Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Machine check handler. | |
e9eee03e | 3 | * |
1da177e4 | 4 | * K8 parts Copyright 2002,2003 Andi Kleen, SuSE Labs. |
d88203d1 TG |
5 | * Rest from unknown author(s). |
6 | * 2004 Andi Kleen. Rewrote most of it. | |
b79109c3 AK |
7 | * Copyright 2008 Intel Corporation |
8 | * Author: Andi Kleen | |
1da177e4 | 9 | */ |
e9eee03e IM |
10 | #include <linux/thread_info.h> |
11 | #include <linux/capability.h> | |
12 | #include <linux/miscdevice.h> | |
ccc3c319 | 13 | #include <linux/interrupt.h> |
e9eee03e IM |
14 | #include <linux/ratelimit.h> |
15 | #include <linux/kallsyms.h> | |
16 | #include <linux/rcupdate.h> | |
e9eee03e | 17 | #include <linux/kobject.h> |
14a02530 | 18 | #include <linux/uaccess.h> |
e9eee03e IM |
19 | #include <linux/kdebug.h> |
20 | #include <linux/kernel.h> | |
21 | #include <linux/percpu.h> | |
1da177e4 | 22 | #include <linux/string.h> |
1da177e4 | 23 | #include <linux/sysdev.h> |
3c079792 | 24 | #include <linux/delay.h> |
8c566ef5 | 25 | #include <linux/ctype.h> |
e9eee03e | 26 | #include <linux/sched.h> |
0d7482e3 | 27 | #include <linux/sysfs.h> |
e9eee03e IM |
28 | #include <linux/types.h> |
29 | #include <linux/init.h> | |
30 | #include <linux/kmod.h> | |
31 | #include <linux/poll.h> | |
3c079792 | 32 | #include <linux/nmi.h> |
e9eee03e | 33 | #include <linux/cpu.h> |
14a02530 | 34 | #include <linux/smp.h> |
e9eee03e | 35 | #include <linux/fs.h> |
9b1beaf2 | 36 | #include <linux/mm.h> |
e9eee03e | 37 | |
d88203d1 | 38 | #include <asm/processor.h> |
ccc3c319 AK |
39 | #include <asm/hw_irq.h> |
40 | #include <asm/apic.h> | |
e02e68d3 | 41 | #include <asm/idle.h> |
ccc3c319 | 42 | #include <asm/ipi.h> |
e9eee03e IM |
43 | #include <asm/mce.h> |
44 | #include <asm/msr.h> | |
1da177e4 | 45 | |
bd19a5e6 | 46 | #include "mce-internal.h" |
711c2e48 | 47 | |
5d727926 AK |
48 | /* Handle unconfigured int18 (should never happen) */ |
49 | static void unexpected_machine_check(struct pt_regs *regs, long error_code) | |
50 | { | |
51 | printk(KERN_ERR "CPU#%d: Unexpected int18 (Machine Check).\n", | |
52 | smp_processor_id()); | |
53 | } | |
54 | ||
55 | /* Call the installed machine check handler for this CPU setup. */ | |
56 | void (*machine_check_vector)(struct pt_regs *, long error_code) = | |
57 | unexpected_machine_check; | |
04b2b1a4 | 58 | |
4e5b3e69 | 59 | int mce_disabled __read_mostly; |
04b2b1a4 | 60 | |
e9eee03e | 61 | #define MISC_MCELOG_MINOR 227 |
0d7482e3 | 62 | |
3c079792 AK |
63 | #define SPINUNIT 100 /* 100ns */ |
64 | ||
553f265f AK |
65 | atomic_t mce_entry; |
66 | ||
01ca79f1 AK |
67 | DEFINE_PER_CPU(unsigned, mce_exception_count); |
68 | ||
bd78432c TH |
69 | /* |
70 | * Tolerant levels: | |
71 | * 0: always panic on uncorrected errors, log corrected errors | |
72 | * 1: panic or SIGBUS on uncorrected errors, log corrected errors | |
73 | * 2: SIGBUS or log uncorrected errors (if possible), log corrected errors | |
74 | * 3: never panic or SIGBUS, log all errors (for testing only) | |
75 | */ | |
4e5b3e69 HS |
76 | static int tolerant __read_mostly = 1; |
77 | static int banks __read_mostly; | |
78 | static u64 *bank __read_mostly; | |
79 | static int rip_msr __read_mostly; | |
80 | static int mce_bootlog __read_mostly = -1; | |
81 | static int monarch_timeout __read_mostly = -1; | |
82 | static int mce_panic_timeout __read_mostly; | |
83 | static int mce_dont_log_ce __read_mostly; | |
84 | int mce_cmci_disabled __read_mostly; | |
85 | int mce_ignore_ce __read_mostly; | |
86 | int mce_ser __read_mostly; | |
a98f0dd3 | 87 | |
1020bcbc HS |
88 | /* User mode helper program triggered by machine check event */ |
89 | static unsigned long mce_need_notify; | |
90 | static char mce_helper[128]; | |
91 | static char *mce_helper_argv[2] = { mce_helper, NULL }; | |
1da177e4 | 92 | |
06b7a7a5 AK |
93 | static unsigned long dont_init_banks; |
94 | ||
e02e68d3 | 95 | static DECLARE_WAIT_QUEUE_HEAD(mce_wait); |
3c079792 AK |
96 | static DEFINE_PER_CPU(struct mce, mces_seen); |
97 | static int cpu_missing; | |
98 | ||
e02e68d3 | 99 | |
ee031c31 AK |
100 | /* MCA banks polled by the period polling timer for corrected events */ |
101 | DEFINE_PER_CPU(mce_banks_t, mce_poll_banks) = { | |
102 | [0 ... BITS_TO_LONGS(MAX_NR_BANKS)-1] = ~0UL | |
103 | }; | |
104 | ||
06b7a7a5 AK |
105 | static inline int skip_bank_init(int i) |
106 | { | |
107 | return i < BITS_PER_LONG && test_bit(i, &dont_init_banks); | |
108 | } | |
109 | ||
9b1beaf2 AK |
110 | static DEFINE_PER_CPU(struct work_struct, mce_work); |
111 | ||
b5f2fa4e AK |
112 | /* Do initial initialization of a struct mce */ |
113 | void mce_setup(struct mce *m) | |
114 | { | |
115 | memset(m, 0, sizeof(struct mce)); | |
d620c67f | 116 | m->cpu = m->extcpu = smp_processor_id(); |
b5f2fa4e | 117 | rdtscll(m->tsc); |
8ee08347 AK |
118 | /* We hope get_seconds stays lockless */ |
119 | m->time = get_seconds(); | |
120 | m->cpuvendor = boot_cpu_data.x86_vendor; | |
121 | m->cpuid = cpuid_eax(1); | |
122 | #ifdef CONFIG_SMP | |
123 | m->socketid = cpu_data(m->extcpu).phys_proc_id; | |
124 | #endif | |
125 | m->apicid = cpu_data(m->extcpu).initial_apicid; | |
126 | rdmsrl(MSR_IA32_MCG_CAP, m->mcgcap); | |
b5f2fa4e AK |
127 | } |
128 | ||
ea149b36 AK |
129 | DEFINE_PER_CPU(struct mce, injectm); |
130 | EXPORT_PER_CPU_SYMBOL_GPL(injectm); | |
131 | ||
1da177e4 LT |
132 | /* |
133 | * Lockless MCE logging infrastructure. | |
134 | * This avoids deadlocks on printk locks without having to break locks. Also | |
135 | * separate MCEs from kernel messages to avoid bogus bug reports. | |
136 | */ | |
137 | ||
231fd906 | 138 | static struct mce_log mcelog = { |
f6fb0ac0 AK |
139 | .signature = MCE_LOG_SIGNATURE, |
140 | .len = MCE_LOG_LEN, | |
141 | .recordlen = sizeof(struct mce), | |
d88203d1 | 142 | }; |
1da177e4 LT |
143 | |
144 | void mce_log(struct mce *mce) | |
145 | { | |
146 | unsigned next, entry; | |
e9eee03e | 147 | |
1da177e4 | 148 | mce->finished = 0; |
7644143c | 149 | wmb(); |
1da177e4 LT |
150 | for (;;) { |
151 | entry = rcu_dereference(mcelog.next); | |
673242c1 | 152 | for (;;) { |
e9eee03e IM |
153 | /* |
154 | * When the buffer fills up discard new entries. | |
155 | * Assume that the earlier errors are the more | |
156 | * interesting ones: | |
157 | */ | |
673242c1 | 158 | if (entry >= MCE_LOG_LEN) { |
14a02530 HS |
159 | set_bit(MCE_OVERFLOW, |
160 | (unsigned long *)&mcelog.flags); | |
673242c1 AK |
161 | return; |
162 | } | |
e9eee03e | 163 | /* Old left over entry. Skip: */ |
673242c1 AK |
164 | if (mcelog.entry[entry].finished) { |
165 | entry++; | |
166 | continue; | |
167 | } | |
7644143c | 168 | break; |
1da177e4 | 169 | } |
1da177e4 LT |
170 | smp_rmb(); |
171 | next = entry + 1; | |
172 | if (cmpxchg(&mcelog.next, entry, next) == entry) | |
173 | break; | |
174 | } | |
175 | memcpy(mcelog.entry + entry, mce, sizeof(struct mce)); | |
7644143c | 176 | wmb(); |
1da177e4 | 177 | mcelog.entry[entry].finished = 1; |
7644143c | 178 | wmb(); |
1da177e4 | 179 | |
a0189c70 | 180 | mce->finished = 1; |
1020bcbc | 181 | set_bit(0, &mce_need_notify); |
1da177e4 LT |
182 | } |
183 | ||
77e26cca | 184 | static void print_mce(struct mce *m) |
1da177e4 | 185 | { |
86503560 | 186 | printk(KERN_EMERG |
1da177e4 | 187 | "CPU %d: Machine Check Exception: %16Lx Bank %d: %016Lx\n", |
d620c67f | 188 | m->extcpu, m->mcgstatus, m->bank, m->status); |
65ea5b03 | 189 | if (m->ip) { |
d88203d1 | 190 | printk(KERN_EMERG "RIP%s %02x:<%016Lx> ", |
1da177e4 | 191 | !(m->mcgstatus & MCG_STATUS_EIPV) ? " !INEXACT!" : "", |
65ea5b03 | 192 | m->cs, m->ip); |
1da177e4 | 193 | if (m->cs == __KERNEL_CS) |
65ea5b03 | 194 | print_symbol("{%s}", m->ip); |
1da177e4 LT |
195 | printk("\n"); |
196 | } | |
f6d1826d | 197 | printk(KERN_EMERG "TSC %llx ", m->tsc); |
1da177e4 | 198 | if (m->addr) |
f6d1826d | 199 | printk("ADDR %llx ", m->addr); |
1da177e4 | 200 | if (m->misc) |
f6d1826d | 201 | printk("MISC %llx ", m->misc); |
1da177e4 | 202 | printk("\n"); |
8ee08347 AK |
203 | printk(KERN_EMERG "PROCESSOR %u:%x TIME %llu SOCKET %u APIC %x\n", |
204 | m->cpuvendor, m->cpuid, m->time, m->socketid, | |
205 | m->apicid); | |
86503560 AK |
206 | } |
207 | ||
77e26cca HS |
208 | static void print_mce_head(void) |
209 | { | |
210 | printk(KERN_EMERG "\n" KERN_EMERG "HARDWARE ERROR\n"); | |
211 | } | |
212 | ||
86503560 AK |
213 | static void print_mce_tail(void) |
214 | { | |
215 | printk(KERN_EMERG "This is not a software problem!\n" | |
216 | KERN_EMERG "Run through mcelog --ascii to decode and contact your hardware vendor\n"); | |
1da177e4 LT |
217 | } |
218 | ||
f94b61c2 AK |
219 | #define PANIC_TIMEOUT 5 /* 5 seconds */ |
220 | ||
221 | static atomic_t mce_paniced; | |
222 | ||
223 | /* Panic in progress. Enable interrupts and wait for final IPI */ | |
224 | static void wait_for_panic(void) | |
225 | { | |
226 | long timeout = PANIC_TIMEOUT*USEC_PER_SEC; | |
227 | preempt_disable(); | |
228 | local_irq_enable(); | |
229 | while (timeout-- > 0) | |
230 | udelay(1); | |
29b0f591 AK |
231 | if (panic_timeout == 0) |
232 | panic_timeout = mce_panic_timeout; | |
f94b61c2 AK |
233 | panic("Panicing machine check CPU died"); |
234 | } | |
235 | ||
bd19a5e6 | 236 | static void mce_panic(char *msg, struct mce *final, char *exp) |
d88203d1 | 237 | { |
1da177e4 | 238 | int i; |
e02e68d3 | 239 | |
f94b61c2 AK |
240 | /* |
241 | * Make sure only one CPU runs in machine check panic | |
242 | */ | |
a95436e4 | 243 | if (atomic_inc_return(&mce_paniced) > 1) |
f94b61c2 AK |
244 | wait_for_panic(); |
245 | barrier(); | |
246 | ||
d896a940 AK |
247 | bust_spinlocks(1); |
248 | console_verbose(); | |
77e26cca | 249 | print_mce_head(); |
a0189c70 | 250 | /* First print corrected ones that are still unlogged */ |
1da177e4 | 251 | for (i = 0; i < MCE_LOG_LEN; i++) { |
a0189c70 | 252 | struct mce *m = &mcelog.entry[i]; |
77e26cca HS |
253 | if (!(m->status & MCI_STATUS_VAL)) |
254 | continue; | |
255 | if (!(m->status & MCI_STATUS_UC)) | |
256 | print_mce(m); | |
a0189c70 AK |
257 | } |
258 | /* Now print uncorrected but with the final one last */ | |
259 | for (i = 0; i < MCE_LOG_LEN; i++) { | |
260 | struct mce *m = &mcelog.entry[i]; | |
261 | if (!(m->status & MCI_STATUS_VAL)) | |
1da177e4 | 262 | continue; |
77e26cca HS |
263 | if (!(m->status & MCI_STATUS_UC)) |
264 | continue; | |
a0189c70 | 265 | if (!final || memcmp(m, final, sizeof(struct mce))) |
77e26cca | 266 | print_mce(m); |
1da177e4 | 267 | } |
a0189c70 | 268 | if (final) |
77e26cca | 269 | print_mce(final); |
3c079792 AK |
270 | if (cpu_missing) |
271 | printk(KERN_EMERG "Some CPUs didn't answer in synchronization\n"); | |
86503560 | 272 | print_mce_tail(); |
bd19a5e6 AK |
273 | if (exp) |
274 | printk(KERN_EMERG "Machine check: %s\n", exp); | |
29b0f591 AK |
275 | if (panic_timeout == 0) |
276 | panic_timeout = mce_panic_timeout; | |
e02e68d3 | 277 | panic(msg); |
d88203d1 | 278 | } |
1da177e4 | 279 | |
ea149b36 AK |
280 | /* Support code for software error injection */ |
281 | ||
282 | static int msr_to_offset(u32 msr) | |
283 | { | |
284 | unsigned bank = __get_cpu_var(injectm.bank); | |
285 | if (msr == rip_msr) | |
286 | return offsetof(struct mce, ip); | |
287 | if (msr == MSR_IA32_MC0_STATUS + bank*4) | |
288 | return offsetof(struct mce, status); | |
289 | if (msr == MSR_IA32_MC0_ADDR + bank*4) | |
290 | return offsetof(struct mce, addr); | |
291 | if (msr == MSR_IA32_MC0_MISC + bank*4) | |
292 | return offsetof(struct mce, misc); | |
293 | if (msr == MSR_IA32_MCG_STATUS) | |
294 | return offsetof(struct mce, mcgstatus); | |
295 | return -1; | |
296 | } | |
297 | ||
5f8c1a54 AK |
298 | /* MSR access wrappers used for error injection */ |
299 | static u64 mce_rdmsrl(u32 msr) | |
300 | { | |
301 | u64 v; | |
ea149b36 AK |
302 | if (__get_cpu_var(injectm).finished) { |
303 | int offset = msr_to_offset(msr); | |
304 | if (offset < 0) | |
305 | return 0; | |
306 | return *(u64 *)((char *)&__get_cpu_var(injectm) + offset); | |
307 | } | |
5f8c1a54 AK |
308 | rdmsrl(msr, v); |
309 | return v; | |
310 | } | |
311 | ||
312 | static void mce_wrmsrl(u32 msr, u64 v) | |
313 | { | |
ea149b36 AK |
314 | if (__get_cpu_var(injectm).finished) { |
315 | int offset = msr_to_offset(msr); | |
316 | if (offset >= 0) | |
317 | *(u64 *)((char *)&__get_cpu_var(injectm) + offset) = v; | |
318 | return; | |
319 | } | |
5f8c1a54 AK |
320 | wrmsrl(msr, v); |
321 | } | |
322 | ||
9b1beaf2 AK |
323 | /* |
324 | * Simple lockless ring to communicate PFNs from the exception handler with the | |
325 | * process context work function. This is vastly simplified because there's | |
326 | * only a single reader and a single writer. | |
327 | */ | |
328 | #define MCE_RING_SIZE 16 /* we use one entry less */ | |
329 | ||
330 | struct mce_ring { | |
331 | unsigned short start; | |
332 | unsigned short end; | |
333 | unsigned long ring[MCE_RING_SIZE]; | |
334 | }; | |
335 | static DEFINE_PER_CPU(struct mce_ring, mce_ring); | |
336 | ||
337 | /* Runs with CPU affinity in workqueue */ | |
338 | static int mce_ring_empty(void) | |
339 | { | |
340 | struct mce_ring *r = &__get_cpu_var(mce_ring); | |
341 | ||
342 | return r->start == r->end; | |
343 | } | |
344 | ||
345 | static int mce_ring_get(unsigned long *pfn) | |
346 | { | |
347 | struct mce_ring *r; | |
348 | int ret = 0; | |
349 | ||
350 | *pfn = 0; | |
351 | get_cpu(); | |
352 | r = &__get_cpu_var(mce_ring); | |
353 | if (r->start == r->end) | |
354 | goto out; | |
355 | *pfn = r->ring[r->start]; | |
356 | r->start = (r->start + 1) % MCE_RING_SIZE; | |
357 | ret = 1; | |
358 | out: | |
359 | put_cpu(); | |
360 | return ret; | |
361 | } | |
362 | ||
363 | /* Always runs in MCE context with preempt off */ | |
364 | static int mce_ring_add(unsigned long pfn) | |
365 | { | |
366 | struct mce_ring *r = &__get_cpu_var(mce_ring); | |
367 | unsigned next; | |
368 | ||
369 | next = (r->end + 1) % MCE_RING_SIZE; | |
370 | if (next == r->start) | |
371 | return -1; | |
372 | r->ring[r->end] = pfn; | |
373 | wmb(); | |
374 | r->end = next; | |
375 | return 0; | |
376 | } | |
377 | ||
88ccbedd | 378 | int mce_available(struct cpuinfo_x86 *c) |
1da177e4 | 379 | { |
04b2b1a4 | 380 | if (mce_disabled) |
5b4408fd | 381 | return 0; |
3d1712c9 | 382 | return cpu_has(c, X86_FEATURE_MCE) && cpu_has(c, X86_FEATURE_MCA); |
1da177e4 LT |
383 | } |
384 | ||
9b1beaf2 AK |
385 | static void mce_schedule_work(void) |
386 | { | |
387 | if (!mce_ring_empty()) { | |
388 | struct work_struct *work = &__get_cpu_var(mce_work); | |
389 | if (!work_pending(work)) | |
390 | schedule_work(work); | |
391 | } | |
392 | } | |
393 | ||
1b2797dc HY |
394 | /* |
395 | * Get the address of the instruction at the time of the machine check | |
396 | * error. | |
397 | */ | |
94ad8474 AK |
398 | static inline void mce_get_rip(struct mce *m, struct pt_regs *regs) |
399 | { | |
1b2797dc HY |
400 | |
401 | if (regs && (m->mcgstatus & (MCG_STATUS_RIPV|MCG_STATUS_EIPV))) { | |
65ea5b03 | 402 | m->ip = regs->ip; |
94ad8474 AK |
403 | m->cs = regs->cs; |
404 | } else { | |
65ea5b03 | 405 | m->ip = 0; |
94ad8474 AK |
406 | m->cs = 0; |
407 | } | |
1b2797dc | 408 | if (rip_msr) |
5f8c1a54 | 409 | m->ip = mce_rdmsrl(rip_msr); |
94ad8474 AK |
410 | } |
411 | ||
ccc3c319 AK |
412 | #ifdef CONFIG_X86_LOCAL_APIC |
413 | /* | |
414 | * Called after interrupts have been reenabled again | |
415 | * when a MCE happened during an interrupts off region | |
416 | * in the kernel. | |
417 | */ | |
418 | asmlinkage void smp_mce_self_interrupt(struct pt_regs *regs) | |
419 | { | |
420 | ack_APIC_irq(); | |
421 | exit_idle(); | |
422 | irq_enter(); | |
9ff36ee9 | 423 | mce_notify_irq(); |
9b1beaf2 | 424 | mce_schedule_work(); |
ccc3c319 AK |
425 | irq_exit(); |
426 | } | |
427 | #endif | |
428 | ||
429 | static void mce_report_event(struct pt_regs *regs) | |
430 | { | |
431 | if (regs->flags & (X86_VM_MASK|X86_EFLAGS_IF)) { | |
9ff36ee9 | 432 | mce_notify_irq(); |
9b1beaf2 AK |
433 | /* |
434 | * Triggering the work queue here is just an insurance | |
435 | * policy in case the syscall exit notify handler | |
436 | * doesn't run soon enough or ends up running on the | |
437 | * wrong CPU (can happen when audit sleeps) | |
438 | */ | |
439 | mce_schedule_work(); | |
ccc3c319 AK |
440 | return; |
441 | } | |
442 | ||
443 | #ifdef CONFIG_X86_LOCAL_APIC | |
444 | /* | |
445 | * Without APIC do not notify. The event will be picked | |
446 | * up eventually. | |
447 | */ | |
448 | if (!cpu_has_apic) | |
449 | return; | |
450 | ||
451 | /* | |
452 | * When interrupts are disabled we cannot use | |
453 | * kernel services safely. Trigger an self interrupt | |
454 | * through the APIC to instead do the notification | |
455 | * after interrupts are reenabled again. | |
456 | */ | |
457 | apic->send_IPI_self(MCE_SELF_VECTOR); | |
458 | ||
459 | /* | |
460 | * Wait for idle afterwards again so that we don't leave the | |
461 | * APIC in a non idle state because the normal APIC writes | |
462 | * cannot exclude us. | |
463 | */ | |
464 | apic_wait_icr_idle(); | |
465 | #endif | |
466 | } | |
467 | ||
ca84f696 AK |
468 | DEFINE_PER_CPU(unsigned, mce_poll_count); |
469 | ||
d88203d1 | 470 | /* |
b79109c3 AK |
471 | * Poll for corrected events or events that happened before reset. |
472 | * Those are just logged through /dev/mcelog. | |
473 | * | |
474 | * This is executed in standard interrupt context. | |
ed7290d0 AK |
475 | * |
476 | * Note: spec recommends to panic for fatal unsignalled | |
477 | * errors here. However this would be quite problematic -- | |
478 | * we would need to reimplement the Monarch handling and | |
479 | * it would mess up the exclusion between exception handler | |
480 | * and poll hander -- * so we skip this for now. | |
481 | * These cases should not happen anyways, or only when the CPU | |
482 | * is already totally * confused. In this case it's likely it will | |
483 | * not fully execute the machine check handler either. | |
b79109c3 | 484 | */ |
ee031c31 | 485 | void machine_check_poll(enum mcp_flags flags, mce_banks_t *b) |
b79109c3 AK |
486 | { |
487 | struct mce m; | |
488 | int i; | |
489 | ||
ca84f696 AK |
490 | __get_cpu_var(mce_poll_count)++; |
491 | ||
b79109c3 AK |
492 | mce_setup(&m); |
493 | ||
5f8c1a54 | 494 | m.mcgstatus = mce_rdmsrl(MSR_IA32_MCG_STATUS); |
b79109c3 | 495 | for (i = 0; i < banks; i++) { |
ee031c31 | 496 | if (!bank[i] || !test_bit(i, *b)) |
b79109c3 AK |
497 | continue; |
498 | ||
499 | m.misc = 0; | |
500 | m.addr = 0; | |
501 | m.bank = i; | |
502 | m.tsc = 0; | |
503 | ||
504 | barrier(); | |
5f8c1a54 | 505 | m.status = mce_rdmsrl(MSR_IA32_MC0_STATUS + i*4); |
b79109c3 AK |
506 | if (!(m.status & MCI_STATUS_VAL)) |
507 | continue; | |
508 | ||
509 | /* | |
ed7290d0 AK |
510 | * Uncorrected or signalled events are handled by the exception |
511 | * handler when it is enabled, so don't process those here. | |
b79109c3 AK |
512 | * |
513 | * TBD do the same check for MCI_STATUS_EN here? | |
514 | */ | |
ed7290d0 AK |
515 | if (!(flags & MCP_UC) && |
516 | (m.status & (mce_ser ? MCI_STATUS_S : MCI_STATUS_UC))) | |
b79109c3 AK |
517 | continue; |
518 | ||
519 | if (m.status & MCI_STATUS_MISCV) | |
5f8c1a54 | 520 | m.misc = mce_rdmsrl(MSR_IA32_MC0_MISC + i*4); |
b79109c3 | 521 | if (m.status & MCI_STATUS_ADDRV) |
5f8c1a54 | 522 | m.addr = mce_rdmsrl(MSR_IA32_MC0_ADDR + i*4); |
b79109c3 AK |
523 | |
524 | if (!(flags & MCP_TIMESTAMP)) | |
525 | m.tsc = 0; | |
526 | /* | |
527 | * Don't get the IP here because it's unlikely to | |
528 | * have anything to do with the actual error location. | |
529 | */ | |
62fdac59 | 530 | if (!(flags & MCP_DONTLOG) && !mce_dont_log_ce) { |
5679af4c AK |
531 | mce_log(&m); |
532 | add_taint(TAINT_MACHINE_CHECK); | |
533 | } | |
b79109c3 AK |
534 | |
535 | /* | |
536 | * Clear state for this bank. | |
537 | */ | |
5f8c1a54 | 538 | mce_wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0); |
b79109c3 AK |
539 | } |
540 | ||
541 | /* | |
542 | * Don't clear MCG_STATUS here because it's only defined for | |
543 | * exceptions. | |
544 | */ | |
88921be3 AK |
545 | |
546 | sync_core(); | |
b79109c3 | 547 | } |
ea149b36 | 548 | EXPORT_SYMBOL_GPL(machine_check_poll); |
b79109c3 | 549 | |
bd19a5e6 AK |
550 | /* |
551 | * Do a quick check if any of the events requires a panic. | |
552 | * This decides if we keep the events around or clear them. | |
553 | */ | |
554 | static int mce_no_way_out(struct mce *m, char **msg) | |
555 | { | |
556 | int i; | |
557 | ||
558 | for (i = 0; i < banks; i++) { | |
559 | m->status = mce_rdmsrl(MSR_IA32_MC0_STATUS + i*4); | |
560 | if (mce_severity(m, tolerant, msg) >= MCE_PANIC_SEVERITY) | |
561 | return 1; | |
562 | } | |
563 | return 0; | |
564 | } | |
565 | ||
3c079792 AK |
566 | /* |
567 | * Variable to establish order between CPUs while scanning. | |
568 | * Each CPU spins initially until executing is equal its number. | |
569 | */ | |
570 | static atomic_t mce_executing; | |
571 | ||
572 | /* | |
573 | * Defines order of CPUs on entry. First CPU becomes Monarch. | |
574 | */ | |
575 | static atomic_t mce_callin; | |
576 | ||
577 | /* | |
578 | * Check if a timeout waiting for other CPUs happened. | |
579 | */ | |
580 | static int mce_timed_out(u64 *t) | |
581 | { | |
582 | /* | |
583 | * The others already did panic for some reason. | |
584 | * Bail out like in a timeout. | |
585 | * rmb() to tell the compiler that system_state | |
586 | * might have been modified by someone else. | |
587 | */ | |
588 | rmb(); | |
589 | if (atomic_read(&mce_paniced)) | |
590 | wait_for_panic(); | |
591 | if (!monarch_timeout) | |
592 | goto out; | |
593 | if ((s64)*t < SPINUNIT) { | |
594 | /* CHECKME: Make panic default for 1 too? */ | |
595 | if (tolerant < 1) | |
596 | mce_panic("Timeout synchronizing machine check over CPUs", | |
597 | NULL, NULL); | |
598 | cpu_missing = 1; | |
599 | return 1; | |
600 | } | |
601 | *t -= SPINUNIT; | |
602 | out: | |
603 | touch_nmi_watchdog(); | |
604 | return 0; | |
605 | } | |
606 | ||
607 | /* | |
608 | * The Monarch's reign. The Monarch is the CPU who entered | |
609 | * the machine check handler first. It waits for the others to | |
610 | * raise the exception too and then grades them. When any | |
611 | * error is fatal panic. Only then let the others continue. | |
612 | * | |
613 | * The other CPUs entering the MCE handler will be controlled by the | |
614 | * Monarch. They are called Subjects. | |
615 | * | |
616 | * This way we prevent any potential data corruption in a unrecoverable case | |
617 | * and also makes sure always all CPU's errors are examined. | |
618 | * | |
619 | * Also this detects the case of an machine check event coming from outer | |
620 | * space (not detected by any CPUs) In this case some external agent wants | |
621 | * us to shut down, so panic too. | |
622 | * | |
623 | * The other CPUs might still decide to panic if the handler happens | |
624 | * in a unrecoverable place, but in this case the system is in a semi-stable | |
625 | * state and won't corrupt anything by itself. It's ok to let the others | |
626 | * continue for a bit first. | |
627 | * | |
628 | * All the spin loops have timeouts; when a timeout happens a CPU | |
629 | * typically elects itself to be Monarch. | |
630 | */ | |
631 | static void mce_reign(void) | |
632 | { | |
633 | int cpu; | |
634 | struct mce *m = NULL; | |
635 | int global_worst = 0; | |
636 | char *msg = NULL; | |
637 | char *nmsg = NULL; | |
638 | ||
639 | /* | |
640 | * This CPU is the Monarch and the other CPUs have run | |
641 | * through their handlers. | |
642 | * Grade the severity of the errors of all the CPUs. | |
643 | */ | |
644 | for_each_possible_cpu(cpu) { | |
645 | int severity = mce_severity(&per_cpu(mces_seen, cpu), tolerant, | |
646 | &nmsg); | |
647 | if (severity > global_worst) { | |
648 | msg = nmsg; | |
649 | global_worst = severity; | |
650 | m = &per_cpu(mces_seen, cpu); | |
651 | } | |
652 | } | |
653 | ||
654 | /* | |
655 | * Cannot recover? Panic here then. | |
656 | * This dumps all the mces in the log buffer and stops the | |
657 | * other CPUs. | |
658 | */ | |
659 | if (m && global_worst >= MCE_PANIC_SEVERITY && tolerant < 3) | |
ac960375 | 660 | mce_panic("Fatal Machine check", m, msg); |
3c079792 AK |
661 | |
662 | /* | |
663 | * For UC somewhere we let the CPU who detects it handle it. | |
664 | * Also must let continue the others, otherwise the handling | |
665 | * CPU could deadlock on a lock. | |
666 | */ | |
667 | ||
668 | /* | |
669 | * No machine check event found. Must be some external | |
670 | * source or one CPU is hung. Panic. | |
671 | */ | |
672 | if (!m && tolerant < 3) | |
673 | mce_panic("Machine check from unknown source", NULL, NULL); | |
674 | ||
675 | /* | |
676 | * Now clear all the mces_seen so that they don't reappear on | |
677 | * the next mce. | |
678 | */ | |
679 | for_each_possible_cpu(cpu) | |
680 | memset(&per_cpu(mces_seen, cpu), 0, sizeof(struct mce)); | |
681 | } | |
682 | ||
683 | static atomic_t global_nwo; | |
684 | ||
685 | /* | |
686 | * Start of Monarch synchronization. This waits until all CPUs have | |
687 | * entered the exception handler and then determines if any of them | |
688 | * saw a fatal event that requires panic. Then it executes them | |
689 | * in the entry order. | |
690 | * TBD double check parallel CPU hotunplug | |
691 | */ | |
7fb06fc9 | 692 | static int mce_start(int *no_way_out) |
3c079792 | 693 | { |
7fb06fc9 | 694 | int order; |
3c079792 AK |
695 | int cpus = num_online_cpus(); |
696 | u64 timeout = (u64)monarch_timeout * NSEC_PER_USEC; | |
697 | ||
7fb06fc9 HS |
698 | if (!timeout) |
699 | return -1; | |
3c079792 | 700 | |
7fb06fc9 | 701 | atomic_add(*no_way_out, &global_nwo); |
184e1fdf HY |
702 | /* |
703 | * global_nwo should be updated before mce_callin | |
704 | */ | |
705 | smp_wmb(); | |
a95436e4 | 706 | order = atomic_inc_return(&mce_callin); |
3c079792 AK |
707 | |
708 | /* | |
709 | * Wait for everyone. | |
710 | */ | |
711 | while (atomic_read(&mce_callin) != cpus) { | |
712 | if (mce_timed_out(&timeout)) { | |
713 | atomic_set(&global_nwo, 0); | |
7fb06fc9 | 714 | return -1; |
3c079792 AK |
715 | } |
716 | ndelay(SPINUNIT); | |
717 | } | |
718 | ||
184e1fdf HY |
719 | /* |
720 | * mce_callin should be read before global_nwo | |
721 | */ | |
722 | smp_rmb(); | |
3c079792 | 723 | |
7fb06fc9 HS |
724 | if (order == 1) { |
725 | /* | |
726 | * Monarch: Starts executing now, the others wait. | |
727 | */ | |
3c079792 | 728 | atomic_set(&mce_executing, 1); |
7fb06fc9 HS |
729 | } else { |
730 | /* | |
731 | * Subject: Now start the scanning loop one by one in | |
732 | * the original callin order. | |
733 | * This way when there are any shared banks it will be | |
734 | * only seen by one CPU before cleared, avoiding duplicates. | |
735 | */ | |
736 | while (atomic_read(&mce_executing) < order) { | |
737 | if (mce_timed_out(&timeout)) { | |
738 | atomic_set(&global_nwo, 0); | |
739 | return -1; | |
740 | } | |
741 | ndelay(SPINUNIT); | |
742 | } | |
3c079792 AK |
743 | } |
744 | ||
745 | /* | |
7fb06fc9 | 746 | * Cache the global no_way_out state. |
3c079792 | 747 | */ |
7fb06fc9 HS |
748 | *no_way_out = atomic_read(&global_nwo); |
749 | ||
750 | return order; | |
3c079792 AK |
751 | } |
752 | ||
753 | /* | |
754 | * Synchronize between CPUs after main scanning loop. | |
755 | * This invokes the bulk of the Monarch processing. | |
756 | */ | |
757 | static int mce_end(int order) | |
758 | { | |
759 | int ret = -1; | |
760 | u64 timeout = (u64)monarch_timeout * NSEC_PER_USEC; | |
761 | ||
762 | if (!timeout) | |
763 | goto reset; | |
764 | if (order < 0) | |
765 | goto reset; | |
766 | ||
767 | /* | |
768 | * Allow others to run. | |
769 | */ | |
770 | atomic_inc(&mce_executing); | |
771 | ||
772 | if (order == 1) { | |
773 | /* CHECKME: Can this race with a parallel hotplug? */ | |
774 | int cpus = num_online_cpus(); | |
775 | ||
776 | /* | |
777 | * Monarch: Wait for everyone to go through their scanning | |
778 | * loops. | |
779 | */ | |
780 | while (atomic_read(&mce_executing) <= cpus) { | |
781 | if (mce_timed_out(&timeout)) | |
782 | goto reset; | |
783 | ndelay(SPINUNIT); | |
784 | } | |
785 | ||
786 | mce_reign(); | |
787 | barrier(); | |
788 | ret = 0; | |
789 | } else { | |
790 | /* | |
791 | * Subject: Wait for Monarch to finish. | |
792 | */ | |
793 | while (atomic_read(&mce_executing) != 0) { | |
794 | if (mce_timed_out(&timeout)) | |
795 | goto reset; | |
796 | ndelay(SPINUNIT); | |
797 | } | |
798 | ||
799 | /* | |
800 | * Don't reset anything. That's done by the Monarch. | |
801 | */ | |
802 | return 0; | |
803 | } | |
804 | ||
805 | /* | |
806 | * Reset all global state. | |
807 | */ | |
808 | reset: | |
809 | atomic_set(&global_nwo, 0); | |
810 | atomic_set(&mce_callin, 0); | |
811 | barrier(); | |
812 | ||
813 | /* | |
814 | * Let others run again. | |
815 | */ | |
816 | atomic_set(&mce_executing, 0); | |
817 | return ret; | |
818 | } | |
819 | ||
9b1beaf2 AK |
820 | /* |
821 | * Check if the address reported by the CPU is in a format we can parse. | |
822 | * It would be possible to add code for most other cases, but all would | |
823 | * be somewhat complicated (e.g. segment offset would require an instruction | |
824 | * parser). So only support physical addresses upto page granuality for now. | |
825 | */ | |
826 | static int mce_usable_address(struct mce *m) | |
827 | { | |
828 | if (!(m->status & MCI_STATUS_MISCV) || !(m->status & MCI_STATUS_ADDRV)) | |
829 | return 0; | |
830 | if ((m->misc & 0x3f) > PAGE_SHIFT) | |
831 | return 0; | |
832 | if (((m->misc >> 6) & 7) != MCM_ADDR_PHYS) | |
833 | return 0; | |
834 | return 1; | |
835 | } | |
836 | ||
3c079792 AK |
837 | static void mce_clear_state(unsigned long *toclear) |
838 | { | |
839 | int i; | |
840 | ||
841 | for (i = 0; i < banks; i++) { | |
842 | if (test_bit(i, toclear)) | |
843 | mce_wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0); | |
844 | } | |
845 | } | |
846 | ||
b79109c3 AK |
847 | /* |
848 | * The actual machine check handler. This only handles real | |
849 | * exceptions when something got corrupted coming in through int 18. | |
850 | * | |
851 | * This is executed in NMI context not subject to normal locking rules. This | |
852 | * implies that most kernel services cannot be safely used. Don't even | |
853 | * think about putting a printk in there! | |
3c079792 AK |
854 | * |
855 | * On Intel systems this is entered on all CPUs in parallel through | |
856 | * MCE broadcast. However some CPUs might be broken beyond repair, | |
857 | * so be always careful when synchronizing with others. | |
1da177e4 | 858 | */ |
e9eee03e | 859 | void do_machine_check(struct pt_regs *regs, long error_code) |
1da177e4 | 860 | { |
3c079792 | 861 | struct mce m, *final; |
1da177e4 | 862 | int i; |
3c079792 AK |
863 | int worst = 0; |
864 | int severity; | |
865 | /* | |
866 | * Establish sequential order between the CPUs entering the machine | |
867 | * check handler. | |
868 | */ | |
7fb06fc9 | 869 | int order; |
bd78432c TH |
870 | /* |
871 | * If no_way_out gets set, there is no safe way to recover from this | |
872 | * MCE. If tolerant is cranked up, we'll try anyway. | |
873 | */ | |
874 | int no_way_out = 0; | |
875 | /* | |
876 | * If kill_it gets set, there might be a way to recover from this | |
877 | * error. | |
878 | */ | |
879 | int kill_it = 0; | |
b79109c3 | 880 | DECLARE_BITMAP(toclear, MAX_NR_BANKS); |
bd19a5e6 | 881 | char *msg = "Unknown"; |
1da177e4 | 882 | |
553f265f AK |
883 | atomic_inc(&mce_entry); |
884 | ||
01ca79f1 AK |
885 | __get_cpu_var(mce_exception_count)++; |
886 | ||
b79109c3 | 887 | if (notify_die(DIE_NMI, "machine check", regs, error_code, |
22f5991c | 888 | 18, SIGKILL) == NOTIFY_STOP) |
32561696 | 889 | goto out; |
b79109c3 | 890 | if (!banks) |
32561696 | 891 | goto out; |
1da177e4 | 892 | |
b5f2fa4e AK |
893 | mce_setup(&m); |
894 | ||
5f8c1a54 | 895 | m.mcgstatus = mce_rdmsrl(MSR_IA32_MCG_STATUS); |
bd19a5e6 | 896 | no_way_out = mce_no_way_out(&m, &msg); |
d88203d1 | 897 | |
3c079792 AK |
898 | final = &__get_cpu_var(mces_seen); |
899 | *final = m; | |
900 | ||
1da177e4 LT |
901 | barrier(); |
902 | ||
ed7290d0 AK |
903 | /* |
904 | * When no restart IP must always kill or panic. | |
905 | */ | |
906 | if (!(m.mcgstatus & MCG_STATUS_RIPV)) | |
907 | kill_it = 1; | |
908 | ||
3c079792 AK |
909 | /* |
910 | * Go through all the banks in exclusion of the other CPUs. | |
911 | * This way we don't report duplicated events on shared banks | |
912 | * because the first one to see it will clear it. | |
913 | */ | |
7fb06fc9 | 914 | order = mce_start(&no_way_out); |
1da177e4 | 915 | for (i = 0; i < banks; i++) { |
b79109c3 | 916 | __clear_bit(i, toclear); |
0d7482e3 | 917 | if (!bank[i]) |
1da177e4 | 918 | continue; |
d88203d1 TG |
919 | |
920 | m.misc = 0; | |
1da177e4 LT |
921 | m.addr = 0; |
922 | m.bank = i; | |
1da177e4 | 923 | |
5f8c1a54 | 924 | m.status = mce_rdmsrl(MSR_IA32_MC0_STATUS + i*4); |
1da177e4 LT |
925 | if ((m.status & MCI_STATUS_VAL) == 0) |
926 | continue; | |
927 | ||
b79109c3 | 928 | /* |
ed7290d0 AK |
929 | * Non uncorrected or non signaled errors are handled by |
930 | * machine_check_poll. Leave them alone, unless this panics. | |
b79109c3 | 931 | */ |
ed7290d0 AK |
932 | if (!(m.status & (mce_ser ? MCI_STATUS_S : MCI_STATUS_UC)) && |
933 | !no_way_out) | |
b79109c3 AK |
934 | continue; |
935 | ||
936 | /* | |
937 | * Set taint even when machine check was not enabled. | |
938 | */ | |
939 | add_taint(TAINT_MACHINE_CHECK); | |
940 | ||
ed7290d0 | 941 | severity = mce_severity(&m, tolerant, NULL); |
b79109c3 | 942 | |
ed7290d0 AK |
943 | /* |
944 | * When machine check was for corrected handler don't touch, | |
945 | * unless we're panicing. | |
946 | */ | |
947 | if (severity == MCE_KEEP_SEVERITY && !no_way_out) | |
948 | continue; | |
949 | __set_bit(i, toclear); | |
950 | if (severity == MCE_NO_SEVERITY) { | |
b79109c3 AK |
951 | /* |
952 | * Machine check event was not enabled. Clear, but | |
953 | * ignore. | |
954 | */ | |
955 | continue; | |
1da177e4 LT |
956 | } |
957 | ||
ed7290d0 AK |
958 | /* |
959 | * Kill on action required. | |
960 | */ | |
961 | if (severity == MCE_AR_SEVERITY) | |
962 | kill_it = 1; | |
963 | ||
1da177e4 | 964 | if (m.status & MCI_STATUS_MISCV) |
5f8c1a54 | 965 | m.misc = mce_rdmsrl(MSR_IA32_MC0_MISC + i*4); |
1da177e4 | 966 | if (m.status & MCI_STATUS_ADDRV) |
5f8c1a54 | 967 | m.addr = mce_rdmsrl(MSR_IA32_MC0_ADDR + i*4); |
1da177e4 | 968 | |
9b1beaf2 AK |
969 | /* |
970 | * Action optional error. Queue address for later processing. | |
971 | * When the ring overflows we just ignore the AO error. | |
972 | * RED-PEN add some logging mechanism when | |
973 | * usable_address or mce_add_ring fails. | |
974 | * RED-PEN don't ignore overflow for tolerant == 0 | |
975 | */ | |
976 | if (severity == MCE_AO_SEVERITY && mce_usable_address(&m)) | |
977 | mce_ring_add(m.addr >> PAGE_SHIFT); | |
978 | ||
94ad8474 | 979 | mce_get_rip(&m, regs); |
b79109c3 | 980 | mce_log(&m); |
1da177e4 | 981 | |
3c079792 AK |
982 | if (severity > worst) { |
983 | *final = m; | |
984 | worst = severity; | |
1da177e4 | 985 | } |
1da177e4 LT |
986 | } |
987 | ||
3c079792 AK |
988 | if (!no_way_out) |
989 | mce_clear_state(toclear); | |
990 | ||
e9eee03e | 991 | /* |
3c079792 AK |
992 | * Do most of the synchronization with other CPUs. |
993 | * When there's any problem use only local no_way_out state. | |
e9eee03e | 994 | */ |
3c079792 AK |
995 | if (mce_end(order) < 0) |
996 | no_way_out = worst >= MCE_PANIC_SEVERITY; | |
bd78432c TH |
997 | |
998 | /* | |
999 | * If we have decided that we just CAN'T continue, and the user | |
e9eee03e | 1000 | * has not set tolerant to an insane level, give up and die. |
3c079792 AK |
1001 | * |
1002 | * This is mainly used in the case when the system doesn't | |
1003 | * support MCE broadcasting or it has been disabled. | |
bd78432c TH |
1004 | */ |
1005 | if (no_way_out && tolerant < 3) | |
ac960375 | 1006 | mce_panic("Fatal machine check on current CPU", final, msg); |
bd78432c TH |
1007 | |
1008 | /* | |
1009 | * If the error seems to be unrecoverable, something should be | |
1010 | * done. Try to kill as little as possible. If we can kill just | |
1011 | * one task, do that. If the user has set the tolerance very | |
1012 | * high, don't try to do anything at all. | |
1013 | */ | |
bd78432c | 1014 | |
ed7290d0 AK |
1015 | if (kill_it && tolerant < 3) |
1016 | force_sig(SIGBUS, current); | |
1da177e4 | 1017 | |
e02e68d3 TH |
1018 | /* notify userspace ASAP */ |
1019 | set_thread_flag(TIF_MCE_NOTIFY); | |
1020 | ||
3c079792 AK |
1021 | if (worst > 0) |
1022 | mce_report_event(regs); | |
5f8c1a54 | 1023 | mce_wrmsrl(MSR_IA32_MCG_STATUS, 0); |
32561696 | 1024 | out: |
553f265f | 1025 | atomic_dec(&mce_entry); |
88921be3 | 1026 | sync_core(); |
1da177e4 | 1027 | } |
ea149b36 | 1028 | EXPORT_SYMBOL_GPL(do_machine_check); |
1da177e4 | 1029 | |
9b1beaf2 AK |
1030 | /* dummy to break dependency. actual code is in mm/memory-failure.c */ |
1031 | void __attribute__((weak)) memory_failure(unsigned long pfn, int vector) | |
1032 | { | |
1033 | printk(KERN_ERR "Action optional memory failure at %lx ignored\n", pfn); | |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * Called after mce notification in process context. This code | |
1038 | * is allowed to sleep. Call the high level VM handler to process | |
1039 | * any corrupted pages. | |
1040 | * Assume that the work queue code only calls this one at a time | |
1041 | * per CPU. | |
1042 | * Note we don't disable preemption, so this code might run on the wrong | |
1043 | * CPU. In this case the event is picked up by the scheduled work queue. | |
1044 | * This is merely a fast path to expedite processing in some common | |
1045 | * cases. | |
1046 | */ | |
1047 | void mce_notify_process(void) | |
1048 | { | |
1049 | unsigned long pfn; | |
1050 | mce_notify_irq(); | |
1051 | while (mce_ring_get(&pfn)) | |
1052 | memory_failure(pfn, MCE_VECTOR); | |
1053 | } | |
1054 | ||
1055 | static void mce_process_work(struct work_struct *dummy) | |
1056 | { | |
1057 | mce_notify_process(); | |
1058 | } | |
1059 | ||
15d5f839 DZ |
1060 | #ifdef CONFIG_X86_MCE_INTEL |
1061 | /*** | |
1062 | * mce_log_therm_throt_event - Logs the thermal throttling event to mcelog | |
676b1855 | 1063 | * @cpu: The CPU on which the event occurred. |
15d5f839 DZ |
1064 | * @status: Event status information |
1065 | * | |
1066 | * This function should be called by the thermal interrupt after the | |
1067 | * event has been processed and the decision was made to log the event | |
1068 | * further. | |
1069 | * | |
1070 | * The status parameter will be saved to the 'status' field of 'struct mce' | |
1071 | * and historically has been the register value of the | |
1072 | * MSR_IA32_THERMAL_STATUS (Intel) msr. | |
1073 | */ | |
b5f2fa4e | 1074 | void mce_log_therm_throt_event(__u64 status) |
15d5f839 DZ |
1075 | { |
1076 | struct mce m; | |
1077 | ||
b5f2fa4e | 1078 | mce_setup(&m); |
15d5f839 DZ |
1079 | m.bank = MCE_THERMAL_BANK; |
1080 | m.status = status; | |
15d5f839 DZ |
1081 | mce_log(&m); |
1082 | } | |
1083 | #endif /* CONFIG_X86_MCE_INTEL */ | |
1084 | ||
1da177e4 | 1085 | /* |
8a336b0a TH |
1086 | * Periodic polling timer for "silent" machine check errors. If the |
1087 | * poller finds an MCE, poll 2x faster. When the poller finds no more | |
1088 | * errors, poll 2x slower (up to check_interval seconds). | |
1da177e4 | 1089 | */ |
1da177e4 | 1090 | static int check_interval = 5 * 60; /* 5 minutes */ |
e9eee03e | 1091 | |
6298c512 | 1092 | static DEFINE_PER_CPU(int, next_interval); /* in jiffies */ |
52d168e2 | 1093 | static DEFINE_PER_CPU(struct timer_list, mce_timer); |
1da177e4 | 1094 | |
52d168e2 | 1095 | static void mcheck_timer(unsigned long data) |
1da177e4 | 1096 | { |
52d168e2 | 1097 | struct timer_list *t = &per_cpu(mce_timer, data); |
6298c512 | 1098 | int *n; |
52d168e2 AK |
1099 | |
1100 | WARN_ON(smp_processor_id() != data); | |
1101 | ||
e9eee03e | 1102 | if (mce_available(¤t_cpu_data)) { |
ee031c31 AK |
1103 | machine_check_poll(MCP_TIMESTAMP, |
1104 | &__get_cpu_var(mce_poll_banks)); | |
e9eee03e | 1105 | } |
1da177e4 LT |
1106 | |
1107 | /* | |
e02e68d3 TH |
1108 | * Alert userspace if needed. If we logged an MCE, reduce the |
1109 | * polling interval, otherwise increase the polling interval. | |
1da177e4 | 1110 | */ |
6298c512 | 1111 | n = &__get_cpu_var(next_interval); |
9ff36ee9 | 1112 | if (mce_notify_irq()) |
6298c512 | 1113 | *n = max(*n/2, HZ/100); |
14a02530 | 1114 | else |
6298c512 | 1115 | *n = min(*n*2, (int)round_jiffies_relative(check_interval*HZ)); |
e02e68d3 | 1116 | |
6298c512 | 1117 | t->expires = jiffies + *n; |
52d168e2 | 1118 | add_timer(t); |
e02e68d3 TH |
1119 | } |
1120 | ||
9bd98405 AK |
1121 | static void mce_do_trigger(struct work_struct *work) |
1122 | { | |
1020bcbc | 1123 | call_usermodehelper(mce_helper, mce_helper_argv, NULL, UMH_NO_WAIT); |
9bd98405 AK |
1124 | } |
1125 | ||
1126 | static DECLARE_WORK(mce_trigger_work, mce_do_trigger); | |
1127 | ||
e02e68d3 | 1128 | /* |
9bd98405 AK |
1129 | * Notify the user(s) about new machine check events. |
1130 | * Can be called from interrupt context, but not from machine check/NMI | |
1131 | * context. | |
e02e68d3 | 1132 | */ |
9ff36ee9 | 1133 | int mce_notify_irq(void) |
e02e68d3 | 1134 | { |
8457c84d AK |
1135 | /* Not more than two messages every minute */ |
1136 | static DEFINE_RATELIMIT_STATE(ratelimit, 60*HZ, 2); | |
1137 | ||
e02e68d3 | 1138 | clear_thread_flag(TIF_MCE_NOTIFY); |
e9eee03e | 1139 | |
1020bcbc | 1140 | if (test_and_clear_bit(0, &mce_need_notify)) { |
e02e68d3 | 1141 | wake_up_interruptible(&mce_wait); |
9bd98405 AK |
1142 | |
1143 | /* | |
1144 | * There is no risk of missing notifications because | |
1145 | * work_pending is always cleared before the function is | |
1146 | * executed. | |
1147 | */ | |
1020bcbc | 1148 | if (mce_helper[0] && !work_pending(&mce_trigger_work)) |
9bd98405 | 1149 | schedule_work(&mce_trigger_work); |
e02e68d3 | 1150 | |
8457c84d | 1151 | if (__ratelimit(&ratelimit)) |
8a336b0a | 1152 | printk(KERN_INFO "Machine check events logged\n"); |
e02e68d3 TH |
1153 | |
1154 | return 1; | |
1da177e4 | 1155 | } |
e02e68d3 TH |
1156 | return 0; |
1157 | } | |
9ff36ee9 | 1158 | EXPORT_SYMBOL_GPL(mce_notify_irq); |
8a336b0a | 1159 | |
d88203d1 | 1160 | /* |
1da177e4 LT |
1161 | * Initialize Machine Checks for a CPU. |
1162 | */ | |
0d7482e3 | 1163 | static int mce_cap_init(void) |
1da177e4 | 1164 | { |
0d7482e3 | 1165 | unsigned b; |
e9eee03e | 1166 | u64 cap; |
1da177e4 LT |
1167 | |
1168 | rdmsrl(MSR_IA32_MCG_CAP, cap); | |
01c6680a TG |
1169 | |
1170 | b = cap & MCG_BANKCNT_MASK; | |
b659294b IM |
1171 | printk(KERN_INFO "mce: CPU supports %d MCE banks\n", b); |
1172 | ||
0d7482e3 AK |
1173 | if (b > MAX_NR_BANKS) { |
1174 | printk(KERN_WARNING | |
1175 | "MCE: Using only %u machine check banks out of %u\n", | |
1176 | MAX_NR_BANKS, b); | |
1177 | b = MAX_NR_BANKS; | |
1178 | } | |
1179 | ||
1180 | /* Don't support asymmetric configurations today */ | |
1181 | WARN_ON(banks != 0 && b != banks); | |
1182 | banks = b; | |
1183 | if (!bank) { | |
1184 | bank = kmalloc(banks * sizeof(u64), GFP_KERNEL); | |
1185 | if (!bank) | |
1186 | return -ENOMEM; | |
1187 | memset(bank, 0xff, banks * sizeof(u64)); | |
1da177e4 | 1188 | } |
0d7482e3 | 1189 | |
94ad8474 | 1190 | /* Use accurate RIP reporting if available. */ |
01c6680a | 1191 | if ((cap & MCG_EXT_P) && MCG_EXT_CNT(cap) >= 9) |
94ad8474 | 1192 | rip_msr = MSR_IA32_MCG_EIP; |
1da177e4 | 1193 | |
ed7290d0 AK |
1194 | if (cap & MCG_SER_P) |
1195 | mce_ser = 1; | |
1196 | ||
0d7482e3 AK |
1197 | return 0; |
1198 | } | |
1199 | ||
8be91105 | 1200 | static void mce_init(void) |
0d7482e3 | 1201 | { |
e9eee03e | 1202 | mce_banks_t all_banks; |
0d7482e3 AK |
1203 | u64 cap; |
1204 | int i; | |
1205 | ||
b79109c3 AK |
1206 | /* |
1207 | * Log the machine checks left over from the previous reset. | |
1208 | */ | |
ee031c31 | 1209 | bitmap_fill(all_banks, MAX_NR_BANKS); |
5679af4c | 1210 | machine_check_poll(MCP_UC|(!mce_bootlog ? MCP_DONTLOG : 0), &all_banks); |
1da177e4 LT |
1211 | |
1212 | set_in_cr4(X86_CR4_MCE); | |
1213 | ||
0d7482e3 | 1214 | rdmsrl(MSR_IA32_MCG_CAP, cap); |
1da177e4 LT |
1215 | if (cap & MCG_CTL_P) |
1216 | wrmsr(MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff); | |
1217 | ||
1218 | for (i = 0; i < banks; i++) { | |
06b7a7a5 AK |
1219 | if (skip_bank_init(i)) |
1220 | continue; | |
0d7482e3 | 1221 | wrmsrl(MSR_IA32_MC0_CTL+4*i, bank[i]); |
1da177e4 | 1222 | wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0); |
d88203d1 | 1223 | } |
1da177e4 LT |
1224 | } |
1225 | ||
1226 | /* Add per CPU specific workarounds here */ | |
ec5b3d32 | 1227 | static void mce_cpu_quirks(struct cpuinfo_x86 *c) |
d88203d1 | 1228 | { |
1da177e4 | 1229 | /* This should be disabled by the BIOS, but isn't always */ |
911f6a7b | 1230 | if (c->x86_vendor == X86_VENDOR_AMD) { |
e9eee03e IM |
1231 | if (c->x86 == 15 && banks > 4) { |
1232 | /* | |
1233 | * disable GART TBL walk error reporting, which | |
1234 | * trips off incorrectly with the IOMMU & 3ware | |
1235 | * & Cerberus: | |
1236 | */ | |
0d7482e3 | 1237 | clear_bit(10, (unsigned long *)&bank[4]); |
e9eee03e IM |
1238 | } |
1239 | if (c->x86 <= 17 && mce_bootlog < 0) { | |
1240 | /* | |
1241 | * Lots of broken BIOS around that don't clear them | |
1242 | * by default and leave crap in there. Don't log: | |
1243 | */ | |
911f6a7b | 1244 | mce_bootlog = 0; |
e9eee03e | 1245 | } |
2e6f694f AK |
1246 | /* |
1247 | * Various K7s with broken bank 0 around. Always disable | |
1248 | * by default. | |
1249 | */ | |
203abd67 | 1250 | if (c->x86 == 6 && banks > 0) |
2e6f694f | 1251 | bank[0] = 0; |
1da177e4 | 1252 | } |
e583538f | 1253 | |
06b7a7a5 AK |
1254 | if (c->x86_vendor == X86_VENDOR_INTEL) { |
1255 | /* | |
1256 | * SDM documents that on family 6 bank 0 should not be written | |
1257 | * because it aliases to another special BIOS controlled | |
1258 | * register. | |
1259 | * But it's not aliased anymore on model 0x1a+ | |
1260 | * Don't ignore bank 0 completely because there could be a | |
1261 | * valid event later, merely don't write CTL0. | |
1262 | */ | |
1263 | ||
1264 | if (c->x86 == 6 && c->x86_model < 0x1A) | |
1265 | __set_bit(0, &dont_init_banks); | |
3c079792 AK |
1266 | |
1267 | /* | |
1268 | * All newer Intel systems support MCE broadcasting. Enable | |
1269 | * synchronization with a one second timeout. | |
1270 | */ | |
1271 | if ((c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xe)) && | |
1272 | monarch_timeout < 0) | |
1273 | monarch_timeout = USEC_PER_SEC; | |
06b7a7a5 | 1274 | } |
3c079792 AK |
1275 | if (monarch_timeout < 0) |
1276 | monarch_timeout = 0; | |
29b0f591 AK |
1277 | if (mce_bootlog != 0) |
1278 | mce_panic_timeout = 30; | |
d88203d1 | 1279 | } |
1da177e4 | 1280 | |
4efc0670 AK |
1281 | static void __cpuinit mce_ancient_init(struct cpuinfo_x86 *c) |
1282 | { | |
1283 | if (c->x86 != 5) | |
1284 | return; | |
1285 | switch (c->x86_vendor) { | |
1286 | case X86_VENDOR_INTEL: | |
c6978369 | 1287 | intel_p5_mcheck_init(c); |
4efc0670 AK |
1288 | break; |
1289 | case X86_VENDOR_CENTAUR: | |
1290 | winchip_mcheck_init(c); | |
1291 | break; | |
1292 | } | |
1293 | } | |
1294 | ||
cc3ca220 | 1295 | static void mce_cpu_features(struct cpuinfo_x86 *c) |
1da177e4 LT |
1296 | { |
1297 | switch (c->x86_vendor) { | |
1298 | case X86_VENDOR_INTEL: | |
1299 | mce_intel_feature_init(c); | |
1300 | break; | |
89b831ef JS |
1301 | case X86_VENDOR_AMD: |
1302 | mce_amd_feature_init(c); | |
1303 | break; | |
1da177e4 LT |
1304 | default: |
1305 | break; | |
1306 | } | |
1307 | } | |
1308 | ||
52d168e2 AK |
1309 | static void mce_init_timer(void) |
1310 | { | |
1311 | struct timer_list *t = &__get_cpu_var(mce_timer); | |
6298c512 | 1312 | int *n = &__get_cpu_var(next_interval); |
52d168e2 | 1313 | |
62fdac59 HS |
1314 | if (mce_ignore_ce) |
1315 | return; | |
1316 | ||
6298c512 AK |
1317 | *n = check_interval * HZ; |
1318 | if (!*n) | |
52d168e2 AK |
1319 | return; |
1320 | setup_timer(t, mcheck_timer, smp_processor_id()); | |
6298c512 | 1321 | t->expires = round_jiffies(jiffies + *n); |
52d168e2 AK |
1322 | add_timer(t); |
1323 | } | |
1324 | ||
d88203d1 | 1325 | /* |
1da177e4 | 1326 | * Called for each booted CPU to set up machine checks. |
e9eee03e | 1327 | * Must be called with preempt off: |
1da177e4 | 1328 | */ |
e6982c67 | 1329 | void __cpuinit mcheck_init(struct cpuinfo_x86 *c) |
1da177e4 | 1330 | { |
4efc0670 AK |
1331 | if (mce_disabled) |
1332 | return; | |
1333 | ||
1334 | mce_ancient_init(c); | |
1335 | ||
5b4408fd | 1336 | if (!mce_available(c)) |
1da177e4 LT |
1337 | return; |
1338 | ||
0d7482e3 | 1339 | if (mce_cap_init() < 0) { |
04b2b1a4 | 1340 | mce_disabled = 1; |
0d7482e3 AK |
1341 | return; |
1342 | } | |
1343 | mce_cpu_quirks(c); | |
1344 | ||
5d727926 AK |
1345 | machine_check_vector = do_machine_check; |
1346 | ||
8be91105 | 1347 | mce_init(); |
1da177e4 | 1348 | mce_cpu_features(c); |
52d168e2 | 1349 | mce_init_timer(); |
9b1beaf2 | 1350 | INIT_WORK(&__get_cpu_var(mce_work), mce_process_work); |
1da177e4 LT |
1351 | } |
1352 | ||
1353 | /* | |
1354 | * Character device to read and clear the MCE log. | |
1355 | */ | |
1356 | ||
f528e7ba | 1357 | static DEFINE_SPINLOCK(mce_state_lock); |
e9eee03e IM |
1358 | static int open_count; /* #times opened */ |
1359 | static int open_exclu; /* already open exclusive? */ | |
f528e7ba TH |
1360 | |
1361 | static int mce_open(struct inode *inode, struct file *file) | |
1362 | { | |
1363 | spin_lock(&mce_state_lock); | |
1364 | ||
1365 | if (open_exclu || (open_count && (file->f_flags & O_EXCL))) { | |
1366 | spin_unlock(&mce_state_lock); | |
e9eee03e | 1367 | |
f528e7ba TH |
1368 | return -EBUSY; |
1369 | } | |
1370 | ||
1371 | if (file->f_flags & O_EXCL) | |
1372 | open_exclu = 1; | |
1373 | open_count++; | |
1374 | ||
1375 | spin_unlock(&mce_state_lock); | |
1376 | ||
bd78432c | 1377 | return nonseekable_open(inode, file); |
f528e7ba TH |
1378 | } |
1379 | ||
1380 | static int mce_release(struct inode *inode, struct file *file) | |
1381 | { | |
1382 | spin_lock(&mce_state_lock); | |
1383 | ||
1384 | open_count--; | |
1385 | open_exclu = 0; | |
1386 | ||
1387 | spin_unlock(&mce_state_lock); | |
1388 | ||
1389 | return 0; | |
1390 | } | |
1391 | ||
d88203d1 TG |
1392 | static void collect_tscs(void *data) |
1393 | { | |
1da177e4 | 1394 | unsigned long *cpu_tsc = (unsigned long *)data; |
d88203d1 | 1395 | |
1da177e4 | 1396 | rdtscll(cpu_tsc[smp_processor_id()]); |
d88203d1 | 1397 | } |
1da177e4 | 1398 | |
e9eee03e IM |
1399 | static DEFINE_MUTEX(mce_read_mutex); |
1400 | ||
d88203d1 TG |
1401 | static ssize_t mce_read(struct file *filp, char __user *ubuf, size_t usize, |
1402 | loff_t *off) | |
1da177e4 | 1403 | { |
e9eee03e | 1404 | char __user *buf = ubuf; |
f0de53bb | 1405 | unsigned long *cpu_tsc; |
ef41df43 | 1406 | unsigned prev, next; |
1da177e4 LT |
1407 | int i, err; |
1408 | ||
6bca67f9 | 1409 | cpu_tsc = kmalloc(nr_cpu_ids * sizeof(long), GFP_KERNEL); |
f0de53bb AK |
1410 | if (!cpu_tsc) |
1411 | return -ENOMEM; | |
1412 | ||
8c8b8859 | 1413 | mutex_lock(&mce_read_mutex); |
1da177e4 LT |
1414 | next = rcu_dereference(mcelog.next); |
1415 | ||
1416 | /* Only supports full reads right now */ | |
d88203d1 | 1417 | if (*off != 0 || usize < MCE_LOG_LEN*sizeof(struct mce)) { |
8c8b8859 | 1418 | mutex_unlock(&mce_read_mutex); |
f0de53bb | 1419 | kfree(cpu_tsc); |
e9eee03e | 1420 | |
1da177e4 LT |
1421 | return -EINVAL; |
1422 | } | |
1423 | ||
1424 | err = 0; | |
ef41df43 HY |
1425 | prev = 0; |
1426 | do { | |
1427 | for (i = prev; i < next; i++) { | |
1428 | unsigned long start = jiffies; | |
1429 | ||
1430 | while (!mcelog.entry[i].finished) { | |
1431 | if (time_after_eq(jiffies, start + 2)) { | |
1432 | memset(mcelog.entry + i, 0, | |
1433 | sizeof(struct mce)); | |
1434 | goto timeout; | |
1435 | } | |
1436 | cpu_relax(); | |
673242c1 | 1437 | } |
ef41df43 HY |
1438 | smp_rmb(); |
1439 | err |= copy_to_user(buf, mcelog.entry + i, | |
1440 | sizeof(struct mce)); | |
1441 | buf += sizeof(struct mce); | |
1442 | timeout: | |
1443 | ; | |
673242c1 | 1444 | } |
1da177e4 | 1445 | |
ef41df43 HY |
1446 | memset(mcelog.entry + prev, 0, |
1447 | (next - prev) * sizeof(struct mce)); | |
1448 | prev = next; | |
1449 | next = cmpxchg(&mcelog.next, prev, 0); | |
1450 | } while (next != prev); | |
1da177e4 | 1451 | |
b2b18660 | 1452 | synchronize_sched(); |
1da177e4 | 1453 | |
d88203d1 TG |
1454 | /* |
1455 | * Collect entries that were still getting written before the | |
1456 | * synchronize. | |
1457 | */ | |
15c8b6c1 | 1458 | on_each_cpu(collect_tscs, cpu_tsc, 1); |
e9eee03e | 1459 | |
d88203d1 TG |
1460 | for (i = next; i < MCE_LOG_LEN; i++) { |
1461 | if (mcelog.entry[i].finished && | |
1462 | mcelog.entry[i].tsc < cpu_tsc[mcelog.entry[i].cpu]) { | |
1463 | err |= copy_to_user(buf, mcelog.entry+i, | |
1464 | sizeof(struct mce)); | |
1da177e4 LT |
1465 | smp_rmb(); |
1466 | buf += sizeof(struct mce); | |
1467 | memset(&mcelog.entry[i], 0, sizeof(struct mce)); | |
1468 | } | |
d88203d1 | 1469 | } |
8c8b8859 | 1470 | mutex_unlock(&mce_read_mutex); |
f0de53bb | 1471 | kfree(cpu_tsc); |
e9eee03e | 1472 | |
d88203d1 | 1473 | return err ? -EFAULT : buf - ubuf; |
1da177e4 LT |
1474 | } |
1475 | ||
e02e68d3 TH |
1476 | static unsigned int mce_poll(struct file *file, poll_table *wait) |
1477 | { | |
1478 | poll_wait(file, &mce_wait, wait); | |
1479 | if (rcu_dereference(mcelog.next)) | |
1480 | return POLLIN | POLLRDNORM; | |
1481 | return 0; | |
1482 | } | |
1483 | ||
c68461b6 | 1484 | static long mce_ioctl(struct file *f, unsigned int cmd, unsigned long arg) |
1da177e4 LT |
1485 | { |
1486 | int __user *p = (int __user *)arg; | |
d88203d1 | 1487 | |
1da177e4 | 1488 | if (!capable(CAP_SYS_ADMIN)) |
d88203d1 | 1489 | return -EPERM; |
e9eee03e | 1490 | |
1da177e4 | 1491 | switch (cmd) { |
d88203d1 | 1492 | case MCE_GET_RECORD_LEN: |
1da177e4 LT |
1493 | return put_user(sizeof(struct mce), p); |
1494 | case MCE_GET_LOG_LEN: | |
d88203d1 | 1495 | return put_user(MCE_LOG_LEN, p); |
1da177e4 LT |
1496 | case MCE_GETCLEAR_FLAGS: { |
1497 | unsigned flags; | |
d88203d1 TG |
1498 | |
1499 | do { | |
1da177e4 | 1500 | flags = mcelog.flags; |
d88203d1 | 1501 | } while (cmpxchg(&mcelog.flags, flags, 0) != flags); |
e9eee03e | 1502 | |
d88203d1 | 1503 | return put_user(flags, p); |
1da177e4 LT |
1504 | } |
1505 | default: | |
d88203d1 TG |
1506 | return -ENOTTY; |
1507 | } | |
1da177e4 LT |
1508 | } |
1509 | ||
a1ff41bf | 1510 | /* Modified in mce-inject.c, so not static or const */ |
ea149b36 | 1511 | struct file_operations mce_chrdev_ops = { |
e9eee03e IM |
1512 | .open = mce_open, |
1513 | .release = mce_release, | |
1514 | .read = mce_read, | |
1515 | .poll = mce_poll, | |
1516 | .unlocked_ioctl = mce_ioctl, | |
1da177e4 | 1517 | }; |
ea149b36 | 1518 | EXPORT_SYMBOL_GPL(mce_chrdev_ops); |
1da177e4 LT |
1519 | |
1520 | static struct miscdevice mce_log_device = { | |
1521 | MISC_MCELOG_MINOR, | |
1522 | "mcelog", | |
1523 | &mce_chrdev_ops, | |
1524 | }; | |
1525 | ||
13503fa9 | 1526 | /* |
62fdac59 HS |
1527 | * mce=off Disables machine check |
1528 | * mce=no_cmci Disables CMCI | |
1529 | * mce=dont_log_ce Clears corrected events silently, no log created for CEs. | |
1530 | * mce=ignore_ce Disables polling and CMCI, corrected events are not cleared. | |
3c079792 AK |
1531 | * mce=TOLERANCELEVEL[,monarchtimeout] (number, see above) |
1532 | * monarchtimeout is how long to wait for other CPUs on machine | |
1533 | * check, or 0 to not wait | |
13503fa9 HS |
1534 | * mce=bootlog Log MCEs from before booting. Disabled by default on AMD. |
1535 | * mce=nobootlog Don't log MCEs from before booting. | |
1536 | */ | |
1da177e4 LT |
1537 | static int __init mcheck_enable(char *str) |
1538 | { | |
4efc0670 AK |
1539 | if (*str == 0) |
1540 | enable_p5_mce(); | |
1541 | if (*str == '=') | |
1542 | str++; | |
1da177e4 | 1543 | if (!strcmp(str, "off")) |
04b2b1a4 | 1544 | mce_disabled = 1; |
62fdac59 HS |
1545 | else if (!strcmp(str, "no_cmci")) |
1546 | mce_cmci_disabled = 1; | |
1547 | else if (!strcmp(str, "dont_log_ce")) | |
1548 | mce_dont_log_ce = 1; | |
1549 | else if (!strcmp(str, "ignore_ce")) | |
1550 | mce_ignore_ce = 1; | |
13503fa9 HS |
1551 | else if (!strcmp(str, "bootlog") || !strcmp(str, "nobootlog")) |
1552 | mce_bootlog = (str[0] == 'b'); | |
3c079792 | 1553 | else if (isdigit(str[0])) { |
8c566ef5 | 1554 | get_option(&str, &tolerant); |
3c079792 AK |
1555 | if (*str == ',') { |
1556 | ++str; | |
1557 | get_option(&str, &monarch_timeout); | |
1558 | } | |
1559 | } else { | |
4efc0670 | 1560 | printk(KERN_INFO "mce argument %s ignored. Please use /sys\n", |
13503fa9 HS |
1561 | str); |
1562 | return 0; | |
1563 | } | |
9b41046c | 1564 | return 1; |
1da177e4 | 1565 | } |
4efc0670 | 1566 | __setup("mce", mcheck_enable); |
1da177e4 | 1567 | |
d88203d1 | 1568 | /* |
1da177e4 | 1569 | * Sysfs support |
d88203d1 | 1570 | */ |
1da177e4 | 1571 | |
973a2dd1 AK |
1572 | /* |
1573 | * Disable machine checks on suspend and shutdown. We can't really handle | |
1574 | * them later. | |
1575 | */ | |
1576 | static int mce_disable(void) | |
1577 | { | |
1578 | int i; | |
1579 | ||
06b7a7a5 AK |
1580 | for (i = 0; i < banks; i++) { |
1581 | if (!skip_bank_init(i)) | |
1582 | wrmsrl(MSR_IA32_MC0_CTL + i*4, 0); | |
1583 | } | |
973a2dd1 AK |
1584 | return 0; |
1585 | } | |
1586 | ||
1587 | static int mce_suspend(struct sys_device *dev, pm_message_t state) | |
1588 | { | |
1589 | return mce_disable(); | |
1590 | } | |
1591 | ||
1592 | static int mce_shutdown(struct sys_device *dev) | |
1593 | { | |
1594 | return mce_disable(); | |
1595 | } | |
1596 | ||
e9eee03e IM |
1597 | /* |
1598 | * On resume clear all MCE state. Don't want to see leftovers from the BIOS. | |
1599 | * Only one CPU is active at this time, the others get re-added later using | |
1600 | * CPU hotplug: | |
1601 | */ | |
1da177e4 LT |
1602 | static int mce_resume(struct sys_device *dev) |
1603 | { | |
8be91105 | 1604 | mce_init(); |
6ec68bff | 1605 | mce_cpu_features(¤t_cpu_data); |
e9eee03e | 1606 | |
1da177e4 LT |
1607 | return 0; |
1608 | } | |
1609 | ||
52d168e2 AK |
1610 | static void mce_cpu_restart(void *data) |
1611 | { | |
1612 | del_timer_sync(&__get_cpu_var(mce_timer)); | |
33edbf02 HS |
1613 | if (!mce_available(¤t_cpu_data)) |
1614 | return; | |
1615 | mce_init(); | |
52d168e2 AK |
1616 | mce_init_timer(); |
1617 | } | |
1618 | ||
1da177e4 | 1619 | /* Reinit MCEs after user configuration changes */ |
d88203d1 TG |
1620 | static void mce_restart(void) |
1621 | { | |
52d168e2 | 1622 | on_each_cpu(mce_cpu_restart, NULL, 1); |
1da177e4 LT |
1623 | } |
1624 | ||
9af43b54 HS |
1625 | /* Toggle features for corrected errors */ |
1626 | static void mce_disable_ce(void *all) | |
1627 | { | |
1628 | if (!mce_available(¤t_cpu_data)) | |
1629 | return; | |
1630 | if (all) | |
1631 | del_timer_sync(&__get_cpu_var(mce_timer)); | |
1632 | cmci_clear(); | |
1633 | } | |
1634 | ||
1635 | static void mce_enable_ce(void *all) | |
1636 | { | |
1637 | if (!mce_available(¤t_cpu_data)) | |
1638 | return; | |
1639 | cmci_reenable(); | |
1640 | cmci_recheck(); | |
1641 | if (all) | |
1642 | mce_init_timer(); | |
1643 | } | |
1644 | ||
1da177e4 | 1645 | static struct sysdev_class mce_sysclass = { |
e9eee03e IM |
1646 | .suspend = mce_suspend, |
1647 | .shutdown = mce_shutdown, | |
1648 | .resume = mce_resume, | |
1649 | .name = "machinecheck", | |
1da177e4 LT |
1650 | }; |
1651 | ||
cb491fca | 1652 | DEFINE_PER_CPU(struct sys_device, mce_dev); |
e9eee03e IM |
1653 | |
1654 | __cpuinitdata | |
1655 | void (*threshold_cpu_callback)(unsigned long action, unsigned int cpu); | |
1da177e4 | 1656 | |
0d7482e3 AK |
1657 | static struct sysdev_attribute *bank_attrs; |
1658 | ||
1659 | static ssize_t show_bank(struct sys_device *s, struct sysdev_attribute *attr, | |
1660 | char *buf) | |
1661 | { | |
1662 | u64 b = bank[attr - bank_attrs]; | |
e9eee03e | 1663 | |
f6d1826d | 1664 | return sprintf(buf, "%llx\n", b); |
0d7482e3 AK |
1665 | } |
1666 | ||
1667 | static ssize_t set_bank(struct sys_device *s, struct sysdev_attribute *attr, | |
9319cec8 | 1668 | const char *buf, size_t size) |
0d7482e3 | 1669 | { |
9319cec8 | 1670 | u64 new; |
e9eee03e | 1671 | |
9319cec8 | 1672 | if (strict_strtoull(buf, 0, &new) < 0) |
0d7482e3 | 1673 | return -EINVAL; |
e9eee03e | 1674 | |
0d7482e3 AK |
1675 | bank[attr - bank_attrs] = new; |
1676 | mce_restart(); | |
e9eee03e | 1677 | |
9319cec8 | 1678 | return size; |
0d7482e3 | 1679 | } |
a98f0dd3 | 1680 | |
e9eee03e IM |
1681 | static ssize_t |
1682 | show_trigger(struct sys_device *s, struct sysdev_attribute *attr, char *buf) | |
a98f0dd3 | 1683 | { |
1020bcbc | 1684 | strcpy(buf, mce_helper); |
a98f0dd3 | 1685 | strcat(buf, "\n"); |
1020bcbc | 1686 | return strlen(mce_helper) + 1; |
a98f0dd3 AK |
1687 | } |
1688 | ||
4a0b2b4d | 1689 | static ssize_t set_trigger(struct sys_device *s, struct sysdev_attribute *attr, |
e9eee03e | 1690 | const char *buf, size_t siz) |
a98f0dd3 AK |
1691 | { |
1692 | char *p; | |
1693 | int len; | |
e9eee03e | 1694 | |
1020bcbc HS |
1695 | strncpy(mce_helper, buf, sizeof(mce_helper)); |
1696 | mce_helper[sizeof(mce_helper)-1] = 0; | |
1697 | len = strlen(mce_helper); | |
1698 | p = strchr(mce_helper, '\n'); | |
e9eee03e IM |
1699 | |
1700 | if (*p) | |
1701 | *p = 0; | |
1702 | ||
a98f0dd3 AK |
1703 | return len; |
1704 | } | |
1705 | ||
9af43b54 HS |
1706 | static ssize_t set_ignore_ce(struct sys_device *s, |
1707 | struct sysdev_attribute *attr, | |
1708 | const char *buf, size_t size) | |
1709 | { | |
1710 | u64 new; | |
1711 | ||
1712 | if (strict_strtoull(buf, 0, &new) < 0) | |
1713 | return -EINVAL; | |
1714 | ||
1715 | if (mce_ignore_ce ^ !!new) { | |
1716 | if (new) { | |
1717 | /* disable ce features */ | |
1718 | on_each_cpu(mce_disable_ce, (void *)1, 1); | |
1719 | mce_ignore_ce = 1; | |
1720 | } else { | |
1721 | /* enable ce features */ | |
1722 | mce_ignore_ce = 0; | |
1723 | on_each_cpu(mce_enable_ce, (void *)1, 1); | |
1724 | } | |
1725 | } | |
1726 | return size; | |
1727 | } | |
1728 | ||
1729 | static ssize_t set_cmci_disabled(struct sys_device *s, | |
1730 | struct sysdev_attribute *attr, | |
1731 | const char *buf, size_t size) | |
1732 | { | |
1733 | u64 new; | |
1734 | ||
1735 | if (strict_strtoull(buf, 0, &new) < 0) | |
1736 | return -EINVAL; | |
1737 | ||
1738 | if (mce_cmci_disabled ^ !!new) { | |
1739 | if (new) { | |
1740 | /* disable cmci */ | |
1741 | on_each_cpu(mce_disable_ce, NULL, 1); | |
1742 | mce_cmci_disabled = 1; | |
1743 | } else { | |
1744 | /* enable cmci */ | |
1745 | mce_cmci_disabled = 0; | |
1746 | on_each_cpu(mce_enable_ce, NULL, 1); | |
1747 | } | |
1748 | } | |
1749 | return size; | |
1750 | } | |
1751 | ||
b56f642d AK |
1752 | static ssize_t store_int_with_restart(struct sys_device *s, |
1753 | struct sysdev_attribute *attr, | |
1754 | const char *buf, size_t size) | |
1755 | { | |
1756 | ssize_t ret = sysdev_store_int(s, attr, buf, size); | |
1757 | mce_restart(); | |
1758 | return ret; | |
1759 | } | |
1760 | ||
a98f0dd3 | 1761 | static SYSDEV_ATTR(trigger, 0644, show_trigger, set_trigger); |
d95d62c0 | 1762 | static SYSDEV_INT_ATTR(tolerant, 0644, tolerant); |
3c079792 | 1763 | static SYSDEV_INT_ATTR(monarch_timeout, 0644, monarch_timeout); |
9af43b54 | 1764 | static SYSDEV_INT_ATTR(dont_log_ce, 0644, mce_dont_log_ce); |
e9eee03e | 1765 | |
b56f642d AK |
1766 | static struct sysdev_ext_attribute attr_check_interval = { |
1767 | _SYSDEV_ATTR(check_interval, 0644, sysdev_show_int, | |
1768 | store_int_with_restart), | |
1769 | &check_interval | |
1770 | }; | |
e9eee03e | 1771 | |
9af43b54 HS |
1772 | static struct sysdev_ext_attribute attr_ignore_ce = { |
1773 | _SYSDEV_ATTR(ignore_ce, 0644, sysdev_show_int, set_ignore_ce), | |
1774 | &mce_ignore_ce | |
1775 | }; | |
1776 | ||
1777 | static struct sysdev_ext_attribute attr_cmci_disabled = { | |
74b602c7 | 1778 | _SYSDEV_ATTR(cmci_disabled, 0644, sysdev_show_int, set_cmci_disabled), |
9af43b54 HS |
1779 | &mce_cmci_disabled |
1780 | }; | |
1781 | ||
cb491fca | 1782 | static struct sysdev_attribute *mce_attrs[] = { |
9af43b54 HS |
1783 | &attr_tolerant.attr, |
1784 | &attr_check_interval.attr, | |
1785 | &attr_trigger, | |
3c079792 | 1786 | &attr_monarch_timeout.attr, |
9af43b54 HS |
1787 | &attr_dont_log_ce.attr, |
1788 | &attr_ignore_ce.attr, | |
1789 | &attr_cmci_disabled.attr, | |
a98f0dd3 AK |
1790 | NULL |
1791 | }; | |
1da177e4 | 1792 | |
cb491fca | 1793 | static cpumask_var_t mce_dev_initialized; |
bae19fe0 | 1794 | |
e9eee03e | 1795 | /* Per cpu sysdev init. All of the cpus still share the same ctrl bank: */ |
91c6d400 | 1796 | static __cpuinit int mce_create_device(unsigned int cpu) |
1da177e4 LT |
1797 | { |
1798 | int err; | |
b1f49f95 | 1799 | int i, j; |
92cb7612 | 1800 | |
90367556 | 1801 | if (!mce_available(&boot_cpu_data)) |
91c6d400 AK |
1802 | return -EIO; |
1803 | ||
cb491fca IM |
1804 | memset(&per_cpu(mce_dev, cpu).kobj, 0, sizeof(struct kobject)); |
1805 | per_cpu(mce_dev, cpu).id = cpu; | |
1806 | per_cpu(mce_dev, cpu).cls = &mce_sysclass; | |
91c6d400 | 1807 | |
cb491fca | 1808 | err = sysdev_register(&per_cpu(mce_dev, cpu)); |
d435d862 AM |
1809 | if (err) |
1810 | return err; | |
1811 | ||
cb491fca IM |
1812 | for (i = 0; mce_attrs[i]; i++) { |
1813 | err = sysdev_create_file(&per_cpu(mce_dev, cpu), mce_attrs[i]); | |
d435d862 AM |
1814 | if (err) |
1815 | goto error; | |
1816 | } | |
b1f49f95 | 1817 | for (j = 0; j < banks; j++) { |
cb491fca | 1818 | err = sysdev_create_file(&per_cpu(mce_dev, cpu), |
b1f49f95 | 1819 | &bank_attrs[j]); |
0d7482e3 AK |
1820 | if (err) |
1821 | goto error2; | |
1822 | } | |
cb491fca | 1823 | cpumask_set_cpu(cpu, mce_dev_initialized); |
91c6d400 | 1824 | |
d435d862 | 1825 | return 0; |
0d7482e3 | 1826 | error2: |
b1f49f95 HS |
1827 | while (--j >= 0) |
1828 | sysdev_remove_file(&per_cpu(mce_dev, cpu), &bank_attrs[j]); | |
d435d862 | 1829 | error: |
cb491fca IM |
1830 | while (--i >= 0) |
1831 | sysdev_remove_file(&per_cpu(mce_dev, cpu), mce_attrs[i]); | |
1832 | ||
1833 | sysdev_unregister(&per_cpu(mce_dev, cpu)); | |
d435d862 | 1834 | |
91c6d400 AK |
1835 | return err; |
1836 | } | |
1837 | ||
2d9cd6c2 | 1838 | static __cpuinit void mce_remove_device(unsigned int cpu) |
91c6d400 | 1839 | { |
73ca5358 SL |
1840 | int i; |
1841 | ||
cb491fca | 1842 | if (!cpumask_test_cpu(cpu, mce_dev_initialized)) |
bae19fe0 AH |
1843 | return; |
1844 | ||
cb491fca IM |
1845 | for (i = 0; mce_attrs[i]; i++) |
1846 | sysdev_remove_file(&per_cpu(mce_dev, cpu), mce_attrs[i]); | |
1847 | ||
0d7482e3 | 1848 | for (i = 0; i < banks; i++) |
cb491fca IM |
1849 | sysdev_remove_file(&per_cpu(mce_dev, cpu), &bank_attrs[i]); |
1850 | ||
1851 | sysdev_unregister(&per_cpu(mce_dev, cpu)); | |
1852 | cpumask_clear_cpu(cpu, mce_dev_initialized); | |
91c6d400 | 1853 | } |
91c6d400 | 1854 | |
d6b75584 | 1855 | /* Make sure there are no machine checks on offlined CPUs. */ |
ec5b3d32 | 1856 | static void mce_disable_cpu(void *h) |
d6b75584 | 1857 | { |
88ccbedd | 1858 | unsigned long action = *(unsigned long *)h; |
cb491fca | 1859 | int i; |
d6b75584 AK |
1860 | |
1861 | if (!mce_available(¤t_cpu_data)) | |
1862 | return; | |
88ccbedd AK |
1863 | if (!(action & CPU_TASKS_FROZEN)) |
1864 | cmci_clear(); | |
06b7a7a5 AK |
1865 | for (i = 0; i < banks; i++) { |
1866 | if (!skip_bank_init(i)) | |
1867 | wrmsrl(MSR_IA32_MC0_CTL + i*4, 0); | |
1868 | } | |
d6b75584 AK |
1869 | } |
1870 | ||
ec5b3d32 | 1871 | static void mce_reenable_cpu(void *h) |
d6b75584 | 1872 | { |
88ccbedd | 1873 | unsigned long action = *(unsigned long *)h; |
e9eee03e | 1874 | int i; |
d6b75584 AK |
1875 | |
1876 | if (!mce_available(¤t_cpu_data)) | |
1877 | return; | |
e9eee03e | 1878 | |
88ccbedd AK |
1879 | if (!(action & CPU_TASKS_FROZEN)) |
1880 | cmci_reenable(); | |
06b7a7a5 AK |
1881 | for (i = 0; i < banks; i++) { |
1882 | if (!skip_bank_init(i)) | |
1883 | wrmsrl(MSR_IA32_MC0_CTL + i*4, bank[i]); | |
1884 | } | |
d6b75584 AK |
1885 | } |
1886 | ||
91c6d400 | 1887 | /* Get notified when a cpu comes on/off. Be hotplug friendly. */ |
e9eee03e IM |
1888 | static int __cpuinit |
1889 | mce_cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
91c6d400 AK |
1890 | { |
1891 | unsigned int cpu = (unsigned long)hcpu; | |
52d168e2 | 1892 | struct timer_list *t = &per_cpu(mce_timer, cpu); |
91c6d400 AK |
1893 | |
1894 | switch (action) { | |
bae19fe0 AH |
1895 | case CPU_ONLINE: |
1896 | case CPU_ONLINE_FROZEN: | |
1897 | mce_create_device(cpu); | |
8735728e RW |
1898 | if (threshold_cpu_callback) |
1899 | threshold_cpu_callback(action, cpu); | |
91c6d400 | 1900 | break; |
91c6d400 | 1901 | case CPU_DEAD: |
8bb78442 | 1902 | case CPU_DEAD_FROZEN: |
8735728e RW |
1903 | if (threshold_cpu_callback) |
1904 | threshold_cpu_callback(action, cpu); | |
91c6d400 AK |
1905 | mce_remove_device(cpu); |
1906 | break; | |
52d168e2 AK |
1907 | case CPU_DOWN_PREPARE: |
1908 | case CPU_DOWN_PREPARE_FROZEN: | |
1909 | del_timer_sync(t); | |
88ccbedd | 1910 | smp_call_function_single(cpu, mce_disable_cpu, &action, 1); |
52d168e2 AK |
1911 | break; |
1912 | case CPU_DOWN_FAILED: | |
1913 | case CPU_DOWN_FAILED_FROZEN: | |
6298c512 AK |
1914 | t->expires = round_jiffies(jiffies + |
1915 | __get_cpu_var(next_interval)); | |
52d168e2 | 1916 | add_timer_on(t, cpu); |
88ccbedd AK |
1917 | smp_call_function_single(cpu, mce_reenable_cpu, &action, 1); |
1918 | break; | |
1919 | case CPU_POST_DEAD: | |
1920 | /* intentionally ignoring frozen here */ | |
1921 | cmci_rediscover(cpu); | |
52d168e2 | 1922 | break; |
91c6d400 | 1923 | } |
bae19fe0 | 1924 | return NOTIFY_OK; |
91c6d400 AK |
1925 | } |
1926 | ||
1e35669d | 1927 | static struct notifier_block mce_cpu_notifier __cpuinitdata = { |
91c6d400 AK |
1928 | .notifier_call = mce_cpu_callback, |
1929 | }; | |
1930 | ||
0d7482e3 AK |
1931 | static __init int mce_init_banks(void) |
1932 | { | |
1933 | int i; | |
1934 | ||
1935 | bank_attrs = kzalloc(sizeof(struct sysdev_attribute) * banks, | |
1936 | GFP_KERNEL); | |
1937 | if (!bank_attrs) | |
1938 | return -ENOMEM; | |
1939 | ||
1940 | for (i = 0; i < banks; i++) { | |
1941 | struct sysdev_attribute *a = &bank_attrs[i]; | |
e9eee03e IM |
1942 | |
1943 | a->attr.name = kasprintf(GFP_KERNEL, "bank%d", i); | |
0d7482e3 AK |
1944 | if (!a->attr.name) |
1945 | goto nomem; | |
e9eee03e IM |
1946 | |
1947 | a->attr.mode = 0644; | |
1948 | a->show = show_bank; | |
1949 | a->store = set_bank; | |
0d7482e3 AK |
1950 | } |
1951 | return 0; | |
1952 | ||
1953 | nomem: | |
1954 | while (--i >= 0) | |
1955 | kfree(bank_attrs[i].attr.name); | |
1956 | kfree(bank_attrs); | |
1957 | bank_attrs = NULL; | |
e9eee03e | 1958 | |
0d7482e3 AK |
1959 | return -ENOMEM; |
1960 | } | |
1961 | ||
91c6d400 AK |
1962 | static __init int mce_init_device(void) |
1963 | { | |
1964 | int err; | |
1965 | int i = 0; | |
1966 | ||
1da177e4 LT |
1967 | if (!mce_available(&boot_cpu_data)) |
1968 | return -EIO; | |
0d7482e3 | 1969 | |
e92fae06 | 1970 | zalloc_cpumask_var(&mce_dev_initialized, GFP_KERNEL); |
996867d0 | 1971 | |
0d7482e3 AK |
1972 | err = mce_init_banks(); |
1973 | if (err) | |
1974 | return err; | |
1975 | ||
1da177e4 | 1976 | err = sysdev_class_register(&mce_sysclass); |
d435d862 AM |
1977 | if (err) |
1978 | return err; | |
91c6d400 AK |
1979 | |
1980 | for_each_online_cpu(i) { | |
d435d862 AM |
1981 | err = mce_create_device(i); |
1982 | if (err) | |
1983 | return err; | |
91c6d400 AK |
1984 | } |
1985 | ||
be6b5a35 | 1986 | register_hotcpu_notifier(&mce_cpu_notifier); |
1da177e4 | 1987 | misc_register(&mce_log_device); |
e9eee03e | 1988 | |
1da177e4 | 1989 | return err; |
1da177e4 | 1990 | } |
91c6d400 | 1991 | |
1da177e4 | 1992 | device_initcall(mce_init_device); |
a988d334 | 1993 | |
d7c3c9a6 AK |
1994 | /* |
1995 | * Old style boot options parsing. Only for compatibility. | |
1996 | */ | |
1997 | static int __init mcheck_disable(char *str) | |
1998 | { | |
1999 | mce_disabled = 1; | |
2000 | return 1; | |
2001 | } | |
2002 | __setup("nomce", mcheck_disable); |