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 | */ |
c767a54b JP |
10 | |
11 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
12 | ||
e9eee03e IM |
13 | #include <linux/thread_info.h> |
14 | #include <linux/capability.h> | |
15 | #include <linux/miscdevice.h> | |
16 | #include <linux/ratelimit.h> | |
17 | #include <linux/kallsyms.h> | |
18 | #include <linux/rcupdate.h> | |
e9eee03e | 19 | #include <linux/kobject.h> |
14a02530 | 20 | #include <linux/uaccess.h> |
e9eee03e IM |
21 | #include <linux/kdebug.h> |
22 | #include <linux/kernel.h> | |
23 | #include <linux/percpu.h> | |
1da177e4 | 24 | #include <linux/string.h> |
8a25a2fd | 25 | #include <linux/device.h> |
f3c6ea1b | 26 | #include <linux/syscore_ops.h> |
3c079792 | 27 | #include <linux/delay.h> |
8c566ef5 | 28 | #include <linux/ctype.h> |
e9eee03e | 29 | #include <linux/sched.h> |
0d7482e3 | 30 | #include <linux/sysfs.h> |
e9eee03e | 31 | #include <linux/types.h> |
5a0e3ad6 | 32 | #include <linux/slab.h> |
e9eee03e IM |
33 | #include <linux/init.h> |
34 | #include <linux/kmod.h> | |
35 | #include <linux/poll.h> | |
3c079792 | 36 | #include <linux/nmi.h> |
e9eee03e | 37 | #include <linux/cpu.h> |
14a02530 | 38 | #include <linux/smp.h> |
e9eee03e | 39 | #include <linux/fs.h> |
9b1beaf2 | 40 | #include <linux/mm.h> |
5be9ed25 | 41 | #include <linux/debugfs.h> |
b77e70bf | 42 | #include <linux/irq_work.h> |
69c60c88 | 43 | #include <linux/export.h> |
e9eee03e | 44 | |
d88203d1 | 45 | #include <asm/processor.h> |
e9eee03e IM |
46 | #include <asm/mce.h> |
47 | #include <asm/msr.h> | |
1da177e4 | 48 | |
bd19a5e6 | 49 | #include "mce-internal.h" |
711c2e48 | 50 | |
93b62c3c | 51 | static DEFINE_MUTEX(mce_chrdev_read_mutex); |
2aa2b50d | 52 | |
f56e8a07 | 53 | #define rcu_dereference_check_mce(p) \ |
ec8c27e0 | 54 | rcu_dereference_index_check((p), \ |
f56e8a07 | 55 | rcu_read_lock_sched_held() || \ |
93b62c3c | 56 | lockdep_is_held(&mce_chrdev_read_mutex)) |
f56e8a07 | 57 | |
8968f9d3 HS |
58 | #define CREATE_TRACE_POINTS |
59 | #include <trace/events/mce.h> | |
60 | ||
4e5b3e69 | 61 | int mce_disabled __read_mostly; |
04b2b1a4 | 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; | |
4e5b3e69 HS |
78 | static int rip_msr __read_mostly; |
79 | static int mce_bootlog __read_mostly = -1; | |
80 | static int monarch_timeout __read_mostly = -1; | |
81 | static int mce_panic_timeout __read_mostly; | |
82 | static int mce_dont_log_ce __read_mostly; | |
83 | int mce_cmci_disabled __read_mostly; | |
84 | int mce_ignore_ce __read_mostly; | |
85 | int mce_ser __read_mostly; | |
a98f0dd3 | 86 | |
cebe1820 AK |
87 | struct mce_bank *mce_banks __read_mostly; |
88 | ||
1020bcbc HS |
89 | /* User mode helper program triggered by machine check event */ |
90 | static unsigned long mce_need_notify; | |
91 | static char mce_helper[128]; | |
92 | static char *mce_helper_argv[2] = { mce_helper, NULL }; | |
1da177e4 | 93 | |
93b62c3c HS |
94 | static DECLARE_WAIT_QUEUE_HEAD(mce_chrdev_wait); |
95 | ||
3c079792 AK |
96 | static DEFINE_PER_CPU(struct mce, mces_seen); |
97 | static int cpu_missing; | |
98 | ||
ee031c31 AK |
99 | /* MCA banks polled by the period polling timer for corrected events */ |
100 | DEFINE_PER_CPU(mce_banks_t, mce_poll_banks) = { | |
101 | [0 ... BITS_TO_LONGS(MAX_NR_BANKS)-1] = ~0UL | |
102 | }; | |
103 | ||
9b1beaf2 AK |
104 | static DEFINE_PER_CPU(struct work_struct, mce_work); |
105 | ||
3653ada5 BP |
106 | /* |
107 | * CPU/chipset specific EDAC code can register a notifier call here to print | |
108 | * MCE errors in a human-readable form. | |
109 | */ | |
110 | ATOMIC_NOTIFIER_HEAD(x86_mce_decoder_chain); | |
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); | |
8ee08347 | 122 | m->socketid = cpu_data(m->extcpu).phys_proc_id; |
8ee08347 AK |
123 | m->apicid = cpu_data(m->extcpu).initial_apicid; |
124 | rdmsrl(MSR_IA32_MCG_CAP, m->mcgcap); | |
b5f2fa4e AK |
125 | } |
126 | ||
ea149b36 AK |
127 | DEFINE_PER_CPU(struct mce, injectm); |
128 | EXPORT_PER_CPU_SYMBOL_GPL(injectm); | |
129 | ||
1da177e4 LT |
130 | /* |
131 | * Lockless MCE logging infrastructure. | |
132 | * This avoids deadlocks on printk locks without having to break locks. Also | |
133 | * separate MCEs from kernel messages to avoid bogus bug reports. | |
134 | */ | |
135 | ||
231fd906 | 136 | static struct mce_log mcelog = { |
f6fb0ac0 AK |
137 | .signature = MCE_LOG_SIGNATURE, |
138 | .len = MCE_LOG_LEN, | |
139 | .recordlen = sizeof(struct mce), | |
d88203d1 | 140 | }; |
1da177e4 LT |
141 | |
142 | void mce_log(struct mce *mce) | |
143 | { | |
144 | unsigned next, entry; | |
f0cb5452 | 145 | int ret = 0; |
e9eee03e | 146 | |
8968f9d3 HS |
147 | /* Emit the trace record: */ |
148 | trace_mce_record(mce); | |
149 | ||
f0cb5452 BP |
150 | ret = atomic_notifier_call_chain(&x86_mce_decoder_chain, 0, mce); |
151 | if (ret == NOTIFY_STOP) | |
152 | return; | |
153 | ||
1da177e4 | 154 | mce->finished = 0; |
7644143c | 155 | wmb(); |
1da177e4 | 156 | for (;;) { |
f56e8a07 | 157 | entry = rcu_dereference_check_mce(mcelog.next); |
673242c1 | 158 | for (;;) { |
696e409d | 159 | |
e9eee03e IM |
160 | /* |
161 | * When the buffer fills up discard new entries. | |
162 | * Assume that the earlier errors are the more | |
163 | * interesting ones: | |
164 | */ | |
673242c1 | 165 | if (entry >= MCE_LOG_LEN) { |
14a02530 HS |
166 | set_bit(MCE_OVERFLOW, |
167 | (unsigned long *)&mcelog.flags); | |
673242c1 AK |
168 | return; |
169 | } | |
e9eee03e | 170 | /* Old left over entry. Skip: */ |
673242c1 AK |
171 | if (mcelog.entry[entry].finished) { |
172 | entry++; | |
173 | continue; | |
174 | } | |
7644143c | 175 | break; |
1da177e4 | 176 | } |
1da177e4 LT |
177 | smp_rmb(); |
178 | next = entry + 1; | |
179 | if (cmpxchg(&mcelog.next, entry, next) == entry) | |
180 | break; | |
181 | } | |
182 | memcpy(mcelog.entry + entry, mce, sizeof(struct mce)); | |
7644143c | 183 | wmb(); |
1da177e4 | 184 | mcelog.entry[entry].finished = 1; |
7644143c | 185 | wmb(); |
1da177e4 | 186 | |
a0189c70 | 187 | mce->finished = 1; |
1020bcbc | 188 | set_bit(0, &mce_need_notify); |
1da177e4 LT |
189 | } |
190 | ||
09371957 BP |
191 | static void drain_mcelog_buffer(void) |
192 | { | |
193 | unsigned int next, i, prev = 0; | |
194 | ||
b11e3d78 | 195 | next = ACCESS_ONCE(mcelog.next); |
09371957 BP |
196 | |
197 | do { | |
198 | struct mce *m; | |
199 | ||
200 | /* drain what was logged during boot */ | |
201 | for (i = prev; i < next; i++) { | |
202 | unsigned long start = jiffies; | |
203 | unsigned retries = 1; | |
204 | ||
205 | m = &mcelog.entry[i]; | |
206 | ||
207 | while (!m->finished) { | |
208 | if (time_after_eq(jiffies, start + 2*retries)) | |
209 | retries++; | |
210 | ||
211 | cpu_relax(); | |
212 | ||
213 | if (!m->finished && retries >= 4) { | |
c767a54b | 214 | pr_err("skipping error being logged currently!\n"); |
09371957 BP |
215 | break; |
216 | } | |
217 | } | |
218 | smp_rmb(); | |
219 | atomic_notifier_call_chain(&x86_mce_decoder_chain, 0, m); | |
220 | } | |
221 | ||
222 | memset(mcelog.entry + prev, 0, (next - prev) * sizeof(*m)); | |
223 | prev = next; | |
224 | next = cmpxchg(&mcelog.next, prev, 0); | |
225 | } while (next != prev); | |
226 | } | |
227 | ||
228 | ||
3653ada5 BP |
229 | void mce_register_decode_chain(struct notifier_block *nb) |
230 | { | |
231 | atomic_notifier_chain_register(&x86_mce_decoder_chain, nb); | |
09371957 | 232 | drain_mcelog_buffer(); |
3653ada5 BP |
233 | } |
234 | EXPORT_SYMBOL_GPL(mce_register_decode_chain); | |
235 | ||
236 | void mce_unregister_decode_chain(struct notifier_block *nb) | |
237 | { | |
238 | atomic_notifier_chain_unregister(&x86_mce_decoder_chain, nb); | |
239 | } | |
240 | EXPORT_SYMBOL_GPL(mce_unregister_decode_chain); | |
241 | ||
77e26cca | 242 | static void print_mce(struct mce *m) |
1da177e4 | 243 | { |
dffa4b2f BP |
244 | int ret = 0; |
245 | ||
a2d7b0d4 | 246 | pr_emerg(HW_ERR "CPU %d: Machine Check Exception: %Lx Bank %d: %016Lx\n", |
d620c67f | 247 | m->extcpu, m->mcgstatus, m->bank, m->status); |
f436f8bb | 248 | |
65ea5b03 | 249 | if (m->ip) { |
a2d7b0d4 | 250 | pr_emerg(HW_ERR "RIP%s %02x:<%016Lx> ", |
f436f8bb IM |
251 | !(m->mcgstatus & MCG_STATUS_EIPV) ? " !INEXACT!" : "", |
252 | m->cs, m->ip); | |
253 | ||
1da177e4 | 254 | if (m->cs == __KERNEL_CS) |
65ea5b03 | 255 | print_symbol("{%s}", m->ip); |
f436f8bb | 256 | pr_cont("\n"); |
1da177e4 | 257 | } |
f436f8bb | 258 | |
a2d7b0d4 | 259 | pr_emerg(HW_ERR "TSC %llx ", m->tsc); |
1da177e4 | 260 | if (m->addr) |
f436f8bb | 261 | pr_cont("ADDR %llx ", m->addr); |
1da177e4 | 262 | if (m->misc) |
f436f8bb | 263 | pr_cont("MISC %llx ", m->misc); |
549d042d | 264 | |
f436f8bb | 265 | pr_cont("\n"); |
506ed6b5 AK |
266 | /* |
267 | * Note this output is parsed by external tools and old fields | |
268 | * should not be changed. | |
269 | */ | |
881e23e5 | 270 | pr_emerg(HW_ERR "PROCESSOR %u:%x TIME %llu SOCKET %u APIC %x microcode %x\n", |
506ed6b5 AK |
271 | m->cpuvendor, m->cpuid, m->time, m->socketid, m->apicid, |
272 | cpu_data(m->extcpu).microcode); | |
f436f8bb IM |
273 | |
274 | /* | |
275 | * Print out human-readable details about the MCE error, | |
fb253195 | 276 | * (if the CPU has an implementation for that) |
f436f8bb | 277 | */ |
dffa4b2f BP |
278 | ret = atomic_notifier_call_chain(&x86_mce_decoder_chain, 0, m); |
279 | if (ret == NOTIFY_STOP) | |
280 | return; | |
281 | ||
282 | pr_emerg_ratelimited(HW_ERR "Run the above through 'mcelog --ascii'\n"); | |
86503560 AK |
283 | } |
284 | ||
f94b61c2 AK |
285 | #define PANIC_TIMEOUT 5 /* 5 seconds */ |
286 | ||
287 | static atomic_t mce_paniced; | |
288 | ||
bf783f9f HY |
289 | static int fake_panic; |
290 | static atomic_t mce_fake_paniced; | |
291 | ||
f94b61c2 AK |
292 | /* Panic in progress. Enable interrupts and wait for final IPI */ |
293 | static void wait_for_panic(void) | |
294 | { | |
295 | long timeout = PANIC_TIMEOUT*USEC_PER_SEC; | |
f436f8bb | 296 | |
f94b61c2 AK |
297 | preempt_disable(); |
298 | local_irq_enable(); | |
299 | while (timeout-- > 0) | |
300 | udelay(1); | |
29b0f591 AK |
301 | if (panic_timeout == 0) |
302 | panic_timeout = mce_panic_timeout; | |
f94b61c2 AK |
303 | panic("Panicing machine check CPU died"); |
304 | } | |
305 | ||
bd19a5e6 | 306 | static void mce_panic(char *msg, struct mce *final, char *exp) |
d88203d1 | 307 | { |
482908b4 | 308 | int i, apei_err = 0; |
e02e68d3 | 309 | |
bf783f9f HY |
310 | if (!fake_panic) { |
311 | /* | |
312 | * Make sure only one CPU runs in machine check panic | |
313 | */ | |
314 | if (atomic_inc_return(&mce_paniced) > 1) | |
315 | wait_for_panic(); | |
316 | barrier(); | |
f94b61c2 | 317 | |
bf783f9f HY |
318 | bust_spinlocks(1); |
319 | console_verbose(); | |
320 | } else { | |
321 | /* Don't log too much for fake panic */ | |
322 | if (atomic_inc_return(&mce_fake_paniced) > 1) | |
323 | return; | |
324 | } | |
a0189c70 | 325 | /* First print corrected ones that are still unlogged */ |
1da177e4 | 326 | for (i = 0; i < MCE_LOG_LEN; i++) { |
a0189c70 | 327 | struct mce *m = &mcelog.entry[i]; |
77e26cca HS |
328 | if (!(m->status & MCI_STATUS_VAL)) |
329 | continue; | |
482908b4 | 330 | if (!(m->status & MCI_STATUS_UC)) { |
77e26cca | 331 | print_mce(m); |
482908b4 HY |
332 | if (!apei_err) |
333 | apei_err = apei_write_mce(m); | |
334 | } | |
a0189c70 AK |
335 | } |
336 | /* Now print uncorrected but with the final one last */ | |
337 | for (i = 0; i < MCE_LOG_LEN; i++) { | |
338 | struct mce *m = &mcelog.entry[i]; | |
339 | if (!(m->status & MCI_STATUS_VAL)) | |
1da177e4 | 340 | continue; |
77e26cca HS |
341 | if (!(m->status & MCI_STATUS_UC)) |
342 | continue; | |
482908b4 | 343 | if (!final || memcmp(m, final, sizeof(struct mce))) { |
77e26cca | 344 | print_mce(m); |
482908b4 HY |
345 | if (!apei_err) |
346 | apei_err = apei_write_mce(m); | |
347 | } | |
1da177e4 | 348 | } |
482908b4 | 349 | if (final) { |
77e26cca | 350 | print_mce(final); |
482908b4 HY |
351 | if (!apei_err) |
352 | apei_err = apei_write_mce(final); | |
353 | } | |
3c079792 | 354 | if (cpu_missing) |
a2d7b0d4 | 355 | pr_emerg(HW_ERR "Some CPUs didn't answer in synchronization\n"); |
bd19a5e6 | 356 | if (exp) |
a2d7b0d4 | 357 | pr_emerg(HW_ERR "Machine check: %s\n", exp); |
bf783f9f HY |
358 | if (!fake_panic) { |
359 | if (panic_timeout == 0) | |
360 | panic_timeout = mce_panic_timeout; | |
361 | panic(msg); | |
362 | } else | |
a2d7b0d4 | 363 | pr_emerg(HW_ERR "Fake kernel panic: %s\n", msg); |
d88203d1 | 364 | } |
1da177e4 | 365 | |
ea149b36 AK |
366 | /* Support code for software error injection */ |
367 | ||
368 | static int msr_to_offset(u32 msr) | |
369 | { | |
0a3aee0d | 370 | unsigned bank = __this_cpu_read(injectm.bank); |
f436f8bb | 371 | |
ea149b36 AK |
372 | if (msr == rip_msr) |
373 | return offsetof(struct mce, ip); | |
a2d32bcb | 374 | if (msr == MSR_IA32_MCx_STATUS(bank)) |
ea149b36 | 375 | return offsetof(struct mce, status); |
a2d32bcb | 376 | if (msr == MSR_IA32_MCx_ADDR(bank)) |
ea149b36 | 377 | return offsetof(struct mce, addr); |
a2d32bcb | 378 | if (msr == MSR_IA32_MCx_MISC(bank)) |
ea149b36 AK |
379 | return offsetof(struct mce, misc); |
380 | if (msr == MSR_IA32_MCG_STATUS) | |
381 | return offsetof(struct mce, mcgstatus); | |
382 | return -1; | |
383 | } | |
384 | ||
5f8c1a54 AK |
385 | /* MSR access wrappers used for error injection */ |
386 | static u64 mce_rdmsrl(u32 msr) | |
387 | { | |
388 | u64 v; | |
11868a2d | 389 | |
0a3aee0d | 390 | if (__this_cpu_read(injectm.finished)) { |
ea149b36 | 391 | int offset = msr_to_offset(msr); |
11868a2d | 392 | |
ea149b36 AK |
393 | if (offset < 0) |
394 | return 0; | |
395 | return *(u64 *)((char *)&__get_cpu_var(injectm) + offset); | |
396 | } | |
11868a2d IM |
397 | |
398 | if (rdmsrl_safe(msr, &v)) { | |
399 | WARN_ONCE(1, "mce: Unable to read msr %d!\n", msr); | |
400 | /* | |
401 | * Return zero in case the access faulted. This should | |
402 | * not happen normally but can happen if the CPU does | |
403 | * something weird, or if the code is buggy. | |
404 | */ | |
405 | v = 0; | |
406 | } | |
407 | ||
5f8c1a54 AK |
408 | return v; |
409 | } | |
410 | ||
411 | static void mce_wrmsrl(u32 msr, u64 v) | |
412 | { | |
0a3aee0d | 413 | if (__this_cpu_read(injectm.finished)) { |
ea149b36 | 414 | int offset = msr_to_offset(msr); |
11868a2d | 415 | |
ea149b36 AK |
416 | if (offset >= 0) |
417 | *(u64 *)((char *)&__get_cpu_var(injectm) + offset) = v; | |
418 | return; | |
419 | } | |
5f8c1a54 AK |
420 | wrmsrl(msr, v); |
421 | } | |
422 | ||
b8325c5b HS |
423 | /* |
424 | * Collect all global (w.r.t. this processor) status about this machine | |
425 | * check into our "mce" struct so that we can use it later to assess | |
426 | * the severity of the problem as we read per-bank specific details. | |
427 | */ | |
428 | static inline void mce_gather_info(struct mce *m, struct pt_regs *regs) | |
429 | { | |
430 | mce_setup(m); | |
431 | ||
432 | m->mcgstatus = mce_rdmsrl(MSR_IA32_MCG_STATUS); | |
433 | if (regs) { | |
434 | /* | |
435 | * Get the address of the instruction at the time of | |
436 | * the machine check error. | |
437 | */ | |
438 | if (m->mcgstatus & (MCG_STATUS_RIPV|MCG_STATUS_EIPV)) { | |
439 | m->ip = regs->ip; | |
440 | m->cs = regs->cs; | |
a129a7c8 AK |
441 | |
442 | /* | |
443 | * When in VM86 mode make the cs look like ring 3 | |
444 | * always. This is a lie, but it's better than passing | |
445 | * the additional vm86 bit around everywhere. | |
446 | */ | |
447 | if (v8086_mode(regs)) | |
448 | m->cs |= 3; | |
b8325c5b HS |
449 | } |
450 | /* Use accurate RIP reporting if available. */ | |
451 | if (rip_msr) | |
452 | m->ip = mce_rdmsrl(rip_msr); | |
453 | } | |
454 | } | |
455 | ||
9b1beaf2 AK |
456 | /* |
457 | * Simple lockless ring to communicate PFNs from the exception handler with the | |
458 | * process context work function. This is vastly simplified because there's | |
459 | * only a single reader and a single writer. | |
460 | */ | |
461 | #define MCE_RING_SIZE 16 /* we use one entry less */ | |
462 | ||
463 | struct mce_ring { | |
464 | unsigned short start; | |
465 | unsigned short end; | |
466 | unsigned long ring[MCE_RING_SIZE]; | |
467 | }; | |
468 | static DEFINE_PER_CPU(struct mce_ring, mce_ring); | |
469 | ||
470 | /* Runs with CPU affinity in workqueue */ | |
471 | static int mce_ring_empty(void) | |
472 | { | |
473 | struct mce_ring *r = &__get_cpu_var(mce_ring); | |
474 | ||
475 | return r->start == r->end; | |
476 | } | |
477 | ||
478 | static int mce_ring_get(unsigned long *pfn) | |
479 | { | |
480 | struct mce_ring *r; | |
481 | int ret = 0; | |
482 | ||
483 | *pfn = 0; | |
484 | get_cpu(); | |
485 | r = &__get_cpu_var(mce_ring); | |
486 | if (r->start == r->end) | |
487 | goto out; | |
488 | *pfn = r->ring[r->start]; | |
489 | r->start = (r->start + 1) % MCE_RING_SIZE; | |
490 | ret = 1; | |
491 | out: | |
492 | put_cpu(); | |
493 | return ret; | |
494 | } | |
495 | ||
496 | /* Always runs in MCE context with preempt off */ | |
497 | static int mce_ring_add(unsigned long pfn) | |
498 | { | |
499 | struct mce_ring *r = &__get_cpu_var(mce_ring); | |
500 | unsigned next; | |
501 | ||
502 | next = (r->end + 1) % MCE_RING_SIZE; | |
503 | if (next == r->start) | |
504 | return -1; | |
505 | r->ring[r->end] = pfn; | |
506 | wmb(); | |
507 | r->end = next; | |
508 | return 0; | |
509 | } | |
510 | ||
88ccbedd | 511 | int mce_available(struct cpuinfo_x86 *c) |
1da177e4 | 512 | { |
04b2b1a4 | 513 | if (mce_disabled) |
5b4408fd | 514 | return 0; |
3d1712c9 | 515 | return cpu_has(c, X86_FEATURE_MCE) && cpu_has(c, X86_FEATURE_MCA); |
1da177e4 LT |
516 | } |
517 | ||
9b1beaf2 AK |
518 | static void mce_schedule_work(void) |
519 | { | |
520 | if (!mce_ring_empty()) { | |
521 | struct work_struct *work = &__get_cpu_var(mce_work); | |
522 | if (!work_pending(work)) | |
523 | schedule_work(work); | |
524 | } | |
525 | } | |
526 | ||
b77e70bf HS |
527 | DEFINE_PER_CPU(struct irq_work, mce_irq_work); |
528 | ||
529 | static void mce_irq_work_cb(struct irq_work *entry) | |
ccc3c319 | 530 | { |
9ff36ee9 | 531 | mce_notify_irq(); |
9b1beaf2 | 532 | mce_schedule_work(); |
ccc3c319 | 533 | } |
ccc3c319 AK |
534 | |
535 | static void mce_report_event(struct pt_regs *regs) | |
536 | { | |
537 | if (regs->flags & (X86_VM_MASK|X86_EFLAGS_IF)) { | |
9ff36ee9 | 538 | mce_notify_irq(); |
9b1beaf2 AK |
539 | /* |
540 | * Triggering the work queue here is just an insurance | |
541 | * policy in case the syscall exit notify handler | |
542 | * doesn't run soon enough or ends up running on the | |
543 | * wrong CPU (can happen when audit sleeps) | |
544 | */ | |
545 | mce_schedule_work(); | |
ccc3c319 AK |
546 | return; |
547 | } | |
548 | ||
b77e70bf | 549 | irq_work_queue(&__get_cpu_var(mce_irq_work)); |
ccc3c319 AK |
550 | } |
551 | ||
85f92694 TL |
552 | /* |
553 | * Read ADDR and MISC registers. | |
554 | */ | |
555 | static void mce_read_aux(struct mce *m, int i) | |
556 | { | |
557 | if (m->status & MCI_STATUS_MISCV) | |
558 | m->misc = mce_rdmsrl(MSR_IA32_MCx_MISC(i)); | |
559 | if (m->status & MCI_STATUS_ADDRV) { | |
560 | m->addr = mce_rdmsrl(MSR_IA32_MCx_ADDR(i)); | |
561 | ||
562 | /* | |
563 | * Mask the reported address by the reported granularity. | |
564 | */ | |
565 | if (mce_ser && (m->status & MCI_STATUS_MISCV)) { | |
566 | u8 shift = MCI_MISC_ADDR_LSB(m->misc); | |
567 | m->addr >>= shift; | |
568 | m->addr <<= shift; | |
569 | } | |
570 | } | |
571 | } | |
572 | ||
ca84f696 AK |
573 | DEFINE_PER_CPU(unsigned, mce_poll_count); |
574 | ||
d88203d1 | 575 | /* |
b79109c3 AK |
576 | * Poll for corrected events or events that happened before reset. |
577 | * Those are just logged through /dev/mcelog. | |
578 | * | |
579 | * This is executed in standard interrupt context. | |
ed7290d0 AK |
580 | * |
581 | * Note: spec recommends to panic for fatal unsignalled | |
582 | * errors here. However this would be quite problematic -- | |
583 | * we would need to reimplement the Monarch handling and | |
584 | * it would mess up the exclusion between exception handler | |
585 | * and poll hander -- * so we skip this for now. | |
586 | * These cases should not happen anyways, or only when the CPU | |
587 | * is already totally * confused. In this case it's likely it will | |
588 | * not fully execute the machine check handler either. | |
b79109c3 | 589 | */ |
ee031c31 | 590 | void machine_check_poll(enum mcp_flags flags, mce_banks_t *b) |
b79109c3 AK |
591 | { |
592 | struct mce m; | |
593 | int i; | |
594 | ||
c6ae41e7 | 595 | this_cpu_inc(mce_poll_count); |
ca84f696 | 596 | |
b8325c5b | 597 | mce_gather_info(&m, NULL); |
b79109c3 | 598 | |
b79109c3 | 599 | for (i = 0; i < banks; i++) { |
cebe1820 | 600 | if (!mce_banks[i].ctl || !test_bit(i, *b)) |
b79109c3 AK |
601 | continue; |
602 | ||
603 | m.misc = 0; | |
604 | m.addr = 0; | |
605 | m.bank = i; | |
606 | m.tsc = 0; | |
607 | ||
608 | barrier(); | |
a2d32bcb | 609 | m.status = mce_rdmsrl(MSR_IA32_MCx_STATUS(i)); |
b79109c3 AK |
610 | if (!(m.status & MCI_STATUS_VAL)) |
611 | continue; | |
612 | ||
613 | /* | |
ed7290d0 AK |
614 | * Uncorrected or signalled events are handled by the exception |
615 | * handler when it is enabled, so don't process those here. | |
b79109c3 AK |
616 | * |
617 | * TBD do the same check for MCI_STATUS_EN here? | |
618 | */ | |
ed7290d0 AK |
619 | if (!(flags & MCP_UC) && |
620 | (m.status & (mce_ser ? MCI_STATUS_S : MCI_STATUS_UC))) | |
b79109c3 AK |
621 | continue; |
622 | ||
85f92694 | 623 | mce_read_aux(&m, i); |
b79109c3 AK |
624 | |
625 | if (!(flags & MCP_TIMESTAMP)) | |
626 | m.tsc = 0; | |
627 | /* | |
628 | * Don't get the IP here because it's unlikely to | |
629 | * have anything to do with the actual error location. | |
630 | */ | |
f0cb5452 | 631 | if (!(flags & MCP_DONTLOG) && !mce_dont_log_ce) |
5679af4c | 632 | mce_log(&m); |
b79109c3 AK |
633 | |
634 | /* | |
635 | * Clear state for this bank. | |
636 | */ | |
a2d32bcb | 637 | mce_wrmsrl(MSR_IA32_MCx_STATUS(i), 0); |
b79109c3 AK |
638 | } |
639 | ||
640 | /* | |
641 | * Don't clear MCG_STATUS here because it's only defined for | |
642 | * exceptions. | |
643 | */ | |
88921be3 AK |
644 | |
645 | sync_core(); | |
b79109c3 | 646 | } |
ea149b36 | 647 | EXPORT_SYMBOL_GPL(machine_check_poll); |
b79109c3 | 648 | |
bd19a5e6 AK |
649 | /* |
650 | * Do a quick check if any of the events requires a panic. | |
651 | * This decides if we keep the events around or clear them. | |
652 | */ | |
95022b8c | 653 | static int mce_no_way_out(struct mce *m, char **msg, unsigned long *validp) |
bd19a5e6 | 654 | { |
95022b8c | 655 | int i, ret = 0; |
bd19a5e6 AK |
656 | |
657 | for (i = 0; i < banks; i++) { | |
a2d32bcb | 658 | m->status = mce_rdmsrl(MSR_IA32_MCx_STATUS(i)); |
95022b8c TL |
659 | if (m->status & MCI_STATUS_VAL) |
660 | __set_bit(i, validp); | |
bd19a5e6 | 661 | if (mce_severity(m, tolerant, msg) >= MCE_PANIC_SEVERITY) |
95022b8c | 662 | ret = 1; |
bd19a5e6 | 663 | } |
95022b8c | 664 | return ret; |
bd19a5e6 AK |
665 | } |
666 | ||
3c079792 AK |
667 | /* |
668 | * Variable to establish order between CPUs while scanning. | |
669 | * Each CPU spins initially until executing is equal its number. | |
670 | */ | |
671 | static atomic_t mce_executing; | |
672 | ||
673 | /* | |
674 | * Defines order of CPUs on entry. First CPU becomes Monarch. | |
675 | */ | |
676 | static atomic_t mce_callin; | |
677 | ||
678 | /* | |
679 | * Check if a timeout waiting for other CPUs happened. | |
680 | */ | |
681 | static int mce_timed_out(u64 *t) | |
682 | { | |
683 | /* | |
684 | * The others already did panic for some reason. | |
685 | * Bail out like in a timeout. | |
686 | * rmb() to tell the compiler that system_state | |
687 | * might have been modified by someone else. | |
688 | */ | |
689 | rmb(); | |
690 | if (atomic_read(&mce_paniced)) | |
691 | wait_for_panic(); | |
692 | if (!monarch_timeout) | |
693 | goto out; | |
694 | if ((s64)*t < SPINUNIT) { | |
695 | /* CHECKME: Make panic default for 1 too? */ | |
696 | if (tolerant < 1) | |
697 | mce_panic("Timeout synchronizing machine check over CPUs", | |
698 | NULL, NULL); | |
699 | cpu_missing = 1; | |
700 | return 1; | |
701 | } | |
702 | *t -= SPINUNIT; | |
703 | out: | |
704 | touch_nmi_watchdog(); | |
705 | return 0; | |
706 | } | |
707 | ||
708 | /* | |
709 | * The Monarch's reign. The Monarch is the CPU who entered | |
710 | * the machine check handler first. It waits for the others to | |
711 | * raise the exception too and then grades them. When any | |
712 | * error is fatal panic. Only then let the others continue. | |
713 | * | |
714 | * The other CPUs entering the MCE handler will be controlled by the | |
715 | * Monarch. They are called Subjects. | |
716 | * | |
717 | * This way we prevent any potential data corruption in a unrecoverable case | |
718 | * and also makes sure always all CPU's errors are examined. | |
719 | * | |
680b6cfd | 720 | * Also this detects the case of a machine check event coming from outer |
3c079792 AK |
721 | * space (not detected by any CPUs) In this case some external agent wants |
722 | * us to shut down, so panic too. | |
723 | * | |
724 | * The other CPUs might still decide to panic if the handler happens | |
725 | * in a unrecoverable place, but in this case the system is in a semi-stable | |
726 | * state and won't corrupt anything by itself. It's ok to let the others | |
727 | * continue for a bit first. | |
728 | * | |
729 | * All the spin loops have timeouts; when a timeout happens a CPU | |
730 | * typically elects itself to be Monarch. | |
731 | */ | |
732 | static void mce_reign(void) | |
733 | { | |
734 | int cpu; | |
735 | struct mce *m = NULL; | |
736 | int global_worst = 0; | |
737 | char *msg = NULL; | |
738 | char *nmsg = NULL; | |
739 | ||
740 | /* | |
741 | * This CPU is the Monarch and the other CPUs have run | |
742 | * through their handlers. | |
743 | * Grade the severity of the errors of all the CPUs. | |
744 | */ | |
745 | for_each_possible_cpu(cpu) { | |
746 | int severity = mce_severity(&per_cpu(mces_seen, cpu), tolerant, | |
747 | &nmsg); | |
748 | if (severity > global_worst) { | |
749 | msg = nmsg; | |
750 | global_worst = severity; | |
751 | m = &per_cpu(mces_seen, cpu); | |
752 | } | |
753 | } | |
754 | ||
755 | /* | |
756 | * Cannot recover? Panic here then. | |
757 | * This dumps all the mces in the log buffer and stops the | |
758 | * other CPUs. | |
759 | */ | |
760 | if (m && global_worst >= MCE_PANIC_SEVERITY && tolerant < 3) | |
ac960375 | 761 | mce_panic("Fatal Machine check", m, msg); |
3c079792 AK |
762 | |
763 | /* | |
764 | * For UC somewhere we let the CPU who detects it handle it. | |
765 | * Also must let continue the others, otherwise the handling | |
766 | * CPU could deadlock on a lock. | |
767 | */ | |
768 | ||
769 | /* | |
770 | * No machine check event found. Must be some external | |
771 | * source or one CPU is hung. Panic. | |
772 | */ | |
680b6cfd | 773 | if (global_worst <= MCE_KEEP_SEVERITY && tolerant < 3) |
3c079792 AK |
774 | mce_panic("Machine check from unknown source", NULL, NULL); |
775 | ||
776 | /* | |
777 | * Now clear all the mces_seen so that they don't reappear on | |
778 | * the next mce. | |
779 | */ | |
780 | for_each_possible_cpu(cpu) | |
781 | memset(&per_cpu(mces_seen, cpu), 0, sizeof(struct mce)); | |
782 | } | |
783 | ||
784 | static atomic_t global_nwo; | |
785 | ||
786 | /* | |
787 | * Start of Monarch synchronization. This waits until all CPUs have | |
788 | * entered the exception handler and then determines if any of them | |
789 | * saw a fatal event that requires panic. Then it executes them | |
790 | * in the entry order. | |
791 | * TBD double check parallel CPU hotunplug | |
792 | */ | |
7fb06fc9 | 793 | static int mce_start(int *no_way_out) |
3c079792 | 794 | { |
7fb06fc9 | 795 | int order; |
3c079792 AK |
796 | int cpus = num_online_cpus(); |
797 | u64 timeout = (u64)monarch_timeout * NSEC_PER_USEC; | |
798 | ||
7fb06fc9 HS |
799 | if (!timeout) |
800 | return -1; | |
3c079792 | 801 | |
7fb06fc9 | 802 | atomic_add(*no_way_out, &global_nwo); |
184e1fdf HY |
803 | /* |
804 | * global_nwo should be updated before mce_callin | |
805 | */ | |
806 | smp_wmb(); | |
a95436e4 | 807 | order = atomic_inc_return(&mce_callin); |
3c079792 AK |
808 | |
809 | /* | |
810 | * Wait for everyone. | |
811 | */ | |
812 | while (atomic_read(&mce_callin) != cpus) { | |
813 | if (mce_timed_out(&timeout)) { | |
814 | atomic_set(&global_nwo, 0); | |
7fb06fc9 | 815 | return -1; |
3c079792 AK |
816 | } |
817 | ndelay(SPINUNIT); | |
818 | } | |
819 | ||
184e1fdf HY |
820 | /* |
821 | * mce_callin should be read before global_nwo | |
822 | */ | |
823 | smp_rmb(); | |
3c079792 | 824 | |
7fb06fc9 HS |
825 | if (order == 1) { |
826 | /* | |
827 | * Monarch: Starts executing now, the others wait. | |
828 | */ | |
3c079792 | 829 | atomic_set(&mce_executing, 1); |
7fb06fc9 HS |
830 | } else { |
831 | /* | |
832 | * Subject: Now start the scanning loop one by one in | |
833 | * the original callin order. | |
834 | * This way when there are any shared banks it will be | |
835 | * only seen by one CPU before cleared, avoiding duplicates. | |
836 | */ | |
837 | while (atomic_read(&mce_executing) < order) { | |
838 | if (mce_timed_out(&timeout)) { | |
839 | atomic_set(&global_nwo, 0); | |
840 | return -1; | |
841 | } | |
842 | ndelay(SPINUNIT); | |
843 | } | |
3c079792 AK |
844 | } |
845 | ||
846 | /* | |
7fb06fc9 | 847 | * Cache the global no_way_out state. |
3c079792 | 848 | */ |
7fb06fc9 HS |
849 | *no_way_out = atomic_read(&global_nwo); |
850 | ||
851 | return order; | |
3c079792 AK |
852 | } |
853 | ||
854 | /* | |
855 | * Synchronize between CPUs after main scanning loop. | |
856 | * This invokes the bulk of the Monarch processing. | |
857 | */ | |
858 | static int mce_end(int order) | |
859 | { | |
860 | int ret = -1; | |
861 | u64 timeout = (u64)monarch_timeout * NSEC_PER_USEC; | |
862 | ||
863 | if (!timeout) | |
864 | goto reset; | |
865 | if (order < 0) | |
866 | goto reset; | |
867 | ||
868 | /* | |
869 | * Allow others to run. | |
870 | */ | |
871 | atomic_inc(&mce_executing); | |
872 | ||
873 | if (order == 1) { | |
874 | /* CHECKME: Can this race with a parallel hotplug? */ | |
875 | int cpus = num_online_cpus(); | |
876 | ||
877 | /* | |
878 | * Monarch: Wait for everyone to go through their scanning | |
879 | * loops. | |
880 | */ | |
881 | while (atomic_read(&mce_executing) <= cpus) { | |
882 | if (mce_timed_out(&timeout)) | |
883 | goto reset; | |
884 | ndelay(SPINUNIT); | |
885 | } | |
886 | ||
887 | mce_reign(); | |
888 | barrier(); | |
889 | ret = 0; | |
890 | } else { | |
891 | /* | |
892 | * Subject: Wait for Monarch to finish. | |
893 | */ | |
894 | while (atomic_read(&mce_executing) != 0) { | |
895 | if (mce_timed_out(&timeout)) | |
896 | goto reset; | |
897 | ndelay(SPINUNIT); | |
898 | } | |
899 | ||
900 | /* | |
901 | * Don't reset anything. That's done by the Monarch. | |
902 | */ | |
903 | return 0; | |
904 | } | |
905 | ||
906 | /* | |
907 | * Reset all global state. | |
908 | */ | |
909 | reset: | |
910 | atomic_set(&global_nwo, 0); | |
911 | atomic_set(&mce_callin, 0); | |
912 | barrier(); | |
913 | ||
914 | /* | |
915 | * Let others run again. | |
916 | */ | |
917 | atomic_set(&mce_executing, 0); | |
918 | return ret; | |
919 | } | |
920 | ||
9b1beaf2 AK |
921 | /* |
922 | * Check if the address reported by the CPU is in a format we can parse. | |
923 | * It would be possible to add code for most other cases, but all would | |
924 | * be somewhat complicated (e.g. segment offset would require an instruction | |
0d2eb44f | 925 | * parser). So only support physical addresses up to page granuality for now. |
9b1beaf2 AK |
926 | */ |
927 | static int mce_usable_address(struct mce *m) | |
928 | { | |
929 | if (!(m->status & MCI_STATUS_MISCV) || !(m->status & MCI_STATUS_ADDRV)) | |
930 | return 0; | |
2b90e77e | 931 | if (MCI_MISC_ADDR_LSB(m->misc) > PAGE_SHIFT) |
9b1beaf2 | 932 | return 0; |
2b90e77e | 933 | if (MCI_MISC_ADDR_MODE(m->misc) != MCI_MISC_ADDR_PHYS) |
9b1beaf2 AK |
934 | return 0; |
935 | return 1; | |
936 | } | |
937 | ||
3c079792 AK |
938 | static void mce_clear_state(unsigned long *toclear) |
939 | { | |
940 | int i; | |
941 | ||
942 | for (i = 0; i < banks; i++) { | |
943 | if (test_bit(i, toclear)) | |
a2d32bcb | 944 | mce_wrmsrl(MSR_IA32_MCx_STATUS(i), 0); |
3c079792 AK |
945 | } |
946 | } | |
947 | ||
af104e39 TL |
948 | /* |
949 | * Need to save faulting physical address associated with a process | |
950 | * in the machine check handler some place where we can grab it back | |
951 | * later in mce_notify_process() | |
952 | */ | |
953 | #define MCE_INFO_MAX 16 | |
954 | ||
955 | struct mce_info { | |
956 | atomic_t inuse; | |
957 | struct task_struct *t; | |
958 | __u64 paddr; | |
dad1743e | 959 | int restartable; |
af104e39 TL |
960 | } mce_info[MCE_INFO_MAX]; |
961 | ||
dad1743e | 962 | static void mce_save_info(__u64 addr, int c) |
af104e39 TL |
963 | { |
964 | struct mce_info *mi; | |
965 | ||
966 | for (mi = mce_info; mi < &mce_info[MCE_INFO_MAX]; mi++) { | |
967 | if (atomic_cmpxchg(&mi->inuse, 0, 1) == 0) { | |
968 | mi->t = current; | |
969 | mi->paddr = addr; | |
dad1743e | 970 | mi->restartable = c; |
af104e39 TL |
971 | return; |
972 | } | |
973 | } | |
974 | ||
975 | mce_panic("Too many concurrent recoverable errors", NULL, NULL); | |
976 | } | |
977 | ||
978 | static struct mce_info *mce_find_info(void) | |
979 | { | |
980 | struct mce_info *mi; | |
981 | ||
982 | for (mi = mce_info; mi < &mce_info[MCE_INFO_MAX]; mi++) | |
983 | if (atomic_read(&mi->inuse) && mi->t == current) | |
984 | return mi; | |
985 | return NULL; | |
986 | } | |
987 | ||
988 | static void mce_clear_info(struct mce_info *mi) | |
989 | { | |
990 | atomic_set(&mi->inuse, 0); | |
991 | } | |
992 | ||
b79109c3 AK |
993 | /* |
994 | * The actual machine check handler. This only handles real | |
995 | * exceptions when something got corrupted coming in through int 18. | |
996 | * | |
997 | * This is executed in NMI context not subject to normal locking rules. This | |
998 | * implies that most kernel services cannot be safely used. Don't even | |
999 | * think about putting a printk in there! | |
3c079792 AK |
1000 | * |
1001 | * On Intel systems this is entered on all CPUs in parallel through | |
1002 | * MCE broadcast. However some CPUs might be broken beyond repair, | |
1003 | * so be always careful when synchronizing with others. | |
1da177e4 | 1004 | */ |
e9eee03e | 1005 | void do_machine_check(struct pt_regs *regs, long error_code) |
1da177e4 | 1006 | { |
3c079792 | 1007 | struct mce m, *final; |
1da177e4 | 1008 | int i; |
3c079792 AK |
1009 | int worst = 0; |
1010 | int severity; | |
1011 | /* | |
1012 | * Establish sequential order between the CPUs entering the machine | |
1013 | * check handler. | |
1014 | */ | |
7fb06fc9 | 1015 | int order; |
bd78432c TH |
1016 | /* |
1017 | * If no_way_out gets set, there is no safe way to recover from this | |
1018 | * MCE. If tolerant is cranked up, we'll try anyway. | |
1019 | */ | |
1020 | int no_way_out = 0; | |
1021 | /* | |
1022 | * If kill_it gets set, there might be a way to recover from this | |
1023 | * error. | |
1024 | */ | |
1025 | int kill_it = 0; | |
b79109c3 | 1026 | DECLARE_BITMAP(toclear, MAX_NR_BANKS); |
95022b8c | 1027 | DECLARE_BITMAP(valid_banks, MAX_NR_BANKS); |
bd19a5e6 | 1028 | char *msg = "Unknown"; |
1da177e4 | 1029 | |
553f265f AK |
1030 | atomic_inc(&mce_entry); |
1031 | ||
c6ae41e7 | 1032 | this_cpu_inc(mce_exception_count); |
01ca79f1 | 1033 | |
b79109c3 | 1034 | if (!banks) |
32561696 | 1035 | goto out; |
1da177e4 | 1036 | |
b8325c5b | 1037 | mce_gather_info(&m, regs); |
b5f2fa4e | 1038 | |
3c079792 AK |
1039 | final = &__get_cpu_var(mces_seen); |
1040 | *final = m; | |
1041 | ||
95022b8c TL |
1042 | memset(valid_banks, 0, sizeof(valid_banks)); |
1043 | no_way_out = mce_no_way_out(&m, &msg, valid_banks); | |
680b6cfd | 1044 | |
1da177e4 LT |
1045 | barrier(); |
1046 | ||
ed7290d0 | 1047 | /* |
a8c321fb TL |
1048 | * When no restart IP might need to kill or panic. |
1049 | * Assume the worst for now, but if we find the | |
1050 | * severity is MCE_AR_SEVERITY we have other options. | |
ed7290d0 AK |
1051 | */ |
1052 | if (!(m.mcgstatus & MCG_STATUS_RIPV)) | |
1053 | kill_it = 1; | |
1054 | ||
3c079792 AK |
1055 | /* |
1056 | * Go through all the banks in exclusion of the other CPUs. | |
1057 | * This way we don't report duplicated events on shared banks | |
1058 | * because the first one to see it will clear it. | |
1059 | */ | |
7fb06fc9 | 1060 | order = mce_start(&no_way_out); |
1da177e4 | 1061 | for (i = 0; i < banks; i++) { |
b79109c3 | 1062 | __clear_bit(i, toclear); |
95022b8c TL |
1063 | if (!test_bit(i, valid_banks)) |
1064 | continue; | |
cebe1820 | 1065 | if (!mce_banks[i].ctl) |
1da177e4 | 1066 | continue; |
d88203d1 TG |
1067 | |
1068 | m.misc = 0; | |
1da177e4 LT |
1069 | m.addr = 0; |
1070 | m.bank = i; | |
1da177e4 | 1071 | |
a2d32bcb | 1072 | m.status = mce_rdmsrl(MSR_IA32_MCx_STATUS(i)); |
1da177e4 LT |
1073 | if ((m.status & MCI_STATUS_VAL) == 0) |
1074 | continue; | |
1075 | ||
b79109c3 | 1076 | /* |
ed7290d0 AK |
1077 | * Non uncorrected or non signaled errors are handled by |
1078 | * machine_check_poll. Leave them alone, unless this panics. | |
b79109c3 | 1079 | */ |
ed7290d0 AK |
1080 | if (!(m.status & (mce_ser ? MCI_STATUS_S : MCI_STATUS_UC)) && |
1081 | !no_way_out) | |
b79109c3 AK |
1082 | continue; |
1083 | ||
1084 | /* | |
1085 | * Set taint even when machine check was not enabled. | |
1086 | */ | |
1087 | add_taint(TAINT_MACHINE_CHECK); | |
1088 | ||
ed7290d0 | 1089 | severity = mce_severity(&m, tolerant, NULL); |
b79109c3 | 1090 | |
ed7290d0 AK |
1091 | /* |
1092 | * When machine check was for corrected handler don't touch, | |
1093 | * unless we're panicing. | |
1094 | */ | |
1095 | if (severity == MCE_KEEP_SEVERITY && !no_way_out) | |
1096 | continue; | |
1097 | __set_bit(i, toclear); | |
1098 | if (severity == MCE_NO_SEVERITY) { | |
b79109c3 AK |
1099 | /* |
1100 | * Machine check event was not enabled. Clear, but | |
1101 | * ignore. | |
1102 | */ | |
1103 | continue; | |
1da177e4 LT |
1104 | } |
1105 | ||
85f92694 | 1106 | mce_read_aux(&m, i); |
1da177e4 | 1107 | |
9b1beaf2 AK |
1108 | /* |
1109 | * Action optional error. Queue address for later processing. | |
1110 | * When the ring overflows we just ignore the AO error. | |
1111 | * RED-PEN add some logging mechanism when | |
1112 | * usable_address or mce_add_ring fails. | |
1113 | * RED-PEN don't ignore overflow for tolerant == 0 | |
1114 | */ | |
1115 | if (severity == MCE_AO_SEVERITY && mce_usable_address(&m)) | |
1116 | mce_ring_add(m.addr >> PAGE_SHIFT); | |
1117 | ||
b79109c3 | 1118 | mce_log(&m); |
1da177e4 | 1119 | |
3c079792 AK |
1120 | if (severity > worst) { |
1121 | *final = m; | |
1122 | worst = severity; | |
1da177e4 | 1123 | } |
1da177e4 LT |
1124 | } |
1125 | ||
a8c321fb TL |
1126 | /* mce_clear_state will clear *final, save locally for use later */ |
1127 | m = *final; | |
1128 | ||
3c079792 AK |
1129 | if (!no_way_out) |
1130 | mce_clear_state(toclear); | |
1131 | ||
e9eee03e | 1132 | /* |
3c079792 AK |
1133 | * Do most of the synchronization with other CPUs. |
1134 | * When there's any problem use only local no_way_out state. | |
e9eee03e | 1135 | */ |
3c079792 AK |
1136 | if (mce_end(order) < 0) |
1137 | no_way_out = worst >= MCE_PANIC_SEVERITY; | |
bd78432c TH |
1138 | |
1139 | /* | |
a8c321fb TL |
1140 | * At insane "tolerant" levels we take no action. Otherwise |
1141 | * we only die if we have no other choice. For less serious | |
1142 | * issues we try to recover, or limit damage to the current | |
1143 | * process. | |
bd78432c | 1144 | */ |
a8c321fb TL |
1145 | if (tolerant < 3) { |
1146 | if (no_way_out) | |
1147 | mce_panic("Fatal machine check on current CPU", &m, msg); | |
1148 | if (worst == MCE_AR_SEVERITY) { | |
1149 | /* schedule action before return to userland */ | |
dad1743e | 1150 | mce_save_info(m.addr, m.mcgstatus & MCG_STATUS_RIPV); |
a8c321fb TL |
1151 | set_thread_flag(TIF_MCE_NOTIFY); |
1152 | } else if (kill_it) { | |
1153 | force_sig(SIGBUS, current); | |
1154 | } | |
1155 | } | |
e02e68d3 | 1156 | |
3c079792 AK |
1157 | if (worst > 0) |
1158 | mce_report_event(regs); | |
5f8c1a54 | 1159 | mce_wrmsrl(MSR_IA32_MCG_STATUS, 0); |
32561696 | 1160 | out: |
553f265f | 1161 | atomic_dec(&mce_entry); |
88921be3 | 1162 | sync_core(); |
1da177e4 | 1163 | } |
ea149b36 | 1164 | EXPORT_SYMBOL_GPL(do_machine_check); |
1da177e4 | 1165 | |
cd42f4a3 TL |
1166 | #ifndef CONFIG_MEMORY_FAILURE |
1167 | int memory_failure(unsigned long pfn, int vector, int flags) | |
9b1beaf2 | 1168 | { |
a8c321fb TL |
1169 | /* mce_severity() should not hand us an ACTION_REQUIRED error */ |
1170 | BUG_ON(flags & MF_ACTION_REQUIRED); | |
c767a54b JP |
1171 | pr_err("Uncorrected memory error in page 0x%lx ignored\n" |
1172 | "Rebuild kernel with CONFIG_MEMORY_FAILURE=y for smarter handling\n", | |
1173 | pfn); | |
cd42f4a3 TL |
1174 | |
1175 | return 0; | |
9b1beaf2 | 1176 | } |
cd42f4a3 | 1177 | #endif |
9b1beaf2 AK |
1178 | |
1179 | /* | |
a8c321fb TL |
1180 | * Called in process context that interrupted by MCE and marked with |
1181 | * TIF_MCE_NOTIFY, just before returning to erroneous userland. | |
1182 | * This code is allowed to sleep. | |
1183 | * Attempt possible recovery such as calling the high level VM handler to | |
1184 | * process any corrupted pages, and kill/signal current process if required. | |
1185 | * Action required errors are handled here. | |
9b1beaf2 AK |
1186 | */ |
1187 | void mce_notify_process(void) | |
1188 | { | |
1189 | unsigned long pfn; | |
a8c321fb | 1190 | struct mce_info *mi = mce_find_info(); |
6751ed65 | 1191 | int flags = MF_ACTION_REQUIRED; |
a8c321fb TL |
1192 | |
1193 | if (!mi) | |
1194 | mce_panic("Lost physical address for unconsumed uncorrectable error", NULL, NULL); | |
1195 | pfn = mi->paddr >> PAGE_SHIFT; | |
1196 | ||
1197 | clear_thread_flag(TIF_MCE_NOTIFY); | |
1198 | ||
1199 | pr_err("Uncorrected hardware memory error in user-access at %llx", | |
1200 | mi->paddr); | |
dad1743e TL |
1201 | /* |
1202 | * We must call memory_failure() here even if the current process is | |
1203 | * doomed. We still need to mark the page as poisoned and alert any | |
1204 | * other users of the page. | |
1205 | */ | |
6751ed65 TL |
1206 | if (!mi->restartable) |
1207 | flags |= MF_MUST_KILL; | |
1208 | if (memory_failure(pfn, MCE_VECTOR, flags) < 0) { | |
a8c321fb TL |
1209 | pr_err("Memory error not recovered"); |
1210 | force_sig(SIGBUS, current); | |
1211 | } | |
1212 | mce_clear_info(mi); | |
9b1beaf2 AK |
1213 | } |
1214 | ||
a8c321fb TL |
1215 | /* |
1216 | * Action optional processing happens here (picking up | |
1217 | * from the list of faulting pages that do_machine_check() | |
1218 | * placed into the "ring"). | |
1219 | */ | |
9b1beaf2 AK |
1220 | static void mce_process_work(struct work_struct *dummy) |
1221 | { | |
a8c321fb TL |
1222 | unsigned long pfn; |
1223 | ||
1224 | while (mce_ring_get(&pfn)) | |
1225 | memory_failure(pfn, MCE_VECTOR, 0); | |
9b1beaf2 AK |
1226 | } |
1227 | ||
15d5f839 DZ |
1228 | #ifdef CONFIG_X86_MCE_INTEL |
1229 | /*** | |
1230 | * mce_log_therm_throt_event - Logs the thermal throttling event to mcelog | |
676b1855 | 1231 | * @cpu: The CPU on which the event occurred. |
15d5f839 DZ |
1232 | * @status: Event status information |
1233 | * | |
1234 | * This function should be called by the thermal interrupt after the | |
1235 | * event has been processed and the decision was made to log the event | |
1236 | * further. | |
1237 | * | |
1238 | * The status parameter will be saved to the 'status' field of 'struct mce' | |
1239 | * and historically has been the register value of the | |
1240 | * MSR_IA32_THERMAL_STATUS (Intel) msr. | |
1241 | */ | |
b5f2fa4e | 1242 | void mce_log_therm_throt_event(__u64 status) |
15d5f839 DZ |
1243 | { |
1244 | struct mce m; | |
1245 | ||
b5f2fa4e | 1246 | mce_setup(&m); |
15d5f839 DZ |
1247 | m.bank = MCE_THERMAL_BANK; |
1248 | m.status = status; | |
15d5f839 DZ |
1249 | mce_log(&m); |
1250 | } | |
1251 | #endif /* CONFIG_X86_MCE_INTEL */ | |
1252 | ||
1da177e4 | 1253 | /* |
8a336b0a TH |
1254 | * Periodic polling timer for "silent" machine check errors. If the |
1255 | * poller finds an MCE, poll 2x faster. When the poller finds no more | |
1256 | * errors, poll 2x slower (up to check_interval seconds). | |
1da177e4 | 1257 | */ |
82f7af09 | 1258 | static unsigned long check_interval = 5 * 60; /* 5 minutes */ |
e9eee03e | 1259 | |
82f7af09 | 1260 | static DEFINE_PER_CPU(unsigned long, mce_next_interval); /* in jiffies */ |
52d168e2 | 1261 | static DEFINE_PER_CPU(struct timer_list, mce_timer); |
1da177e4 | 1262 | |
82f7af09 | 1263 | static void mce_timer_fn(unsigned long data) |
1da177e4 | 1264 | { |
82f7af09 TG |
1265 | struct timer_list *t = &__get_cpu_var(mce_timer); |
1266 | unsigned long iv; | |
52d168e2 AK |
1267 | |
1268 | WARN_ON(smp_processor_id() != data); | |
1269 | ||
7b543a53 | 1270 | if (mce_available(__this_cpu_ptr(&cpu_info))) { |
ee031c31 AK |
1271 | machine_check_poll(MCP_TIMESTAMP, |
1272 | &__get_cpu_var(mce_poll_banks)); | |
e9eee03e | 1273 | } |
1da177e4 LT |
1274 | |
1275 | /* | |
e02e68d3 TH |
1276 | * Alert userspace if needed. If we logged an MCE, reduce the |
1277 | * polling interval, otherwise increase the polling interval. | |
1da177e4 | 1278 | */ |
82f7af09 | 1279 | iv = __this_cpu_read(mce_next_interval); |
9ff36ee9 | 1280 | if (mce_notify_irq()) |
958fb3c5 | 1281 | iv = max(iv / 2, (unsigned long) HZ/100); |
14a02530 | 1282 | else |
82f7af09 TG |
1283 | iv = min(iv * 2, round_jiffies_relative(check_interval * HZ)); |
1284 | __this_cpu_write(mce_next_interval, iv); | |
e02e68d3 | 1285 | |
82f7af09 | 1286 | t->expires = jiffies + iv; |
5be6066a | 1287 | add_timer_on(t, smp_processor_id()); |
e02e68d3 TH |
1288 | } |
1289 | ||
9aaef96f HS |
1290 | /* Must not be called in IRQ context where del_timer_sync() can deadlock */ |
1291 | static void mce_timer_delete_all(void) | |
1292 | { | |
1293 | int cpu; | |
1294 | ||
1295 | for_each_online_cpu(cpu) | |
1296 | del_timer_sync(&per_cpu(mce_timer, cpu)); | |
1297 | } | |
1298 | ||
9bd98405 AK |
1299 | static void mce_do_trigger(struct work_struct *work) |
1300 | { | |
1020bcbc | 1301 | call_usermodehelper(mce_helper, mce_helper_argv, NULL, UMH_NO_WAIT); |
9bd98405 AK |
1302 | } |
1303 | ||
1304 | static DECLARE_WORK(mce_trigger_work, mce_do_trigger); | |
1305 | ||
e02e68d3 | 1306 | /* |
9bd98405 AK |
1307 | * Notify the user(s) about new machine check events. |
1308 | * Can be called from interrupt context, but not from machine check/NMI | |
1309 | * context. | |
e02e68d3 | 1310 | */ |
9ff36ee9 | 1311 | int mce_notify_irq(void) |
e02e68d3 | 1312 | { |
8457c84d AK |
1313 | /* Not more than two messages every minute */ |
1314 | static DEFINE_RATELIMIT_STATE(ratelimit, 60*HZ, 2); | |
1315 | ||
1020bcbc | 1316 | if (test_and_clear_bit(0, &mce_need_notify)) { |
93b62c3c HS |
1317 | /* wake processes polling /dev/mcelog */ |
1318 | wake_up_interruptible(&mce_chrdev_wait); | |
9bd98405 AK |
1319 | |
1320 | /* | |
1321 | * There is no risk of missing notifications because | |
1322 | * work_pending is always cleared before the function is | |
1323 | * executed. | |
1324 | */ | |
1020bcbc | 1325 | if (mce_helper[0] && !work_pending(&mce_trigger_work)) |
9bd98405 | 1326 | schedule_work(&mce_trigger_work); |
e02e68d3 | 1327 | |
8457c84d | 1328 | if (__ratelimit(&ratelimit)) |
a2d7b0d4 | 1329 | pr_info(HW_ERR "Machine check events logged\n"); |
e02e68d3 TH |
1330 | |
1331 | return 1; | |
1da177e4 | 1332 | } |
e02e68d3 TH |
1333 | return 0; |
1334 | } | |
9ff36ee9 | 1335 | EXPORT_SYMBOL_GPL(mce_notify_irq); |
8a336b0a | 1336 | |
cffd377e | 1337 | static int __cpuinit __mcheck_cpu_mce_banks_init(void) |
cebe1820 AK |
1338 | { |
1339 | int i; | |
1340 | ||
1341 | mce_banks = kzalloc(banks * sizeof(struct mce_bank), GFP_KERNEL); | |
1342 | if (!mce_banks) | |
1343 | return -ENOMEM; | |
1344 | for (i = 0; i < banks; i++) { | |
1345 | struct mce_bank *b = &mce_banks[i]; | |
11868a2d | 1346 | |
cebe1820 AK |
1347 | b->ctl = -1ULL; |
1348 | b->init = 1; | |
1349 | } | |
1350 | return 0; | |
1351 | } | |
1352 | ||
d88203d1 | 1353 | /* |
1da177e4 LT |
1354 | * Initialize Machine Checks for a CPU. |
1355 | */ | |
5e09954a | 1356 | static int __cpuinit __mcheck_cpu_cap_init(void) |
1da177e4 | 1357 | { |
0d7482e3 | 1358 | unsigned b; |
e9eee03e | 1359 | u64 cap; |
1da177e4 LT |
1360 | |
1361 | rdmsrl(MSR_IA32_MCG_CAP, cap); | |
01c6680a TG |
1362 | |
1363 | b = cap & MCG_BANKCNT_MASK; | |
93ae5012 | 1364 | if (!banks) |
c767a54b | 1365 | pr_info("CPU supports %d MCE banks\n", b); |
b659294b | 1366 | |
0d7482e3 | 1367 | if (b > MAX_NR_BANKS) { |
c767a54b | 1368 | pr_warn("Using only %u machine check banks out of %u\n", |
0d7482e3 AK |
1369 | MAX_NR_BANKS, b); |
1370 | b = MAX_NR_BANKS; | |
1371 | } | |
1372 | ||
1373 | /* Don't support asymmetric configurations today */ | |
1374 | WARN_ON(banks != 0 && b != banks); | |
1375 | banks = b; | |
cebe1820 | 1376 | if (!mce_banks) { |
cffd377e | 1377 | int err = __mcheck_cpu_mce_banks_init(); |
11868a2d | 1378 | |
cebe1820 AK |
1379 | if (err) |
1380 | return err; | |
1da177e4 | 1381 | } |
0d7482e3 | 1382 | |
94ad8474 | 1383 | /* Use accurate RIP reporting if available. */ |
01c6680a | 1384 | if ((cap & MCG_EXT_P) && MCG_EXT_CNT(cap) >= 9) |
94ad8474 | 1385 | rip_msr = MSR_IA32_MCG_EIP; |
1da177e4 | 1386 | |
ed7290d0 AK |
1387 | if (cap & MCG_SER_P) |
1388 | mce_ser = 1; | |
1389 | ||
0d7482e3 AK |
1390 | return 0; |
1391 | } | |
1392 | ||
5e09954a | 1393 | static void __mcheck_cpu_init_generic(void) |
0d7482e3 | 1394 | { |
e9eee03e | 1395 | mce_banks_t all_banks; |
0d7482e3 AK |
1396 | u64 cap; |
1397 | int i; | |
1398 | ||
b79109c3 AK |
1399 | /* |
1400 | * Log the machine checks left over from the previous reset. | |
1401 | */ | |
ee031c31 | 1402 | bitmap_fill(all_banks, MAX_NR_BANKS); |
5679af4c | 1403 | machine_check_poll(MCP_UC|(!mce_bootlog ? MCP_DONTLOG : 0), &all_banks); |
1da177e4 LT |
1404 | |
1405 | set_in_cr4(X86_CR4_MCE); | |
1406 | ||
0d7482e3 | 1407 | rdmsrl(MSR_IA32_MCG_CAP, cap); |
1da177e4 LT |
1408 | if (cap & MCG_CTL_P) |
1409 | wrmsr(MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff); | |
1410 | ||
1411 | for (i = 0; i < banks; i++) { | |
cebe1820 | 1412 | struct mce_bank *b = &mce_banks[i]; |
11868a2d | 1413 | |
cebe1820 | 1414 | if (!b->init) |
06b7a7a5 | 1415 | continue; |
a2d32bcb AK |
1416 | wrmsrl(MSR_IA32_MCx_CTL(i), b->ctl); |
1417 | wrmsrl(MSR_IA32_MCx_STATUS(i), 0); | |
d88203d1 | 1418 | } |
1da177e4 LT |
1419 | } |
1420 | ||
1421 | /* Add per CPU specific workarounds here */ | |
5e09954a | 1422 | static int __cpuinit __mcheck_cpu_apply_quirks(struct cpuinfo_x86 *c) |
d88203d1 | 1423 | { |
e412cd25 | 1424 | if (c->x86_vendor == X86_VENDOR_UNKNOWN) { |
c767a54b | 1425 | pr_info("unknown CPU type - not enabling MCE support\n"); |
e412cd25 IM |
1426 | return -EOPNOTSUPP; |
1427 | } | |
1428 | ||
1da177e4 | 1429 | /* This should be disabled by the BIOS, but isn't always */ |
911f6a7b | 1430 | if (c->x86_vendor == X86_VENDOR_AMD) { |
e9eee03e IM |
1431 | if (c->x86 == 15 && banks > 4) { |
1432 | /* | |
1433 | * disable GART TBL walk error reporting, which | |
1434 | * trips off incorrectly with the IOMMU & 3ware | |
1435 | * & Cerberus: | |
1436 | */ | |
cebe1820 | 1437 | clear_bit(10, (unsigned long *)&mce_banks[4].ctl); |
e9eee03e IM |
1438 | } |
1439 | if (c->x86 <= 17 && mce_bootlog < 0) { | |
1440 | /* | |
1441 | * Lots of broken BIOS around that don't clear them | |
1442 | * by default and leave crap in there. Don't log: | |
1443 | */ | |
911f6a7b | 1444 | mce_bootlog = 0; |
e9eee03e | 1445 | } |
2e6f694f AK |
1446 | /* |
1447 | * Various K7s with broken bank 0 around. Always disable | |
1448 | * by default. | |
1449 | */ | |
203abd67 | 1450 | if (c->x86 == 6 && banks > 0) |
cebe1820 | 1451 | mce_banks[0].ctl = 0; |
575203b4 BP |
1452 | |
1453 | /* | |
1454 | * Turn off MC4_MISC thresholding banks on those models since | |
1455 | * they're not supported there. | |
1456 | */ | |
1457 | if (c->x86 == 0x15 && | |
1458 | (c->x86_model >= 0x10 && c->x86_model <= 0x1f)) { | |
1459 | int i; | |
1460 | u64 val, hwcr; | |
1461 | bool need_toggle; | |
1462 | u32 msrs[] = { | |
1463 | 0x00000413, /* MC4_MISC0 */ | |
1464 | 0xc0000408, /* MC4_MISC1 */ | |
1465 | }; | |
1466 | ||
1467 | rdmsrl(MSR_K7_HWCR, hwcr); | |
1468 | ||
1469 | /* McStatusWrEn has to be set */ | |
1470 | need_toggle = !(hwcr & BIT(18)); | |
1471 | ||
1472 | if (need_toggle) | |
1473 | wrmsrl(MSR_K7_HWCR, hwcr | BIT(18)); | |
1474 | ||
1475 | for (i = 0; i < ARRAY_SIZE(msrs); i++) { | |
1476 | rdmsrl(msrs[i], val); | |
1477 | ||
1478 | /* CntP bit set? */ | |
80f03361 BP |
1479 | if (val & BIT_64(62)) { |
1480 | val &= ~BIT_64(62); | |
1481 | wrmsrl(msrs[i], val); | |
575203b4 BP |
1482 | } |
1483 | } | |
1484 | ||
1485 | /* restore old settings */ | |
1486 | if (need_toggle) | |
1487 | wrmsrl(MSR_K7_HWCR, hwcr); | |
1488 | } | |
1da177e4 | 1489 | } |
e583538f | 1490 | |
06b7a7a5 AK |
1491 | if (c->x86_vendor == X86_VENDOR_INTEL) { |
1492 | /* | |
1493 | * SDM documents that on family 6 bank 0 should not be written | |
1494 | * because it aliases to another special BIOS controlled | |
1495 | * register. | |
1496 | * But it's not aliased anymore on model 0x1a+ | |
1497 | * Don't ignore bank 0 completely because there could be a | |
1498 | * valid event later, merely don't write CTL0. | |
1499 | */ | |
1500 | ||
cebe1820 AK |
1501 | if (c->x86 == 6 && c->x86_model < 0x1A && banks > 0) |
1502 | mce_banks[0].init = 0; | |
3c079792 AK |
1503 | |
1504 | /* | |
1505 | * All newer Intel systems support MCE broadcasting. Enable | |
1506 | * synchronization with a one second timeout. | |
1507 | */ | |
1508 | if ((c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xe)) && | |
1509 | monarch_timeout < 0) | |
1510 | monarch_timeout = USEC_PER_SEC; | |
c7f6fa44 | 1511 | |
e412cd25 IM |
1512 | /* |
1513 | * There are also broken BIOSes on some Pentium M and | |
1514 | * earlier systems: | |
1515 | */ | |
1516 | if (c->x86 == 6 && c->x86_model <= 13 && mce_bootlog < 0) | |
c7f6fa44 | 1517 | mce_bootlog = 0; |
06b7a7a5 | 1518 | } |
3c079792 AK |
1519 | if (monarch_timeout < 0) |
1520 | monarch_timeout = 0; | |
29b0f591 AK |
1521 | if (mce_bootlog != 0) |
1522 | mce_panic_timeout = 30; | |
e412cd25 IM |
1523 | |
1524 | return 0; | |
d88203d1 | 1525 | } |
1da177e4 | 1526 | |
3a97fc34 | 1527 | static int __cpuinit __mcheck_cpu_ancient_init(struct cpuinfo_x86 *c) |
4efc0670 AK |
1528 | { |
1529 | if (c->x86 != 5) | |
3a97fc34 HS |
1530 | return 0; |
1531 | ||
4efc0670 AK |
1532 | switch (c->x86_vendor) { |
1533 | case X86_VENDOR_INTEL: | |
c6978369 | 1534 | intel_p5_mcheck_init(c); |
3a97fc34 | 1535 | return 1; |
4efc0670 AK |
1536 | break; |
1537 | case X86_VENDOR_CENTAUR: | |
1538 | winchip_mcheck_init(c); | |
3a97fc34 | 1539 | return 1; |
4efc0670 AK |
1540 | break; |
1541 | } | |
3a97fc34 HS |
1542 | |
1543 | return 0; | |
4efc0670 AK |
1544 | } |
1545 | ||
5e09954a | 1546 | static void __mcheck_cpu_init_vendor(struct cpuinfo_x86 *c) |
1da177e4 LT |
1547 | { |
1548 | switch (c->x86_vendor) { | |
1549 | case X86_VENDOR_INTEL: | |
1550 | mce_intel_feature_init(c); | |
1551 | break; | |
89b831ef JS |
1552 | case X86_VENDOR_AMD: |
1553 | mce_amd_feature_init(c); | |
1554 | break; | |
1da177e4 LT |
1555 | default: |
1556 | break; | |
1557 | } | |
1558 | } | |
1559 | ||
5e09954a | 1560 | static void __mcheck_cpu_init_timer(void) |
52d168e2 AK |
1561 | { |
1562 | struct timer_list *t = &__get_cpu_var(mce_timer); | |
1a87fc1e | 1563 | unsigned long iv = check_interval * HZ; |
52d168e2 | 1564 | |
82f7af09 | 1565 | setup_timer(t, mce_timer_fn, smp_processor_id()); |
bc09effa | 1566 | |
62fdac59 HS |
1567 | if (mce_ignore_ce) |
1568 | return; | |
1569 | ||
82f7af09 TG |
1570 | __this_cpu_write(mce_next_interval, iv); |
1571 | if (!iv) | |
52d168e2 | 1572 | return; |
82f7af09 | 1573 | t->expires = round_jiffies(jiffies + iv); |
5be6066a | 1574 | add_timer_on(t, smp_processor_id()); |
52d168e2 AK |
1575 | } |
1576 | ||
9eda8cb3 AK |
1577 | /* Handle unconfigured int18 (should never happen) */ |
1578 | static void unexpected_machine_check(struct pt_regs *regs, long error_code) | |
1579 | { | |
c767a54b | 1580 | pr_err("CPU#%d: Unexpected int18 (Machine Check)\n", |
9eda8cb3 AK |
1581 | smp_processor_id()); |
1582 | } | |
1583 | ||
1584 | /* Call the installed machine check handler for this CPU setup. */ | |
1585 | void (*machine_check_vector)(struct pt_regs *, long error_code) = | |
1586 | unexpected_machine_check; | |
1587 | ||
d88203d1 | 1588 | /* |
1da177e4 | 1589 | * Called for each booted CPU to set up machine checks. |
e9eee03e | 1590 | * Must be called with preempt off: |
1da177e4 | 1591 | */ |
5e09954a | 1592 | void __cpuinit mcheck_cpu_init(struct cpuinfo_x86 *c) |
1da177e4 | 1593 | { |
4efc0670 AK |
1594 | if (mce_disabled) |
1595 | return; | |
1596 | ||
3a97fc34 HS |
1597 | if (__mcheck_cpu_ancient_init(c)) |
1598 | return; | |
4efc0670 | 1599 | |
5b4408fd | 1600 | if (!mce_available(c)) |
1da177e4 LT |
1601 | return; |
1602 | ||
5e09954a | 1603 | if (__mcheck_cpu_cap_init() < 0 || __mcheck_cpu_apply_quirks(c) < 0) { |
04b2b1a4 | 1604 | mce_disabled = 1; |
0d7482e3 AK |
1605 | return; |
1606 | } | |
0d7482e3 | 1607 | |
5d727926 AK |
1608 | machine_check_vector = do_machine_check; |
1609 | ||
5e09954a BP |
1610 | __mcheck_cpu_init_generic(); |
1611 | __mcheck_cpu_init_vendor(c); | |
1612 | __mcheck_cpu_init_timer(); | |
9b1beaf2 | 1613 | INIT_WORK(&__get_cpu_var(mce_work), mce_process_work); |
b77e70bf | 1614 | init_irq_work(&__get_cpu_var(mce_irq_work), &mce_irq_work_cb); |
1da177e4 LT |
1615 | } |
1616 | ||
1617 | /* | |
93b62c3c | 1618 | * mce_chrdev: Character device /dev/mcelog to read and clear the MCE log. |
1da177e4 LT |
1619 | */ |
1620 | ||
93b62c3c HS |
1621 | static DEFINE_SPINLOCK(mce_chrdev_state_lock); |
1622 | static int mce_chrdev_open_count; /* #times opened */ | |
1623 | static int mce_chrdev_open_exclu; /* already open exclusive? */ | |
f528e7ba | 1624 | |
93b62c3c | 1625 | static int mce_chrdev_open(struct inode *inode, struct file *file) |
f528e7ba | 1626 | { |
93b62c3c | 1627 | spin_lock(&mce_chrdev_state_lock); |
f528e7ba | 1628 | |
93b62c3c HS |
1629 | if (mce_chrdev_open_exclu || |
1630 | (mce_chrdev_open_count && (file->f_flags & O_EXCL))) { | |
1631 | spin_unlock(&mce_chrdev_state_lock); | |
e9eee03e | 1632 | |
f528e7ba TH |
1633 | return -EBUSY; |
1634 | } | |
1635 | ||
1636 | if (file->f_flags & O_EXCL) | |
93b62c3c HS |
1637 | mce_chrdev_open_exclu = 1; |
1638 | mce_chrdev_open_count++; | |
f528e7ba | 1639 | |
93b62c3c | 1640 | spin_unlock(&mce_chrdev_state_lock); |
f528e7ba | 1641 | |
bd78432c | 1642 | return nonseekable_open(inode, file); |
f528e7ba TH |
1643 | } |
1644 | ||
93b62c3c | 1645 | static int mce_chrdev_release(struct inode *inode, struct file *file) |
f528e7ba | 1646 | { |
93b62c3c | 1647 | spin_lock(&mce_chrdev_state_lock); |
f528e7ba | 1648 | |
93b62c3c HS |
1649 | mce_chrdev_open_count--; |
1650 | mce_chrdev_open_exclu = 0; | |
f528e7ba | 1651 | |
93b62c3c | 1652 | spin_unlock(&mce_chrdev_state_lock); |
f528e7ba TH |
1653 | |
1654 | return 0; | |
1655 | } | |
1656 | ||
d88203d1 TG |
1657 | static void collect_tscs(void *data) |
1658 | { | |
1da177e4 | 1659 | unsigned long *cpu_tsc = (unsigned long *)data; |
d88203d1 | 1660 | |
1da177e4 | 1661 | rdtscll(cpu_tsc[smp_processor_id()]); |
d88203d1 | 1662 | } |
1da177e4 | 1663 | |
482908b4 HY |
1664 | static int mce_apei_read_done; |
1665 | ||
1666 | /* Collect MCE record of previous boot in persistent storage via APEI ERST. */ | |
1667 | static int __mce_read_apei(char __user **ubuf, size_t usize) | |
1668 | { | |
1669 | int rc; | |
1670 | u64 record_id; | |
1671 | struct mce m; | |
1672 | ||
1673 | if (usize < sizeof(struct mce)) | |
1674 | return -EINVAL; | |
1675 | ||
1676 | rc = apei_read_mce(&m, &record_id); | |
1677 | /* Error or no more MCE record */ | |
1678 | if (rc <= 0) { | |
1679 | mce_apei_read_done = 1; | |
fadd85f1 NH |
1680 | /* |
1681 | * When ERST is disabled, mce_chrdev_read() should return | |
1682 | * "no record" instead of "no device." | |
1683 | */ | |
1684 | if (rc == -ENODEV) | |
1685 | return 0; | |
482908b4 HY |
1686 | return rc; |
1687 | } | |
1688 | rc = -EFAULT; | |
1689 | if (copy_to_user(*ubuf, &m, sizeof(struct mce))) | |
1690 | return rc; | |
1691 | /* | |
1692 | * In fact, we should have cleared the record after that has | |
1693 | * been flushed to the disk or sent to network in | |
1694 | * /sbin/mcelog, but we have no interface to support that now, | |
1695 | * so just clear it to avoid duplication. | |
1696 | */ | |
1697 | rc = apei_clear_mce(record_id); | |
1698 | if (rc) { | |
1699 | mce_apei_read_done = 1; | |
1700 | return rc; | |
1701 | } | |
1702 | *ubuf += sizeof(struct mce); | |
1703 | ||
1704 | return 0; | |
1705 | } | |
1706 | ||
93b62c3c HS |
1707 | static ssize_t mce_chrdev_read(struct file *filp, char __user *ubuf, |
1708 | size_t usize, loff_t *off) | |
1da177e4 | 1709 | { |
e9eee03e | 1710 | char __user *buf = ubuf; |
f0de53bb | 1711 | unsigned long *cpu_tsc; |
ef41df43 | 1712 | unsigned prev, next; |
1da177e4 LT |
1713 | int i, err; |
1714 | ||
6bca67f9 | 1715 | cpu_tsc = kmalloc(nr_cpu_ids * sizeof(long), GFP_KERNEL); |
f0de53bb AK |
1716 | if (!cpu_tsc) |
1717 | return -ENOMEM; | |
1718 | ||
93b62c3c | 1719 | mutex_lock(&mce_chrdev_read_mutex); |
482908b4 HY |
1720 | |
1721 | if (!mce_apei_read_done) { | |
1722 | err = __mce_read_apei(&buf, usize); | |
1723 | if (err || buf != ubuf) | |
1724 | goto out; | |
1725 | } | |
1726 | ||
f56e8a07 | 1727 | next = rcu_dereference_check_mce(mcelog.next); |
1da177e4 LT |
1728 | |
1729 | /* Only supports full reads right now */ | |
482908b4 HY |
1730 | err = -EINVAL; |
1731 | if (*off != 0 || usize < MCE_LOG_LEN*sizeof(struct mce)) | |
1732 | goto out; | |
1da177e4 LT |
1733 | |
1734 | err = 0; | |
ef41df43 HY |
1735 | prev = 0; |
1736 | do { | |
1737 | for (i = prev; i < next; i++) { | |
1738 | unsigned long start = jiffies; | |
559faa6b | 1739 | struct mce *m = &mcelog.entry[i]; |
ef41df43 | 1740 | |
559faa6b | 1741 | while (!m->finished) { |
ef41df43 | 1742 | if (time_after_eq(jiffies, start + 2)) { |
559faa6b | 1743 | memset(m, 0, sizeof(*m)); |
ef41df43 HY |
1744 | goto timeout; |
1745 | } | |
1746 | cpu_relax(); | |
673242c1 | 1747 | } |
ef41df43 | 1748 | smp_rmb(); |
559faa6b HS |
1749 | err |= copy_to_user(buf, m, sizeof(*m)); |
1750 | buf += sizeof(*m); | |
ef41df43 HY |
1751 | timeout: |
1752 | ; | |
673242c1 | 1753 | } |
1da177e4 | 1754 | |
ef41df43 HY |
1755 | memset(mcelog.entry + prev, 0, |
1756 | (next - prev) * sizeof(struct mce)); | |
1757 | prev = next; | |
1758 | next = cmpxchg(&mcelog.next, prev, 0); | |
1759 | } while (next != prev); | |
1da177e4 | 1760 | |
b2b18660 | 1761 | synchronize_sched(); |
1da177e4 | 1762 | |
d88203d1 TG |
1763 | /* |
1764 | * Collect entries that were still getting written before the | |
1765 | * synchronize. | |
1766 | */ | |
15c8b6c1 | 1767 | on_each_cpu(collect_tscs, cpu_tsc, 1); |
e9eee03e | 1768 | |
d88203d1 | 1769 | for (i = next; i < MCE_LOG_LEN; i++) { |
559faa6b HS |
1770 | struct mce *m = &mcelog.entry[i]; |
1771 | ||
1772 | if (m->finished && m->tsc < cpu_tsc[m->cpu]) { | |
1773 | err |= copy_to_user(buf, m, sizeof(*m)); | |
1da177e4 | 1774 | smp_rmb(); |
559faa6b HS |
1775 | buf += sizeof(*m); |
1776 | memset(m, 0, sizeof(*m)); | |
1da177e4 | 1777 | } |
d88203d1 | 1778 | } |
482908b4 HY |
1779 | |
1780 | if (err) | |
1781 | err = -EFAULT; | |
1782 | ||
1783 | out: | |
93b62c3c | 1784 | mutex_unlock(&mce_chrdev_read_mutex); |
f0de53bb | 1785 | kfree(cpu_tsc); |
e9eee03e | 1786 | |
482908b4 | 1787 | return err ? err : buf - ubuf; |
1da177e4 LT |
1788 | } |
1789 | ||
93b62c3c | 1790 | static unsigned int mce_chrdev_poll(struct file *file, poll_table *wait) |
e02e68d3 | 1791 | { |
93b62c3c | 1792 | poll_wait(file, &mce_chrdev_wait, wait); |
a4dd9925 | 1793 | if (rcu_access_index(mcelog.next)) |
e02e68d3 | 1794 | return POLLIN | POLLRDNORM; |
482908b4 HY |
1795 | if (!mce_apei_read_done && apei_check_mce()) |
1796 | return POLLIN | POLLRDNORM; | |
e02e68d3 TH |
1797 | return 0; |
1798 | } | |
1799 | ||
93b62c3c HS |
1800 | static long mce_chrdev_ioctl(struct file *f, unsigned int cmd, |
1801 | unsigned long arg) | |
1da177e4 LT |
1802 | { |
1803 | int __user *p = (int __user *)arg; | |
d88203d1 | 1804 | |
1da177e4 | 1805 | if (!capable(CAP_SYS_ADMIN)) |
d88203d1 | 1806 | return -EPERM; |
e9eee03e | 1807 | |
1da177e4 | 1808 | switch (cmd) { |
d88203d1 | 1809 | case MCE_GET_RECORD_LEN: |
1da177e4 LT |
1810 | return put_user(sizeof(struct mce), p); |
1811 | case MCE_GET_LOG_LEN: | |
d88203d1 | 1812 | return put_user(MCE_LOG_LEN, p); |
1da177e4 LT |
1813 | case MCE_GETCLEAR_FLAGS: { |
1814 | unsigned flags; | |
d88203d1 TG |
1815 | |
1816 | do { | |
1da177e4 | 1817 | flags = mcelog.flags; |
d88203d1 | 1818 | } while (cmpxchg(&mcelog.flags, flags, 0) != flags); |
e9eee03e | 1819 | |
d88203d1 | 1820 | return put_user(flags, p); |
1da177e4 LT |
1821 | } |
1822 | default: | |
d88203d1 TG |
1823 | return -ENOTTY; |
1824 | } | |
1da177e4 LT |
1825 | } |
1826 | ||
66f5ddf3 LT |
1827 | static ssize_t (*mce_write)(struct file *filp, const char __user *ubuf, |
1828 | size_t usize, loff_t *off); | |
1829 | ||
1830 | void register_mce_write_callback(ssize_t (*fn)(struct file *filp, | |
1831 | const char __user *ubuf, | |
1832 | size_t usize, loff_t *off)) | |
1833 | { | |
1834 | mce_write = fn; | |
1835 | } | |
1836 | EXPORT_SYMBOL_GPL(register_mce_write_callback); | |
1837 | ||
1838 | ssize_t mce_chrdev_write(struct file *filp, const char __user *ubuf, | |
1839 | size_t usize, loff_t *off) | |
1840 | { | |
1841 | if (mce_write) | |
1842 | return mce_write(filp, ubuf, usize, off); | |
1843 | else | |
1844 | return -EINVAL; | |
1845 | } | |
1846 | ||
1847 | static const struct file_operations mce_chrdev_ops = { | |
93b62c3c HS |
1848 | .open = mce_chrdev_open, |
1849 | .release = mce_chrdev_release, | |
1850 | .read = mce_chrdev_read, | |
66f5ddf3 | 1851 | .write = mce_chrdev_write, |
93b62c3c HS |
1852 | .poll = mce_chrdev_poll, |
1853 | .unlocked_ioctl = mce_chrdev_ioctl, | |
1854 | .llseek = no_llseek, | |
1da177e4 LT |
1855 | }; |
1856 | ||
93b62c3c | 1857 | static struct miscdevice mce_chrdev_device = { |
1da177e4 LT |
1858 | MISC_MCELOG_MINOR, |
1859 | "mcelog", | |
1860 | &mce_chrdev_ops, | |
1861 | }; | |
1862 | ||
13503fa9 | 1863 | /* |
62fdac59 HS |
1864 | * mce=off Disables machine check |
1865 | * mce=no_cmci Disables CMCI | |
1866 | * mce=dont_log_ce Clears corrected events silently, no log created for CEs. | |
1867 | * mce=ignore_ce Disables polling and CMCI, corrected events are not cleared. | |
3c079792 AK |
1868 | * mce=TOLERANCELEVEL[,monarchtimeout] (number, see above) |
1869 | * monarchtimeout is how long to wait for other CPUs on machine | |
1870 | * check, or 0 to not wait | |
13503fa9 HS |
1871 | * mce=bootlog Log MCEs from before booting. Disabled by default on AMD. |
1872 | * mce=nobootlog Don't log MCEs from before booting. | |
1873 | */ | |
1da177e4 LT |
1874 | static int __init mcheck_enable(char *str) |
1875 | { | |
e3346fc4 | 1876 | if (*str == 0) { |
4efc0670 | 1877 | enable_p5_mce(); |
e3346fc4 BZ |
1878 | return 1; |
1879 | } | |
4efc0670 AK |
1880 | if (*str == '=') |
1881 | str++; | |
1da177e4 | 1882 | if (!strcmp(str, "off")) |
04b2b1a4 | 1883 | mce_disabled = 1; |
62fdac59 HS |
1884 | else if (!strcmp(str, "no_cmci")) |
1885 | mce_cmci_disabled = 1; | |
1886 | else if (!strcmp(str, "dont_log_ce")) | |
1887 | mce_dont_log_ce = 1; | |
1888 | else if (!strcmp(str, "ignore_ce")) | |
1889 | mce_ignore_ce = 1; | |
13503fa9 HS |
1890 | else if (!strcmp(str, "bootlog") || !strcmp(str, "nobootlog")) |
1891 | mce_bootlog = (str[0] == 'b'); | |
3c079792 | 1892 | else if (isdigit(str[0])) { |
8c566ef5 | 1893 | get_option(&str, &tolerant); |
3c079792 AK |
1894 | if (*str == ',') { |
1895 | ++str; | |
1896 | get_option(&str, &monarch_timeout); | |
1897 | } | |
1898 | } else { | |
c767a54b | 1899 | pr_info("mce argument %s ignored. Please use /sys\n", str); |
13503fa9 HS |
1900 | return 0; |
1901 | } | |
9b41046c | 1902 | return 1; |
1da177e4 | 1903 | } |
4efc0670 | 1904 | __setup("mce", mcheck_enable); |
1da177e4 | 1905 | |
a2202aa2 | 1906 | int __init mcheck_init(void) |
b33a6363 | 1907 | { |
a2202aa2 YW |
1908 | mcheck_intel_therm_init(); |
1909 | ||
b33a6363 BP |
1910 | return 0; |
1911 | } | |
b33a6363 | 1912 | |
d88203d1 | 1913 | /* |
c7cece89 | 1914 | * mce_syscore: PM support |
d88203d1 | 1915 | */ |
1da177e4 | 1916 | |
973a2dd1 AK |
1917 | /* |
1918 | * Disable machine checks on suspend and shutdown. We can't really handle | |
1919 | * them later. | |
1920 | */ | |
5e09954a | 1921 | static int mce_disable_error_reporting(void) |
973a2dd1 AK |
1922 | { |
1923 | int i; | |
1924 | ||
06b7a7a5 | 1925 | for (i = 0; i < banks; i++) { |
cebe1820 | 1926 | struct mce_bank *b = &mce_banks[i]; |
11868a2d | 1927 | |
cebe1820 | 1928 | if (b->init) |
a2d32bcb | 1929 | wrmsrl(MSR_IA32_MCx_CTL(i), 0); |
06b7a7a5 | 1930 | } |
973a2dd1 AK |
1931 | return 0; |
1932 | } | |
1933 | ||
c7cece89 | 1934 | static int mce_syscore_suspend(void) |
973a2dd1 | 1935 | { |
5e09954a | 1936 | return mce_disable_error_reporting(); |
973a2dd1 AK |
1937 | } |
1938 | ||
c7cece89 | 1939 | static void mce_syscore_shutdown(void) |
973a2dd1 | 1940 | { |
f3c6ea1b | 1941 | mce_disable_error_reporting(); |
973a2dd1 AK |
1942 | } |
1943 | ||
e9eee03e IM |
1944 | /* |
1945 | * On resume clear all MCE state. Don't want to see leftovers from the BIOS. | |
1946 | * Only one CPU is active at this time, the others get re-added later using | |
1947 | * CPU hotplug: | |
1948 | */ | |
c7cece89 | 1949 | static void mce_syscore_resume(void) |
1da177e4 | 1950 | { |
5e09954a | 1951 | __mcheck_cpu_init_generic(); |
7b543a53 | 1952 | __mcheck_cpu_init_vendor(__this_cpu_ptr(&cpu_info)); |
1da177e4 LT |
1953 | } |
1954 | ||
f3c6ea1b | 1955 | static struct syscore_ops mce_syscore_ops = { |
c7cece89 HS |
1956 | .suspend = mce_syscore_suspend, |
1957 | .shutdown = mce_syscore_shutdown, | |
1958 | .resume = mce_syscore_resume, | |
f3c6ea1b RW |
1959 | }; |
1960 | ||
c7cece89 | 1961 | /* |
8a25a2fd | 1962 | * mce_device: Sysfs support |
c7cece89 HS |
1963 | */ |
1964 | ||
52d168e2 AK |
1965 | static void mce_cpu_restart(void *data) |
1966 | { | |
7b543a53 | 1967 | if (!mce_available(__this_cpu_ptr(&cpu_info))) |
33edbf02 | 1968 | return; |
5e09954a BP |
1969 | __mcheck_cpu_init_generic(); |
1970 | __mcheck_cpu_init_timer(); | |
52d168e2 AK |
1971 | } |
1972 | ||
1da177e4 | 1973 | /* Reinit MCEs after user configuration changes */ |
d88203d1 TG |
1974 | static void mce_restart(void) |
1975 | { | |
9aaef96f | 1976 | mce_timer_delete_all(); |
52d168e2 | 1977 | on_each_cpu(mce_cpu_restart, NULL, 1); |
1da177e4 LT |
1978 | } |
1979 | ||
9af43b54 | 1980 | /* Toggle features for corrected errors */ |
9aaef96f | 1981 | static void mce_disable_cmci(void *data) |
9af43b54 | 1982 | { |
7b543a53 | 1983 | if (!mce_available(__this_cpu_ptr(&cpu_info))) |
9af43b54 | 1984 | return; |
9af43b54 HS |
1985 | cmci_clear(); |
1986 | } | |
1987 | ||
1988 | static void mce_enable_ce(void *all) | |
1989 | { | |
7b543a53 | 1990 | if (!mce_available(__this_cpu_ptr(&cpu_info))) |
9af43b54 HS |
1991 | return; |
1992 | cmci_reenable(); | |
1993 | cmci_recheck(); | |
1994 | if (all) | |
5e09954a | 1995 | __mcheck_cpu_init_timer(); |
9af43b54 HS |
1996 | } |
1997 | ||
8a25a2fd | 1998 | static struct bus_type mce_subsys = { |
e9eee03e | 1999 | .name = "machinecheck", |
8a25a2fd | 2000 | .dev_name = "machinecheck", |
1da177e4 LT |
2001 | }; |
2002 | ||
d6126ef5 | 2003 | DEFINE_PER_CPU(struct device *, mce_device); |
e9eee03e IM |
2004 | |
2005 | __cpuinitdata | |
2006 | void (*threshold_cpu_callback)(unsigned long action, unsigned int cpu); | |
1da177e4 | 2007 | |
8a25a2fd | 2008 | static inline struct mce_bank *attr_to_bank(struct device_attribute *attr) |
cebe1820 AK |
2009 | { |
2010 | return container_of(attr, struct mce_bank, attr); | |
2011 | } | |
0d7482e3 | 2012 | |
8a25a2fd | 2013 | static ssize_t show_bank(struct device *s, struct device_attribute *attr, |
0d7482e3 AK |
2014 | char *buf) |
2015 | { | |
cebe1820 | 2016 | return sprintf(buf, "%llx\n", attr_to_bank(attr)->ctl); |
0d7482e3 AK |
2017 | } |
2018 | ||
8a25a2fd | 2019 | static ssize_t set_bank(struct device *s, struct device_attribute *attr, |
9319cec8 | 2020 | const char *buf, size_t size) |
0d7482e3 | 2021 | { |
9319cec8 | 2022 | u64 new; |
e9eee03e | 2023 | |
9319cec8 | 2024 | if (strict_strtoull(buf, 0, &new) < 0) |
0d7482e3 | 2025 | return -EINVAL; |
e9eee03e | 2026 | |
cebe1820 | 2027 | attr_to_bank(attr)->ctl = new; |
0d7482e3 | 2028 | mce_restart(); |
e9eee03e | 2029 | |
9319cec8 | 2030 | return size; |
0d7482e3 | 2031 | } |
a98f0dd3 | 2032 | |
e9eee03e | 2033 | static ssize_t |
8a25a2fd | 2034 | show_trigger(struct device *s, struct device_attribute *attr, char *buf) |
a98f0dd3 | 2035 | { |
1020bcbc | 2036 | strcpy(buf, mce_helper); |
a98f0dd3 | 2037 | strcat(buf, "\n"); |
1020bcbc | 2038 | return strlen(mce_helper) + 1; |
a98f0dd3 AK |
2039 | } |
2040 | ||
8a25a2fd | 2041 | static ssize_t set_trigger(struct device *s, struct device_attribute *attr, |
e9eee03e | 2042 | const char *buf, size_t siz) |
a98f0dd3 AK |
2043 | { |
2044 | char *p; | |
e9eee03e | 2045 | |
1020bcbc HS |
2046 | strncpy(mce_helper, buf, sizeof(mce_helper)); |
2047 | mce_helper[sizeof(mce_helper)-1] = 0; | |
1020bcbc | 2048 | p = strchr(mce_helper, '\n'); |
e9eee03e | 2049 | |
e9084ec9 | 2050 | if (p) |
e9eee03e IM |
2051 | *p = 0; |
2052 | ||
e9084ec9 | 2053 | return strlen(mce_helper) + !!p; |
a98f0dd3 AK |
2054 | } |
2055 | ||
8a25a2fd KS |
2056 | static ssize_t set_ignore_ce(struct device *s, |
2057 | struct device_attribute *attr, | |
9af43b54 HS |
2058 | const char *buf, size_t size) |
2059 | { | |
2060 | u64 new; | |
2061 | ||
2062 | if (strict_strtoull(buf, 0, &new) < 0) | |
2063 | return -EINVAL; | |
2064 | ||
2065 | if (mce_ignore_ce ^ !!new) { | |
2066 | if (new) { | |
2067 | /* disable ce features */ | |
9aaef96f HS |
2068 | mce_timer_delete_all(); |
2069 | on_each_cpu(mce_disable_cmci, NULL, 1); | |
9af43b54 HS |
2070 | mce_ignore_ce = 1; |
2071 | } else { | |
2072 | /* enable ce features */ | |
2073 | mce_ignore_ce = 0; | |
2074 | on_each_cpu(mce_enable_ce, (void *)1, 1); | |
2075 | } | |
2076 | } | |
2077 | return size; | |
2078 | } | |
2079 | ||
8a25a2fd KS |
2080 | static ssize_t set_cmci_disabled(struct device *s, |
2081 | struct device_attribute *attr, | |
9af43b54 HS |
2082 | const char *buf, size_t size) |
2083 | { | |
2084 | u64 new; | |
2085 | ||
2086 | if (strict_strtoull(buf, 0, &new) < 0) | |
2087 | return -EINVAL; | |
2088 | ||
2089 | if (mce_cmci_disabled ^ !!new) { | |
2090 | if (new) { | |
2091 | /* disable cmci */ | |
9aaef96f | 2092 | on_each_cpu(mce_disable_cmci, NULL, 1); |
9af43b54 HS |
2093 | mce_cmci_disabled = 1; |
2094 | } else { | |
2095 | /* enable cmci */ | |
2096 | mce_cmci_disabled = 0; | |
2097 | on_each_cpu(mce_enable_ce, NULL, 1); | |
2098 | } | |
2099 | } | |
2100 | return size; | |
2101 | } | |
2102 | ||
8a25a2fd KS |
2103 | static ssize_t store_int_with_restart(struct device *s, |
2104 | struct device_attribute *attr, | |
b56f642d AK |
2105 | const char *buf, size_t size) |
2106 | { | |
8a25a2fd | 2107 | ssize_t ret = device_store_int(s, attr, buf, size); |
b56f642d AK |
2108 | mce_restart(); |
2109 | return ret; | |
2110 | } | |
2111 | ||
8a25a2fd KS |
2112 | static DEVICE_ATTR(trigger, 0644, show_trigger, set_trigger); |
2113 | static DEVICE_INT_ATTR(tolerant, 0644, tolerant); | |
2114 | static DEVICE_INT_ATTR(monarch_timeout, 0644, monarch_timeout); | |
2115 | static DEVICE_INT_ATTR(dont_log_ce, 0644, mce_dont_log_ce); | |
e9eee03e | 2116 | |
8a25a2fd KS |
2117 | static struct dev_ext_attribute dev_attr_check_interval = { |
2118 | __ATTR(check_interval, 0644, device_show_int, store_int_with_restart), | |
b56f642d AK |
2119 | &check_interval |
2120 | }; | |
e9eee03e | 2121 | |
8a25a2fd KS |
2122 | static struct dev_ext_attribute dev_attr_ignore_ce = { |
2123 | __ATTR(ignore_ce, 0644, device_show_int, set_ignore_ce), | |
9af43b54 HS |
2124 | &mce_ignore_ce |
2125 | }; | |
2126 | ||
8a25a2fd KS |
2127 | static struct dev_ext_attribute dev_attr_cmci_disabled = { |
2128 | __ATTR(cmci_disabled, 0644, device_show_int, set_cmci_disabled), | |
9af43b54 HS |
2129 | &mce_cmci_disabled |
2130 | }; | |
2131 | ||
8a25a2fd KS |
2132 | static struct device_attribute *mce_device_attrs[] = { |
2133 | &dev_attr_tolerant.attr, | |
2134 | &dev_attr_check_interval.attr, | |
2135 | &dev_attr_trigger, | |
2136 | &dev_attr_monarch_timeout.attr, | |
2137 | &dev_attr_dont_log_ce.attr, | |
2138 | &dev_attr_ignore_ce.attr, | |
2139 | &dev_attr_cmci_disabled.attr, | |
a98f0dd3 AK |
2140 | NULL |
2141 | }; | |
1da177e4 | 2142 | |
8a25a2fd | 2143 | static cpumask_var_t mce_device_initialized; |
bae19fe0 | 2144 | |
e032d807 GKH |
2145 | static void mce_device_release(struct device *dev) |
2146 | { | |
2147 | kfree(dev); | |
2148 | } | |
2149 | ||
8a25a2fd KS |
2150 | /* Per cpu device init. All of the cpus still share the same ctrl bank: */ |
2151 | static __cpuinit int mce_device_create(unsigned int cpu) | |
1da177e4 | 2152 | { |
e032d807 | 2153 | struct device *dev; |
1da177e4 | 2154 | int err; |
b1f49f95 | 2155 | int i, j; |
92cb7612 | 2156 | |
90367556 | 2157 | if (!mce_available(&boot_cpu_data)) |
91c6d400 AK |
2158 | return -EIO; |
2159 | ||
e032d807 GKH |
2160 | dev = kzalloc(sizeof *dev, GFP_KERNEL); |
2161 | if (!dev) | |
2162 | return -ENOMEM; | |
8a25a2fd KS |
2163 | dev->id = cpu; |
2164 | dev->bus = &mce_subsys; | |
e032d807 | 2165 | dev->release = &mce_device_release; |
91c6d400 | 2166 | |
8a25a2fd | 2167 | err = device_register(dev); |
d435d862 AM |
2168 | if (err) |
2169 | return err; | |
2170 | ||
8a25a2fd KS |
2171 | for (i = 0; mce_device_attrs[i]; i++) { |
2172 | err = device_create_file(dev, mce_device_attrs[i]); | |
d435d862 AM |
2173 | if (err) |
2174 | goto error; | |
2175 | } | |
b1f49f95 | 2176 | for (j = 0; j < banks; j++) { |
8a25a2fd | 2177 | err = device_create_file(dev, &mce_banks[j].attr); |
0d7482e3 AK |
2178 | if (err) |
2179 | goto error2; | |
2180 | } | |
8a25a2fd | 2181 | cpumask_set_cpu(cpu, mce_device_initialized); |
d6126ef5 | 2182 | per_cpu(mce_device, cpu) = dev; |
91c6d400 | 2183 | |
d435d862 | 2184 | return 0; |
0d7482e3 | 2185 | error2: |
b1f49f95 | 2186 | while (--j >= 0) |
8a25a2fd | 2187 | device_remove_file(dev, &mce_banks[j].attr); |
d435d862 | 2188 | error: |
cb491fca | 2189 | while (--i >= 0) |
8a25a2fd | 2190 | device_remove_file(dev, mce_device_attrs[i]); |
cb491fca | 2191 | |
8a25a2fd | 2192 | device_unregister(dev); |
d435d862 | 2193 | |
91c6d400 AK |
2194 | return err; |
2195 | } | |
2196 | ||
8a25a2fd | 2197 | static __cpuinit void mce_device_remove(unsigned int cpu) |
91c6d400 | 2198 | { |
d6126ef5 | 2199 | struct device *dev = per_cpu(mce_device, cpu); |
73ca5358 SL |
2200 | int i; |
2201 | ||
8a25a2fd | 2202 | if (!cpumask_test_cpu(cpu, mce_device_initialized)) |
bae19fe0 AH |
2203 | return; |
2204 | ||
8a25a2fd KS |
2205 | for (i = 0; mce_device_attrs[i]; i++) |
2206 | device_remove_file(dev, mce_device_attrs[i]); | |
cb491fca | 2207 | |
0d7482e3 | 2208 | for (i = 0; i < banks; i++) |
8a25a2fd | 2209 | device_remove_file(dev, &mce_banks[i].attr); |
cb491fca | 2210 | |
8a25a2fd KS |
2211 | device_unregister(dev); |
2212 | cpumask_clear_cpu(cpu, mce_device_initialized); | |
d6126ef5 | 2213 | per_cpu(mce_device, cpu) = NULL; |
91c6d400 | 2214 | } |
91c6d400 | 2215 | |
d6b75584 | 2216 | /* Make sure there are no machine checks on offlined CPUs. */ |
767df1bd | 2217 | static void __cpuinit mce_disable_cpu(void *h) |
d6b75584 | 2218 | { |
88ccbedd | 2219 | unsigned long action = *(unsigned long *)h; |
cb491fca | 2220 | int i; |
d6b75584 | 2221 | |
7b543a53 | 2222 | if (!mce_available(__this_cpu_ptr(&cpu_info))) |
d6b75584 | 2223 | return; |
767df1bd | 2224 | |
88ccbedd AK |
2225 | if (!(action & CPU_TASKS_FROZEN)) |
2226 | cmci_clear(); | |
06b7a7a5 | 2227 | for (i = 0; i < banks; i++) { |
cebe1820 | 2228 | struct mce_bank *b = &mce_banks[i]; |
11868a2d | 2229 | |
cebe1820 | 2230 | if (b->init) |
a2d32bcb | 2231 | wrmsrl(MSR_IA32_MCx_CTL(i), 0); |
06b7a7a5 | 2232 | } |
d6b75584 AK |
2233 | } |
2234 | ||
767df1bd | 2235 | static void __cpuinit mce_reenable_cpu(void *h) |
d6b75584 | 2236 | { |
88ccbedd | 2237 | unsigned long action = *(unsigned long *)h; |
e9eee03e | 2238 | int i; |
d6b75584 | 2239 | |
7b543a53 | 2240 | if (!mce_available(__this_cpu_ptr(&cpu_info))) |
d6b75584 | 2241 | return; |
e9eee03e | 2242 | |
88ccbedd AK |
2243 | if (!(action & CPU_TASKS_FROZEN)) |
2244 | cmci_reenable(); | |
06b7a7a5 | 2245 | for (i = 0; i < banks; i++) { |
cebe1820 | 2246 | struct mce_bank *b = &mce_banks[i]; |
11868a2d | 2247 | |
cebe1820 | 2248 | if (b->init) |
a2d32bcb | 2249 | wrmsrl(MSR_IA32_MCx_CTL(i), b->ctl); |
06b7a7a5 | 2250 | } |
d6b75584 AK |
2251 | } |
2252 | ||
91c6d400 | 2253 | /* Get notified when a cpu comes on/off. Be hotplug friendly. */ |
e9eee03e IM |
2254 | static int __cpuinit |
2255 | mce_cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
91c6d400 AK |
2256 | { |
2257 | unsigned int cpu = (unsigned long)hcpu; | |
52d168e2 | 2258 | struct timer_list *t = &per_cpu(mce_timer, cpu); |
91c6d400 AK |
2259 | |
2260 | switch (action) { | |
bae19fe0 AH |
2261 | case CPU_ONLINE: |
2262 | case CPU_ONLINE_FROZEN: | |
8a25a2fd | 2263 | mce_device_create(cpu); |
8735728e RW |
2264 | if (threshold_cpu_callback) |
2265 | threshold_cpu_callback(action, cpu); | |
91c6d400 | 2266 | break; |
91c6d400 | 2267 | case CPU_DEAD: |
8bb78442 | 2268 | case CPU_DEAD_FROZEN: |
8735728e RW |
2269 | if (threshold_cpu_callback) |
2270 | threshold_cpu_callback(action, cpu); | |
8a25a2fd | 2271 | mce_device_remove(cpu); |
91c6d400 | 2272 | break; |
52d168e2 AK |
2273 | case CPU_DOWN_PREPARE: |
2274 | case CPU_DOWN_PREPARE_FROZEN: | |
2275 | del_timer_sync(t); | |
88ccbedd | 2276 | smp_call_function_single(cpu, mce_disable_cpu, &action, 1); |
52d168e2 AK |
2277 | break; |
2278 | case CPU_DOWN_FAILED: | |
2279 | case CPU_DOWN_FAILED_FROZEN: | |
fe5ed91d HS |
2280 | if (!mce_ignore_ce && check_interval) { |
2281 | t->expires = round_jiffies(jiffies + | |
82f7af09 | 2282 | per_cpu(mce_next_interval, cpu)); |
fe5ed91d HS |
2283 | add_timer_on(t, cpu); |
2284 | } | |
88ccbedd AK |
2285 | smp_call_function_single(cpu, mce_reenable_cpu, &action, 1); |
2286 | break; | |
2287 | case CPU_POST_DEAD: | |
2288 | /* intentionally ignoring frozen here */ | |
2289 | cmci_rediscover(cpu); | |
52d168e2 | 2290 | break; |
91c6d400 | 2291 | } |
bae19fe0 | 2292 | return NOTIFY_OK; |
91c6d400 AK |
2293 | } |
2294 | ||
1e35669d | 2295 | static struct notifier_block mce_cpu_notifier __cpuinitdata = { |
91c6d400 AK |
2296 | .notifier_call = mce_cpu_callback, |
2297 | }; | |
2298 | ||
cebe1820 | 2299 | static __init void mce_init_banks(void) |
0d7482e3 AK |
2300 | { |
2301 | int i; | |
2302 | ||
0d7482e3 | 2303 | for (i = 0; i < banks; i++) { |
cebe1820 | 2304 | struct mce_bank *b = &mce_banks[i]; |
8a25a2fd | 2305 | struct device_attribute *a = &b->attr; |
e9eee03e | 2306 | |
a07e4156 | 2307 | sysfs_attr_init(&a->attr); |
cebe1820 AK |
2308 | a->attr.name = b->attrname; |
2309 | snprintf(b->attrname, ATTR_LEN, "bank%d", i); | |
e9eee03e IM |
2310 | |
2311 | a->attr.mode = 0644; | |
2312 | a->show = show_bank; | |
2313 | a->store = set_bank; | |
0d7482e3 | 2314 | } |
0d7482e3 AK |
2315 | } |
2316 | ||
5e09954a | 2317 | static __init int mcheck_init_device(void) |
91c6d400 AK |
2318 | { |
2319 | int err; | |
2320 | int i = 0; | |
2321 | ||
1da177e4 LT |
2322 | if (!mce_available(&boot_cpu_data)) |
2323 | return -EIO; | |
0d7482e3 | 2324 | |
8a25a2fd | 2325 | zalloc_cpumask_var(&mce_device_initialized, GFP_KERNEL); |
996867d0 | 2326 | |
cebe1820 | 2327 | mce_init_banks(); |
0d7482e3 | 2328 | |
8a25a2fd | 2329 | err = subsys_system_register(&mce_subsys, NULL); |
d435d862 AM |
2330 | if (err) |
2331 | return err; | |
91c6d400 AK |
2332 | |
2333 | for_each_online_cpu(i) { | |
8a25a2fd | 2334 | err = mce_device_create(i); |
d435d862 AM |
2335 | if (err) |
2336 | return err; | |
91c6d400 AK |
2337 | } |
2338 | ||
f3c6ea1b | 2339 | register_syscore_ops(&mce_syscore_ops); |
be6b5a35 | 2340 | register_hotcpu_notifier(&mce_cpu_notifier); |
93b62c3c HS |
2341 | |
2342 | /* register character device /dev/mcelog */ | |
2343 | misc_register(&mce_chrdev_device); | |
e9eee03e | 2344 | |
1da177e4 | 2345 | return err; |
1da177e4 | 2346 | } |
cef12ee5 | 2347 | device_initcall_sync(mcheck_init_device); |
a988d334 | 2348 | |
d7c3c9a6 AK |
2349 | /* |
2350 | * Old style boot options parsing. Only for compatibility. | |
2351 | */ | |
2352 | static int __init mcheck_disable(char *str) | |
2353 | { | |
2354 | mce_disabled = 1; | |
2355 | return 1; | |
2356 | } | |
2357 | __setup("nomce", mcheck_disable); | |
a988d334 | 2358 | |
5be9ed25 HY |
2359 | #ifdef CONFIG_DEBUG_FS |
2360 | struct dentry *mce_get_debugfs_dir(void) | |
a988d334 | 2361 | { |
5be9ed25 | 2362 | static struct dentry *dmce; |
a988d334 | 2363 | |
5be9ed25 HY |
2364 | if (!dmce) |
2365 | dmce = debugfs_create_dir("mce", NULL); | |
a988d334 | 2366 | |
5be9ed25 HY |
2367 | return dmce; |
2368 | } | |
a988d334 | 2369 | |
bf783f9f HY |
2370 | static void mce_reset(void) |
2371 | { | |
2372 | cpu_missing = 0; | |
2373 | atomic_set(&mce_fake_paniced, 0); | |
2374 | atomic_set(&mce_executing, 0); | |
2375 | atomic_set(&mce_callin, 0); | |
2376 | atomic_set(&global_nwo, 0); | |
2377 | } | |
a988d334 | 2378 | |
bf783f9f HY |
2379 | static int fake_panic_get(void *data, u64 *val) |
2380 | { | |
2381 | *val = fake_panic; | |
2382 | return 0; | |
a988d334 IM |
2383 | } |
2384 | ||
bf783f9f | 2385 | static int fake_panic_set(void *data, u64 val) |
a988d334 | 2386 | { |
bf783f9f HY |
2387 | mce_reset(); |
2388 | fake_panic = val; | |
2389 | return 0; | |
a988d334 | 2390 | } |
a988d334 | 2391 | |
bf783f9f HY |
2392 | DEFINE_SIMPLE_ATTRIBUTE(fake_panic_fops, fake_panic_get, |
2393 | fake_panic_set, "%llu\n"); | |
d7c3c9a6 | 2394 | |
5e09954a | 2395 | static int __init mcheck_debugfs_init(void) |
d7c3c9a6 | 2396 | { |
bf783f9f HY |
2397 | struct dentry *dmce, *ffake_panic; |
2398 | ||
2399 | dmce = mce_get_debugfs_dir(); | |
2400 | if (!dmce) | |
2401 | return -ENOMEM; | |
2402 | ffake_panic = debugfs_create_file("fake_panic", 0444, dmce, NULL, | |
2403 | &fake_panic_fops); | |
2404 | if (!ffake_panic) | |
2405 | return -ENOMEM; | |
2406 | ||
2407 | return 0; | |
d7c3c9a6 | 2408 | } |
5e09954a | 2409 | late_initcall(mcheck_debugfs_init); |
5be9ed25 | 2410 | #endif |