Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Machine check handler. | |
3 | * K8 parts Copyright 2002,2003 Andi Kleen, SuSE Labs. | |
4 | * Rest from unknown author(s). | |
5 | * 2004 Andi Kleen. Rewrote most of it. | |
6 | */ | |
7 | ||
8 | #include <linux/init.h> | |
9 | #include <linux/types.h> | |
10 | #include <linux/kernel.h> | |
11 | #include <linux/sched.h> | |
12 | #include <linux/string.h> | |
13 | #include <linux/rcupdate.h> | |
14 | #include <linux/kallsyms.h> | |
15 | #include <linux/sysdev.h> | |
16 | #include <linux/miscdevice.h> | |
17 | #include <linux/fs.h> | |
a9415644 | 18 | #include <linux/capability.h> |
91c6d400 AK |
19 | #include <linux/cpu.h> |
20 | #include <linux/percpu.h> | |
8c566ef5 | 21 | #include <linux/ctype.h> |
a98f0dd3 | 22 | #include <linux/kmod.h> |
1da177e4 LT |
23 | #include <asm/processor.h> |
24 | #include <asm/msr.h> | |
25 | #include <asm/mce.h> | |
26 | #include <asm/kdebug.h> | |
27 | #include <asm/uaccess.h> | |
0a9c3ee7 | 28 | #include <asm/smp.h> |
1da177e4 LT |
29 | |
30 | #define MISC_MCELOG_MINOR 227 | |
73ca5358 | 31 | #define NR_BANKS 6 |
1da177e4 | 32 | |
553f265f AK |
33 | atomic_t mce_entry; |
34 | ||
1da177e4 LT |
35 | static int mce_dont_init; |
36 | ||
37 | /* 0: always panic, 1: panic if deadlock possible, 2: try to avoid panic, | |
38 | 3: never panic or exit (for testing only) */ | |
39 | static int tolerant = 1; | |
40 | static int banks; | |
41 | static unsigned long bank[NR_BANKS] = { [0 ... NR_BANKS-1] = ~0UL }; | |
42 | static unsigned long console_logged; | |
43 | static int notify_user; | |
94ad8474 | 44 | static int rip_msr; |
e583538f | 45 | static int mce_bootlog = 1; |
a98f0dd3 AK |
46 | static atomic_t mce_events; |
47 | ||
48 | static char trigger[128]; | |
49 | static char *trigger_argv[2] = { trigger, NULL }; | |
1da177e4 LT |
50 | |
51 | /* | |
52 | * Lockless MCE logging infrastructure. | |
53 | * This avoids deadlocks on printk locks without having to break locks. Also | |
54 | * separate MCEs from kernel messages to avoid bogus bug reports. | |
55 | */ | |
56 | ||
57 | struct mce_log mcelog = { | |
58 | MCE_LOG_SIGNATURE, | |
59 | MCE_LOG_LEN, | |
60 | }; | |
61 | ||
62 | void mce_log(struct mce *mce) | |
63 | { | |
64 | unsigned next, entry; | |
a98f0dd3 | 65 | atomic_inc(&mce_events); |
1da177e4 | 66 | mce->finished = 0; |
7644143c | 67 | wmb(); |
1da177e4 LT |
68 | for (;;) { |
69 | entry = rcu_dereference(mcelog.next); | |
7644143c MW |
70 | /* The rmb forces the compiler to reload next in each |
71 | iteration */ | |
72 | rmb(); | |
673242c1 AK |
73 | for (;;) { |
74 | /* When the buffer fills up discard new entries. Assume | |
75 | that the earlier errors are the more interesting. */ | |
76 | if (entry >= MCE_LOG_LEN) { | |
77 | set_bit(MCE_OVERFLOW, &mcelog.flags); | |
78 | return; | |
79 | } | |
80 | /* Old left over entry. Skip. */ | |
81 | if (mcelog.entry[entry].finished) { | |
82 | entry++; | |
83 | continue; | |
84 | } | |
7644143c | 85 | break; |
1da177e4 | 86 | } |
1da177e4 LT |
87 | smp_rmb(); |
88 | next = entry + 1; | |
89 | if (cmpxchg(&mcelog.next, entry, next) == entry) | |
90 | break; | |
91 | } | |
92 | memcpy(mcelog.entry + entry, mce, sizeof(struct mce)); | |
7644143c | 93 | wmb(); |
1da177e4 | 94 | mcelog.entry[entry].finished = 1; |
7644143c | 95 | wmb(); |
1da177e4 LT |
96 | |
97 | if (!test_and_set_bit(0, &console_logged)) | |
98 | notify_user = 1; | |
99 | } | |
100 | ||
101 | static void print_mce(struct mce *m) | |
102 | { | |
103 | printk(KERN_EMERG "\n" | |
4855170f | 104 | KERN_EMERG "HARDWARE ERROR\n" |
1da177e4 LT |
105 | KERN_EMERG |
106 | "CPU %d: Machine Check Exception: %16Lx Bank %d: %016Lx\n", | |
107 | m->cpu, m->mcgstatus, m->bank, m->status); | |
108 | if (m->rip) { | |
109 | printk(KERN_EMERG | |
110 | "RIP%s %02x:<%016Lx> ", | |
111 | !(m->mcgstatus & MCG_STATUS_EIPV) ? " !INEXACT!" : "", | |
112 | m->cs, m->rip); | |
113 | if (m->cs == __KERNEL_CS) | |
114 | print_symbol("{%s}", m->rip); | |
115 | printk("\n"); | |
116 | } | |
117 | printk(KERN_EMERG "TSC %Lx ", m->tsc); | |
118 | if (m->addr) | |
119 | printk("ADDR %Lx ", m->addr); | |
120 | if (m->misc) | |
121 | printk("MISC %Lx ", m->misc); | |
122 | printk("\n"); | |
4855170f AK |
123 | printk(KERN_EMERG "This is not a software problem!\n"); |
124 | printk(KERN_EMERG | |
125 | "Run through mcelog --ascii to decode and contact your hardware vendor\n"); | |
1da177e4 LT |
126 | } |
127 | ||
128 | static void mce_panic(char *msg, struct mce *backup, unsigned long start) | |
129 | { | |
130 | int i; | |
131 | oops_begin(); | |
132 | for (i = 0; i < MCE_LOG_LEN; i++) { | |
133 | unsigned long tsc = mcelog.entry[i].tsc; | |
134 | if (time_before(tsc, start)) | |
135 | continue; | |
136 | print_mce(&mcelog.entry[i]); | |
137 | if (backup && mcelog.entry[i].tsc == backup->tsc) | |
138 | backup = NULL; | |
139 | } | |
140 | if (backup) | |
141 | print_mce(backup); | |
142 | if (tolerant >= 3) | |
143 | printk("Fake panic: %s\n", msg); | |
144 | else | |
145 | panic(msg); | |
146 | } | |
147 | ||
148 | static int mce_available(struct cpuinfo_x86 *c) | |
149 | { | |
3d1712c9 | 150 | return cpu_has(c, X86_FEATURE_MCE) && cpu_has(c, X86_FEATURE_MCA); |
1da177e4 LT |
151 | } |
152 | ||
94ad8474 AK |
153 | static inline void mce_get_rip(struct mce *m, struct pt_regs *regs) |
154 | { | |
155 | if (regs && (m->mcgstatus & MCG_STATUS_RIPV)) { | |
156 | m->rip = regs->rip; | |
157 | m->cs = regs->cs; | |
158 | } else { | |
159 | m->rip = 0; | |
160 | m->cs = 0; | |
161 | } | |
162 | if (rip_msr) { | |
163 | /* Assume the RIP in the MSR is exact. Is this true? */ | |
164 | m->mcgstatus |= MCG_STATUS_EIPV; | |
165 | rdmsrl(rip_msr, m->rip); | |
166 | m->cs = 0; | |
167 | } | |
168 | } | |
169 | ||
a98f0dd3 AK |
170 | static void do_mce_trigger(void) |
171 | { | |
172 | static atomic_t mce_logged; | |
173 | int events = atomic_read(&mce_events); | |
174 | if (events != atomic_read(&mce_logged) && trigger[0]) { | |
175 | /* Small race window, but should be harmless. */ | |
176 | atomic_set(&mce_logged, events); | |
177 | call_usermodehelper(trigger, trigger_argv, NULL, -1); | |
178 | } | |
179 | } | |
180 | ||
1da177e4 LT |
181 | /* |
182 | * The actual machine check handler | |
183 | */ | |
184 | ||
185 | void do_machine_check(struct pt_regs * regs, long error_code) | |
186 | { | |
187 | struct mce m, panicm; | |
188 | int nowayout = (tolerant < 1); | |
189 | int kill_it = 0; | |
190 | u64 mcestart = 0; | |
191 | int i; | |
192 | int panicm_found = 0; | |
193 | ||
553f265f AK |
194 | atomic_inc(&mce_entry); |
195 | ||
1da177e4 | 196 | if (regs) |
6e3f3617 | 197 | notify_die(DIE_NMI, "machine check", regs, error_code, 18, SIGKILL); |
1da177e4 | 198 | if (!banks) |
553f265f | 199 | goto out2; |
1da177e4 LT |
200 | |
201 | memset(&m, 0, sizeof(struct mce)); | |
151f8cc1 | 202 | m.cpu = smp_processor_id(); |
1da177e4 LT |
203 | rdmsrl(MSR_IA32_MCG_STATUS, m.mcgstatus); |
204 | if (!(m.mcgstatus & MCG_STATUS_RIPV)) | |
205 | kill_it = 1; | |
206 | ||
207 | rdtscll(mcestart); | |
208 | barrier(); | |
209 | ||
210 | for (i = 0; i < banks; i++) { | |
211 | if (!bank[i]) | |
212 | continue; | |
213 | ||
214 | m.misc = 0; | |
215 | m.addr = 0; | |
216 | m.bank = i; | |
217 | m.tsc = 0; | |
218 | ||
219 | rdmsrl(MSR_IA32_MC0_STATUS + i*4, m.status); | |
220 | if ((m.status & MCI_STATUS_VAL) == 0) | |
221 | continue; | |
222 | ||
223 | if (m.status & MCI_STATUS_EN) { | |
224 | /* In theory _OVER could be a nowayout too, but | |
225 | assume any overflowed errors were no fatal. */ | |
226 | nowayout |= !!(m.status & MCI_STATUS_PCC); | |
227 | kill_it |= !!(m.status & MCI_STATUS_UC); | |
228 | } | |
229 | ||
230 | if (m.status & MCI_STATUS_MISCV) | |
231 | rdmsrl(MSR_IA32_MC0_MISC + i*4, m.misc); | |
232 | if (m.status & MCI_STATUS_ADDRV) | |
233 | rdmsrl(MSR_IA32_MC0_ADDR + i*4, m.addr); | |
234 | ||
94ad8474 | 235 | mce_get_rip(&m, regs); |
d5172f26 | 236 | if (error_code >= 0) |
1da177e4 LT |
237 | rdtscll(m.tsc); |
238 | wrmsrl(MSR_IA32_MC0_STATUS + i*4, 0); | |
d5172f26 AK |
239 | if (error_code != -2) |
240 | mce_log(&m); | |
1da177e4 LT |
241 | |
242 | /* Did this bank cause the exception? */ | |
243 | /* Assume that the bank with uncorrectable errors did it, | |
244 | and that there is only a single one. */ | |
245 | if ((m.status & MCI_STATUS_UC) && (m.status & MCI_STATUS_EN)) { | |
246 | panicm = m; | |
247 | panicm_found = 1; | |
248 | } | |
249 | ||
9f158333 | 250 | add_taint(TAINT_MACHINE_CHECK); |
1da177e4 LT |
251 | } |
252 | ||
253 | /* Never do anything final in the polling timer */ | |
a98f0dd3 AK |
254 | if (!regs) { |
255 | /* Normal interrupt context here. Call trigger for any new | |
256 | events. */ | |
257 | do_mce_trigger(); | |
1da177e4 | 258 | goto out; |
a98f0dd3 | 259 | } |
1da177e4 LT |
260 | |
261 | /* If we didn't find an uncorrectable error, pick | |
262 | the last one (shouldn't happen, just being safe). */ | |
263 | if (!panicm_found) | |
264 | panicm = m; | |
265 | if (nowayout) | |
266 | mce_panic("Machine check", &panicm, mcestart); | |
267 | if (kill_it) { | |
268 | int user_space = 0; | |
269 | ||
270 | if (m.mcgstatus & MCG_STATUS_RIPV) | |
271 | user_space = panicm.rip && (panicm.cs & 3); | |
272 | ||
273 | /* When the machine was in user space and the CPU didn't get | |
274 | confused it's normally not necessary to panic, unless you | |
275 | are paranoid (tolerant == 0) | |
276 | ||
277 | RED-PEN could be more tolerant for MCEs in idle, | |
278 | but most likely they occur at boot anyways, where | |
279 | it is best to just halt the machine. */ | |
280 | if ((!user_space && (panic_on_oops || tolerant < 2)) || | |
281 | (unsigned)current->pid <= 1) | |
282 | mce_panic("Uncorrected machine check", &panicm, mcestart); | |
283 | ||
284 | /* do_exit takes an awful lot of locks and has as | |
285 | slight risk of deadlocking. If you don't want that | |
286 | don't set tolerant >= 2 */ | |
287 | if (tolerant < 3) | |
288 | do_exit(SIGBUS); | |
289 | } | |
290 | ||
291 | out: | |
292 | /* Last thing done in the machine check exception to clear state. */ | |
293 | wrmsrl(MSR_IA32_MCG_STATUS, 0); | |
553f265f AK |
294 | out2: |
295 | atomic_dec(&mce_entry); | |
1da177e4 LT |
296 | } |
297 | ||
15d5f839 DZ |
298 | #ifdef CONFIG_X86_MCE_INTEL |
299 | /*** | |
300 | * mce_log_therm_throt_event - Logs the thermal throttling event to mcelog | |
301 | * @cpu: The CPU on which the event occured. | |
302 | * @status: Event status information | |
303 | * | |
304 | * This function should be called by the thermal interrupt after the | |
305 | * event has been processed and the decision was made to log the event | |
306 | * further. | |
307 | * | |
308 | * The status parameter will be saved to the 'status' field of 'struct mce' | |
309 | * and historically has been the register value of the | |
310 | * MSR_IA32_THERMAL_STATUS (Intel) msr. | |
311 | */ | |
312 | void mce_log_therm_throt_event(unsigned int cpu, __u64 status) | |
313 | { | |
314 | struct mce m; | |
315 | ||
316 | memset(&m, 0, sizeof(m)); | |
317 | m.cpu = cpu; | |
318 | m.bank = MCE_THERMAL_BANK; | |
319 | m.status = status; | |
320 | rdtscll(m.tsc); | |
321 | mce_log(&m); | |
322 | } | |
323 | #endif /* CONFIG_X86_MCE_INTEL */ | |
324 | ||
1da177e4 LT |
325 | /* |
326 | * Periodic polling timer for "silent" machine check errors. | |
327 | */ | |
328 | ||
329 | static int check_interval = 5 * 60; /* 5 minutes */ | |
65f27f38 DH |
330 | static void mcheck_timer(struct work_struct *work); |
331 | static DECLARE_DELAYED_WORK(mcheck_work, mcheck_timer); | |
1da177e4 LT |
332 | |
333 | static void mcheck_check_cpu(void *info) | |
334 | { | |
335 | if (mce_available(¤t_cpu_data)) | |
336 | do_machine_check(NULL, 0); | |
337 | } | |
338 | ||
65f27f38 | 339 | static void mcheck_timer(struct work_struct *work) |
1da177e4 LT |
340 | { |
341 | on_each_cpu(mcheck_check_cpu, NULL, 1, 1); | |
342 | schedule_delayed_work(&mcheck_work, check_interval * HZ); | |
343 | ||
344 | /* | |
345 | * It's ok to read stale data here for notify_user and | |
346 | * console_logged as we'll simply get the updated versions | |
347 | * on the next mcheck_timer execution and atomic operations | |
348 | * on console_logged act as synchronization for notify_user | |
349 | * writes. | |
350 | */ | |
351 | if (notify_user && console_logged) { | |
352 | notify_user = 0; | |
353 | clear_bit(0, &console_logged); | |
354 | printk(KERN_INFO "Machine check events logged\n"); | |
355 | } | |
356 | } | |
357 | ||
358 | ||
359 | static __init int periodic_mcheck_init(void) | |
360 | { | |
361 | if (check_interval) | |
362 | schedule_delayed_work(&mcheck_work, check_interval*HZ); | |
363 | return 0; | |
364 | } | |
365 | __initcall(periodic_mcheck_init); | |
366 | ||
367 | ||
368 | /* | |
369 | * Initialize Machine Checks for a CPU. | |
370 | */ | |
371 | static void mce_init(void *dummy) | |
372 | { | |
373 | u64 cap; | |
374 | int i; | |
375 | ||
376 | rdmsrl(MSR_IA32_MCG_CAP, cap); | |
377 | banks = cap & 0xff; | |
378 | if (banks > NR_BANKS) { | |
379 | printk(KERN_INFO "MCE: warning: using only %d banks\n", banks); | |
380 | banks = NR_BANKS; | |
381 | } | |
94ad8474 AK |
382 | /* Use accurate RIP reporting if available. */ |
383 | if ((cap & (1<<9)) && ((cap >> 16) & 0xff) >= 9) | |
384 | rip_msr = MSR_IA32_MCG_EIP; | |
1da177e4 LT |
385 | |
386 | /* Log the machine checks left over from the previous reset. | |
387 | This also clears all registers */ | |
d5172f26 | 388 | do_machine_check(NULL, mce_bootlog ? -1 : -2); |
1da177e4 LT |
389 | |
390 | set_in_cr4(X86_CR4_MCE); | |
391 | ||
392 | if (cap & MCG_CTL_P) | |
393 | wrmsr(MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff); | |
394 | ||
395 | for (i = 0; i < banks; i++) { | |
396 | wrmsrl(MSR_IA32_MC0_CTL+4*i, bank[i]); | |
397 | wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0); | |
398 | } | |
399 | } | |
400 | ||
401 | /* Add per CPU specific workarounds here */ | |
e6982c67 | 402 | static void __cpuinit mce_cpu_quirks(struct cpuinfo_x86 *c) |
1da177e4 LT |
403 | { |
404 | /* This should be disabled by the BIOS, but isn't always */ | |
405 | if (c->x86_vendor == X86_VENDOR_AMD && c->x86 == 15) { | |
406 | /* disable GART TBL walk error reporting, which trips off | |
407 | incorrectly with the IOMMU & 3ware & Cerberus. */ | |
408 | clear_bit(10, &bank[4]); | |
e583538f AK |
409 | /* Lots of broken BIOS around that don't clear them |
410 | by default and leave crap in there. Don't log. */ | |
411 | mce_bootlog = 0; | |
1da177e4 | 412 | } |
e583538f | 413 | |
1da177e4 LT |
414 | } |
415 | ||
e6982c67 | 416 | static void __cpuinit mce_cpu_features(struct cpuinfo_x86 *c) |
1da177e4 LT |
417 | { |
418 | switch (c->x86_vendor) { | |
419 | case X86_VENDOR_INTEL: | |
420 | mce_intel_feature_init(c); | |
421 | break; | |
89b831ef JS |
422 | case X86_VENDOR_AMD: |
423 | mce_amd_feature_init(c); | |
424 | break; | |
1da177e4 LT |
425 | default: |
426 | break; | |
427 | } | |
428 | } | |
429 | ||
430 | /* | |
431 | * Called for each booted CPU to set up machine checks. | |
432 | * Must be called with preempt off. | |
433 | */ | |
e6982c67 | 434 | void __cpuinit mcheck_init(struct cpuinfo_x86 *c) |
1da177e4 | 435 | { |
7ded5689 | 436 | static cpumask_t mce_cpus = CPU_MASK_NONE; |
1da177e4 LT |
437 | |
438 | mce_cpu_quirks(c); | |
439 | ||
440 | if (mce_dont_init || | |
441 | cpu_test_and_set(smp_processor_id(), mce_cpus) || | |
442 | !mce_available(c)) | |
443 | return; | |
444 | ||
445 | mce_init(NULL); | |
446 | mce_cpu_features(c); | |
447 | } | |
448 | ||
449 | /* | |
450 | * Character device to read and clear the MCE log. | |
451 | */ | |
452 | ||
453 | static void collect_tscs(void *data) | |
454 | { | |
455 | unsigned long *cpu_tsc = (unsigned long *)data; | |
456 | rdtscll(cpu_tsc[smp_processor_id()]); | |
457 | } | |
458 | ||
459 | static ssize_t mce_read(struct file *filp, char __user *ubuf, size_t usize, loff_t *off) | |
460 | { | |
f0de53bb | 461 | unsigned long *cpu_tsc; |
1da177e4 LT |
462 | static DECLARE_MUTEX(mce_read_sem); |
463 | unsigned next; | |
464 | char __user *buf = ubuf; | |
465 | int i, err; | |
466 | ||
f0de53bb AK |
467 | cpu_tsc = kmalloc(NR_CPUS * sizeof(long), GFP_KERNEL); |
468 | if (!cpu_tsc) | |
469 | return -ENOMEM; | |
470 | ||
1da177e4 LT |
471 | down(&mce_read_sem); |
472 | next = rcu_dereference(mcelog.next); | |
473 | ||
474 | /* Only supports full reads right now */ | |
475 | if (*off != 0 || usize < MCE_LOG_LEN*sizeof(struct mce)) { | |
476 | up(&mce_read_sem); | |
f0de53bb | 477 | kfree(cpu_tsc); |
1da177e4 LT |
478 | return -EINVAL; |
479 | } | |
480 | ||
481 | err = 0; | |
673242c1 AK |
482 | for (i = 0; i < next; i++) { |
483 | unsigned long start = jiffies; | |
484 | while (!mcelog.entry[i].finished) { | |
485 | if (!time_before(jiffies, start + 2)) { | |
486 | memset(mcelog.entry + i,0, sizeof(struct mce)); | |
487 | continue; | |
488 | } | |
489 | cpu_relax(); | |
490 | } | |
1da177e4 LT |
491 | smp_rmb(); |
492 | err |= copy_to_user(buf, mcelog.entry + i, sizeof(struct mce)); | |
493 | buf += sizeof(struct mce); | |
494 | } | |
495 | ||
496 | memset(mcelog.entry, 0, next * sizeof(struct mce)); | |
497 | mcelog.next = 0; | |
498 | ||
b2b18660 | 499 | synchronize_sched(); |
1da177e4 LT |
500 | |
501 | /* Collect entries that were still getting written before the synchronize. */ | |
502 | ||
503 | on_each_cpu(collect_tscs, cpu_tsc, 1, 1); | |
504 | for (i = next; i < MCE_LOG_LEN; i++) { | |
505 | if (mcelog.entry[i].finished && | |
506 | mcelog.entry[i].tsc < cpu_tsc[mcelog.entry[i].cpu]) { | |
507 | err |= copy_to_user(buf, mcelog.entry+i, sizeof(struct mce)); | |
508 | smp_rmb(); | |
509 | buf += sizeof(struct mce); | |
510 | memset(&mcelog.entry[i], 0, sizeof(struct mce)); | |
511 | } | |
512 | } | |
513 | up(&mce_read_sem); | |
f0de53bb | 514 | kfree(cpu_tsc); |
1da177e4 LT |
515 | return err ? -EFAULT : buf - ubuf; |
516 | } | |
517 | ||
518 | static int mce_ioctl(struct inode *i, struct file *f,unsigned int cmd, unsigned long arg) | |
519 | { | |
520 | int __user *p = (int __user *)arg; | |
521 | if (!capable(CAP_SYS_ADMIN)) | |
522 | return -EPERM; | |
523 | switch (cmd) { | |
524 | case MCE_GET_RECORD_LEN: | |
525 | return put_user(sizeof(struct mce), p); | |
526 | case MCE_GET_LOG_LEN: | |
527 | return put_user(MCE_LOG_LEN, p); | |
528 | case MCE_GETCLEAR_FLAGS: { | |
529 | unsigned flags; | |
530 | do { | |
531 | flags = mcelog.flags; | |
532 | } while (cmpxchg(&mcelog.flags, flags, 0) != flags); | |
533 | return put_user(flags, p); | |
534 | } | |
535 | default: | |
536 | return -ENOTTY; | |
537 | } | |
538 | } | |
539 | ||
5dfe4c96 | 540 | static const struct file_operations mce_chrdev_ops = { |
1da177e4 LT |
541 | .read = mce_read, |
542 | .ioctl = mce_ioctl, | |
543 | }; | |
544 | ||
545 | static struct miscdevice mce_log_device = { | |
546 | MISC_MCELOG_MINOR, | |
547 | "mcelog", | |
548 | &mce_chrdev_ops, | |
549 | }; | |
550 | ||
551 | /* | |
552 | * Old style boot options parsing. Only for compatibility. | |
553 | */ | |
554 | ||
555 | static int __init mcheck_disable(char *str) | |
556 | { | |
557 | mce_dont_init = 1; | |
9b41046c | 558 | return 1; |
1da177e4 LT |
559 | } |
560 | ||
561 | /* mce=off disables machine check. Note you can reenable it later | |
d5172f26 | 562 | using sysfs. |
8c566ef5 | 563 | mce=TOLERANCELEVEL (number, see above) |
e583538f AK |
564 | mce=bootlog Log MCEs from before booting. Disabled by default on AMD. |
565 | mce=nobootlog Don't log MCEs from before booting. */ | |
1da177e4 LT |
566 | static int __init mcheck_enable(char *str) |
567 | { | |
d5172f26 AK |
568 | if (*str == '=') |
569 | str++; | |
1da177e4 LT |
570 | if (!strcmp(str, "off")) |
571 | mce_dont_init = 1; | |
e583538f AK |
572 | else if (!strcmp(str, "bootlog") || !strcmp(str,"nobootlog")) |
573 | mce_bootlog = str[0] == 'b'; | |
8c566ef5 AK |
574 | else if (isdigit(str[0])) |
575 | get_option(&str, &tolerant); | |
1da177e4 LT |
576 | else |
577 | printk("mce= argument %s ignored. Please use /sys", str); | |
9b41046c | 578 | return 1; |
1da177e4 LT |
579 | } |
580 | ||
581 | __setup("nomce", mcheck_disable); | |
582 | __setup("mce", mcheck_enable); | |
583 | ||
584 | /* | |
585 | * Sysfs support | |
586 | */ | |
587 | ||
413588c7 AK |
588 | /* On resume clear all MCE state. Don't want to see leftovers from the BIOS. |
589 | Only one CPU is active at this time, the others get readded later using | |
590 | CPU hotplug. */ | |
1da177e4 LT |
591 | static int mce_resume(struct sys_device *dev) |
592 | { | |
413588c7 | 593 | mce_init(NULL); |
1da177e4 LT |
594 | return 0; |
595 | } | |
596 | ||
597 | /* Reinit MCEs after user configuration changes */ | |
598 | static void mce_restart(void) | |
599 | { | |
600 | if (check_interval) | |
601 | cancel_delayed_work(&mcheck_work); | |
602 | /* Timer race is harmless here */ | |
603 | on_each_cpu(mce_init, NULL, 1, 1); | |
604 | if (check_interval) | |
605 | schedule_delayed_work(&mcheck_work, check_interval*HZ); | |
606 | } | |
607 | ||
608 | static struct sysdev_class mce_sysclass = { | |
609 | .resume = mce_resume, | |
610 | set_kset_name("machinecheck"), | |
611 | }; | |
612 | ||
fff2e89f | 613 | DEFINE_PER_CPU(struct sys_device, device_mce); |
1da177e4 LT |
614 | |
615 | /* Why are there no generic functions for this? */ | |
616 | #define ACCESSOR(name, var, start) \ | |
617 | static ssize_t show_ ## name(struct sys_device *s, char *buf) { \ | |
618 | return sprintf(buf, "%lx\n", (unsigned long)var); \ | |
619 | } \ | |
620 | static ssize_t set_ ## name(struct sys_device *s,const char *buf,size_t siz) { \ | |
621 | char *end; \ | |
622 | unsigned long new = simple_strtoul(buf, &end, 0); \ | |
623 | if (end == buf) return -EINVAL; \ | |
624 | var = new; \ | |
625 | start; \ | |
626 | return end-buf; \ | |
627 | } \ | |
628 | static SYSDEV_ATTR(name, 0644, show_ ## name, set_ ## name); | |
629 | ||
a98f0dd3 | 630 | /* TBD should generate these dynamically based on number of available banks */ |
1da177e4 LT |
631 | ACCESSOR(bank0ctl,bank[0],mce_restart()) |
632 | ACCESSOR(bank1ctl,bank[1],mce_restart()) | |
633 | ACCESSOR(bank2ctl,bank[2],mce_restart()) | |
634 | ACCESSOR(bank3ctl,bank[3],mce_restart()) | |
635 | ACCESSOR(bank4ctl,bank[4],mce_restart()) | |
73ca5358 | 636 | ACCESSOR(bank5ctl,bank[5],mce_restart()) |
a98f0dd3 AK |
637 | |
638 | static ssize_t show_trigger(struct sys_device *s, char *buf) | |
639 | { | |
640 | strcpy(buf, trigger); | |
641 | strcat(buf, "\n"); | |
642 | return strlen(trigger) + 1; | |
643 | } | |
644 | ||
645 | static ssize_t set_trigger(struct sys_device *s,const char *buf,size_t siz) | |
646 | { | |
647 | char *p; | |
648 | int len; | |
649 | strncpy(trigger, buf, sizeof(trigger)); | |
650 | trigger[sizeof(trigger)-1] = 0; | |
651 | len = strlen(trigger); | |
652 | p = strchr(trigger, '\n'); | |
653 | if (*p) *p = 0; | |
654 | return len; | |
655 | } | |
656 | ||
657 | static SYSDEV_ATTR(trigger, 0644, show_trigger, set_trigger); | |
1da177e4 LT |
658 | ACCESSOR(tolerant,tolerant,) |
659 | ACCESSOR(check_interval,check_interval,mce_restart()) | |
a98f0dd3 AK |
660 | static struct sysdev_attribute *mce_attributes[] = { |
661 | &attr_bank0ctl, &attr_bank1ctl, &attr_bank2ctl, | |
662 | &attr_bank3ctl, &attr_bank4ctl, &attr_bank5ctl, | |
663 | &attr_tolerant, &attr_check_interval, &attr_trigger, | |
664 | NULL | |
665 | }; | |
1da177e4 | 666 | |
91c6d400 AK |
667 | /* Per cpu sysdev init. All of the cpus still share the same ctl bank */ |
668 | static __cpuinit int mce_create_device(unsigned int cpu) | |
1da177e4 LT |
669 | { |
670 | int err; | |
73ca5358 | 671 | int i; |
91c6d400 AK |
672 | if (!mce_available(&cpu_data[cpu])) |
673 | return -EIO; | |
674 | ||
675 | per_cpu(device_mce,cpu).id = cpu; | |
676 | per_cpu(device_mce,cpu).cls = &mce_sysclass; | |
677 | ||
678 | err = sysdev_register(&per_cpu(device_mce,cpu)); | |
679 | ||
680 | if (!err) { | |
a98f0dd3 | 681 | for (i = 0; mce_attributes[i]; i++) |
73ca5358 | 682 | sysdev_create_file(&per_cpu(device_mce,cpu), |
a98f0dd3 | 683 | mce_attributes[i]); |
91c6d400 AK |
684 | } |
685 | return err; | |
686 | } | |
687 | ||
be6b5a35 | 688 | static void mce_remove_device(unsigned int cpu) |
91c6d400 | 689 | { |
73ca5358 SL |
690 | int i; |
691 | ||
a98f0dd3 | 692 | for (i = 0; mce_attributes[i]; i++) |
73ca5358 | 693 | sysdev_remove_file(&per_cpu(device_mce,cpu), |
a98f0dd3 | 694 | mce_attributes[i]); |
91c6d400 | 695 | sysdev_unregister(&per_cpu(device_mce,cpu)); |
d4c45718 | 696 | memset(&per_cpu(device_mce, cpu).kobj, 0, sizeof(struct kobject)); |
91c6d400 | 697 | } |
91c6d400 AK |
698 | |
699 | /* Get notified when a cpu comes on/off. Be hotplug friendly. */ | |
be6b5a35 | 700 | static int |
91c6d400 AK |
701 | mce_cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) |
702 | { | |
703 | unsigned int cpu = (unsigned long)hcpu; | |
704 | ||
705 | switch (action) { | |
706 | case CPU_ONLINE: | |
707 | mce_create_device(cpu); | |
708 | break; | |
91c6d400 AK |
709 | case CPU_DEAD: |
710 | mce_remove_device(cpu); | |
711 | break; | |
91c6d400 AK |
712 | } |
713 | return NOTIFY_OK; | |
714 | } | |
715 | ||
be6b5a35 | 716 | static struct notifier_block mce_cpu_notifier = { |
91c6d400 AK |
717 | .notifier_call = mce_cpu_callback, |
718 | }; | |
719 | ||
720 | static __init int mce_init_device(void) | |
721 | { | |
722 | int err; | |
723 | int i = 0; | |
724 | ||
1da177e4 LT |
725 | if (!mce_available(&boot_cpu_data)) |
726 | return -EIO; | |
727 | err = sysdev_class_register(&mce_sysclass); | |
91c6d400 AK |
728 | |
729 | for_each_online_cpu(i) { | |
730 | mce_create_device(i); | |
731 | } | |
732 | ||
be6b5a35 | 733 | register_hotcpu_notifier(&mce_cpu_notifier); |
1da177e4 LT |
734 | misc_register(&mce_log_device); |
735 | return err; | |
1da177e4 | 736 | } |
91c6d400 | 737 | |
1da177e4 | 738 | device_initcall(mce_init_device); |