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 ME |
26 | #include <linux/types.h> |
27 | ||
28 | #include <asm/processor.h> | |
29 | #include <asm/machdep.h> | |
c0ce7d08 | 30 | #include <asm/kexec.h> |
cc532915 ME |
31 | #include <asm/kdump.h> |
32 | #include <asm/lmb.h> | |
33 | #include <asm/firmware.h> | |
f6cc82fc | 34 | #include <asm/smp.h> |
496b010e MN |
35 | #include <asm/system.h> |
36 | #include <asm/setjmp.h> | |
cc532915 ME |
37 | |
38 | #ifdef DEBUG | |
39 | #include <asm/udbg.h> | |
40 | #define DBG(fmt...) udbg_printf(fmt) | |
41 | #else | |
42 | #define DBG(fmt...) | |
43 | #endif | |
44 | ||
45 | /* This keeps a track of which one is crashing cpu. */ | |
46 | int crashing_cpu = -1; | |
c0ce7d08 | 47 | static cpumask_t cpus_in_crash = CPU_MASK_NONE; |
b6f35b49 | 48 | cpumask_t cpus_in_sr = CPU_MASK_NONE; |
cc532915 | 49 | |
496b010e MN |
50 | #define CRASH_HANDLER_MAX 1 |
51 | /* NULL terminated list of shutdown handles */ | |
52 | static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX+1]; | |
53 | static DEFINE_SPINLOCK(crash_handlers_lock); | |
54 | ||
cc532915 | 55 | #ifdef CONFIG_SMP |
c0ce7d08 | 56 | static atomic_t enter_on_soft_reset = ATOMIC_INIT(0); |
cc532915 ME |
57 | |
58 | void crash_ipi_callback(struct pt_regs *regs) | |
59 | { | |
60 | int cpu = smp_processor_id(); | |
61 | ||
cc532915 ME |
62 | if (!cpu_online(cpu)) |
63 | return; | |
64 | ||
d04c56f7 | 65 | hard_irq_disable(); |
c0ce7d08 | 66 | if (!cpu_isset(cpu, cpus_in_crash)) |
85916f81 | 67 | crash_save_cpu(regs, cpu); |
c0ce7d08 | 68 | cpu_set(cpu, cpus_in_crash); |
cc532915 | 69 | |
c0ce7d08 DW |
70 | /* |
71 | * Entered via soft-reset - could be the kdump | |
72 | * process is invoked using soft-reset or user activated | |
73 | * it if some CPU did not respond to an IPI. | |
74 | * For soft-reset, the secondary CPU can enter this func | |
75 | * twice. 1 - using IPI, and 2. soft-reset. | |
76 | * Tell the kexec CPU that entered via soft-reset and ready | |
77 | * to go down. | |
78 | */ | |
79 | if (cpu_isset(cpu, cpus_in_sr)) { | |
80 | cpu_clear(cpu, cpus_in_sr); | |
81 | atomic_inc(&enter_on_soft_reset); | |
82 | } | |
83 | ||
84 | /* | |
85 | * Starting the kdump boot. | |
86 | * This barrier is needed to make sure that all CPUs are stopped. | |
87 | * If not, soft-reset will be invoked to bring other CPUs. | |
88 | */ | |
89 | while (!cpu_isset(crashing_cpu, cpus_in_crash)) | |
90 | cpu_relax(); | |
91 | ||
92 | if (ppc_md.kexec_cpu_down) | |
93 | ppc_md.kexec_cpu_down(1, 1); | |
b6f35b49 ME |
94 | |
95 | #ifdef CONFIG_PPC64 | |
cc532915 | 96 | kexec_smp_wait(); |
b6f35b49 ME |
97 | #else |
98 | for (;;); /* FIXME */ | |
99 | #endif | |
100 | ||
cc532915 ME |
101 | /* NOTREACHED */ |
102 | } | |
103 | ||
c0ce7d08 DW |
104 | /* |
105 | * Wait until all CPUs are entered via soft-reset. | |
106 | */ | |
107 | static void crash_soft_reset_check(int cpu) | |
108 | { | |
109 | unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */ | |
110 | ||
111 | cpu_clear(cpu, cpus_in_sr); | |
112 | while (atomic_read(&enter_on_soft_reset) != ncpus) | |
113 | cpu_relax(); | |
114 | } | |
115 | ||
116 | ||
117 | static void crash_kexec_prepare_cpus(int cpu) | |
cc532915 ME |
118 | { |
119 | unsigned int msecs; | |
120 | ||
c0ce7d08 | 121 | unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */ |
cc532915 ME |
122 | |
123 | crash_send_ipi(crash_ipi_callback); | |
124 | smp_wmb(); | |
125 | ||
126 | /* | |
127 | * FIXME: Until we will have the way to stop other CPUSs reliabally, | |
128 | * the crash CPU will send an IPI and wait for other CPUs to | |
c0ce7d08 | 129 | * respond. |
01aaed9d | 130 | * Delay of at least 10 seconds. |
cc532915 | 131 | */ |
c0ce7d08 | 132 | printk(KERN_EMERG "Sending IPI to other cpus...\n"); |
01aaed9d | 133 | msecs = 10000; |
c0ce7d08 DW |
134 | while ((cpus_weight(cpus_in_crash) < ncpus) && (--msecs > 0)) { |
135 | cpu_relax(); | |
cc532915 ME |
136 | mdelay(1); |
137 | } | |
138 | ||
139 | /* Would it be better to replace the trap vector here? */ | |
140 | ||
141 | /* | |
142 | * FIXME: In case if we do not get all CPUs, one possibility: ask the | |
143 | * user to do soft reset such that we get all. | |
c0ce7d08 DW |
144 | * Soft-reset will be used until better mechanism is implemented. |
145 | */ | |
146 | if (cpus_weight(cpus_in_crash) < ncpus) { | |
147 | printk(KERN_EMERG "done waiting: %d cpu(s) not responding\n", | |
148 | ncpus - cpus_weight(cpus_in_crash)); | |
149 | printk(KERN_EMERG "Activate soft-reset to stop other cpu(s)\n"); | |
150 | cpus_in_sr = CPU_MASK_NONE; | |
151 | atomic_set(&enter_on_soft_reset, 0); | |
152 | while (cpus_weight(cpus_in_crash) < ncpus) | |
153 | cpu_relax(); | |
154 | } | |
155 | /* | |
156 | * Make sure all CPUs are entered via soft-reset if the kdump is | |
157 | * invoked using soft-reset. | |
cc532915 | 158 | */ |
c0ce7d08 DW |
159 | if (cpu_isset(cpu, cpus_in_sr)) |
160 | crash_soft_reset_check(cpu); | |
cc532915 ME |
161 | /* Leave the IPI callback set */ |
162 | } | |
c0ce7d08 DW |
163 | |
164 | /* | |
165 | * This function will be called by secondary cpus or by kexec cpu | |
166 | * if soft-reset is activated to stop some CPUs. | |
167 | */ | |
168 | void crash_kexec_secondary(struct pt_regs *regs) | |
169 | { | |
170 | int cpu = smp_processor_id(); | |
171 | unsigned long flags; | |
172 | int msecs = 5; | |
173 | ||
174 | local_irq_save(flags); | |
175 | /* Wait 5ms if the kexec CPU is not entered yet. */ | |
176 | while (crashing_cpu < 0) { | |
177 | if (--msecs < 0) { | |
178 | /* | |
179 | * Either kdump image is not loaded or | |
180 | * kdump process is not started - Probably xmon | |
181 | * exited using 'x'(exit and recover) or | |
182 | * kexec_should_crash() failed for all running tasks. | |
183 | */ | |
184 | cpu_clear(cpu, cpus_in_sr); | |
185 | local_irq_restore(flags); | |
186 | return; | |
187 | } | |
188 | mdelay(1); | |
189 | cpu_relax(); | |
190 | } | |
191 | if (cpu == crashing_cpu) { | |
192 | /* | |
193 | * Panic CPU will enter this func only via soft-reset. | |
194 | * Wait until all secondary CPUs entered and | |
195 | * then start kexec boot. | |
196 | */ | |
197 | crash_soft_reset_check(cpu); | |
198 | cpu_set(crashing_cpu, cpus_in_crash); | |
199 | if (ppc_md.kexec_cpu_down) | |
200 | ppc_md.kexec_cpu_down(1, 0); | |
201 | machine_kexec(kexec_crash_image); | |
202 | /* NOTREACHED */ | |
203 | } | |
204 | crash_ipi_callback(regs); | |
205 | } | |
206 | ||
cc532915 | 207 | #else |
c0ce7d08 | 208 | static void crash_kexec_prepare_cpus(int cpu) |
cc532915 ME |
209 | { |
210 | /* | |
211 | * move the secondarys to us so that we can copy | |
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 |
223 | void crash_kexec_secondary(struct pt_regs *regs) |
224 | { | |
225 | cpus_in_sr = CPU_MASK_NONE; | |
226 | } | |
cc532915 | 227 | #endif |
8d2655e6 AD |
228 | #ifdef CONFIG_SPU_BASE |
229 | ||
230 | #include <asm/spu.h> | |
231 | #include <asm/spu_priv1.h> | |
232 | ||
233 | struct crash_spu_info { | |
234 | struct spu *spu; | |
235 | u32 saved_spu_runcntl_RW; | |
236 | u32 saved_spu_status_R; | |
237 | u32 saved_spu_npc_RW; | |
238 | u64 saved_mfc_sr1_RW; | |
239 | u64 saved_mfc_dar; | |
240 | u64 saved_mfc_dsisr; | |
241 | }; | |
242 | ||
243 | #define CRASH_NUM_SPUS 16 /* Enough for current hardware */ | |
244 | static struct crash_spu_info crash_spu_info[CRASH_NUM_SPUS]; | |
245 | ||
246 | static void crash_kexec_stop_spus(void) | |
247 | { | |
248 | struct spu *spu; | |
249 | int i; | |
250 | u64 tmp; | |
251 | ||
252 | for (i = 0; i < CRASH_NUM_SPUS; i++) { | |
253 | if (!crash_spu_info[i].spu) | |
254 | continue; | |
255 | ||
256 | spu = crash_spu_info[i].spu; | |
257 | ||
258 | crash_spu_info[i].saved_spu_runcntl_RW = | |
259 | in_be32(&spu->problem->spu_runcntl_RW); | |
260 | crash_spu_info[i].saved_spu_status_R = | |
261 | in_be32(&spu->problem->spu_status_R); | |
262 | crash_spu_info[i].saved_spu_npc_RW = | |
263 | in_be32(&spu->problem->spu_npc_RW); | |
264 | ||
265 | crash_spu_info[i].saved_mfc_dar = spu_mfc_dar_get(spu); | |
266 | crash_spu_info[i].saved_mfc_dsisr = spu_mfc_dsisr_get(spu); | |
267 | tmp = spu_mfc_sr1_get(spu); | |
268 | crash_spu_info[i].saved_mfc_sr1_RW = tmp; | |
269 | ||
270 | tmp &= ~MFC_STATE1_MASTER_RUN_CONTROL_MASK; | |
271 | spu_mfc_sr1_set(spu, tmp); | |
272 | ||
273 | __delay(200); | |
274 | } | |
275 | } | |
276 | ||
277 | void crash_register_spus(struct list_head *list) | |
278 | { | |
279 | struct spu *spu; | |
280 | ||
281 | list_for_each_entry(spu, list, full_list) { | |
282 | if (WARN_ON(spu->number >= CRASH_NUM_SPUS)) | |
283 | continue; | |
284 | ||
285 | crash_spu_info[spu->number].spu = spu; | |
286 | } | |
287 | } | |
288 | ||
289 | #else | |
290 | static inline void crash_kexec_stop_spus(void) | |
291 | { | |
292 | } | |
293 | #endif /* CONFIG_SPU_BASE */ | |
cc532915 | 294 | |
496b010e MN |
295 | /* |
296 | * Register a function to be called on shutdown. Only use this if you | |
297 | * can't reset your device in the second kernel. | |
298 | */ | |
299 | int crash_shutdown_register(crash_shutdown_t handler) | |
300 | { | |
301 | unsigned int i, rc; | |
302 | ||
303 | spin_lock(&crash_handlers_lock); | |
304 | for (i = 0 ; i < CRASH_HANDLER_MAX; i++) | |
305 | if (!crash_shutdown_handles[i]) { | |
306 | /* Insert handle at first empty entry */ | |
307 | crash_shutdown_handles[i] = handler; | |
308 | rc = 0; | |
309 | break; | |
310 | } | |
311 | ||
312 | if (i == CRASH_HANDLER_MAX) { | |
313 | printk(KERN_ERR "Crash shutdown handles full, " | |
314 | "not registered.\n"); | |
315 | rc = 1; | |
316 | } | |
317 | ||
318 | spin_unlock(&crash_handlers_lock); | |
319 | return rc; | |
320 | } | |
321 | EXPORT_SYMBOL(crash_shutdown_register); | |
322 | ||
323 | int crash_shutdown_unregister(crash_shutdown_t handler) | |
324 | { | |
325 | unsigned int i, rc; | |
326 | ||
327 | spin_lock(&crash_handlers_lock); | |
328 | for (i = 0 ; i < CRASH_HANDLER_MAX; i++) | |
329 | if (crash_shutdown_handles[i] == handler) | |
330 | break; | |
331 | ||
332 | if (i == CRASH_HANDLER_MAX) { | |
333 | printk(KERN_ERR "Crash shutdown handle not found\n"); | |
334 | rc = 1; | |
335 | } else { | |
336 | /* Shift handles down */ | |
337 | for (; crash_shutdown_handles[i]; i++) | |
338 | crash_shutdown_handles[i] = | |
339 | crash_shutdown_handles[i+1]; | |
340 | rc = 0; | |
341 | } | |
342 | ||
343 | spin_unlock(&crash_handlers_lock); | |
344 | return rc; | |
345 | } | |
346 | EXPORT_SYMBOL(crash_shutdown_unregister); | |
347 | ||
348 | static unsigned long crash_shutdown_buf[JMP_BUF_LEN]; | |
349 | ||
350 | static int handle_fault(struct pt_regs *regs) | |
351 | { | |
352 | longjmp(crash_shutdown_buf, 1); | |
353 | return 0; | |
354 | } | |
355 | ||
cc532915 ME |
356 | void default_machine_crash_shutdown(struct pt_regs *regs) |
357 | { | |
496b010e MN |
358 | unsigned int i; |
359 | int (*old_handler)(struct pt_regs *regs); | |
360 | ||
d6c1a908 | 361 | |
cc532915 ME |
362 | /* |
363 | * This function is only called after the system | |
f18190bd | 364 | * has panicked or is otherwise in a critical state. |
cc532915 ME |
365 | * The minimum amount of code to allow a kexec'd kernel |
366 | * to run successfully needs to happen here. | |
367 | * | |
368 | * In practice this means stopping other cpus in | |
369 | * an SMP system. | |
370 | * The kernel is broken so disable interrupts. | |
371 | */ | |
d04c56f7 | 372 | hard_irq_disable(); |
cc532915 | 373 | |
496b010e MN |
374 | for_each_irq(i) { |
375 | struct irq_desc *desc = irq_desc + i; | |
d6c1a908 ME |
376 | |
377 | if (desc->status & IRQ_INPROGRESS) | |
496b010e | 378 | desc->chip->eoi(i); |
d6c1a908 ME |
379 | |
380 | if (!(desc->status & IRQ_DISABLED)) | |
496b010e MN |
381 | desc->chip->disable(i); |
382 | } | |
383 | ||
384 | /* | |
385 | * Call registered shutdown routines savely. Swap out | |
386 | * __debugger_fault_handler, and replace on exit. | |
387 | */ | |
388 | old_handler = __debugger_fault_handler; | |
389 | __debugger_fault_handler = handle_fault; | |
390 | for (i = 0; crash_shutdown_handles[i]; i++) { | |
391 | if (setjmp(crash_shutdown_buf) == 0) { | |
392 | /* | |
393 | * Insert syncs and delay to ensure | |
394 | * instructions in the dangerous region don't | |
395 | * leak away from this protected region. | |
396 | */ | |
397 | asm volatile("sync; isync"); | |
398 | /* dangerous region */ | |
399 | crash_shutdown_handles[i](); | |
400 | asm volatile("sync; isync"); | |
401 | } | |
d6c1a908 | 402 | } |
496b010e | 403 | __debugger_fault_handler = old_handler; |
d6c1a908 | 404 | |
cc532915 ME |
405 | /* |
406 | * Make a note of crashing cpu. Will be used in machine_kexec | |
407 | * such that another IPI will not be sent. | |
408 | */ | |
409 | crashing_cpu = smp_processor_id(); | |
85916f81 | 410 | crash_save_cpu(regs, crashing_cpu); |
c0ce7d08 DW |
411 | crash_kexec_prepare_cpus(crashing_cpu); |
412 | cpu_set(crashing_cpu, cpus_in_crash); | |
8d2655e6 | 413 | crash_kexec_stop_spus(); |
c0ce7d08 DW |
414 | if (ppc_md.kexec_cpu_down) |
415 | ppc_md.kexec_cpu_down(1, 0); | |
cc532915 | 416 | } |