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> | |
91c6d400 AK |
18 | #include <linux/cpu.h> |
19 | #include <linux/percpu.h> | |
8c566ef5 | 20 | #include <linux/ctype.h> |
1da177e4 LT |
21 | #include <asm/processor.h> |
22 | #include <asm/msr.h> | |
23 | #include <asm/mce.h> | |
24 | #include <asm/kdebug.h> | |
25 | #include <asm/uaccess.h> | |
26 | ||
27 | #define MISC_MCELOG_MINOR 227 | |
28 | #define NR_BANKS 5 | |
29 | ||
30 | static int mce_dont_init; | |
31 | ||
32 | /* 0: always panic, 1: panic if deadlock possible, 2: try to avoid panic, | |
33 | 3: never panic or exit (for testing only) */ | |
34 | static int tolerant = 1; | |
35 | static int banks; | |
36 | static unsigned long bank[NR_BANKS] = { [0 ... NR_BANKS-1] = ~0UL }; | |
37 | static unsigned long console_logged; | |
38 | static int notify_user; | |
94ad8474 | 39 | static int rip_msr; |
d5172f26 | 40 | static int mce_bootlog; |
1da177e4 LT |
41 | |
42 | /* | |
43 | * Lockless MCE logging infrastructure. | |
44 | * This avoids deadlocks on printk locks without having to break locks. Also | |
45 | * separate MCEs from kernel messages to avoid bogus bug reports. | |
46 | */ | |
47 | ||
48 | struct mce_log mcelog = { | |
49 | MCE_LOG_SIGNATURE, | |
50 | MCE_LOG_LEN, | |
51 | }; | |
52 | ||
53 | void mce_log(struct mce *mce) | |
54 | { | |
55 | unsigned next, entry; | |
56 | mce->finished = 0; | |
57 | smp_wmb(); | |
58 | for (;;) { | |
59 | entry = rcu_dereference(mcelog.next); | |
673242c1 AK |
60 | for (;;) { |
61 | /* When the buffer fills up discard new entries. Assume | |
62 | that the earlier errors are the more interesting. */ | |
63 | if (entry >= MCE_LOG_LEN) { | |
64 | set_bit(MCE_OVERFLOW, &mcelog.flags); | |
65 | return; | |
66 | } | |
67 | /* Old left over entry. Skip. */ | |
68 | if (mcelog.entry[entry].finished) { | |
69 | entry++; | |
70 | continue; | |
71 | } | |
1da177e4 | 72 | } |
1da177e4 LT |
73 | smp_rmb(); |
74 | next = entry + 1; | |
75 | if (cmpxchg(&mcelog.next, entry, next) == entry) | |
76 | break; | |
77 | } | |
78 | memcpy(mcelog.entry + entry, mce, sizeof(struct mce)); | |
79 | smp_wmb(); | |
80 | mcelog.entry[entry].finished = 1; | |
81 | smp_wmb(); | |
82 | ||
83 | if (!test_and_set_bit(0, &console_logged)) | |
84 | notify_user = 1; | |
85 | } | |
86 | ||
87 | static void print_mce(struct mce *m) | |
88 | { | |
89 | printk(KERN_EMERG "\n" | |
90 | KERN_EMERG | |
91 | "CPU %d: Machine Check Exception: %16Lx Bank %d: %016Lx\n", | |
92 | m->cpu, m->mcgstatus, m->bank, m->status); | |
93 | if (m->rip) { | |
94 | printk(KERN_EMERG | |
95 | "RIP%s %02x:<%016Lx> ", | |
96 | !(m->mcgstatus & MCG_STATUS_EIPV) ? " !INEXACT!" : "", | |
97 | m->cs, m->rip); | |
98 | if (m->cs == __KERNEL_CS) | |
99 | print_symbol("{%s}", m->rip); | |
100 | printk("\n"); | |
101 | } | |
102 | printk(KERN_EMERG "TSC %Lx ", m->tsc); | |
103 | if (m->addr) | |
104 | printk("ADDR %Lx ", m->addr); | |
105 | if (m->misc) | |
106 | printk("MISC %Lx ", m->misc); | |
107 | printk("\n"); | |
108 | } | |
109 | ||
110 | static void mce_panic(char *msg, struct mce *backup, unsigned long start) | |
111 | { | |
112 | int i; | |
113 | oops_begin(); | |
114 | for (i = 0; i < MCE_LOG_LEN; i++) { | |
115 | unsigned long tsc = mcelog.entry[i].tsc; | |
116 | if (time_before(tsc, start)) | |
117 | continue; | |
118 | print_mce(&mcelog.entry[i]); | |
119 | if (backup && mcelog.entry[i].tsc == backup->tsc) | |
120 | backup = NULL; | |
121 | } | |
122 | if (backup) | |
123 | print_mce(backup); | |
124 | if (tolerant >= 3) | |
125 | printk("Fake panic: %s\n", msg); | |
126 | else | |
127 | panic(msg); | |
128 | } | |
129 | ||
130 | static int mce_available(struct cpuinfo_x86 *c) | |
131 | { | |
132 | return test_bit(X86_FEATURE_MCE, &c->x86_capability) && | |
133 | test_bit(X86_FEATURE_MCA, &c->x86_capability); | |
134 | } | |
135 | ||
94ad8474 AK |
136 | static inline void mce_get_rip(struct mce *m, struct pt_regs *regs) |
137 | { | |
138 | if (regs && (m->mcgstatus & MCG_STATUS_RIPV)) { | |
139 | m->rip = regs->rip; | |
140 | m->cs = regs->cs; | |
141 | } else { | |
142 | m->rip = 0; | |
143 | m->cs = 0; | |
144 | } | |
145 | if (rip_msr) { | |
146 | /* Assume the RIP in the MSR is exact. Is this true? */ | |
147 | m->mcgstatus |= MCG_STATUS_EIPV; | |
148 | rdmsrl(rip_msr, m->rip); | |
149 | m->cs = 0; | |
150 | } | |
151 | } | |
152 | ||
1da177e4 LT |
153 | /* |
154 | * The actual machine check handler | |
155 | */ | |
156 | ||
157 | void do_machine_check(struct pt_regs * regs, long error_code) | |
158 | { | |
159 | struct mce m, panicm; | |
160 | int nowayout = (tolerant < 1); | |
161 | int kill_it = 0; | |
162 | u64 mcestart = 0; | |
163 | int i; | |
164 | int panicm_found = 0; | |
165 | ||
166 | if (regs) | |
167 | notify_die(DIE_NMI, "machine check", regs, error_code, 255, SIGKILL); | |
168 | if (!banks) | |
169 | return; | |
170 | ||
171 | memset(&m, 0, sizeof(struct mce)); | |
172 | m.cpu = hard_smp_processor_id(); | |
173 | rdmsrl(MSR_IA32_MCG_STATUS, m.mcgstatus); | |
174 | if (!(m.mcgstatus & MCG_STATUS_RIPV)) | |
175 | kill_it = 1; | |
176 | ||
177 | rdtscll(mcestart); | |
178 | barrier(); | |
179 | ||
180 | for (i = 0; i < banks; i++) { | |
181 | if (!bank[i]) | |
182 | continue; | |
183 | ||
184 | m.misc = 0; | |
185 | m.addr = 0; | |
186 | m.bank = i; | |
187 | m.tsc = 0; | |
188 | ||
189 | rdmsrl(MSR_IA32_MC0_STATUS + i*4, m.status); | |
190 | if ((m.status & MCI_STATUS_VAL) == 0) | |
191 | continue; | |
192 | ||
193 | if (m.status & MCI_STATUS_EN) { | |
194 | /* In theory _OVER could be a nowayout too, but | |
195 | assume any overflowed errors were no fatal. */ | |
196 | nowayout |= !!(m.status & MCI_STATUS_PCC); | |
197 | kill_it |= !!(m.status & MCI_STATUS_UC); | |
198 | } | |
199 | ||
200 | if (m.status & MCI_STATUS_MISCV) | |
201 | rdmsrl(MSR_IA32_MC0_MISC + i*4, m.misc); | |
202 | if (m.status & MCI_STATUS_ADDRV) | |
203 | rdmsrl(MSR_IA32_MC0_ADDR + i*4, m.addr); | |
204 | ||
94ad8474 | 205 | mce_get_rip(&m, regs); |
d5172f26 | 206 | if (error_code >= 0) |
1da177e4 LT |
207 | rdtscll(m.tsc); |
208 | wrmsrl(MSR_IA32_MC0_STATUS + i*4, 0); | |
d5172f26 AK |
209 | if (error_code != -2) |
210 | mce_log(&m); | |
1da177e4 LT |
211 | |
212 | /* Did this bank cause the exception? */ | |
213 | /* Assume that the bank with uncorrectable errors did it, | |
214 | and that there is only a single one. */ | |
215 | if ((m.status & MCI_STATUS_UC) && (m.status & MCI_STATUS_EN)) { | |
216 | panicm = m; | |
217 | panicm_found = 1; | |
218 | } | |
219 | ||
220 | tainted |= TAINT_MACHINE_CHECK; | |
221 | } | |
222 | ||
223 | /* Never do anything final in the polling timer */ | |
224 | if (!regs) | |
225 | goto out; | |
226 | ||
227 | /* If we didn't find an uncorrectable error, pick | |
228 | the last one (shouldn't happen, just being safe). */ | |
229 | if (!panicm_found) | |
230 | panicm = m; | |
231 | if (nowayout) | |
232 | mce_panic("Machine check", &panicm, mcestart); | |
233 | if (kill_it) { | |
234 | int user_space = 0; | |
235 | ||
236 | if (m.mcgstatus & MCG_STATUS_RIPV) | |
237 | user_space = panicm.rip && (panicm.cs & 3); | |
238 | ||
239 | /* When the machine was in user space and the CPU didn't get | |
240 | confused it's normally not necessary to panic, unless you | |
241 | are paranoid (tolerant == 0) | |
242 | ||
243 | RED-PEN could be more tolerant for MCEs in idle, | |
244 | but most likely they occur at boot anyways, where | |
245 | it is best to just halt the machine. */ | |
246 | if ((!user_space && (panic_on_oops || tolerant < 2)) || | |
247 | (unsigned)current->pid <= 1) | |
248 | mce_panic("Uncorrected machine check", &panicm, mcestart); | |
249 | ||
250 | /* do_exit takes an awful lot of locks and has as | |
251 | slight risk of deadlocking. If you don't want that | |
252 | don't set tolerant >= 2 */ | |
253 | if (tolerant < 3) | |
254 | do_exit(SIGBUS); | |
255 | } | |
256 | ||
257 | out: | |
258 | /* Last thing done in the machine check exception to clear state. */ | |
259 | wrmsrl(MSR_IA32_MCG_STATUS, 0); | |
260 | } | |
261 | ||
262 | /* | |
263 | * Periodic polling timer for "silent" machine check errors. | |
264 | */ | |
265 | ||
266 | static int check_interval = 5 * 60; /* 5 minutes */ | |
267 | static void mcheck_timer(void *data); | |
268 | static DECLARE_WORK(mcheck_work, mcheck_timer, NULL); | |
269 | ||
270 | static void mcheck_check_cpu(void *info) | |
271 | { | |
272 | if (mce_available(¤t_cpu_data)) | |
273 | do_machine_check(NULL, 0); | |
274 | } | |
275 | ||
276 | static void mcheck_timer(void *data) | |
277 | { | |
278 | on_each_cpu(mcheck_check_cpu, NULL, 1, 1); | |
279 | schedule_delayed_work(&mcheck_work, check_interval * HZ); | |
280 | ||
281 | /* | |
282 | * It's ok to read stale data here for notify_user and | |
283 | * console_logged as we'll simply get the updated versions | |
284 | * on the next mcheck_timer execution and atomic operations | |
285 | * on console_logged act as synchronization for notify_user | |
286 | * writes. | |
287 | */ | |
288 | if (notify_user && console_logged) { | |
289 | notify_user = 0; | |
290 | clear_bit(0, &console_logged); | |
291 | printk(KERN_INFO "Machine check events logged\n"); | |
292 | } | |
293 | } | |
294 | ||
295 | ||
296 | static __init int periodic_mcheck_init(void) | |
297 | { | |
298 | if (check_interval) | |
299 | schedule_delayed_work(&mcheck_work, check_interval*HZ); | |
300 | return 0; | |
301 | } | |
302 | __initcall(periodic_mcheck_init); | |
303 | ||
304 | ||
305 | /* | |
306 | * Initialize Machine Checks for a CPU. | |
307 | */ | |
308 | static void mce_init(void *dummy) | |
309 | { | |
310 | u64 cap; | |
311 | int i; | |
312 | ||
313 | rdmsrl(MSR_IA32_MCG_CAP, cap); | |
314 | banks = cap & 0xff; | |
315 | if (banks > NR_BANKS) { | |
316 | printk(KERN_INFO "MCE: warning: using only %d banks\n", banks); | |
317 | banks = NR_BANKS; | |
318 | } | |
94ad8474 AK |
319 | /* Use accurate RIP reporting if available. */ |
320 | if ((cap & (1<<9)) && ((cap >> 16) & 0xff) >= 9) | |
321 | rip_msr = MSR_IA32_MCG_EIP; | |
1da177e4 LT |
322 | |
323 | /* Log the machine checks left over from the previous reset. | |
324 | This also clears all registers */ | |
d5172f26 | 325 | do_machine_check(NULL, mce_bootlog ? -1 : -2); |
1da177e4 LT |
326 | |
327 | set_in_cr4(X86_CR4_MCE); | |
328 | ||
329 | if (cap & MCG_CTL_P) | |
330 | wrmsr(MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff); | |
331 | ||
332 | for (i = 0; i < banks; i++) { | |
333 | wrmsrl(MSR_IA32_MC0_CTL+4*i, bank[i]); | |
334 | wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0); | |
335 | } | |
336 | } | |
337 | ||
338 | /* Add per CPU specific workarounds here */ | |
e6982c67 | 339 | static void __cpuinit mce_cpu_quirks(struct cpuinfo_x86 *c) |
1da177e4 LT |
340 | { |
341 | /* This should be disabled by the BIOS, but isn't always */ | |
342 | if (c->x86_vendor == X86_VENDOR_AMD && c->x86 == 15) { | |
343 | /* disable GART TBL walk error reporting, which trips off | |
344 | incorrectly with the IOMMU & 3ware & Cerberus. */ | |
345 | clear_bit(10, &bank[4]); | |
346 | } | |
347 | } | |
348 | ||
e6982c67 | 349 | static void __cpuinit mce_cpu_features(struct cpuinfo_x86 *c) |
1da177e4 LT |
350 | { |
351 | switch (c->x86_vendor) { | |
352 | case X86_VENDOR_INTEL: | |
353 | mce_intel_feature_init(c); | |
354 | break; | |
355 | default: | |
356 | break; | |
357 | } | |
358 | } | |
359 | ||
360 | /* | |
361 | * Called for each booted CPU to set up machine checks. | |
362 | * Must be called with preempt off. | |
363 | */ | |
e6982c67 | 364 | void __cpuinit mcheck_init(struct cpuinfo_x86 *c) |
1da177e4 LT |
365 | { |
366 | static cpumask_t mce_cpus __initdata = CPU_MASK_NONE; | |
367 | ||
368 | mce_cpu_quirks(c); | |
369 | ||
370 | if (mce_dont_init || | |
371 | cpu_test_and_set(smp_processor_id(), mce_cpus) || | |
372 | !mce_available(c)) | |
373 | return; | |
374 | ||
375 | mce_init(NULL); | |
376 | mce_cpu_features(c); | |
377 | } | |
378 | ||
379 | /* | |
380 | * Character device to read and clear the MCE log. | |
381 | */ | |
382 | ||
383 | static void collect_tscs(void *data) | |
384 | { | |
385 | unsigned long *cpu_tsc = (unsigned long *)data; | |
386 | rdtscll(cpu_tsc[smp_processor_id()]); | |
387 | } | |
388 | ||
389 | static ssize_t mce_read(struct file *filp, char __user *ubuf, size_t usize, loff_t *off) | |
390 | { | |
f0de53bb | 391 | unsigned long *cpu_tsc; |
1da177e4 LT |
392 | static DECLARE_MUTEX(mce_read_sem); |
393 | unsigned next; | |
394 | char __user *buf = ubuf; | |
395 | int i, err; | |
396 | ||
f0de53bb AK |
397 | cpu_tsc = kmalloc(NR_CPUS * sizeof(long), GFP_KERNEL); |
398 | if (!cpu_tsc) | |
399 | return -ENOMEM; | |
400 | ||
1da177e4 LT |
401 | down(&mce_read_sem); |
402 | next = rcu_dereference(mcelog.next); | |
403 | ||
404 | /* Only supports full reads right now */ | |
405 | if (*off != 0 || usize < MCE_LOG_LEN*sizeof(struct mce)) { | |
406 | up(&mce_read_sem); | |
f0de53bb | 407 | kfree(cpu_tsc); |
1da177e4 LT |
408 | return -EINVAL; |
409 | } | |
410 | ||
411 | err = 0; | |
673242c1 AK |
412 | for (i = 0; i < next; i++) { |
413 | unsigned long start = jiffies; | |
414 | while (!mcelog.entry[i].finished) { | |
415 | if (!time_before(jiffies, start + 2)) { | |
416 | memset(mcelog.entry + i,0, sizeof(struct mce)); | |
417 | continue; | |
418 | } | |
419 | cpu_relax(); | |
420 | } | |
1da177e4 LT |
421 | smp_rmb(); |
422 | err |= copy_to_user(buf, mcelog.entry + i, sizeof(struct mce)); | |
423 | buf += sizeof(struct mce); | |
424 | } | |
425 | ||
426 | memset(mcelog.entry, 0, next * sizeof(struct mce)); | |
427 | mcelog.next = 0; | |
428 | ||
b2b18660 | 429 | synchronize_sched(); |
1da177e4 LT |
430 | |
431 | /* Collect entries that were still getting written before the synchronize. */ | |
432 | ||
433 | on_each_cpu(collect_tscs, cpu_tsc, 1, 1); | |
434 | for (i = next; i < MCE_LOG_LEN; i++) { | |
435 | if (mcelog.entry[i].finished && | |
436 | mcelog.entry[i].tsc < cpu_tsc[mcelog.entry[i].cpu]) { | |
437 | err |= copy_to_user(buf, mcelog.entry+i, sizeof(struct mce)); | |
438 | smp_rmb(); | |
439 | buf += sizeof(struct mce); | |
440 | memset(&mcelog.entry[i], 0, sizeof(struct mce)); | |
441 | } | |
442 | } | |
443 | up(&mce_read_sem); | |
f0de53bb | 444 | kfree(cpu_tsc); |
1da177e4 LT |
445 | return err ? -EFAULT : buf - ubuf; |
446 | } | |
447 | ||
448 | static int mce_ioctl(struct inode *i, struct file *f,unsigned int cmd, unsigned long arg) | |
449 | { | |
450 | int __user *p = (int __user *)arg; | |
451 | if (!capable(CAP_SYS_ADMIN)) | |
452 | return -EPERM; | |
453 | switch (cmd) { | |
454 | case MCE_GET_RECORD_LEN: | |
455 | return put_user(sizeof(struct mce), p); | |
456 | case MCE_GET_LOG_LEN: | |
457 | return put_user(MCE_LOG_LEN, p); | |
458 | case MCE_GETCLEAR_FLAGS: { | |
459 | unsigned flags; | |
460 | do { | |
461 | flags = mcelog.flags; | |
462 | } while (cmpxchg(&mcelog.flags, flags, 0) != flags); | |
463 | return put_user(flags, p); | |
464 | } | |
465 | default: | |
466 | return -ENOTTY; | |
467 | } | |
468 | } | |
469 | ||
470 | static struct file_operations mce_chrdev_ops = { | |
471 | .read = mce_read, | |
472 | .ioctl = mce_ioctl, | |
473 | }; | |
474 | ||
475 | static struct miscdevice mce_log_device = { | |
476 | MISC_MCELOG_MINOR, | |
477 | "mcelog", | |
478 | &mce_chrdev_ops, | |
479 | }; | |
480 | ||
481 | /* | |
482 | * Old style boot options parsing. Only for compatibility. | |
483 | */ | |
484 | ||
485 | static int __init mcheck_disable(char *str) | |
486 | { | |
487 | mce_dont_init = 1; | |
488 | return 0; | |
489 | } | |
490 | ||
491 | /* mce=off disables machine check. Note you can reenable it later | |
d5172f26 | 492 | using sysfs. |
8c566ef5 | 493 | mce=TOLERANCELEVEL (number, see above) |
d5172f26 AK |
494 | mce=bootlog Log MCEs from before booting. Disabled by default to work |
495 | around buggy BIOS that leave bogus MCEs. */ | |
1da177e4 LT |
496 | static int __init mcheck_enable(char *str) |
497 | { | |
d5172f26 AK |
498 | if (*str == '=') |
499 | str++; | |
1da177e4 LT |
500 | if (!strcmp(str, "off")) |
501 | mce_dont_init = 1; | |
d5172f26 AK |
502 | else if (!strcmp(str, "bootlog")) |
503 | mce_bootlog = 1; | |
8c566ef5 AK |
504 | else if (isdigit(str[0])) |
505 | get_option(&str, &tolerant); | |
1da177e4 LT |
506 | else |
507 | printk("mce= argument %s ignored. Please use /sys", str); | |
508 | return 0; | |
509 | } | |
510 | ||
511 | __setup("nomce", mcheck_disable); | |
512 | __setup("mce", mcheck_enable); | |
513 | ||
514 | /* | |
515 | * Sysfs support | |
516 | */ | |
517 | ||
518 | /* On resume clear all MCE state. Don't want to see leftovers from the BIOS. */ | |
519 | static int mce_resume(struct sys_device *dev) | |
520 | { | |
521 | on_each_cpu(mce_init, NULL, 1, 1); | |
522 | return 0; | |
523 | } | |
524 | ||
525 | /* Reinit MCEs after user configuration changes */ | |
526 | static void mce_restart(void) | |
527 | { | |
528 | if (check_interval) | |
529 | cancel_delayed_work(&mcheck_work); | |
530 | /* Timer race is harmless here */ | |
531 | on_each_cpu(mce_init, NULL, 1, 1); | |
532 | if (check_interval) | |
533 | schedule_delayed_work(&mcheck_work, check_interval*HZ); | |
534 | } | |
535 | ||
536 | static struct sysdev_class mce_sysclass = { | |
537 | .resume = mce_resume, | |
538 | set_kset_name("machinecheck"), | |
539 | }; | |
540 | ||
91c6d400 | 541 | static DEFINE_PER_CPU(struct sys_device, device_mce); |
1da177e4 LT |
542 | |
543 | /* Why are there no generic functions for this? */ | |
544 | #define ACCESSOR(name, var, start) \ | |
545 | static ssize_t show_ ## name(struct sys_device *s, char *buf) { \ | |
546 | return sprintf(buf, "%lx\n", (unsigned long)var); \ | |
547 | } \ | |
548 | static ssize_t set_ ## name(struct sys_device *s,const char *buf,size_t siz) { \ | |
549 | char *end; \ | |
550 | unsigned long new = simple_strtoul(buf, &end, 0); \ | |
551 | if (end == buf) return -EINVAL; \ | |
552 | var = new; \ | |
553 | start; \ | |
554 | return end-buf; \ | |
555 | } \ | |
556 | static SYSDEV_ATTR(name, 0644, show_ ## name, set_ ## name); | |
557 | ||
558 | ACCESSOR(bank0ctl,bank[0],mce_restart()) | |
559 | ACCESSOR(bank1ctl,bank[1],mce_restart()) | |
560 | ACCESSOR(bank2ctl,bank[2],mce_restart()) | |
561 | ACCESSOR(bank3ctl,bank[3],mce_restart()) | |
562 | ACCESSOR(bank4ctl,bank[4],mce_restart()) | |
563 | ACCESSOR(tolerant,tolerant,) | |
564 | ACCESSOR(check_interval,check_interval,mce_restart()) | |
565 | ||
91c6d400 AK |
566 | /* Per cpu sysdev init. All of the cpus still share the same ctl bank */ |
567 | static __cpuinit int mce_create_device(unsigned int cpu) | |
1da177e4 LT |
568 | { |
569 | int err; | |
91c6d400 AK |
570 | if (!mce_available(&cpu_data[cpu])) |
571 | return -EIO; | |
572 | ||
573 | per_cpu(device_mce,cpu).id = cpu; | |
574 | per_cpu(device_mce,cpu).cls = &mce_sysclass; | |
575 | ||
576 | err = sysdev_register(&per_cpu(device_mce,cpu)); | |
577 | ||
578 | if (!err) { | |
579 | sysdev_create_file(&per_cpu(device_mce,cpu), &attr_bank0ctl); | |
580 | sysdev_create_file(&per_cpu(device_mce,cpu), &attr_bank1ctl); | |
581 | sysdev_create_file(&per_cpu(device_mce,cpu), &attr_bank2ctl); | |
582 | sysdev_create_file(&per_cpu(device_mce,cpu), &attr_bank3ctl); | |
583 | sysdev_create_file(&per_cpu(device_mce,cpu), &attr_bank4ctl); | |
584 | sysdev_create_file(&per_cpu(device_mce,cpu), &attr_tolerant); | |
585 | sysdev_create_file(&per_cpu(device_mce,cpu), &attr_check_interval); | |
586 | } | |
587 | return err; | |
588 | } | |
589 | ||
590 | #ifdef CONFIG_HOTPLUG_CPU | |
591 | static __cpuinit void mce_remove_device(unsigned int cpu) | |
592 | { | |
593 | sysdev_remove_file(&per_cpu(device_mce,cpu), &attr_bank0ctl); | |
594 | sysdev_remove_file(&per_cpu(device_mce,cpu), &attr_bank1ctl); | |
595 | sysdev_remove_file(&per_cpu(device_mce,cpu), &attr_bank2ctl); | |
596 | sysdev_remove_file(&per_cpu(device_mce,cpu), &attr_bank3ctl); | |
597 | sysdev_remove_file(&per_cpu(device_mce,cpu), &attr_bank4ctl); | |
598 | sysdev_remove_file(&per_cpu(device_mce,cpu), &attr_tolerant); | |
599 | sysdev_remove_file(&per_cpu(device_mce,cpu), &attr_check_interval); | |
600 | sysdev_unregister(&per_cpu(device_mce,cpu)); | |
601 | } | |
602 | #endif | |
603 | ||
604 | /* Get notified when a cpu comes on/off. Be hotplug friendly. */ | |
605 | static __cpuinit int | |
606 | mce_cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
607 | { | |
608 | unsigned int cpu = (unsigned long)hcpu; | |
609 | ||
610 | switch (action) { | |
611 | case CPU_ONLINE: | |
612 | mce_create_device(cpu); | |
613 | break; | |
614 | #ifdef CONFIG_HOTPLUG_CPU | |
615 | case CPU_DEAD: | |
616 | mce_remove_device(cpu); | |
617 | break; | |
618 | #endif | |
619 | } | |
620 | return NOTIFY_OK; | |
621 | } | |
622 | ||
623 | static struct notifier_block mce_cpu_notifier = { | |
624 | .notifier_call = mce_cpu_callback, | |
625 | }; | |
626 | ||
627 | static __init int mce_init_device(void) | |
628 | { | |
629 | int err; | |
630 | int i = 0; | |
631 | ||
1da177e4 LT |
632 | if (!mce_available(&boot_cpu_data)) |
633 | return -EIO; | |
634 | err = sysdev_class_register(&mce_sysclass); | |
91c6d400 AK |
635 | |
636 | for_each_online_cpu(i) { | |
637 | mce_create_device(i); | |
638 | } | |
639 | ||
640 | register_cpu_notifier(&mce_cpu_notifier); | |
1da177e4 LT |
641 | misc_register(&mce_log_device); |
642 | return err; | |
1da177e4 | 643 | } |
91c6d400 | 644 | |
1da177e4 | 645 | device_initcall(mce_init_device); |