Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Machine check handler. | |
e9eee03e | 3 | * |
1da177e4 | 4 | * K8 parts Copyright 2002,2003 Andi Kleen, SuSE Labs. |
d88203d1 TG |
5 | * Rest from unknown author(s). |
6 | * 2004 Andi Kleen. Rewrote most of it. | |
b79109c3 AK |
7 | * Copyright 2008 Intel Corporation |
8 | * Author: Andi Kleen | |
1da177e4 | 9 | */ |
e9eee03e IM |
10 | #include <linux/thread_info.h> |
11 | #include <linux/capability.h> | |
12 | #include <linux/miscdevice.h> | |
13 | #include <linux/ratelimit.h> | |
14 | #include <linux/kallsyms.h> | |
15 | #include <linux/rcupdate.h> | |
e9eee03e | 16 | #include <linux/kobject.h> |
14a02530 | 17 | #include <linux/uaccess.h> |
e9eee03e IM |
18 | #include <linux/kdebug.h> |
19 | #include <linux/kernel.h> | |
20 | #include <linux/percpu.h> | |
1da177e4 | 21 | #include <linux/string.h> |
1da177e4 | 22 | #include <linux/sysdev.h> |
8c566ef5 | 23 | #include <linux/ctype.h> |
e9eee03e | 24 | #include <linux/sched.h> |
0d7482e3 | 25 | #include <linux/sysfs.h> |
e9eee03e IM |
26 | #include <linux/types.h> |
27 | #include <linux/init.h> | |
28 | #include <linux/kmod.h> | |
29 | #include <linux/poll.h> | |
30 | #include <linux/cpu.h> | |
14a02530 | 31 | #include <linux/smp.h> |
e9eee03e IM |
32 | #include <linux/fs.h> |
33 | ||
d88203d1 | 34 | #include <asm/processor.h> |
e02e68d3 | 35 | #include <asm/idle.h> |
e9eee03e IM |
36 | #include <asm/mce.h> |
37 | #include <asm/msr.h> | |
1da177e4 | 38 | |
711c2e48 IM |
39 | #include "mce.h" |
40 | ||
5d727926 AK |
41 | /* Handle unconfigured int18 (should never happen) */ |
42 | static void unexpected_machine_check(struct pt_regs *regs, long error_code) | |
43 | { | |
44 | printk(KERN_ERR "CPU#%d: Unexpected int18 (Machine Check).\n", | |
45 | smp_processor_id()); | |
46 | } | |
47 | ||
48 | /* Call the installed machine check handler for this CPU setup. */ | |
49 | void (*machine_check_vector)(struct pt_regs *, long error_code) = | |
50 | unexpected_machine_check; | |
04b2b1a4 AK |
51 | |
52 | int mce_disabled; | |
53 | ||
4efc0670 | 54 | #ifdef CONFIG_X86_NEW_MCE |
711c2e48 | 55 | |
e9eee03e | 56 | #define MISC_MCELOG_MINOR 227 |
0d7482e3 | 57 | |
553f265f AK |
58 | atomic_t mce_entry; |
59 | ||
bd78432c TH |
60 | /* |
61 | * Tolerant levels: | |
62 | * 0: always panic on uncorrected errors, log corrected errors | |
63 | * 1: panic or SIGBUS on uncorrected errors, log corrected errors | |
64 | * 2: SIGBUS or log uncorrected errors (if possible), log corrected errors | |
65 | * 3: never panic or SIGBUS, log all errors (for testing only) | |
66 | */ | |
e9eee03e IM |
67 | static int tolerant = 1; |
68 | static int banks; | |
69 | static u64 *bank; | |
70 | static unsigned long notify_user; | |
71 | static int rip_msr; | |
72 | static int mce_bootlog = -1; | |
a98f0dd3 | 73 | |
e9eee03e IM |
74 | static char trigger[128]; |
75 | static char *trigger_argv[2] = { trigger, NULL }; | |
1da177e4 | 76 | |
06b7a7a5 AK |
77 | static unsigned long dont_init_banks; |
78 | ||
e02e68d3 TH |
79 | static DECLARE_WAIT_QUEUE_HEAD(mce_wait); |
80 | ||
ee031c31 AK |
81 | /* MCA banks polled by the period polling timer for corrected events */ |
82 | DEFINE_PER_CPU(mce_banks_t, mce_poll_banks) = { | |
83 | [0 ... BITS_TO_LONGS(MAX_NR_BANKS)-1] = ~0UL | |
84 | }; | |
85 | ||
06b7a7a5 AK |
86 | static inline int skip_bank_init(int i) |
87 | { | |
88 | return i < BITS_PER_LONG && test_bit(i, &dont_init_banks); | |
89 | } | |
90 | ||
b5f2fa4e AK |
91 | /* Do initial initialization of a struct mce */ |
92 | void mce_setup(struct mce *m) | |
93 | { | |
94 | memset(m, 0, sizeof(struct mce)); | |
95 | m->cpu = smp_processor_id(); | |
96 | rdtscll(m->tsc); | |
97 | } | |
98 | ||
ea149b36 AK |
99 | DEFINE_PER_CPU(struct mce, injectm); |
100 | EXPORT_PER_CPU_SYMBOL_GPL(injectm); | |
101 | ||
1da177e4 LT |
102 | /* |
103 | * Lockless MCE logging infrastructure. | |
104 | * This avoids deadlocks on printk locks without having to break locks. Also | |
105 | * separate MCEs from kernel messages to avoid bogus bug reports. | |
106 | */ | |
107 | ||
231fd906 | 108 | static struct mce_log mcelog = { |
1da177e4 LT |
109 | MCE_LOG_SIGNATURE, |
110 | MCE_LOG_LEN, | |
d88203d1 | 111 | }; |
1da177e4 LT |
112 | |
113 | void mce_log(struct mce *mce) | |
114 | { | |
115 | unsigned next, entry; | |
e9eee03e | 116 | |
1da177e4 | 117 | mce->finished = 0; |
7644143c | 118 | wmb(); |
1da177e4 LT |
119 | for (;;) { |
120 | entry = rcu_dereference(mcelog.next); | |
673242c1 | 121 | for (;;) { |
e9eee03e IM |
122 | /* |
123 | * When the buffer fills up discard new entries. | |
124 | * Assume that the earlier errors are the more | |
125 | * interesting ones: | |
126 | */ | |
673242c1 | 127 | if (entry >= MCE_LOG_LEN) { |
14a02530 HS |
128 | set_bit(MCE_OVERFLOW, |
129 | (unsigned long *)&mcelog.flags); | |
673242c1 AK |
130 | return; |
131 | } | |
e9eee03e | 132 | /* Old left over entry. Skip: */ |
673242c1 AK |
133 | if (mcelog.entry[entry].finished) { |
134 | entry++; | |
135 | continue; | |
136 | } | |
7644143c | 137 | break; |
1da177e4 | 138 | } |
1da177e4 LT |
139 | smp_rmb(); |
140 | next = entry + 1; | |
141 | if (cmpxchg(&mcelog.next, entry, next) == entry) | |
142 | break; | |
143 | } | |
144 | memcpy(mcelog.entry + entry, mce, sizeof(struct mce)); | |
7644143c | 145 | wmb(); |
1da177e4 | 146 | mcelog.entry[entry].finished = 1; |
7644143c | 147 | wmb(); |
1da177e4 | 148 | |
e02e68d3 | 149 | set_bit(0, ¬ify_user); |
1da177e4 LT |
150 | } |
151 | ||
152 | static void print_mce(struct mce *m) | |
153 | { | |
154 | printk(KERN_EMERG "\n" | |
4855170f | 155 | KERN_EMERG "HARDWARE ERROR\n" |
1da177e4 LT |
156 | KERN_EMERG |
157 | "CPU %d: Machine Check Exception: %16Lx Bank %d: %016Lx\n", | |
158 | m->cpu, m->mcgstatus, m->bank, m->status); | |
65ea5b03 | 159 | if (m->ip) { |
d88203d1 | 160 | printk(KERN_EMERG "RIP%s %02x:<%016Lx> ", |
1da177e4 | 161 | !(m->mcgstatus & MCG_STATUS_EIPV) ? " !INEXACT!" : "", |
65ea5b03 | 162 | m->cs, m->ip); |
1da177e4 | 163 | if (m->cs == __KERNEL_CS) |
65ea5b03 | 164 | print_symbol("{%s}", m->ip); |
1da177e4 LT |
165 | printk("\n"); |
166 | } | |
f6d1826d | 167 | printk(KERN_EMERG "TSC %llx ", m->tsc); |
1da177e4 | 168 | if (m->addr) |
f6d1826d | 169 | printk("ADDR %llx ", m->addr); |
1da177e4 | 170 | if (m->misc) |
f6d1826d | 171 | printk("MISC %llx ", m->misc); |
1da177e4 | 172 | printk("\n"); |
4855170f | 173 | printk(KERN_EMERG "This is not a software problem!\n"); |
d88203d1 TG |
174 | printk(KERN_EMERG "Run through mcelog --ascii to decode " |
175 | "and contact your hardware vendor\n"); | |
1da177e4 LT |
176 | } |
177 | ||
3cde5c8c | 178 | static void mce_panic(char *msg, struct mce *backup, u64 start) |
d88203d1 | 179 | { |
1da177e4 | 180 | int i; |
e02e68d3 | 181 | |
d896a940 AK |
182 | bust_spinlocks(1); |
183 | console_verbose(); | |
1da177e4 | 184 | for (i = 0; i < MCE_LOG_LEN; i++) { |
3cde5c8c | 185 | u64 tsc = mcelog.entry[i].tsc; |
d88203d1 | 186 | |
3cde5c8c | 187 | if ((s64)(tsc - start) < 0) |
1da177e4 | 188 | continue; |
d88203d1 | 189 | print_mce(&mcelog.entry[i]); |
1da177e4 LT |
190 | if (backup && mcelog.entry[i].tsc == backup->tsc) |
191 | backup = NULL; | |
192 | } | |
193 | if (backup) | |
194 | print_mce(backup); | |
e02e68d3 | 195 | panic(msg); |
d88203d1 | 196 | } |
1da177e4 | 197 | |
ea149b36 AK |
198 | /* Support code for software error injection */ |
199 | ||
200 | static int msr_to_offset(u32 msr) | |
201 | { | |
202 | unsigned bank = __get_cpu_var(injectm.bank); | |
203 | if (msr == rip_msr) | |
204 | return offsetof(struct mce, ip); | |
205 | if (msr == MSR_IA32_MC0_STATUS + bank*4) | |
206 | return offsetof(struct mce, status); | |
207 | if (msr == MSR_IA32_MC0_ADDR + bank*4) | |
208 | return offsetof(struct mce, addr); | |
209 | if (msr == MSR_IA32_MC0_MISC + bank*4) | |
210 | return offsetof(struct mce, misc); | |
211 | if (msr == MSR_IA32_MCG_STATUS) | |
212 | return offsetof(struct mce, mcgstatus); | |
213 | return -1; | |
214 | } | |
215 | ||
5f8c1a54 AK |
216 | /* MSR access wrappers used for error injection */ |
217 | static u64 mce_rdmsrl(u32 msr) | |
218 | { | |
219 | u64 v; | |
ea149b36 AK |
220 | if (__get_cpu_var(injectm).finished) { |
221 | int offset = msr_to_offset(msr); | |
222 | if (offset < 0) | |
223 | return 0; | |
224 | return *(u64 *)((char *)&__get_cpu_var(injectm) + offset); | |
225 | } | |
5f8c1a54 AK |
226 | rdmsrl(msr, v); |
227 | return v; | |
228 | } | |
229 | ||
230 | static void mce_wrmsrl(u32 msr, u64 v) | |
231 | { | |
ea149b36 AK |
232 | if (__get_cpu_var(injectm).finished) { |
233 | int offset = msr_to_offset(msr); | |
234 | if (offset >= 0) | |
235 | *(u64 *)((char *)&__get_cpu_var(injectm) + offset) = v; | |
236 | return; | |
237 | } | |
5f8c1a54 AK |
238 | wrmsrl(msr, v); |
239 | } | |
240 | ||
88ccbedd | 241 | int mce_available(struct cpuinfo_x86 *c) |
1da177e4 | 242 | { |
04b2b1a4 | 243 | if (mce_disabled) |
5b4408fd | 244 | return 0; |
3d1712c9 | 245 | return cpu_has(c, X86_FEATURE_MCE) && cpu_has(c, X86_FEATURE_MCA); |
1da177e4 LT |
246 | } |
247 | ||
94ad8474 AK |
248 | static inline void mce_get_rip(struct mce *m, struct pt_regs *regs) |
249 | { | |
250 | if (regs && (m->mcgstatus & MCG_STATUS_RIPV)) { | |
65ea5b03 | 251 | m->ip = regs->ip; |
94ad8474 AK |
252 | m->cs = regs->cs; |
253 | } else { | |
65ea5b03 | 254 | m->ip = 0; |
94ad8474 AK |
255 | m->cs = 0; |
256 | } | |
257 | if (rip_msr) { | |
258 | /* Assume the RIP in the MSR is exact. Is this true? */ | |
259 | m->mcgstatus |= MCG_STATUS_EIPV; | |
5f8c1a54 | 260 | m->ip = mce_rdmsrl(rip_msr); |
94ad8474 AK |
261 | m->cs = 0; |
262 | } | |
263 | } | |
264 | ||
d88203d1 | 265 | /* |
b79109c3 AK |
266 | * Poll for corrected events or events that happened before reset. |
267 | * Those are just logged through /dev/mcelog. | |
268 | * | |
269 | * This is executed in standard interrupt context. | |
270 | */ | |
ee031c31 | 271 | void machine_check_poll(enum mcp_flags flags, mce_banks_t *b) |
b79109c3 AK |
272 | { |
273 | struct mce m; | |
274 | int i; | |
275 | ||
276 | mce_setup(&m); | |
277 | ||
5f8c1a54 | 278 | m.mcgstatus = mce_rdmsrl(MSR_IA32_MCG_STATUS); |
b79109c3 | 279 | for (i = 0; i < banks; i++) { |
ee031c31 | 280 | if (!bank[i] || !test_bit(i, *b)) |
b79109c3 AK |
281 | continue; |
282 | ||
283 | m.misc = 0; | |
284 | m.addr = 0; | |
285 | m.bank = i; | |
286 | m.tsc = 0; | |
287 | ||
288 | barrier(); | |
5f8c1a54 | 289 | m.status = mce_rdmsrl(MSR_IA32_MC0_STATUS + i*4); |
b79109c3 AK |
290 | if (!(m.status & MCI_STATUS_VAL)) |
291 | continue; | |
292 | ||
293 | /* | |
294 | * Uncorrected events are handled by the exception handler | |
295 | * when it is enabled. But when the exception is disabled log | |
296 | * everything. | |
297 | * | |
298 | * TBD do the same check for MCI_STATUS_EN here? | |
299 | */ | |
300 | if ((m.status & MCI_STATUS_UC) && !(flags & MCP_UC)) | |
301 | continue; | |
302 | ||
303 | if (m.status & MCI_STATUS_MISCV) | |
5f8c1a54 | 304 | m.misc = mce_rdmsrl(MSR_IA32_MC0_MISC + i*4); |
b79109c3 | 305 | if (m.status & MCI_STATUS_ADDRV) |
5f8c1a54 | 306 | m.addr = mce_rdmsrl(MSR_IA32_MC0_ADDR + i*4); |
b79109c3 AK |
307 | |
308 | if (!(flags & MCP_TIMESTAMP)) | |
309 | m.tsc = 0; | |
310 | /* | |
311 | * Don't get the IP here because it's unlikely to | |
312 | * have anything to do with the actual error location. | |
313 | */ | |
5679af4c AK |
314 | if (!(flags & MCP_DONTLOG)) { |
315 | mce_log(&m); | |
316 | add_taint(TAINT_MACHINE_CHECK); | |
317 | } | |
b79109c3 AK |
318 | |
319 | /* | |
320 | * Clear state for this bank. | |
321 | */ | |
5f8c1a54 | 322 | mce_wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0); |
b79109c3 AK |
323 | } |
324 | ||
325 | /* | |
326 | * Don't clear MCG_STATUS here because it's only defined for | |
327 | * exceptions. | |
328 | */ | |
88921be3 AK |
329 | |
330 | sync_core(); | |
b79109c3 | 331 | } |
ea149b36 | 332 | EXPORT_SYMBOL_GPL(machine_check_poll); |
b79109c3 AK |
333 | |
334 | /* | |
335 | * The actual machine check handler. This only handles real | |
336 | * exceptions when something got corrupted coming in through int 18. | |
337 | * | |
338 | * This is executed in NMI context not subject to normal locking rules. This | |
339 | * implies that most kernel services cannot be safely used. Don't even | |
340 | * think about putting a printk in there! | |
1da177e4 | 341 | */ |
e9eee03e | 342 | void do_machine_check(struct pt_regs *regs, long error_code) |
1da177e4 LT |
343 | { |
344 | struct mce m, panicm; | |
e9eee03e | 345 | int panicm_found = 0; |
1da177e4 LT |
346 | u64 mcestart = 0; |
347 | int i; | |
bd78432c TH |
348 | /* |
349 | * If no_way_out gets set, there is no safe way to recover from this | |
350 | * MCE. If tolerant is cranked up, we'll try anyway. | |
351 | */ | |
352 | int no_way_out = 0; | |
353 | /* | |
354 | * If kill_it gets set, there might be a way to recover from this | |
355 | * error. | |
356 | */ | |
357 | int kill_it = 0; | |
b79109c3 | 358 | DECLARE_BITMAP(toclear, MAX_NR_BANKS); |
1da177e4 | 359 | |
553f265f AK |
360 | atomic_inc(&mce_entry); |
361 | ||
b79109c3 | 362 | if (notify_die(DIE_NMI, "machine check", regs, error_code, |
22f5991c | 363 | 18, SIGKILL) == NOTIFY_STOP) |
32561696 | 364 | goto out; |
b79109c3 | 365 | if (!banks) |
32561696 | 366 | goto out; |
1da177e4 | 367 | |
b5f2fa4e AK |
368 | mce_setup(&m); |
369 | ||
5f8c1a54 | 370 | m.mcgstatus = mce_rdmsrl(MSR_IA32_MCG_STATUS); |
e9eee03e | 371 | |
bd78432c | 372 | /* if the restart IP is not valid, we're done for */ |
1da177e4 | 373 | if (!(m.mcgstatus & MCG_STATUS_RIPV)) |
bd78432c | 374 | no_way_out = 1; |
d88203d1 | 375 | |
1da177e4 LT |
376 | rdtscll(mcestart); |
377 | barrier(); | |
378 | ||
379 | for (i = 0; i < banks; i++) { | |
b79109c3 | 380 | __clear_bit(i, toclear); |
0d7482e3 | 381 | if (!bank[i]) |
1da177e4 | 382 | continue; |
d88203d1 TG |
383 | |
384 | m.misc = 0; | |
1da177e4 LT |
385 | m.addr = 0; |
386 | m.bank = i; | |
1da177e4 | 387 | |
5f8c1a54 | 388 | m.status = mce_rdmsrl(MSR_IA32_MC0_STATUS + i*4); |
1da177e4 LT |
389 | if ((m.status & MCI_STATUS_VAL) == 0) |
390 | continue; | |
391 | ||
b79109c3 AK |
392 | /* |
393 | * Non uncorrected errors are handled by machine_check_poll | |
394 | * Leave them alone. | |
395 | */ | |
396 | if ((m.status & MCI_STATUS_UC) == 0) | |
397 | continue; | |
398 | ||
399 | /* | |
400 | * Set taint even when machine check was not enabled. | |
401 | */ | |
402 | add_taint(TAINT_MACHINE_CHECK); | |
403 | ||
404 | __set_bit(i, toclear); | |
405 | ||
1da177e4 | 406 | if (m.status & MCI_STATUS_EN) { |
bd78432c TH |
407 | /* if PCC was set, there's no way out */ |
408 | no_way_out |= !!(m.status & MCI_STATUS_PCC); | |
409 | /* | |
410 | * If this error was uncorrectable and there was | |
411 | * an overflow, we're in trouble. If no overflow, | |
412 | * we might get away with just killing a task. | |
413 | */ | |
414 | if (m.status & MCI_STATUS_UC) { | |
415 | if (tolerant < 1 || m.status & MCI_STATUS_OVER) | |
416 | no_way_out = 1; | |
417 | kill_it = 1; | |
418 | } | |
b79109c3 AK |
419 | } else { |
420 | /* | |
421 | * Machine check event was not enabled. Clear, but | |
422 | * ignore. | |
423 | */ | |
424 | continue; | |
1da177e4 LT |
425 | } |
426 | ||
427 | if (m.status & MCI_STATUS_MISCV) | |
5f8c1a54 | 428 | m.misc = mce_rdmsrl(MSR_IA32_MC0_MISC + i*4); |
1da177e4 | 429 | if (m.status & MCI_STATUS_ADDRV) |
5f8c1a54 | 430 | m.addr = mce_rdmsrl(MSR_IA32_MC0_ADDR + i*4); |
1da177e4 | 431 | |
94ad8474 | 432 | mce_get_rip(&m, regs); |
b79109c3 | 433 | mce_log(&m); |
1da177e4 | 434 | |
e9eee03e IM |
435 | /* |
436 | * Did this bank cause the exception? | |
437 | * | |
438 | * Assume that the bank with uncorrectable errors did it, | |
439 | * and that there is only a single one: | |
440 | */ | |
441 | if ((m.status & MCI_STATUS_UC) && | |
442 | (m.status & MCI_STATUS_EN)) { | |
1da177e4 LT |
443 | panicm = m; |
444 | panicm_found = 1; | |
445 | } | |
1da177e4 LT |
446 | } |
447 | ||
e9eee03e IM |
448 | /* |
449 | * If we didn't find an uncorrectable error, pick | |
450 | * the last one (shouldn't happen, just being safe). | |
451 | */ | |
1da177e4 LT |
452 | if (!panicm_found) |
453 | panicm = m; | |
bd78432c TH |
454 | |
455 | /* | |
456 | * If we have decided that we just CAN'T continue, and the user | |
e9eee03e | 457 | * has not set tolerant to an insane level, give up and die. |
bd78432c TH |
458 | */ |
459 | if (no_way_out && tolerant < 3) | |
1da177e4 | 460 | mce_panic("Machine check", &panicm, mcestart); |
bd78432c TH |
461 | |
462 | /* | |
463 | * If the error seems to be unrecoverable, something should be | |
464 | * done. Try to kill as little as possible. If we can kill just | |
465 | * one task, do that. If the user has set the tolerance very | |
466 | * high, don't try to do anything at all. | |
467 | */ | |
468 | if (kill_it && tolerant < 3) { | |
1da177e4 LT |
469 | int user_space = 0; |
470 | ||
bd78432c TH |
471 | /* |
472 | * If the EIPV bit is set, it means the saved IP is the | |
473 | * instruction which caused the MCE. | |
474 | */ | |
475 | if (m.mcgstatus & MCG_STATUS_EIPV) | |
65ea5b03 | 476 | user_space = panicm.ip && (panicm.cs & 3); |
bd78432c TH |
477 | |
478 | /* | |
479 | * If we know that the error was in user space, send a | |
480 | * SIGBUS. Otherwise, panic if tolerance is low. | |
481 | * | |
380851bc | 482 | * force_sig() takes an awful lot of locks and has a slight |
bd78432c TH |
483 | * risk of deadlocking. |
484 | */ | |
485 | if (user_space) { | |
380851bc | 486 | force_sig(SIGBUS, current); |
bd78432c TH |
487 | } else if (panic_on_oops || tolerant < 2) { |
488 | mce_panic("Uncorrected machine check", | |
489 | &panicm, mcestart); | |
490 | } | |
1da177e4 LT |
491 | } |
492 | ||
e02e68d3 TH |
493 | /* notify userspace ASAP */ |
494 | set_thread_flag(TIF_MCE_NOTIFY); | |
495 | ||
bd78432c | 496 | /* the last thing we do is clear state */ |
b79109c3 AK |
497 | for (i = 0; i < banks; i++) { |
498 | if (test_bit(i, toclear)) | |
5f8c1a54 | 499 | mce_wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0); |
b79109c3 | 500 | } |
5f8c1a54 | 501 | mce_wrmsrl(MSR_IA32_MCG_STATUS, 0); |
32561696 | 502 | out: |
553f265f | 503 | atomic_dec(&mce_entry); |
88921be3 | 504 | sync_core(); |
1da177e4 | 505 | } |
ea149b36 | 506 | EXPORT_SYMBOL_GPL(do_machine_check); |
1da177e4 | 507 | |
15d5f839 DZ |
508 | #ifdef CONFIG_X86_MCE_INTEL |
509 | /*** | |
510 | * mce_log_therm_throt_event - Logs the thermal throttling event to mcelog | |
676b1855 | 511 | * @cpu: The CPU on which the event occurred. |
15d5f839 DZ |
512 | * @status: Event status information |
513 | * | |
514 | * This function should be called by the thermal interrupt after the | |
515 | * event has been processed and the decision was made to log the event | |
516 | * further. | |
517 | * | |
518 | * The status parameter will be saved to the 'status' field of 'struct mce' | |
519 | * and historically has been the register value of the | |
520 | * MSR_IA32_THERMAL_STATUS (Intel) msr. | |
521 | */ | |
b5f2fa4e | 522 | void mce_log_therm_throt_event(__u64 status) |
15d5f839 DZ |
523 | { |
524 | struct mce m; | |
525 | ||
b5f2fa4e | 526 | mce_setup(&m); |
15d5f839 DZ |
527 | m.bank = MCE_THERMAL_BANK; |
528 | m.status = status; | |
15d5f839 DZ |
529 | mce_log(&m); |
530 | } | |
531 | #endif /* CONFIG_X86_MCE_INTEL */ | |
532 | ||
1da177e4 | 533 | /* |
8a336b0a TH |
534 | * Periodic polling timer for "silent" machine check errors. If the |
535 | * poller finds an MCE, poll 2x faster. When the poller finds no more | |
536 | * errors, poll 2x slower (up to check_interval seconds). | |
1da177e4 | 537 | */ |
1da177e4 | 538 | static int check_interval = 5 * 60; /* 5 minutes */ |
e9eee03e | 539 | |
6298c512 | 540 | static DEFINE_PER_CPU(int, next_interval); /* in jiffies */ |
52d168e2 | 541 | static DEFINE_PER_CPU(struct timer_list, mce_timer); |
1da177e4 | 542 | |
52d168e2 | 543 | static void mcheck_timer(unsigned long data) |
1da177e4 | 544 | { |
52d168e2 | 545 | struct timer_list *t = &per_cpu(mce_timer, data); |
6298c512 | 546 | int *n; |
52d168e2 AK |
547 | |
548 | WARN_ON(smp_processor_id() != data); | |
549 | ||
e9eee03e | 550 | if (mce_available(¤t_cpu_data)) { |
ee031c31 AK |
551 | machine_check_poll(MCP_TIMESTAMP, |
552 | &__get_cpu_var(mce_poll_banks)); | |
e9eee03e | 553 | } |
1da177e4 LT |
554 | |
555 | /* | |
e02e68d3 TH |
556 | * Alert userspace if needed. If we logged an MCE, reduce the |
557 | * polling interval, otherwise increase the polling interval. | |
1da177e4 | 558 | */ |
6298c512 | 559 | n = &__get_cpu_var(next_interval); |
14a02530 | 560 | if (mce_notify_user()) |
6298c512 | 561 | *n = max(*n/2, HZ/100); |
14a02530 | 562 | else |
6298c512 | 563 | *n = min(*n*2, (int)round_jiffies_relative(check_interval*HZ)); |
e02e68d3 | 564 | |
6298c512 | 565 | t->expires = jiffies + *n; |
52d168e2 | 566 | add_timer(t); |
e02e68d3 TH |
567 | } |
568 | ||
9bd98405 AK |
569 | static void mce_do_trigger(struct work_struct *work) |
570 | { | |
571 | call_usermodehelper(trigger, trigger_argv, NULL, UMH_NO_WAIT); | |
572 | } | |
573 | ||
574 | static DECLARE_WORK(mce_trigger_work, mce_do_trigger); | |
575 | ||
e02e68d3 | 576 | /* |
9bd98405 AK |
577 | * Notify the user(s) about new machine check events. |
578 | * Can be called from interrupt context, but not from machine check/NMI | |
579 | * context. | |
e02e68d3 TH |
580 | */ |
581 | int mce_notify_user(void) | |
582 | { | |
8457c84d AK |
583 | /* Not more than two messages every minute */ |
584 | static DEFINE_RATELIMIT_STATE(ratelimit, 60*HZ, 2); | |
585 | ||
e02e68d3 | 586 | clear_thread_flag(TIF_MCE_NOTIFY); |
e9eee03e | 587 | |
e02e68d3 | 588 | if (test_and_clear_bit(0, ¬ify_user)) { |
e02e68d3 | 589 | wake_up_interruptible(&mce_wait); |
9bd98405 AK |
590 | |
591 | /* | |
592 | * There is no risk of missing notifications because | |
593 | * work_pending is always cleared before the function is | |
594 | * executed. | |
595 | */ | |
596 | if (trigger[0] && !work_pending(&mce_trigger_work)) | |
597 | schedule_work(&mce_trigger_work); | |
e02e68d3 | 598 | |
8457c84d | 599 | if (__ratelimit(&ratelimit)) |
8a336b0a | 600 | printk(KERN_INFO "Machine check events logged\n"); |
e02e68d3 TH |
601 | |
602 | return 1; | |
1da177e4 | 603 | } |
e02e68d3 TH |
604 | return 0; |
605 | } | |
ea149b36 | 606 | EXPORT_SYMBOL_GPL(mce_notify_user); |
8a336b0a | 607 | |
d88203d1 | 608 | /* |
1da177e4 LT |
609 | * Initialize Machine Checks for a CPU. |
610 | */ | |
0d7482e3 | 611 | static int mce_cap_init(void) |
1da177e4 | 612 | { |
0d7482e3 | 613 | unsigned b; |
e9eee03e | 614 | u64 cap; |
1da177e4 LT |
615 | |
616 | rdmsrl(MSR_IA32_MCG_CAP, cap); | |
01c6680a TG |
617 | |
618 | b = cap & MCG_BANKCNT_MASK; | |
b659294b IM |
619 | printk(KERN_INFO "mce: CPU supports %d MCE banks\n", b); |
620 | ||
0d7482e3 AK |
621 | if (b > MAX_NR_BANKS) { |
622 | printk(KERN_WARNING | |
623 | "MCE: Using only %u machine check banks out of %u\n", | |
624 | MAX_NR_BANKS, b); | |
625 | b = MAX_NR_BANKS; | |
626 | } | |
627 | ||
628 | /* Don't support asymmetric configurations today */ | |
629 | WARN_ON(banks != 0 && b != banks); | |
630 | banks = b; | |
631 | if (!bank) { | |
632 | bank = kmalloc(banks * sizeof(u64), GFP_KERNEL); | |
633 | if (!bank) | |
634 | return -ENOMEM; | |
635 | memset(bank, 0xff, banks * sizeof(u64)); | |
1da177e4 | 636 | } |
0d7482e3 | 637 | |
94ad8474 | 638 | /* Use accurate RIP reporting if available. */ |
01c6680a | 639 | if ((cap & MCG_EXT_P) && MCG_EXT_CNT(cap) >= 9) |
94ad8474 | 640 | rip_msr = MSR_IA32_MCG_EIP; |
1da177e4 | 641 | |
0d7482e3 AK |
642 | return 0; |
643 | } | |
644 | ||
8be91105 | 645 | static void mce_init(void) |
0d7482e3 | 646 | { |
e9eee03e | 647 | mce_banks_t all_banks; |
0d7482e3 AK |
648 | u64 cap; |
649 | int i; | |
650 | ||
b79109c3 AK |
651 | /* |
652 | * Log the machine checks left over from the previous reset. | |
653 | */ | |
ee031c31 | 654 | bitmap_fill(all_banks, MAX_NR_BANKS); |
5679af4c | 655 | machine_check_poll(MCP_UC|(!mce_bootlog ? MCP_DONTLOG : 0), &all_banks); |
1da177e4 LT |
656 | |
657 | set_in_cr4(X86_CR4_MCE); | |
658 | ||
0d7482e3 | 659 | rdmsrl(MSR_IA32_MCG_CAP, cap); |
1da177e4 LT |
660 | if (cap & MCG_CTL_P) |
661 | wrmsr(MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff); | |
662 | ||
663 | for (i = 0; i < banks; i++) { | |
06b7a7a5 AK |
664 | if (skip_bank_init(i)) |
665 | continue; | |
0d7482e3 | 666 | wrmsrl(MSR_IA32_MC0_CTL+4*i, bank[i]); |
1da177e4 | 667 | wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0); |
d88203d1 | 668 | } |
1da177e4 LT |
669 | } |
670 | ||
671 | /* Add per CPU specific workarounds here */ | |
ec5b3d32 | 672 | static void mce_cpu_quirks(struct cpuinfo_x86 *c) |
d88203d1 | 673 | { |
1da177e4 | 674 | /* This should be disabled by the BIOS, but isn't always */ |
911f6a7b | 675 | if (c->x86_vendor == X86_VENDOR_AMD) { |
e9eee03e IM |
676 | if (c->x86 == 15 && banks > 4) { |
677 | /* | |
678 | * disable GART TBL walk error reporting, which | |
679 | * trips off incorrectly with the IOMMU & 3ware | |
680 | * & Cerberus: | |
681 | */ | |
0d7482e3 | 682 | clear_bit(10, (unsigned long *)&bank[4]); |
e9eee03e IM |
683 | } |
684 | if (c->x86 <= 17 && mce_bootlog < 0) { | |
685 | /* | |
686 | * Lots of broken BIOS around that don't clear them | |
687 | * by default and leave crap in there. Don't log: | |
688 | */ | |
911f6a7b | 689 | mce_bootlog = 0; |
e9eee03e | 690 | } |
2e6f694f AK |
691 | /* |
692 | * Various K7s with broken bank 0 around. Always disable | |
693 | * by default. | |
694 | */ | |
695 | if (c->x86 == 6) | |
696 | bank[0] = 0; | |
1da177e4 | 697 | } |
e583538f | 698 | |
06b7a7a5 AK |
699 | if (c->x86_vendor == X86_VENDOR_INTEL) { |
700 | /* | |
701 | * SDM documents that on family 6 bank 0 should not be written | |
702 | * because it aliases to another special BIOS controlled | |
703 | * register. | |
704 | * But it's not aliased anymore on model 0x1a+ | |
705 | * Don't ignore bank 0 completely because there could be a | |
706 | * valid event later, merely don't write CTL0. | |
707 | */ | |
708 | ||
709 | if (c->x86 == 6 && c->x86_model < 0x1A) | |
710 | __set_bit(0, &dont_init_banks); | |
711 | } | |
d88203d1 | 712 | } |
1da177e4 | 713 | |
4efc0670 AK |
714 | static void __cpuinit mce_ancient_init(struct cpuinfo_x86 *c) |
715 | { | |
716 | if (c->x86 != 5) | |
717 | return; | |
718 | switch (c->x86_vendor) { | |
719 | case X86_VENDOR_INTEL: | |
720 | if (mce_p5_enabled()) | |
721 | intel_p5_mcheck_init(c); | |
722 | break; | |
723 | case X86_VENDOR_CENTAUR: | |
724 | winchip_mcheck_init(c); | |
725 | break; | |
726 | } | |
727 | } | |
728 | ||
cc3ca220 | 729 | static void mce_cpu_features(struct cpuinfo_x86 *c) |
1da177e4 LT |
730 | { |
731 | switch (c->x86_vendor) { | |
732 | case X86_VENDOR_INTEL: | |
733 | mce_intel_feature_init(c); | |
734 | break; | |
89b831ef JS |
735 | case X86_VENDOR_AMD: |
736 | mce_amd_feature_init(c); | |
737 | break; | |
1da177e4 LT |
738 | default: |
739 | break; | |
740 | } | |
741 | } | |
742 | ||
52d168e2 AK |
743 | static void mce_init_timer(void) |
744 | { | |
745 | struct timer_list *t = &__get_cpu_var(mce_timer); | |
6298c512 | 746 | int *n = &__get_cpu_var(next_interval); |
52d168e2 | 747 | |
6298c512 AK |
748 | *n = check_interval * HZ; |
749 | if (!*n) | |
52d168e2 AK |
750 | return; |
751 | setup_timer(t, mcheck_timer, smp_processor_id()); | |
6298c512 | 752 | t->expires = round_jiffies(jiffies + *n); |
52d168e2 AK |
753 | add_timer(t); |
754 | } | |
755 | ||
d88203d1 | 756 | /* |
1da177e4 | 757 | * Called for each booted CPU to set up machine checks. |
e9eee03e | 758 | * Must be called with preempt off: |
1da177e4 | 759 | */ |
e6982c67 | 760 | void __cpuinit mcheck_init(struct cpuinfo_x86 *c) |
1da177e4 | 761 | { |
4efc0670 AK |
762 | if (mce_disabled) |
763 | return; | |
764 | ||
765 | mce_ancient_init(c); | |
766 | ||
5b4408fd | 767 | if (!mce_available(c)) |
1da177e4 LT |
768 | return; |
769 | ||
0d7482e3 | 770 | if (mce_cap_init() < 0) { |
04b2b1a4 | 771 | mce_disabled = 1; |
0d7482e3 AK |
772 | return; |
773 | } | |
774 | mce_cpu_quirks(c); | |
775 | ||
5d727926 AK |
776 | machine_check_vector = do_machine_check; |
777 | ||
8be91105 | 778 | mce_init(); |
1da177e4 | 779 | mce_cpu_features(c); |
52d168e2 | 780 | mce_init_timer(); |
1da177e4 LT |
781 | } |
782 | ||
783 | /* | |
784 | * Character device to read and clear the MCE log. | |
785 | */ | |
786 | ||
f528e7ba | 787 | static DEFINE_SPINLOCK(mce_state_lock); |
e9eee03e IM |
788 | static int open_count; /* #times opened */ |
789 | static int open_exclu; /* already open exclusive? */ | |
f528e7ba TH |
790 | |
791 | static int mce_open(struct inode *inode, struct file *file) | |
792 | { | |
793 | spin_lock(&mce_state_lock); | |
794 | ||
795 | if (open_exclu || (open_count && (file->f_flags & O_EXCL))) { | |
796 | spin_unlock(&mce_state_lock); | |
e9eee03e | 797 | |
f528e7ba TH |
798 | return -EBUSY; |
799 | } | |
800 | ||
801 | if (file->f_flags & O_EXCL) | |
802 | open_exclu = 1; | |
803 | open_count++; | |
804 | ||
805 | spin_unlock(&mce_state_lock); | |
806 | ||
bd78432c | 807 | return nonseekable_open(inode, file); |
f528e7ba TH |
808 | } |
809 | ||
810 | static int mce_release(struct inode *inode, struct file *file) | |
811 | { | |
812 | spin_lock(&mce_state_lock); | |
813 | ||
814 | open_count--; | |
815 | open_exclu = 0; | |
816 | ||
817 | spin_unlock(&mce_state_lock); | |
818 | ||
819 | return 0; | |
820 | } | |
821 | ||
d88203d1 TG |
822 | static void collect_tscs(void *data) |
823 | { | |
1da177e4 | 824 | unsigned long *cpu_tsc = (unsigned long *)data; |
d88203d1 | 825 | |
1da177e4 | 826 | rdtscll(cpu_tsc[smp_processor_id()]); |
d88203d1 | 827 | } |
1da177e4 | 828 | |
e9eee03e IM |
829 | static DEFINE_MUTEX(mce_read_mutex); |
830 | ||
d88203d1 TG |
831 | static ssize_t mce_read(struct file *filp, char __user *ubuf, size_t usize, |
832 | loff_t *off) | |
1da177e4 | 833 | { |
e9eee03e | 834 | char __user *buf = ubuf; |
f0de53bb | 835 | unsigned long *cpu_tsc; |
ef41df43 | 836 | unsigned prev, next; |
1da177e4 LT |
837 | int i, err; |
838 | ||
6bca67f9 | 839 | cpu_tsc = kmalloc(nr_cpu_ids * sizeof(long), GFP_KERNEL); |
f0de53bb AK |
840 | if (!cpu_tsc) |
841 | return -ENOMEM; | |
842 | ||
8c8b8859 | 843 | mutex_lock(&mce_read_mutex); |
1da177e4 LT |
844 | next = rcu_dereference(mcelog.next); |
845 | ||
846 | /* Only supports full reads right now */ | |
d88203d1 | 847 | if (*off != 0 || usize < MCE_LOG_LEN*sizeof(struct mce)) { |
8c8b8859 | 848 | mutex_unlock(&mce_read_mutex); |
f0de53bb | 849 | kfree(cpu_tsc); |
e9eee03e | 850 | |
1da177e4 LT |
851 | return -EINVAL; |
852 | } | |
853 | ||
854 | err = 0; | |
ef41df43 HY |
855 | prev = 0; |
856 | do { | |
857 | for (i = prev; i < next; i++) { | |
858 | unsigned long start = jiffies; | |
859 | ||
860 | while (!mcelog.entry[i].finished) { | |
861 | if (time_after_eq(jiffies, start + 2)) { | |
862 | memset(mcelog.entry + i, 0, | |
863 | sizeof(struct mce)); | |
864 | goto timeout; | |
865 | } | |
866 | cpu_relax(); | |
673242c1 | 867 | } |
ef41df43 HY |
868 | smp_rmb(); |
869 | err |= copy_to_user(buf, mcelog.entry + i, | |
870 | sizeof(struct mce)); | |
871 | buf += sizeof(struct mce); | |
872 | timeout: | |
873 | ; | |
673242c1 | 874 | } |
1da177e4 | 875 | |
ef41df43 HY |
876 | memset(mcelog.entry + prev, 0, |
877 | (next - prev) * sizeof(struct mce)); | |
878 | prev = next; | |
879 | next = cmpxchg(&mcelog.next, prev, 0); | |
880 | } while (next != prev); | |
1da177e4 | 881 | |
b2b18660 | 882 | synchronize_sched(); |
1da177e4 | 883 | |
d88203d1 TG |
884 | /* |
885 | * Collect entries that were still getting written before the | |
886 | * synchronize. | |
887 | */ | |
15c8b6c1 | 888 | on_each_cpu(collect_tscs, cpu_tsc, 1); |
e9eee03e | 889 | |
d88203d1 TG |
890 | for (i = next; i < MCE_LOG_LEN; i++) { |
891 | if (mcelog.entry[i].finished && | |
892 | mcelog.entry[i].tsc < cpu_tsc[mcelog.entry[i].cpu]) { | |
893 | err |= copy_to_user(buf, mcelog.entry+i, | |
894 | sizeof(struct mce)); | |
1da177e4 LT |
895 | smp_rmb(); |
896 | buf += sizeof(struct mce); | |
897 | memset(&mcelog.entry[i], 0, sizeof(struct mce)); | |
898 | } | |
d88203d1 | 899 | } |
8c8b8859 | 900 | mutex_unlock(&mce_read_mutex); |
f0de53bb | 901 | kfree(cpu_tsc); |
e9eee03e | 902 | |
d88203d1 | 903 | return err ? -EFAULT : buf - ubuf; |
1da177e4 LT |
904 | } |
905 | ||
e02e68d3 TH |
906 | static unsigned int mce_poll(struct file *file, poll_table *wait) |
907 | { | |
908 | poll_wait(file, &mce_wait, wait); | |
909 | if (rcu_dereference(mcelog.next)) | |
910 | return POLLIN | POLLRDNORM; | |
911 | return 0; | |
912 | } | |
913 | ||
c68461b6 | 914 | static long mce_ioctl(struct file *f, unsigned int cmd, unsigned long arg) |
1da177e4 LT |
915 | { |
916 | int __user *p = (int __user *)arg; | |
d88203d1 | 917 | |
1da177e4 | 918 | if (!capable(CAP_SYS_ADMIN)) |
d88203d1 | 919 | return -EPERM; |
e9eee03e | 920 | |
1da177e4 | 921 | switch (cmd) { |
d88203d1 | 922 | case MCE_GET_RECORD_LEN: |
1da177e4 LT |
923 | return put_user(sizeof(struct mce), p); |
924 | case MCE_GET_LOG_LEN: | |
d88203d1 | 925 | return put_user(MCE_LOG_LEN, p); |
1da177e4 LT |
926 | case MCE_GETCLEAR_FLAGS: { |
927 | unsigned flags; | |
d88203d1 TG |
928 | |
929 | do { | |
1da177e4 | 930 | flags = mcelog.flags; |
d88203d1 | 931 | } while (cmpxchg(&mcelog.flags, flags, 0) != flags); |
e9eee03e | 932 | |
d88203d1 | 933 | return put_user(flags, p); |
1da177e4 LT |
934 | } |
935 | default: | |
d88203d1 TG |
936 | return -ENOTTY; |
937 | } | |
1da177e4 LT |
938 | } |
939 | ||
a1ff41bf | 940 | /* Modified in mce-inject.c, so not static or const */ |
ea149b36 | 941 | struct file_operations mce_chrdev_ops = { |
e9eee03e IM |
942 | .open = mce_open, |
943 | .release = mce_release, | |
944 | .read = mce_read, | |
945 | .poll = mce_poll, | |
946 | .unlocked_ioctl = mce_ioctl, | |
1da177e4 | 947 | }; |
ea149b36 | 948 | EXPORT_SYMBOL_GPL(mce_chrdev_ops); |
1da177e4 LT |
949 | |
950 | static struct miscdevice mce_log_device = { | |
951 | MISC_MCELOG_MINOR, | |
952 | "mcelog", | |
953 | &mce_chrdev_ops, | |
954 | }; | |
955 | ||
13503fa9 HS |
956 | /* |
957 | * mce=off disables machine check | |
958 | * mce=TOLERANCELEVEL (number, see above) | |
959 | * mce=bootlog Log MCEs from before booting. Disabled by default on AMD. | |
960 | * mce=nobootlog Don't log MCEs from before booting. | |
961 | */ | |
1da177e4 LT |
962 | static int __init mcheck_enable(char *str) |
963 | { | |
4efc0670 AK |
964 | if (*str == 0) |
965 | enable_p5_mce(); | |
966 | if (*str == '=') | |
967 | str++; | |
1da177e4 | 968 | if (!strcmp(str, "off")) |
04b2b1a4 | 969 | mce_disabled = 1; |
13503fa9 HS |
970 | else if (!strcmp(str, "bootlog") || !strcmp(str, "nobootlog")) |
971 | mce_bootlog = (str[0] == 'b'); | |
8c566ef5 AK |
972 | else if (isdigit(str[0])) |
973 | get_option(&str, &tolerant); | |
13503fa9 | 974 | else { |
4efc0670 | 975 | printk(KERN_INFO "mce argument %s ignored. Please use /sys\n", |
13503fa9 HS |
976 | str); |
977 | return 0; | |
978 | } | |
9b41046c | 979 | return 1; |
1da177e4 | 980 | } |
4efc0670 | 981 | __setup("mce", mcheck_enable); |
1da177e4 | 982 | |
d88203d1 | 983 | /* |
1da177e4 | 984 | * Sysfs support |
d88203d1 | 985 | */ |
1da177e4 | 986 | |
973a2dd1 AK |
987 | /* |
988 | * Disable machine checks on suspend and shutdown. We can't really handle | |
989 | * them later. | |
990 | */ | |
991 | static int mce_disable(void) | |
992 | { | |
993 | int i; | |
994 | ||
06b7a7a5 AK |
995 | for (i = 0; i < banks; i++) { |
996 | if (!skip_bank_init(i)) | |
997 | wrmsrl(MSR_IA32_MC0_CTL + i*4, 0); | |
998 | } | |
973a2dd1 AK |
999 | return 0; |
1000 | } | |
1001 | ||
1002 | static int mce_suspend(struct sys_device *dev, pm_message_t state) | |
1003 | { | |
1004 | return mce_disable(); | |
1005 | } | |
1006 | ||
1007 | static int mce_shutdown(struct sys_device *dev) | |
1008 | { | |
1009 | return mce_disable(); | |
1010 | } | |
1011 | ||
e9eee03e IM |
1012 | /* |
1013 | * On resume clear all MCE state. Don't want to see leftovers from the BIOS. | |
1014 | * Only one CPU is active at this time, the others get re-added later using | |
1015 | * CPU hotplug: | |
1016 | */ | |
1da177e4 LT |
1017 | static int mce_resume(struct sys_device *dev) |
1018 | { | |
8be91105 | 1019 | mce_init(); |
6ec68bff | 1020 | mce_cpu_features(¤t_cpu_data); |
e9eee03e | 1021 | |
1da177e4 LT |
1022 | return 0; |
1023 | } | |
1024 | ||
52d168e2 AK |
1025 | static void mce_cpu_restart(void *data) |
1026 | { | |
1027 | del_timer_sync(&__get_cpu_var(mce_timer)); | |
1028 | if (mce_available(¤t_cpu_data)) | |
8be91105 | 1029 | mce_init(); |
52d168e2 AK |
1030 | mce_init_timer(); |
1031 | } | |
1032 | ||
1da177e4 | 1033 | /* Reinit MCEs after user configuration changes */ |
d88203d1 TG |
1034 | static void mce_restart(void) |
1035 | { | |
52d168e2 | 1036 | on_each_cpu(mce_cpu_restart, NULL, 1); |
1da177e4 LT |
1037 | } |
1038 | ||
1039 | static struct sysdev_class mce_sysclass = { | |
e9eee03e IM |
1040 | .suspend = mce_suspend, |
1041 | .shutdown = mce_shutdown, | |
1042 | .resume = mce_resume, | |
1043 | .name = "machinecheck", | |
1da177e4 LT |
1044 | }; |
1045 | ||
cb491fca | 1046 | DEFINE_PER_CPU(struct sys_device, mce_dev); |
e9eee03e IM |
1047 | |
1048 | __cpuinitdata | |
1049 | void (*threshold_cpu_callback)(unsigned long action, unsigned int cpu); | |
1da177e4 | 1050 | |
0d7482e3 AK |
1051 | static struct sysdev_attribute *bank_attrs; |
1052 | ||
1053 | static ssize_t show_bank(struct sys_device *s, struct sysdev_attribute *attr, | |
1054 | char *buf) | |
1055 | { | |
1056 | u64 b = bank[attr - bank_attrs]; | |
e9eee03e | 1057 | |
f6d1826d | 1058 | return sprintf(buf, "%llx\n", b); |
0d7482e3 AK |
1059 | } |
1060 | ||
1061 | static ssize_t set_bank(struct sys_device *s, struct sysdev_attribute *attr, | |
9319cec8 | 1062 | const char *buf, size_t size) |
0d7482e3 | 1063 | { |
9319cec8 | 1064 | u64 new; |
e9eee03e | 1065 | |
9319cec8 | 1066 | if (strict_strtoull(buf, 0, &new) < 0) |
0d7482e3 | 1067 | return -EINVAL; |
e9eee03e | 1068 | |
0d7482e3 AK |
1069 | bank[attr - bank_attrs] = new; |
1070 | mce_restart(); | |
e9eee03e | 1071 | |
9319cec8 | 1072 | return size; |
0d7482e3 | 1073 | } |
a98f0dd3 | 1074 | |
e9eee03e IM |
1075 | static ssize_t |
1076 | show_trigger(struct sys_device *s, struct sysdev_attribute *attr, char *buf) | |
a98f0dd3 AK |
1077 | { |
1078 | strcpy(buf, trigger); | |
1079 | strcat(buf, "\n"); | |
1080 | return strlen(trigger) + 1; | |
1081 | } | |
1082 | ||
4a0b2b4d | 1083 | static ssize_t set_trigger(struct sys_device *s, struct sysdev_attribute *attr, |
e9eee03e | 1084 | const char *buf, size_t siz) |
a98f0dd3 AK |
1085 | { |
1086 | char *p; | |
1087 | int len; | |
e9eee03e | 1088 | |
a98f0dd3 AK |
1089 | strncpy(trigger, buf, sizeof(trigger)); |
1090 | trigger[sizeof(trigger)-1] = 0; | |
1091 | len = strlen(trigger); | |
1092 | p = strchr(trigger, '\n'); | |
e9eee03e IM |
1093 | |
1094 | if (*p) | |
1095 | *p = 0; | |
1096 | ||
a98f0dd3 AK |
1097 | return len; |
1098 | } | |
1099 | ||
b56f642d AK |
1100 | static ssize_t store_int_with_restart(struct sys_device *s, |
1101 | struct sysdev_attribute *attr, | |
1102 | const char *buf, size_t size) | |
1103 | { | |
1104 | ssize_t ret = sysdev_store_int(s, attr, buf, size); | |
1105 | mce_restart(); | |
1106 | return ret; | |
1107 | } | |
1108 | ||
a98f0dd3 | 1109 | static SYSDEV_ATTR(trigger, 0644, show_trigger, set_trigger); |
d95d62c0 | 1110 | static SYSDEV_INT_ATTR(tolerant, 0644, tolerant); |
e9eee03e | 1111 | |
b56f642d AK |
1112 | static struct sysdev_ext_attribute attr_check_interval = { |
1113 | _SYSDEV_ATTR(check_interval, 0644, sysdev_show_int, | |
1114 | store_int_with_restart), | |
1115 | &check_interval | |
1116 | }; | |
e9eee03e | 1117 | |
cb491fca | 1118 | static struct sysdev_attribute *mce_attrs[] = { |
b56f642d | 1119 | &attr_tolerant.attr, &attr_check_interval.attr, &attr_trigger, |
a98f0dd3 AK |
1120 | NULL |
1121 | }; | |
1da177e4 | 1122 | |
cb491fca | 1123 | static cpumask_var_t mce_dev_initialized; |
bae19fe0 | 1124 | |
e9eee03e | 1125 | /* Per cpu sysdev init. All of the cpus still share the same ctrl bank: */ |
91c6d400 | 1126 | static __cpuinit int mce_create_device(unsigned int cpu) |
1da177e4 LT |
1127 | { |
1128 | int err; | |
73ca5358 | 1129 | int i; |
92cb7612 | 1130 | |
90367556 | 1131 | if (!mce_available(&boot_cpu_data)) |
91c6d400 AK |
1132 | return -EIO; |
1133 | ||
cb491fca IM |
1134 | memset(&per_cpu(mce_dev, cpu).kobj, 0, sizeof(struct kobject)); |
1135 | per_cpu(mce_dev, cpu).id = cpu; | |
1136 | per_cpu(mce_dev, cpu).cls = &mce_sysclass; | |
91c6d400 | 1137 | |
cb491fca | 1138 | err = sysdev_register(&per_cpu(mce_dev, cpu)); |
d435d862 AM |
1139 | if (err) |
1140 | return err; | |
1141 | ||
cb491fca IM |
1142 | for (i = 0; mce_attrs[i]; i++) { |
1143 | err = sysdev_create_file(&per_cpu(mce_dev, cpu), mce_attrs[i]); | |
d435d862 AM |
1144 | if (err) |
1145 | goto error; | |
1146 | } | |
0d7482e3 | 1147 | for (i = 0; i < banks; i++) { |
cb491fca | 1148 | err = sysdev_create_file(&per_cpu(mce_dev, cpu), |
0d7482e3 AK |
1149 | &bank_attrs[i]); |
1150 | if (err) | |
1151 | goto error2; | |
1152 | } | |
cb491fca | 1153 | cpumask_set_cpu(cpu, mce_dev_initialized); |
91c6d400 | 1154 | |
d435d862 | 1155 | return 0; |
0d7482e3 | 1156 | error2: |
cb491fca IM |
1157 | while (--i >= 0) |
1158 | sysdev_remove_file(&per_cpu(mce_dev, cpu), &bank_attrs[i]); | |
d435d862 | 1159 | error: |
cb491fca IM |
1160 | while (--i >= 0) |
1161 | sysdev_remove_file(&per_cpu(mce_dev, cpu), mce_attrs[i]); | |
1162 | ||
1163 | sysdev_unregister(&per_cpu(mce_dev, cpu)); | |
d435d862 | 1164 | |
91c6d400 AK |
1165 | return err; |
1166 | } | |
1167 | ||
2d9cd6c2 | 1168 | static __cpuinit void mce_remove_device(unsigned int cpu) |
91c6d400 | 1169 | { |
73ca5358 SL |
1170 | int i; |
1171 | ||
cb491fca | 1172 | if (!cpumask_test_cpu(cpu, mce_dev_initialized)) |
bae19fe0 AH |
1173 | return; |
1174 | ||
cb491fca IM |
1175 | for (i = 0; mce_attrs[i]; i++) |
1176 | sysdev_remove_file(&per_cpu(mce_dev, cpu), mce_attrs[i]); | |
1177 | ||
0d7482e3 | 1178 | for (i = 0; i < banks; i++) |
cb491fca IM |
1179 | sysdev_remove_file(&per_cpu(mce_dev, cpu), &bank_attrs[i]); |
1180 | ||
1181 | sysdev_unregister(&per_cpu(mce_dev, cpu)); | |
1182 | cpumask_clear_cpu(cpu, mce_dev_initialized); | |
91c6d400 | 1183 | } |
91c6d400 | 1184 | |
d6b75584 | 1185 | /* Make sure there are no machine checks on offlined CPUs. */ |
ec5b3d32 | 1186 | static void mce_disable_cpu(void *h) |
d6b75584 | 1187 | { |
88ccbedd | 1188 | unsigned long action = *(unsigned long *)h; |
cb491fca | 1189 | int i; |
d6b75584 AK |
1190 | |
1191 | if (!mce_available(¤t_cpu_data)) | |
1192 | return; | |
88ccbedd AK |
1193 | if (!(action & CPU_TASKS_FROZEN)) |
1194 | cmci_clear(); | |
06b7a7a5 AK |
1195 | for (i = 0; i < banks; i++) { |
1196 | if (!skip_bank_init(i)) | |
1197 | wrmsrl(MSR_IA32_MC0_CTL + i*4, 0); | |
1198 | } | |
d6b75584 AK |
1199 | } |
1200 | ||
ec5b3d32 | 1201 | static void mce_reenable_cpu(void *h) |
d6b75584 | 1202 | { |
88ccbedd | 1203 | unsigned long action = *(unsigned long *)h; |
e9eee03e | 1204 | int i; |
d6b75584 AK |
1205 | |
1206 | if (!mce_available(¤t_cpu_data)) | |
1207 | return; | |
e9eee03e | 1208 | |
88ccbedd AK |
1209 | if (!(action & CPU_TASKS_FROZEN)) |
1210 | cmci_reenable(); | |
06b7a7a5 AK |
1211 | for (i = 0; i < banks; i++) { |
1212 | if (!skip_bank_init(i)) | |
1213 | wrmsrl(MSR_IA32_MC0_CTL + i*4, bank[i]); | |
1214 | } | |
d6b75584 AK |
1215 | } |
1216 | ||
91c6d400 | 1217 | /* Get notified when a cpu comes on/off. Be hotplug friendly. */ |
e9eee03e IM |
1218 | static int __cpuinit |
1219 | mce_cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
91c6d400 AK |
1220 | { |
1221 | unsigned int cpu = (unsigned long)hcpu; | |
52d168e2 | 1222 | struct timer_list *t = &per_cpu(mce_timer, cpu); |
91c6d400 AK |
1223 | |
1224 | switch (action) { | |
bae19fe0 AH |
1225 | case CPU_ONLINE: |
1226 | case CPU_ONLINE_FROZEN: | |
1227 | mce_create_device(cpu); | |
8735728e RW |
1228 | if (threshold_cpu_callback) |
1229 | threshold_cpu_callback(action, cpu); | |
91c6d400 | 1230 | break; |
91c6d400 | 1231 | case CPU_DEAD: |
8bb78442 | 1232 | case CPU_DEAD_FROZEN: |
8735728e RW |
1233 | if (threshold_cpu_callback) |
1234 | threshold_cpu_callback(action, cpu); | |
91c6d400 AK |
1235 | mce_remove_device(cpu); |
1236 | break; | |
52d168e2 AK |
1237 | case CPU_DOWN_PREPARE: |
1238 | case CPU_DOWN_PREPARE_FROZEN: | |
1239 | del_timer_sync(t); | |
88ccbedd | 1240 | smp_call_function_single(cpu, mce_disable_cpu, &action, 1); |
52d168e2 AK |
1241 | break; |
1242 | case CPU_DOWN_FAILED: | |
1243 | case CPU_DOWN_FAILED_FROZEN: | |
6298c512 AK |
1244 | t->expires = round_jiffies(jiffies + |
1245 | __get_cpu_var(next_interval)); | |
52d168e2 | 1246 | add_timer_on(t, cpu); |
88ccbedd AK |
1247 | smp_call_function_single(cpu, mce_reenable_cpu, &action, 1); |
1248 | break; | |
1249 | case CPU_POST_DEAD: | |
1250 | /* intentionally ignoring frozen here */ | |
1251 | cmci_rediscover(cpu); | |
52d168e2 | 1252 | break; |
91c6d400 | 1253 | } |
bae19fe0 | 1254 | return NOTIFY_OK; |
91c6d400 AK |
1255 | } |
1256 | ||
1e35669d | 1257 | static struct notifier_block mce_cpu_notifier __cpuinitdata = { |
91c6d400 AK |
1258 | .notifier_call = mce_cpu_callback, |
1259 | }; | |
1260 | ||
0d7482e3 AK |
1261 | static __init int mce_init_banks(void) |
1262 | { | |
1263 | int i; | |
1264 | ||
1265 | bank_attrs = kzalloc(sizeof(struct sysdev_attribute) * banks, | |
1266 | GFP_KERNEL); | |
1267 | if (!bank_attrs) | |
1268 | return -ENOMEM; | |
1269 | ||
1270 | for (i = 0; i < banks; i++) { | |
1271 | struct sysdev_attribute *a = &bank_attrs[i]; | |
e9eee03e IM |
1272 | |
1273 | a->attr.name = kasprintf(GFP_KERNEL, "bank%d", i); | |
0d7482e3 AK |
1274 | if (!a->attr.name) |
1275 | goto nomem; | |
e9eee03e IM |
1276 | |
1277 | a->attr.mode = 0644; | |
1278 | a->show = show_bank; | |
1279 | a->store = set_bank; | |
0d7482e3 AK |
1280 | } |
1281 | return 0; | |
1282 | ||
1283 | nomem: | |
1284 | while (--i >= 0) | |
1285 | kfree(bank_attrs[i].attr.name); | |
1286 | kfree(bank_attrs); | |
1287 | bank_attrs = NULL; | |
e9eee03e | 1288 | |
0d7482e3 AK |
1289 | return -ENOMEM; |
1290 | } | |
1291 | ||
91c6d400 AK |
1292 | static __init int mce_init_device(void) |
1293 | { | |
1294 | int err; | |
1295 | int i = 0; | |
1296 | ||
1da177e4 LT |
1297 | if (!mce_available(&boot_cpu_data)) |
1298 | return -EIO; | |
0d7482e3 | 1299 | |
cb491fca | 1300 | alloc_cpumask_var(&mce_dev_initialized, GFP_KERNEL); |
996867d0 | 1301 | |
0d7482e3 AK |
1302 | err = mce_init_banks(); |
1303 | if (err) | |
1304 | return err; | |
1305 | ||
1da177e4 | 1306 | err = sysdev_class_register(&mce_sysclass); |
d435d862 AM |
1307 | if (err) |
1308 | return err; | |
91c6d400 AK |
1309 | |
1310 | for_each_online_cpu(i) { | |
d435d862 AM |
1311 | err = mce_create_device(i); |
1312 | if (err) | |
1313 | return err; | |
91c6d400 AK |
1314 | } |
1315 | ||
be6b5a35 | 1316 | register_hotcpu_notifier(&mce_cpu_notifier); |
1da177e4 | 1317 | misc_register(&mce_log_device); |
e9eee03e | 1318 | |
1da177e4 | 1319 | return err; |
1da177e4 | 1320 | } |
91c6d400 | 1321 | |
1da177e4 | 1322 | device_initcall(mce_init_device); |
a988d334 | 1323 | |
4efc0670 | 1324 | #else /* CONFIG_X86_OLD_MCE: */ |
a988d334 | 1325 | |
a988d334 IM |
1326 | int nr_mce_banks; |
1327 | EXPORT_SYMBOL_GPL(nr_mce_banks); /* non-fatal.o */ | |
1328 | ||
a988d334 IM |
1329 | /* This has to be run for each processor */ |
1330 | void mcheck_init(struct cpuinfo_x86 *c) | |
1331 | { | |
1332 | if (mce_disabled == 1) | |
1333 | return; | |
1334 | ||
1335 | switch (c->x86_vendor) { | |
1336 | case X86_VENDOR_AMD: | |
1337 | amd_mcheck_init(c); | |
1338 | break; | |
1339 | ||
1340 | case X86_VENDOR_INTEL: | |
1341 | if (c->x86 == 5) | |
1342 | intel_p5_mcheck_init(c); | |
1343 | if (c->x86 == 6) | |
1344 | intel_p6_mcheck_init(c); | |
1345 | if (c->x86 == 15) | |
1346 | intel_p4_mcheck_init(c); | |
1347 | break; | |
1348 | ||
1349 | case X86_VENDOR_CENTAUR: | |
1350 | if (c->x86 == 5) | |
1351 | winchip_mcheck_init(c); | |
1352 | break; | |
1353 | ||
1354 | default: | |
1355 | break; | |
1356 | } | |
b659294b | 1357 | printk(KERN_INFO "mce: CPU supports %d MCE banks\n", nr_mce_banks); |
a988d334 IM |
1358 | } |
1359 | ||
a988d334 IM |
1360 | static int __init mcheck_enable(char *str) |
1361 | { | |
1362 | mce_disabled = -1; | |
1363 | return 1; | |
1364 | } | |
1365 | ||
a988d334 IM |
1366 | __setup("mce", mcheck_enable); |
1367 | ||
d7c3c9a6 AK |
1368 | #endif /* CONFIG_X86_OLD_MCE */ |
1369 | ||
1370 | /* | |
1371 | * Old style boot options parsing. Only for compatibility. | |
1372 | */ | |
1373 | static int __init mcheck_disable(char *str) | |
1374 | { | |
1375 | mce_disabled = 1; | |
1376 | return 1; | |
1377 | } | |
1378 | __setup("nomce", mcheck_disable); |