? */ | |
141 | ||
142 | /* | |
143 | * FIXME: In case if we do not get all CPUs, one possibility: ask the | |
144 | * user to do soft reset such that we get all. | |
c0ce7d08 DW |
145 | * Soft-reset will be used until better mechanism is implemented. |
146 | */ | |
104699c0 | 147 | if (cpumask_weight(&cpus_in_crash) < ncpus) { |
c0ce7d08 | 148 | printk(KERN_EMERG "done waiting: %d cpu(s) not responding\n", |
104699c0 | 149 | ncpus - cpumask_weight(&cpus_in_crash)); |
c0ce7d08 | 150 | printk(KERN_EMERG "Activate soft-reset to stop other cpu(s)\n"); |
104699c0 | 151 | cpumask_clear(&cpus_in_sr); |
c0ce7d08 | 152 | atomic_set(&enter_on_soft_reset, 0); |
104699c0 | 153 | while (cpumask_weight(&cpus_in_crash) < ncpus) |
c0ce7d08 DW |
154 | cpu_relax(); |
155 | } | |
156 | /* | |
157 | * Make sure all CPUs are entered via soft-reset if the kdump is | |
158 | * invoked using soft-reset. | |
cc532915 | 159 | */ |
104699c0 | 160 | if (cpumask_test_cpu(cpu, &cpus_in_sr)) |
c0ce7d08 | 161 | crash_soft_reset_check(cpu); |
cc532915 ME |
162 | /* Leave the IPI callback set */ |
163 | } | |
c0ce7d08 DW |
164 | |
165 | /* | |
166 | * This function will be called by secondary cpus or by kexec cpu | |
167 | * if soft-reset is activated to stop some CPUs. | |
168 | */ | |
169 | void crash_kexec_secondary(struct pt_regs *regs) | |
170 | { | |
171 | int cpu = smp_processor_id(); | |
172 | unsigned long flags; | |
173 | int msecs = 5; | |
174 | ||
175 | local_irq_save(flags); | |
176 | /* Wait 5ms if the kexec CPU is not entered yet. */ | |
177 | while (crashing_cpu < 0) { | |
178 | if (--msecs < 0) { | |
179 | /* | |
180 | * Either kdump image is not loaded or | |
181 | * kdump process is not started - Probably xmon | |
182 | * exited using 'x'(exit and recover) or | |
183 | * kexec_should_crash() failed for all running tasks. | |
184 | */ | |
104699c0 | 185 | cpumask_clear_cpu(cpu, &cpus_in_sr); |
c0ce7d08 DW |
186 | local_irq_restore(flags); |
187 | return; | |
188 | } | |
189 | mdelay(1); | |
190 | cpu_relax(); | |
191 | } | |
192 | if (cpu == crashing_cpu) { | |
193 | /* | |
194 | * Panic CPU will enter this func only via soft-reset. | |
195 | * Wait until all secondary CPUs entered and | |
196 | * then start kexec boot. | |
197 | */ | |
198 | crash_soft_reset_check(cpu); | |
104699c0 | 199 | cpumask_set_cpu(crashing_cpu, &cpus_in_crash); |
c0ce7d08 DW |
200 | if (ppc_md.kexec_cpu_down) |
201 | ppc_md.kexec_cpu_down(1, 0); | |
202 | machine_kexec(kexec_crash_image); | |
203 | /* NOTREACHED */ | |
204 | } | |
205 | crash_ipi_callback(regs); | |
206 | } | |
207 | ||
7c7a81b5 | 208 | #else /* ! CONFIG_SMP */ |
7c7a81b5 | 209 | |
c0ce7d08 | 210 | static void crash_kexec_prepare_cpus(int cpu) |
cc532915 ME |
211 | { |
212 | /* | |
213 | * move the secondarys to us so that we can copy | |
214 | * the new kernel 0-0x100 safely | |
215 | * | |
216 | * do this if kexec in setup.c ? | |
217 | */ | |
b6f35b49 | 218 | #ifdef CONFIG_PPC64 |
cc532915 | 219 | smp_release_cpus(); |
b6f35b49 ME |
220 | #else |
221 | /* FIXME */ | |
222 | #endif | |
cc532915 ME |
223 | } |
224 | ||
c0ce7d08 DW |
225 | void crash_kexec_secondary(struct pt_regs *regs) |
226 | { | |
104699c0 | 227 | cpumask_clear(&cpus_in_sr); |
c0ce7d08 | 228 | } |
7c7a81b5 | 229 | #endif /* CONFIG_SMP */ |
cc532915 | 230 | |
7707e411 BH |
231 | /* wait for all the CPUs to hit real mode but timeout if they don't come in */ |
232 | #if defined(CONFIG_SMP) && defined(CONFIG_PPC_STD_MMU_64) | |
233 | static void crash_kexec_wait_realmode(int cpu) | |
234 | { | |
235 | unsigned int msecs; | |
236 | int i; | |
237 | ||
238 | msecs = 10000; | |
bd9e5eef | 239 | for (i=0; i < nr_cpu_ids && msecs > 0; i++) { |
7707e411 BH |
240 | if (i == cpu) |
241 | continue; | |
242 | ||
243 | while (paca[i].kexec_state < KEXEC_STATE_REAL_MODE) { | |
244 | barrier(); | |
63f21a56 | 245 | if (!cpu_possible(i) || !cpu_online(i) || (msecs <= 0)) |
7707e411 | 246 | break; |
7707e411 BH |
247 | msecs--; |
248 | mdelay(1); | |
249 | } | |
250 | } | |
251 | mb(); | |
252 | } | |
253 | #else | |
254 | static inline void crash_kexec_wait_realmode(int cpu) {} | |
255 | #endif /* CONFIG_SMP && CONFIG_PPC_STD_MMU_64 */ | |
256 | ||
496b010e MN |
257 | /* |
258 | * Register a function to be called on shutdown. Only use this if you | |
259 | * can't reset your device in the second kernel. | |
260 | */ | |
261 | int crash_shutdown_register(crash_shutdown_t handler) | |
262 | { | |
263 | unsigned int i, rc; | |
264 | ||
265 | spin_lock(&crash_handlers_lock); | |
266 | for (i = 0 ; i < CRASH_HANDLER_MAX; i++) | |
267 | if (!crash_shutdown_handles[i]) { | |
268 | /* Insert handle at first empty entry */ | |
269 | crash_shutdown_handles[i] = handler; | |
270 | rc = 0; | |
271 | break; | |
272 | } | |
273 | ||
274 | if (i == CRASH_HANDLER_MAX) { | |
275 | printk(KERN_ERR "Crash shutdown handles full, " | |
276 | "not registered.\n"); | |
277 | rc = 1; | |
278 | } | |
279 | ||
280 | spin_unlock(&crash_handlers_lock); | |
281 | return rc; | |
282 | } | |
283 | EXPORT_SYMBOL(crash_shutdown_register); | |
284 | ||
285 | int crash_shutdown_unregister(crash_shutdown_t handler) | |
286 | { | |
287 | unsigned int i, rc; | |
288 | ||
289 | spin_lock(&crash_handlers_lock); | |
290 | for (i = 0 ; i < CRASH_HANDLER_MAX; i++) | |
291 | if (crash_shutdown_handles[i] == handler) | |
292 | break; | |
293 | ||
294 | if (i == CRASH_HANDLER_MAX) { | |
295 | printk(KERN_ERR "Crash shutdown handle not found\n"); | |
296 | rc = 1; | |
297 | } else { | |
298 | /* Shift handles down */ | |
299 | for (; crash_shutdown_handles[i]; i++) | |
300 | crash_shutdown_handles[i] = | |
301 | crash_shutdown_handles[i+1]; | |
302 | rc = 0; | |
303 | } | |
304 | ||
305 | spin_unlock(&crash_handlers_lock); | |
306 | return rc; | |
307 | } | |
308 | EXPORT_SYMBOL(crash_shutdown_unregister); | |
309 | ||
310 | static unsigned long crash_shutdown_buf[JMP_BUF_LEN]; | |
06440794 | 311 | static int crash_shutdown_cpu = -1; |
496b010e MN |
312 | |
313 | static int handle_fault(struct pt_regs *regs) | |
314 | { | |
06440794 AB |
315 | if (crash_shutdown_cpu == smp_processor_id()) |
316 | longjmp(crash_shutdown_buf, 1); | |
496b010e MN |
317 | return 0; |
318 | } | |
319 | ||
cc532915 ME |
320 | void default_machine_crash_shutdown(struct pt_regs *regs) |
321 | { | |
496b010e MN |
322 | unsigned int i; |
323 | int (*old_handler)(struct pt_regs *regs); | |
324 | ||
d6c1a908 | 325 | |
cc532915 ME |
326 | /* |
327 | * This function is only called after the system | |
f18190bd | 328 | * has panicked or is otherwise in a critical state. |
cc532915 ME |
329 | * The minimum amount of code to allow a kexec'd kernel |
330 | * to run successfully needs to happen here. | |
331 | * | |
332 | * In practice this means stopping other cpus in | |
333 | * an SMP system. | |
334 | * The kernel is broken so disable interrupts. | |
335 | */ | |
d04c56f7 | 336 | hard_irq_disable(); |
cc532915 | 337 | |
249ec228 AB |
338 | /* |
339 | * Make a note of crashing cpu. Will be used in machine_kexec | |
340 | * such that another IPI will not be sent. | |
341 | */ | |
342 | crashing_cpu = smp_processor_id(); | |
343 | crash_save_cpu(regs, crashing_cpu); | |
344 | crash_kexec_prepare_cpus(crashing_cpu); | |
104699c0 | 345 | cpumask_set_cpu(crashing_cpu, &cpus_in_crash); |
249ec228 | 346 | crash_kexec_wait_realmode(crashing_cpu); |
249ec228 | 347 | |
c71635d2 | 348 | machine_kexec_mask_interrupts(); |
496b010e MN |
349 | |
350 | /* | |
351 | * Call registered shutdown routines savely. Swap out | |
352 | * __debugger_fault_handler, and replace on exit. | |
353 | */ | |
354 | old_handler = __debugger_fault_handler; | |
355 | __debugger_fault_handler = handle_fault; | |
06440794 | 356 | crash_shutdown_cpu = smp_processor_id(); |
496b010e MN |
357 | for (i = 0; crash_shutdown_handles[i]; i++) { |
358 | if (setjmp(crash_shutdown_buf) == 0) { | |
359 | /* | |
360 | * Insert syncs and delay to ensure | |
361 | * instructions in the dangerous region don't | |
362 | * leak away from this protected region. | |
363 | */ | |
364 | asm volatile("sync; isync"); | |
365 | /* dangerous region */ | |
366 | crash_shutdown_handles[i](); | |
367 | asm volatile("sync; isync"); | |
368 | } | |
d6c1a908 | 369 | } |
06440794 | 370 | crash_shutdown_cpu = -1; |
496b010e | 371 | __debugger_fault_handler = old_handler; |
d6c1a908 | 372 | |
c0ce7d08 DW |
373 | if (ppc_md.kexec_cpu_down) |
374 | ppc_md.kexec_cpu_down(1, 0); | |
cc532915 | 375 | } |