Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Machine check handler. | |
3 | * K8 parts Copyright 2002,2003 Andi Kleen, SuSE Labs. | |
d88203d1 TG |
4 | * Rest from unknown author(s). |
5 | * 2004 Andi Kleen. Rewrote most of it. | |
1da177e4 LT |
6 | */ |
7 | ||
8 | #include <linux/init.h> | |
9 | #include <linux/types.h> | |
10 | #include <linux/kernel.h> | |
11 | #include <linux/sched.h> | |
38c4c97c | 12 | #include <linux/smp_lock.h> |
1da177e4 LT |
13 | #include <linux/string.h> |
14 | #include <linux/rcupdate.h> | |
15 | #include <linux/kallsyms.h> | |
16 | #include <linux/sysdev.h> | |
17 | #include <linux/miscdevice.h> | |
18 | #include <linux/fs.h> | |
a9415644 | 19 | #include <linux/capability.h> |
91c6d400 AK |
20 | #include <linux/cpu.h> |
21 | #include <linux/percpu.h> | |
e02e68d3 TH |
22 | #include <linux/poll.h> |
23 | #include <linux/thread_info.h> | |
8c566ef5 | 24 | #include <linux/ctype.h> |
a98f0dd3 | 25 | #include <linux/kmod.h> |
1eeb66a1 | 26 | #include <linux/kdebug.h> |
d88203d1 | 27 | #include <asm/processor.h> |
1da177e4 LT |
28 | #include <asm/msr.h> |
29 | #include <asm/mce.h> | |
1da177e4 | 30 | #include <asm/uaccess.h> |
0a9c3ee7 | 31 | #include <asm/smp.h> |
e02e68d3 | 32 | #include <asm/idle.h> |
1da177e4 LT |
33 | |
34 | #define MISC_MCELOG_MINOR 227 | |
8edc5cc5 | 35 | #define NR_SYSFS_BANKS 6 |
1da177e4 | 36 | |
553f265f AK |
37 | atomic_t mce_entry; |
38 | ||
1da177e4 LT |
39 | static int mce_dont_init; |
40 | ||
bd78432c TH |
41 | /* |
42 | * Tolerant levels: | |
43 | * 0: always panic on uncorrected errors, log corrected errors | |
44 | * 1: panic or SIGBUS on uncorrected errors, log corrected errors | |
45 | * 2: SIGBUS or log uncorrected errors (if possible), log corrected errors | |
46 | * 3: never panic or SIGBUS, log all errors (for testing only) | |
47 | */ | |
1da177e4 LT |
48 | static int tolerant = 1; |
49 | static int banks; | |
8edc5cc5 | 50 | static unsigned long bank[NR_SYSFS_BANKS] = { [0 ... NR_SYSFS_BANKS-1] = ~0UL }; |
e02e68d3 | 51 | static unsigned long notify_user; |
94ad8474 | 52 | static int rip_msr; |
911f6a7b | 53 | static int mce_bootlog = -1; |
a98f0dd3 AK |
54 | static atomic_t mce_events; |
55 | ||
56 | static char trigger[128]; | |
57 | static char *trigger_argv[2] = { trigger, NULL }; | |
1da177e4 | 58 | |
e02e68d3 TH |
59 | static DECLARE_WAIT_QUEUE_HEAD(mce_wait); |
60 | ||
1da177e4 LT |
61 | /* |
62 | * Lockless MCE logging infrastructure. | |
63 | * This avoids deadlocks on printk locks without having to break locks. Also | |
64 | * separate MCEs from kernel messages to avoid bogus bug reports. | |
65 | */ | |
66 | ||
231fd906 | 67 | static struct mce_log mcelog = { |
1da177e4 LT |
68 | MCE_LOG_SIGNATURE, |
69 | MCE_LOG_LEN, | |
d88203d1 | 70 | }; |
1da177e4 LT |
71 | |
72 | void mce_log(struct mce *mce) | |
73 | { | |
74 | unsigned next, entry; | |
a98f0dd3 | 75 | atomic_inc(&mce_events); |
1da177e4 | 76 | mce->finished = 0; |
7644143c | 77 | wmb(); |
1da177e4 LT |
78 | for (;;) { |
79 | entry = rcu_dereference(mcelog.next); | |
673242c1 AK |
80 | for (;;) { |
81 | /* When the buffer fills up discard new entries. Assume | |
82 | that the earlier errors are the more interesting. */ | |
83 | if (entry >= MCE_LOG_LEN) { | |
53756d37 | 84 | set_bit(MCE_OVERFLOW, (unsigned long *)&mcelog.flags); |
673242c1 AK |
85 | return; |
86 | } | |
87 | /* Old left over entry. Skip. */ | |
88 | if (mcelog.entry[entry].finished) { | |
89 | entry++; | |
90 | continue; | |
91 | } | |
7644143c | 92 | break; |
1da177e4 | 93 | } |
1da177e4 LT |
94 | smp_rmb(); |
95 | next = entry + 1; | |
96 | if (cmpxchg(&mcelog.next, entry, next) == entry) | |
97 | break; | |
98 | } | |
99 | memcpy(mcelog.entry + entry, mce, sizeof(struct mce)); | |
7644143c | 100 | wmb(); |
1da177e4 | 101 | mcelog.entry[entry].finished = 1; |
7644143c | 102 | wmb(); |
1da177e4 | 103 | |
e02e68d3 | 104 | set_bit(0, ¬ify_user); |
1da177e4 LT |
105 | } |
106 | ||
107 | static void print_mce(struct mce *m) | |
108 | { | |
109 | printk(KERN_EMERG "\n" | |
4855170f | 110 | KERN_EMERG "HARDWARE ERROR\n" |
1da177e4 LT |
111 | KERN_EMERG |
112 | "CPU %d: Machine Check Exception: %16Lx Bank %d: %016Lx\n", | |
113 | m->cpu, m->mcgstatus, m->bank, m->status); | |
65ea5b03 | 114 | if (m->ip) { |
d88203d1 | 115 | printk(KERN_EMERG "RIP%s %02x:<%016Lx> ", |
1da177e4 | 116 | !(m->mcgstatus & MCG_STATUS_EIPV) ? " !INEXACT!" : "", |
65ea5b03 | 117 | m->cs, m->ip); |
1da177e4 | 118 | if (m->cs == __KERNEL_CS) |
65ea5b03 | 119 | print_symbol("{%s}", m->ip); |
1da177e4 LT |
120 | printk("\n"); |
121 | } | |
d88203d1 | 122 | printk(KERN_EMERG "TSC %Lx ", m->tsc); |
1da177e4 LT |
123 | if (m->addr) |
124 | printk("ADDR %Lx ", m->addr); | |
125 | if (m->misc) | |
d88203d1 | 126 | printk("MISC %Lx ", m->misc); |
1da177e4 | 127 | printk("\n"); |
4855170f | 128 | printk(KERN_EMERG "This is not a software problem!\n"); |
d88203d1 TG |
129 | printk(KERN_EMERG "Run through mcelog --ascii to decode " |
130 | "and contact your hardware vendor\n"); | |
1da177e4 LT |
131 | } |
132 | ||
133 | static void mce_panic(char *msg, struct mce *backup, unsigned long start) | |
d88203d1 | 134 | { |
1da177e4 | 135 | int i; |
e02e68d3 | 136 | |
1da177e4 LT |
137 | oops_begin(); |
138 | for (i = 0; i < MCE_LOG_LEN; i++) { | |
139 | unsigned long tsc = mcelog.entry[i].tsc; | |
d88203d1 | 140 | |
1da177e4 LT |
141 | if (time_before(tsc, start)) |
142 | continue; | |
d88203d1 | 143 | print_mce(&mcelog.entry[i]); |
1da177e4 LT |
144 | if (backup && mcelog.entry[i].tsc == backup->tsc) |
145 | backup = NULL; | |
146 | } | |
147 | if (backup) | |
148 | print_mce(backup); | |
e02e68d3 | 149 | panic(msg); |
d88203d1 | 150 | } |
1da177e4 LT |
151 | |
152 | static int mce_available(struct cpuinfo_x86 *c) | |
153 | { | |
3d1712c9 | 154 | return cpu_has(c, X86_FEATURE_MCE) && cpu_has(c, X86_FEATURE_MCA); |
1da177e4 LT |
155 | } |
156 | ||
94ad8474 AK |
157 | static inline void mce_get_rip(struct mce *m, struct pt_regs *regs) |
158 | { | |
159 | if (regs && (m->mcgstatus & MCG_STATUS_RIPV)) { | |
65ea5b03 | 160 | m->ip = regs->ip; |
94ad8474 AK |
161 | m->cs = regs->cs; |
162 | } else { | |
65ea5b03 | 163 | m->ip = 0; |
94ad8474 AK |
164 | m->cs = 0; |
165 | } | |
166 | if (rip_msr) { | |
167 | /* Assume the RIP in the MSR is exact. Is this true? */ | |
168 | m->mcgstatus |= MCG_STATUS_EIPV; | |
65ea5b03 | 169 | rdmsrl(rip_msr, m->ip); |
94ad8474 AK |
170 | m->cs = 0; |
171 | } | |
172 | } | |
173 | ||
d88203d1 | 174 | /* |
1da177e4 LT |
175 | * The actual machine check handler |
176 | */ | |
1da177e4 LT |
177 | void do_machine_check(struct pt_regs * regs, long error_code) |
178 | { | |
179 | struct mce m, panicm; | |
1da177e4 LT |
180 | u64 mcestart = 0; |
181 | int i; | |
182 | int panicm_found = 0; | |
bd78432c TH |
183 | /* |
184 | * If no_way_out gets set, there is no safe way to recover from this | |
185 | * MCE. If tolerant is cranked up, we'll try anyway. | |
186 | */ | |
187 | int no_way_out = 0; | |
188 | /* | |
189 | * If kill_it gets set, there might be a way to recover from this | |
190 | * error. | |
191 | */ | |
192 | int kill_it = 0; | |
1da177e4 | 193 | |
553f265f AK |
194 | atomic_inc(&mce_entry); |
195 | ||
22f5991c JB |
196 | if ((regs |
197 | && notify_die(DIE_NMI, "machine check", regs, error_code, | |
198 | 18, SIGKILL) == NOTIFY_STOP) | |
199 | || !banks) | |
553f265f | 200 | goto out2; |
1da177e4 LT |
201 | |
202 | memset(&m, 0, sizeof(struct mce)); | |
151f8cc1 | 203 | m.cpu = smp_processor_id(); |
1da177e4 | 204 | rdmsrl(MSR_IA32_MCG_STATUS, m.mcgstatus); |
bd78432c | 205 | /* if the restart IP is not valid, we're done for */ |
1da177e4 | 206 | if (!(m.mcgstatus & MCG_STATUS_RIPV)) |
bd78432c | 207 | no_way_out = 1; |
d88203d1 | 208 | |
1da177e4 LT |
209 | rdtscll(mcestart); |
210 | barrier(); | |
211 | ||
212 | for (i = 0; i < banks; i++) { | |
8edc5cc5 | 213 | if (i < NR_SYSFS_BANKS && !bank[i]) |
1da177e4 | 214 | continue; |
d88203d1 TG |
215 | |
216 | m.misc = 0; | |
1da177e4 LT |
217 | m.addr = 0; |
218 | m.bank = i; | |
219 | m.tsc = 0; | |
220 | ||
221 | rdmsrl(MSR_IA32_MC0_STATUS + i*4, m.status); | |
222 | if ((m.status & MCI_STATUS_VAL) == 0) | |
223 | continue; | |
224 | ||
225 | if (m.status & MCI_STATUS_EN) { | |
bd78432c TH |
226 | /* if PCC was set, there's no way out */ |
227 | no_way_out |= !!(m.status & MCI_STATUS_PCC); | |
228 | /* | |
229 | * If this error was uncorrectable and there was | |
230 | * an overflow, we're in trouble. If no overflow, | |
231 | * we might get away with just killing a task. | |
232 | */ | |
233 | if (m.status & MCI_STATUS_UC) { | |
234 | if (tolerant < 1 || m.status & MCI_STATUS_OVER) | |
235 | no_way_out = 1; | |
236 | kill_it = 1; | |
237 | } | |
1da177e4 LT |
238 | } |
239 | ||
240 | if (m.status & MCI_STATUS_MISCV) | |
241 | rdmsrl(MSR_IA32_MC0_MISC + i*4, m.misc); | |
242 | if (m.status & MCI_STATUS_ADDRV) | |
243 | rdmsrl(MSR_IA32_MC0_ADDR + i*4, m.addr); | |
244 | ||
94ad8474 | 245 | mce_get_rip(&m, regs); |
d5172f26 | 246 | if (error_code >= 0) |
1da177e4 | 247 | rdtscll(m.tsc); |
d5172f26 AK |
248 | if (error_code != -2) |
249 | mce_log(&m); | |
1da177e4 LT |
250 | |
251 | /* Did this bank cause the exception? */ | |
252 | /* Assume that the bank with uncorrectable errors did it, | |
253 | and that there is only a single one. */ | |
254 | if ((m.status & MCI_STATUS_UC) && (m.status & MCI_STATUS_EN)) { | |
255 | panicm = m; | |
256 | panicm_found = 1; | |
257 | } | |
258 | ||
9f158333 | 259 | add_taint(TAINT_MACHINE_CHECK); |
1da177e4 LT |
260 | } |
261 | ||
262 | /* Never do anything final in the polling timer */ | |
e02e68d3 | 263 | if (!regs) |
1da177e4 LT |
264 | goto out; |
265 | ||
266 | /* If we didn't find an uncorrectable error, pick | |
267 | the last one (shouldn't happen, just being safe). */ | |
268 | if (!panicm_found) | |
269 | panicm = m; | |
bd78432c TH |
270 | |
271 | /* | |
272 | * If we have decided that we just CAN'T continue, and the user | |
273 | * has not set tolerant to an insane level, give up and die. | |
274 | */ | |
275 | if (no_way_out && tolerant < 3) | |
1da177e4 | 276 | mce_panic("Machine check", &panicm, mcestart); |
bd78432c TH |
277 | |
278 | /* | |
279 | * If the error seems to be unrecoverable, something should be | |
280 | * done. Try to kill as little as possible. If we can kill just | |
281 | * one task, do that. If the user has set the tolerance very | |
282 | * high, don't try to do anything at all. | |
283 | */ | |
284 | if (kill_it && tolerant < 3) { | |
1da177e4 LT |
285 | int user_space = 0; |
286 | ||
bd78432c TH |
287 | /* |
288 | * If the EIPV bit is set, it means the saved IP is the | |
289 | * instruction which caused the MCE. | |
290 | */ | |
291 | if (m.mcgstatus & MCG_STATUS_EIPV) | |
65ea5b03 | 292 | user_space = panicm.ip && (panicm.cs & 3); |
bd78432c TH |
293 | |
294 | /* | |
295 | * If we know that the error was in user space, send a | |
296 | * SIGBUS. Otherwise, panic if tolerance is low. | |
297 | * | |
380851bc | 298 | * force_sig() takes an awful lot of locks and has a slight |
bd78432c TH |
299 | * risk of deadlocking. |
300 | */ | |
301 | if (user_space) { | |
380851bc | 302 | force_sig(SIGBUS, current); |
bd78432c TH |
303 | } else if (panic_on_oops || tolerant < 2) { |
304 | mce_panic("Uncorrected machine check", | |
305 | &panicm, mcestart); | |
306 | } | |
1da177e4 LT |
307 | } |
308 | ||
e02e68d3 TH |
309 | /* notify userspace ASAP */ |
310 | set_thread_flag(TIF_MCE_NOTIFY); | |
311 | ||
1da177e4 | 312 | out: |
bd78432c TH |
313 | /* the last thing we do is clear state */ |
314 | for (i = 0; i < banks; i++) | |
315 | wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0); | |
1da177e4 | 316 | wrmsrl(MSR_IA32_MCG_STATUS, 0); |
553f265f AK |
317 | out2: |
318 | atomic_dec(&mce_entry); | |
1da177e4 LT |
319 | } |
320 | ||
15d5f839 DZ |
321 | #ifdef CONFIG_X86_MCE_INTEL |
322 | /*** | |
323 | * mce_log_therm_throt_event - Logs the thermal throttling event to mcelog | |
676b1855 | 324 | * @cpu: The CPU on which the event occurred. |
15d5f839 DZ |
325 | * @status: Event status information |
326 | * | |
327 | * This function should be called by the thermal interrupt after the | |
328 | * event has been processed and the decision was made to log the event | |
329 | * further. | |
330 | * | |
331 | * The status parameter will be saved to the 'status' field of 'struct mce' | |
332 | * and historically has been the register value of the | |
333 | * MSR_IA32_THERMAL_STATUS (Intel) msr. | |
334 | */ | |
335 | void mce_log_therm_throt_event(unsigned int cpu, __u64 status) | |
336 | { | |
337 | struct mce m; | |
338 | ||
339 | memset(&m, 0, sizeof(m)); | |
340 | m.cpu = cpu; | |
341 | m.bank = MCE_THERMAL_BANK; | |
342 | m.status = status; | |
343 | rdtscll(m.tsc); | |
344 | mce_log(&m); | |
345 | } | |
346 | #endif /* CONFIG_X86_MCE_INTEL */ | |
347 | ||
1da177e4 | 348 | /* |
8a336b0a TH |
349 | * Periodic polling timer for "silent" machine check errors. If the |
350 | * poller finds an MCE, poll 2x faster. When the poller finds no more | |
351 | * errors, poll 2x slower (up to check_interval seconds). | |
1da177e4 LT |
352 | */ |
353 | ||
354 | static int check_interval = 5 * 60; /* 5 minutes */ | |
8a336b0a | 355 | static int next_interval; /* in jiffies */ |
65f27f38 DH |
356 | static void mcheck_timer(struct work_struct *work); |
357 | static DECLARE_DELAYED_WORK(mcheck_work, mcheck_timer); | |
1da177e4 LT |
358 | |
359 | static void mcheck_check_cpu(void *info) | |
360 | { | |
361 | if (mce_available(¤t_cpu_data)) | |
362 | do_machine_check(NULL, 0); | |
363 | } | |
364 | ||
65f27f38 | 365 | static void mcheck_timer(struct work_struct *work) |
1da177e4 | 366 | { |
15c8b6c1 | 367 | on_each_cpu(mcheck_check_cpu, NULL, 1); |
1da177e4 LT |
368 | |
369 | /* | |
e02e68d3 TH |
370 | * Alert userspace if needed. If we logged an MCE, reduce the |
371 | * polling interval, otherwise increase the polling interval. | |
1da177e4 | 372 | */ |
e02e68d3 TH |
373 | if (mce_notify_user()) { |
374 | next_interval = max(next_interval/2, HZ/100); | |
375 | } else { | |
d88203d1 | 376 | next_interval = min(next_interval * 2, |
22293e58 | 377 | (int)round_jiffies_relative(check_interval*HZ)); |
e02e68d3 TH |
378 | } |
379 | ||
380 | schedule_delayed_work(&mcheck_work, next_interval); | |
381 | } | |
382 | ||
9bd98405 AK |
383 | static void mce_do_trigger(struct work_struct *work) |
384 | { | |
385 | call_usermodehelper(trigger, trigger_argv, NULL, UMH_NO_WAIT); | |
386 | } | |
387 | ||
388 | static DECLARE_WORK(mce_trigger_work, mce_do_trigger); | |
389 | ||
e02e68d3 | 390 | /* |
9bd98405 AK |
391 | * Notify the user(s) about new machine check events. |
392 | * Can be called from interrupt context, but not from machine check/NMI | |
393 | * context. | |
e02e68d3 TH |
394 | */ |
395 | int mce_notify_user(void) | |
396 | { | |
397 | clear_thread_flag(TIF_MCE_NOTIFY); | |
398 | if (test_and_clear_bit(0, ¬ify_user)) { | |
8a336b0a TH |
399 | static unsigned long last_print; |
400 | unsigned long now = jiffies; | |
401 | ||
e02e68d3 | 402 | wake_up_interruptible(&mce_wait); |
9bd98405 AK |
403 | |
404 | /* | |
405 | * There is no risk of missing notifications because | |
406 | * work_pending is always cleared before the function is | |
407 | * executed. | |
408 | */ | |
409 | if (trigger[0] && !work_pending(&mce_trigger_work)) | |
410 | schedule_work(&mce_trigger_work); | |
e02e68d3 | 411 | |
8a336b0a TH |
412 | if (time_after_eq(now, last_print + (check_interval*HZ))) { |
413 | last_print = now; | |
414 | printk(KERN_INFO "Machine check events logged\n"); | |
415 | } | |
e02e68d3 TH |
416 | |
417 | return 1; | |
1da177e4 | 418 | } |
e02e68d3 TH |
419 | return 0; |
420 | } | |
8a336b0a | 421 | |
e02e68d3 TH |
422 | /* see if the idle task needs to notify userspace */ |
423 | static int | |
424 | mce_idle_callback(struct notifier_block *nfb, unsigned long action, void *junk) | |
425 | { | |
426 | /* IDLE_END should be safe - interrupts are back on */ | |
427 | if (action == IDLE_END && test_thread_flag(TIF_MCE_NOTIFY)) | |
428 | mce_notify_user(); | |
429 | ||
430 | return NOTIFY_OK; | |
1da177e4 LT |
431 | } |
432 | ||
e02e68d3 TH |
433 | static struct notifier_block mce_idle_notifier = { |
434 | .notifier_call = mce_idle_callback, | |
435 | }; | |
1da177e4 LT |
436 | |
437 | static __init int periodic_mcheck_init(void) | |
d88203d1 | 438 | { |
8a336b0a TH |
439 | next_interval = check_interval * HZ; |
440 | if (next_interval) | |
22293e58 VP |
441 | schedule_delayed_work(&mcheck_work, |
442 | round_jiffies_relative(next_interval)); | |
e02e68d3 | 443 | idle_notifier_register(&mce_idle_notifier); |
1da177e4 | 444 | return 0; |
d88203d1 | 445 | } |
1da177e4 LT |
446 | __initcall(periodic_mcheck_init); |
447 | ||
448 | ||
d88203d1 | 449 | /* |
1da177e4 LT |
450 | * Initialize Machine Checks for a CPU. |
451 | */ | |
452 | static void mce_init(void *dummy) | |
453 | { | |
454 | u64 cap; | |
455 | int i; | |
456 | ||
457 | rdmsrl(MSR_IA32_MCG_CAP, cap); | |
458 | banks = cap & 0xff; | |
8edc5cc5 | 459 | if (banks > MCE_EXTENDED_BANK) { |
b4b3bd96 | 460 | banks = MCE_EXTENDED_BANK; |
8edc5cc5 VP |
461 | printk(KERN_INFO "MCE: warning: using only %d banks\n", |
462 | MCE_EXTENDED_BANK); | |
1da177e4 | 463 | } |
94ad8474 AK |
464 | /* Use accurate RIP reporting if available. */ |
465 | if ((cap & (1<<9)) && ((cap >> 16) & 0xff) >= 9) | |
466 | rip_msr = MSR_IA32_MCG_EIP; | |
1da177e4 LT |
467 | |
468 | /* Log the machine checks left over from the previous reset. | |
469 | This also clears all registers */ | |
d5172f26 | 470 | do_machine_check(NULL, mce_bootlog ? -1 : -2); |
1da177e4 LT |
471 | |
472 | set_in_cr4(X86_CR4_MCE); | |
473 | ||
474 | if (cap & MCG_CTL_P) | |
475 | wrmsr(MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff); | |
476 | ||
477 | for (i = 0; i < banks; i++) { | |
2d144e63 VP |
478 | if (i < NR_SYSFS_BANKS) |
479 | wrmsrl(MSR_IA32_MC0_CTL+4*i, bank[i]); | |
480 | else | |
481 | wrmsrl(MSR_IA32_MC0_CTL+4*i, ~0UL); | |
482 | ||
1da177e4 | 483 | wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0); |
d88203d1 | 484 | } |
1da177e4 LT |
485 | } |
486 | ||
487 | /* Add per CPU specific workarounds here */ | |
e6982c67 | 488 | static void __cpuinit mce_cpu_quirks(struct cpuinfo_x86 *c) |
d88203d1 | 489 | { |
1da177e4 | 490 | /* This should be disabled by the BIOS, but isn't always */ |
911f6a7b JB |
491 | if (c->x86_vendor == X86_VENDOR_AMD) { |
492 | if(c->x86 == 15) | |
493 | /* disable GART TBL walk error reporting, which trips off | |
494 | incorrectly with the IOMMU & 3ware & Cerberus. */ | |
495 | clear_bit(10, &bank[4]); | |
496 | if(c->x86 <= 17 && mce_bootlog < 0) | |
497 | /* Lots of broken BIOS around that don't clear them | |
498 | by default and leave crap in there. Don't log. */ | |
499 | mce_bootlog = 0; | |
1da177e4 | 500 | } |
e583538f | 501 | |
d88203d1 | 502 | } |
1da177e4 | 503 | |
e6982c67 | 504 | static void __cpuinit mce_cpu_features(struct cpuinfo_x86 *c) |
1da177e4 LT |
505 | { |
506 | switch (c->x86_vendor) { | |
507 | case X86_VENDOR_INTEL: | |
508 | mce_intel_feature_init(c); | |
509 | break; | |
89b831ef JS |
510 | case X86_VENDOR_AMD: |
511 | mce_amd_feature_init(c); | |
512 | break; | |
1da177e4 LT |
513 | default: |
514 | break; | |
515 | } | |
516 | } | |
517 | ||
d88203d1 | 518 | /* |
1da177e4 | 519 | * Called for each booted CPU to set up machine checks. |
d88203d1 | 520 | * Must be called with preempt off. |
1da177e4 | 521 | */ |
e6982c67 | 522 | void __cpuinit mcheck_init(struct cpuinfo_x86 *c) |
1da177e4 | 523 | { |
d88203d1 | 524 | mce_cpu_quirks(c); |
1da177e4 LT |
525 | |
526 | if (mce_dont_init || | |
1da177e4 LT |
527 | !mce_available(c)) |
528 | return; | |
529 | ||
530 | mce_init(NULL); | |
531 | mce_cpu_features(c); | |
532 | } | |
533 | ||
534 | /* | |
535 | * Character device to read and clear the MCE log. | |
536 | */ | |
537 | ||
f528e7ba TH |
538 | static DEFINE_SPINLOCK(mce_state_lock); |
539 | static int open_count; /* #times opened */ | |
540 | static int open_exclu; /* already open exclusive? */ | |
541 | ||
542 | static int mce_open(struct inode *inode, struct file *file) | |
543 | { | |
38c4c97c | 544 | lock_kernel(); |
f528e7ba TH |
545 | spin_lock(&mce_state_lock); |
546 | ||
547 | if (open_exclu || (open_count && (file->f_flags & O_EXCL))) { | |
548 | spin_unlock(&mce_state_lock); | |
38c4c97c | 549 | unlock_kernel(); |
f528e7ba TH |
550 | return -EBUSY; |
551 | } | |
552 | ||
553 | if (file->f_flags & O_EXCL) | |
554 | open_exclu = 1; | |
555 | open_count++; | |
556 | ||
557 | spin_unlock(&mce_state_lock); | |
38c4c97c | 558 | unlock_kernel(); |
f528e7ba | 559 | |
bd78432c | 560 | return nonseekable_open(inode, file); |
f528e7ba TH |
561 | } |
562 | ||
563 | static int mce_release(struct inode *inode, struct file *file) | |
564 | { | |
565 | spin_lock(&mce_state_lock); | |
566 | ||
567 | open_count--; | |
568 | open_exclu = 0; | |
569 | ||
570 | spin_unlock(&mce_state_lock); | |
571 | ||
572 | return 0; | |
573 | } | |
574 | ||
d88203d1 TG |
575 | static void collect_tscs(void *data) |
576 | { | |
1da177e4 | 577 | unsigned long *cpu_tsc = (unsigned long *)data; |
d88203d1 | 578 | |
1da177e4 | 579 | rdtscll(cpu_tsc[smp_processor_id()]); |
d88203d1 | 580 | } |
1da177e4 | 581 | |
d88203d1 TG |
582 | static ssize_t mce_read(struct file *filp, char __user *ubuf, size_t usize, |
583 | loff_t *off) | |
1da177e4 | 584 | { |
f0de53bb | 585 | unsigned long *cpu_tsc; |
8c8b8859 | 586 | static DEFINE_MUTEX(mce_read_mutex); |
1da177e4 LT |
587 | unsigned next; |
588 | char __user *buf = ubuf; | |
589 | int i, err; | |
590 | ||
6bca67f9 | 591 | cpu_tsc = kmalloc(nr_cpu_ids * sizeof(long), GFP_KERNEL); |
f0de53bb AK |
592 | if (!cpu_tsc) |
593 | return -ENOMEM; | |
594 | ||
8c8b8859 | 595 | mutex_lock(&mce_read_mutex); |
1da177e4 LT |
596 | next = rcu_dereference(mcelog.next); |
597 | ||
598 | /* Only supports full reads right now */ | |
d88203d1 | 599 | if (*off != 0 || usize < MCE_LOG_LEN*sizeof(struct mce)) { |
8c8b8859 | 600 | mutex_unlock(&mce_read_mutex); |
f0de53bb | 601 | kfree(cpu_tsc); |
1da177e4 LT |
602 | return -EINVAL; |
603 | } | |
604 | ||
605 | err = 0; | |
d88203d1 | 606 | for (i = 0; i < next; i++) { |
673242c1 | 607 | unsigned long start = jiffies; |
d88203d1 | 608 | |
673242c1 | 609 | while (!mcelog.entry[i].finished) { |
4f84e4be | 610 | if (time_after_eq(jiffies, start + 2)) { |
673242c1 | 611 | memset(mcelog.entry + i,0, sizeof(struct mce)); |
4f84e4be | 612 | goto timeout; |
673242c1 AK |
613 | } |
614 | cpu_relax(); | |
615 | } | |
1da177e4 LT |
616 | smp_rmb(); |
617 | err |= copy_to_user(buf, mcelog.entry + i, sizeof(struct mce)); | |
d88203d1 | 618 | buf += sizeof(struct mce); |
4f84e4be JW |
619 | timeout: |
620 | ; | |
d88203d1 | 621 | } |
1da177e4 LT |
622 | |
623 | memset(mcelog.entry, 0, next * sizeof(struct mce)); | |
624 | mcelog.next = 0; | |
625 | ||
b2b18660 | 626 | synchronize_sched(); |
1da177e4 | 627 | |
d88203d1 TG |
628 | /* |
629 | * Collect entries that were still getting written before the | |
630 | * synchronize. | |
631 | */ | |
15c8b6c1 | 632 | on_each_cpu(collect_tscs, cpu_tsc, 1); |
d88203d1 TG |
633 | for (i = next; i < MCE_LOG_LEN; i++) { |
634 | if (mcelog.entry[i].finished && | |
635 | mcelog.entry[i].tsc < cpu_tsc[mcelog.entry[i].cpu]) { | |
636 | err |= copy_to_user(buf, mcelog.entry+i, | |
637 | sizeof(struct mce)); | |
1da177e4 LT |
638 | smp_rmb(); |
639 | buf += sizeof(struct mce); | |
640 | memset(&mcelog.entry[i], 0, sizeof(struct mce)); | |
641 | } | |
d88203d1 | 642 | } |
8c8b8859 | 643 | mutex_unlock(&mce_read_mutex); |
f0de53bb | 644 | kfree(cpu_tsc); |
d88203d1 | 645 | return err ? -EFAULT : buf - ubuf; |
1da177e4 LT |
646 | } |
647 | ||
e02e68d3 TH |
648 | static unsigned int mce_poll(struct file *file, poll_table *wait) |
649 | { | |
650 | poll_wait(file, &mce_wait, wait); | |
651 | if (rcu_dereference(mcelog.next)) | |
652 | return POLLIN | POLLRDNORM; | |
653 | return 0; | |
654 | } | |
655 | ||
c68461b6 | 656 | static long mce_ioctl(struct file *f, unsigned int cmd, unsigned long arg) |
1da177e4 LT |
657 | { |
658 | int __user *p = (int __user *)arg; | |
d88203d1 | 659 | |
1da177e4 | 660 | if (!capable(CAP_SYS_ADMIN)) |
d88203d1 | 661 | return -EPERM; |
1da177e4 | 662 | switch (cmd) { |
d88203d1 | 663 | case MCE_GET_RECORD_LEN: |
1da177e4 LT |
664 | return put_user(sizeof(struct mce), p); |
665 | case MCE_GET_LOG_LEN: | |
d88203d1 | 666 | return put_user(MCE_LOG_LEN, p); |
1da177e4 LT |
667 | case MCE_GETCLEAR_FLAGS: { |
668 | unsigned flags; | |
d88203d1 TG |
669 | |
670 | do { | |
1da177e4 | 671 | flags = mcelog.flags; |
d88203d1 TG |
672 | } while (cmpxchg(&mcelog.flags, flags, 0) != flags); |
673 | return put_user(flags, p); | |
1da177e4 LT |
674 | } |
675 | default: | |
d88203d1 TG |
676 | return -ENOTTY; |
677 | } | |
1da177e4 LT |
678 | } |
679 | ||
5dfe4c96 | 680 | static const struct file_operations mce_chrdev_ops = { |
f528e7ba TH |
681 | .open = mce_open, |
682 | .release = mce_release, | |
1da177e4 | 683 | .read = mce_read, |
e02e68d3 | 684 | .poll = mce_poll, |
c68461b6 | 685 | .unlocked_ioctl = mce_ioctl, |
1da177e4 LT |
686 | }; |
687 | ||
688 | static struct miscdevice mce_log_device = { | |
689 | MISC_MCELOG_MINOR, | |
690 | "mcelog", | |
691 | &mce_chrdev_ops, | |
692 | }; | |
693 | ||
d88203d1 TG |
694 | /* |
695 | * Old style boot options parsing. Only for compatibility. | |
1da177e4 | 696 | */ |
1da177e4 LT |
697 | static int __init mcheck_disable(char *str) |
698 | { | |
699 | mce_dont_init = 1; | |
9b41046c | 700 | return 1; |
1da177e4 LT |
701 | } |
702 | ||
676b1855 | 703 | /* mce=off disables machine check. Note you can re-enable it later |
d5172f26 | 704 | using sysfs. |
8c566ef5 | 705 | mce=TOLERANCELEVEL (number, see above) |
e583538f AK |
706 | mce=bootlog Log MCEs from before booting. Disabled by default on AMD. |
707 | mce=nobootlog Don't log MCEs from before booting. */ | |
1da177e4 LT |
708 | static int __init mcheck_enable(char *str) |
709 | { | |
710 | if (!strcmp(str, "off")) | |
711 | mce_dont_init = 1; | |
e583538f AK |
712 | else if (!strcmp(str, "bootlog") || !strcmp(str,"nobootlog")) |
713 | mce_bootlog = str[0] == 'b'; | |
8c566ef5 AK |
714 | else if (isdigit(str[0])) |
715 | get_option(&str, &tolerant); | |
1da177e4 | 716 | else |
d88203d1 | 717 | printk("mce= argument %s ignored. Please use /sys", str); |
9b41046c | 718 | return 1; |
1da177e4 LT |
719 | } |
720 | ||
721 | __setup("nomce", mcheck_disable); | |
909dd324 | 722 | __setup("mce=", mcheck_enable); |
1da177e4 | 723 | |
d88203d1 | 724 | /* |
1da177e4 | 725 | * Sysfs support |
d88203d1 | 726 | */ |
1da177e4 | 727 | |
973a2dd1 AK |
728 | /* |
729 | * Disable machine checks on suspend and shutdown. We can't really handle | |
730 | * them later. | |
731 | */ | |
732 | static int mce_disable(void) | |
733 | { | |
734 | int i; | |
735 | ||
736 | for (i = 0; i < banks; i++) | |
737 | wrmsrl(MSR_IA32_MC0_CTL + i*4, 0); | |
738 | return 0; | |
739 | } | |
740 | ||
741 | static int mce_suspend(struct sys_device *dev, pm_message_t state) | |
742 | { | |
743 | return mce_disable(); | |
744 | } | |
745 | ||
746 | static int mce_shutdown(struct sys_device *dev) | |
747 | { | |
748 | return mce_disable(); | |
749 | } | |
750 | ||
413588c7 AK |
751 | /* On resume clear all MCE state. Don't want to see leftovers from the BIOS. |
752 | Only one CPU is active at this time, the others get readded later using | |
753 | CPU hotplug. */ | |
1da177e4 LT |
754 | static int mce_resume(struct sys_device *dev) |
755 | { | |
413588c7 | 756 | mce_init(NULL); |
6ec68bff | 757 | mce_cpu_features(¤t_cpu_data); |
1da177e4 LT |
758 | return 0; |
759 | } | |
760 | ||
761 | /* Reinit MCEs after user configuration changes */ | |
d88203d1 TG |
762 | static void mce_restart(void) |
763 | { | |
8a336b0a | 764 | if (next_interval) |
1da177e4 LT |
765 | cancel_delayed_work(&mcheck_work); |
766 | /* Timer race is harmless here */ | |
15c8b6c1 | 767 | on_each_cpu(mce_init, NULL, 1); |
8a336b0a TH |
768 | next_interval = check_interval * HZ; |
769 | if (next_interval) | |
22293e58 VP |
770 | schedule_delayed_work(&mcheck_work, |
771 | round_jiffies_relative(next_interval)); | |
1da177e4 LT |
772 | } |
773 | ||
774 | static struct sysdev_class mce_sysclass = { | |
973a2dd1 AK |
775 | .suspend = mce_suspend, |
776 | .shutdown = mce_shutdown, | |
1da177e4 | 777 | .resume = mce_resume, |
af5ca3f4 | 778 | .name = "machinecheck", |
1da177e4 LT |
779 | }; |
780 | ||
fff2e89f | 781 | DEFINE_PER_CPU(struct sys_device, device_mce); |
8735728e | 782 | void (*threshold_cpu_callback)(unsigned long action, unsigned int cpu) __cpuinitdata; |
1da177e4 LT |
783 | |
784 | /* Why are there no generic functions for this? */ | |
785 | #define ACCESSOR(name, var, start) \ | |
4a0b2b4d AK |
786 | static ssize_t show_ ## name(struct sys_device *s, \ |
787 | struct sysdev_attribute *attr, \ | |
788 | char *buf) { \ | |
d88203d1 TG |
789 | return sprintf(buf, "%lx\n", (unsigned long)var); \ |
790 | } \ | |
4a0b2b4d AK |
791 | static ssize_t set_ ## name(struct sys_device *s, \ |
792 | struct sysdev_attribute *attr, \ | |
793 | const char *buf, size_t siz) { \ | |
d88203d1 TG |
794 | char *end; \ |
795 | unsigned long new = simple_strtoul(buf, &end, 0); \ | |
796 | if (end == buf) return -EINVAL; \ | |
797 | var = new; \ | |
798 | start; \ | |
799 | return end-buf; \ | |
800 | } \ | |
1da177e4 LT |
801 | static SYSDEV_ATTR(name, 0644, show_ ## name, set_ ## name); |
802 | ||
8edc5cc5 VP |
803 | /* |
804 | * TBD should generate these dynamically based on number of available banks. | |
805 | * Have only 6 contol banks in /sysfs until then. | |
806 | */ | |
1da177e4 LT |
807 | ACCESSOR(bank0ctl,bank[0],mce_restart()) |
808 | ACCESSOR(bank1ctl,bank[1],mce_restart()) | |
809 | ACCESSOR(bank2ctl,bank[2],mce_restart()) | |
810 | ACCESSOR(bank3ctl,bank[3],mce_restart()) | |
811 | ACCESSOR(bank4ctl,bank[4],mce_restart()) | |
73ca5358 | 812 | ACCESSOR(bank5ctl,bank[5],mce_restart()) |
a98f0dd3 | 813 | |
4a0b2b4d AK |
814 | static ssize_t show_trigger(struct sys_device *s, struct sysdev_attribute *attr, |
815 | char *buf) | |
a98f0dd3 AK |
816 | { |
817 | strcpy(buf, trigger); | |
818 | strcat(buf, "\n"); | |
819 | return strlen(trigger) + 1; | |
820 | } | |
821 | ||
4a0b2b4d AK |
822 | static ssize_t set_trigger(struct sys_device *s, struct sysdev_attribute *attr, |
823 | const char *buf,size_t siz) | |
a98f0dd3 AK |
824 | { |
825 | char *p; | |
826 | int len; | |
827 | strncpy(trigger, buf, sizeof(trigger)); | |
828 | trigger[sizeof(trigger)-1] = 0; | |
829 | len = strlen(trigger); | |
830 | p = strchr(trigger, '\n'); | |
831 | if (*p) *p = 0; | |
832 | return len; | |
833 | } | |
834 | ||
835 | static SYSDEV_ATTR(trigger, 0644, show_trigger, set_trigger); | |
d95d62c0 | 836 | static SYSDEV_INT_ATTR(tolerant, 0644, tolerant); |
1da177e4 | 837 | ACCESSOR(check_interval,check_interval,mce_restart()) |
a98f0dd3 AK |
838 | static struct sysdev_attribute *mce_attributes[] = { |
839 | &attr_bank0ctl, &attr_bank1ctl, &attr_bank2ctl, | |
840 | &attr_bank3ctl, &attr_bank4ctl, &attr_bank5ctl, | |
d95d62c0 | 841 | &attr_tolerant.attr, &attr_check_interval, &attr_trigger, |
a98f0dd3 AK |
842 | NULL |
843 | }; | |
1da177e4 | 844 | |
bae19fe0 AH |
845 | static cpumask_t mce_device_initialized = CPU_MASK_NONE; |
846 | ||
91c6d400 AK |
847 | /* Per cpu sysdev init. All of the cpus still share the same ctl bank */ |
848 | static __cpuinit int mce_create_device(unsigned int cpu) | |
1da177e4 LT |
849 | { |
850 | int err; | |
73ca5358 | 851 | int i; |
92cb7612 | 852 | |
90367556 | 853 | if (!mce_available(&boot_cpu_data)) |
91c6d400 AK |
854 | return -EIO; |
855 | ||
d435d862 | 856 | memset(&per_cpu(device_mce, cpu).kobj, 0, sizeof(struct kobject)); |
91c6d400 AK |
857 | per_cpu(device_mce,cpu).id = cpu; |
858 | per_cpu(device_mce,cpu).cls = &mce_sysclass; | |
859 | ||
860 | err = sysdev_register(&per_cpu(device_mce,cpu)); | |
d435d862 AM |
861 | if (err) |
862 | return err; | |
863 | ||
864 | for (i = 0; mce_attributes[i]; i++) { | |
865 | err = sysdev_create_file(&per_cpu(device_mce,cpu), | |
866 | mce_attributes[i]); | |
867 | if (err) | |
868 | goto error; | |
869 | } | |
bae19fe0 | 870 | cpu_set(cpu, mce_device_initialized); |
91c6d400 | 871 | |
d435d862 AM |
872 | return 0; |
873 | error: | |
874 | while (i--) { | |
875 | sysdev_remove_file(&per_cpu(device_mce,cpu), | |
876 | mce_attributes[i]); | |
91c6d400 | 877 | } |
d435d862 AM |
878 | sysdev_unregister(&per_cpu(device_mce,cpu)); |
879 | ||
91c6d400 AK |
880 | return err; |
881 | } | |
882 | ||
2d9cd6c2 | 883 | static __cpuinit void mce_remove_device(unsigned int cpu) |
91c6d400 | 884 | { |
73ca5358 SL |
885 | int i; |
886 | ||
bae19fe0 AH |
887 | if (!cpu_isset(cpu, mce_device_initialized)) |
888 | return; | |
889 | ||
a98f0dd3 | 890 | for (i = 0; mce_attributes[i]; i++) |
73ca5358 | 891 | sysdev_remove_file(&per_cpu(device_mce,cpu), |
a98f0dd3 | 892 | mce_attributes[i]); |
91c6d400 | 893 | sysdev_unregister(&per_cpu(device_mce,cpu)); |
bae19fe0 | 894 | cpu_clear(cpu, mce_device_initialized); |
91c6d400 | 895 | } |
91c6d400 AK |
896 | |
897 | /* Get notified when a cpu comes on/off. Be hotplug friendly. */ | |
1e35669d SR |
898 | static int __cpuinit mce_cpu_callback(struct notifier_block *nfb, |
899 | unsigned long action, void *hcpu) | |
91c6d400 AK |
900 | { |
901 | unsigned int cpu = (unsigned long)hcpu; | |
902 | ||
903 | switch (action) { | |
bae19fe0 AH |
904 | case CPU_ONLINE: |
905 | case CPU_ONLINE_FROZEN: | |
906 | mce_create_device(cpu); | |
8735728e RW |
907 | if (threshold_cpu_callback) |
908 | threshold_cpu_callback(action, cpu); | |
91c6d400 | 909 | break; |
91c6d400 | 910 | case CPU_DEAD: |
8bb78442 | 911 | case CPU_DEAD_FROZEN: |
8735728e RW |
912 | if (threshold_cpu_callback) |
913 | threshold_cpu_callback(action, cpu); | |
91c6d400 AK |
914 | mce_remove_device(cpu); |
915 | break; | |
91c6d400 | 916 | } |
bae19fe0 | 917 | return NOTIFY_OK; |
91c6d400 AK |
918 | } |
919 | ||
1e35669d | 920 | static struct notifier_block mce_cpu_notifier __cpuinitdata = { |
91c6d400 AK |
921 | .notifier_call = mce_cpu_callback, |
922 | }; | |
923 | ||
924 | static __init int mce_init_device(void) | |
925 | { | |
926 | int err; | |
927 | int i = 0; | |
928 | ||
1da177e4 LT |
929 | if (!mce_available(&boot_cpu_data)) |
930 | return -EIO; | |
931 | err = sysdev_class_register(&mce_sysclass); | |
d435d862 AM |
932 | if (err) |
933 | return err; | |
91c6d400 AK |
934 | |
935 | for_each_online_cpu(i) { | |
d435d862 AM |
936 | err = mce_create_device(i); |
937 | if (err) | |
938 | return err; | |
91c6d400 AK |
939 | } |
940 | ||
be6b5a35 | 941 | register_hotcpu_notifier(&mce_cpu_notifier); |
1da177e4 LT |
942 | misc_register(&mce_log_device); |
943 | return err; | |
1da177e4 | 944 | } |
91c6d400 | 945 | |
1da177e4 | 946 | device_initcall(mce_init_device); |