mce_64.c 22.8 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3
/*
 * Machine check handler.
 * K8 parts Copyright 2002,2003 Andi Kleen, SuSE Labs.
4 5
 * Rest from unknown author(s).
 * 2004 Andi Kleen. Rewrote most of it.
L
Linus Torvalds 已提交
6 7 8 9 10 11
 */

#include <linux/init.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/sched.h>
A
Arnd Bergmann 已提交
12
#include <linux/smp_lock.h>
L
Linus Torvalds 已提交
13 14 15 16 17 18
#include <linux/string.h>
#include <linux/rcupdate.h>
#include <linux/kallsyms.h>
#include <linux/sysdev.h>
#include <linux/miscdevice.h>
#include <linux/fs.h>
19
#include <linux/capability.h>
20 21
#include <linux/cpu.h>
#include <linux/percpu.h>
22 23
#include <linux/poll.h>
#include <linux/thread_info.h>
24
#include <linux/ctype.h>
25
#include <linux/kmod.h>
26
#include <linux/kdebug.h>
27
#include <asm/processor.h>
L
Linus Torvalds 已提交
28 29 30
#include <asm/msr.h>
#include <asm/mce.h>
#include <asm/uaccess.h>
31
#include <asm/smp.h>
32
#include <asm/idle.h>
L
Linus Torvalds 已提交
33 34

#define MISC_MCELOG_MINOR 227
35
#define NR_SYSFS_BANKS 6
L
Linus Torvalds 已提交
36

37 38
atomic_t mce_entry;

L
Linus Torvalds 已提交
39 40
static int mce_dont_init;

41 42 43 44 45 46 47
/*
 * Tolerant levels:
 *   0: always panic on uncorrected errors, log corrected errors
 *   1: panic or SIGBUS on uncorrected errors, log corrected errors
 *   2: SIGBUS or log uncorrected errors (if possible), log corrected errors
 *   3: never panic or SIGBUS, log all errors (for testing only)
 */
L
Linus Torvalds 已提交
48 49
static int tolerant = 1;
static int banks;
50
static unsigned long bank[NR_SYSFS_BANKS] = { [0 ... NR_SYSFS_BANKS-1] = ~0UL };
51
static unsigned long notify_user;
52
static int rip_msr;
53
static int mce_bootlog = -1;
54 55 56 57
static atomic_t mce_events;

static char trigger[128];
static char *trigger_argv[2] = { trigger, NULL };
L
Linus Torvalds 已提交
58

59 60
static DECLARE_WAIT_QUEUE_HEAD(mce_wait);

L
Linus Torvalds 已提交
61 62 63 64 65 66
/*
 * Lockless MCE logging infrastructure.
 * This avoids deadlocks on printk locks without having to break locks. Also
 * separate MCEs from kernel messages to avoid bogus bug reports.
 */

67
static struct mce_log mcelog = {
L
Linus Torvalds 已提交
68 69
	MCE_LOG_SIGNATURE,
	MCE_LOG_LEN,
70
};
L
Linus Torvalds 已提交
71 72 73 74

void mce_log(struct mce *mce)
{
	unsigned next, entry;
75
	atomic_inc(&mce_events);
L
Linus Torvalds 已提交
76
	mce->finished = 0;
M
Mike Waychison 已提交
77
	wmb();
L
Linus Torvalds 已提交
78 79
	for (;;) {
		entry = rcu_dereference(mcelog.next);
80 81 82 83
		for (;;) {
			/* When the buffer fills up discard new entries. Assume
			   that the earlier errors are the more interesting. */
			if (entry >= MCE_LOG_LEN) {
84
				set_bit(MCE_OVERFLOW, (unsigned long *)&mcelog.flags);
85 86 87 88 89 90 91
				return;
			}
			/* Old left over entry. Skip. */
			if (mcelog.entry[entry].finished) {
				entry++;
				continue;
			}
M
Mike Waychison 已提交
92
			break;
L
Linus Torvalds 已提交
93 94 95 96 97 98 99
		}
		smp_rmb();
		next = entry + 1;
		if (cmpxchg(&mcelog.next, entry, next) == entry)
			break;
	}
	memcpy(mcelog.entry + entry, mce, sizeof(struct mce));
M
Mike Waychison 已提交
100
	wmb();
L
Linus Torvalds 已提交
101
	mcelog.entry[entry].finished = 1;
M
Mike Waychison 已提交
102
	wmb();
L
Linus Torvalds 已提交
103

104
	set_bit(0, &notify_user);
L
Linus Torvalds 已提交
105 106 107 108 109
}

static void print_mce(struct mce *m)
{
	printk(KERN_EMERG "\n"
110
	       KERN_EMERG "HARDWARE ERROR\n"
L
Linus Torvalds 已提交
111 112 113
	       KERN_EMERG
	       "CPU %d: Machine Check Exception: %16Lx Bank %d: %016Lx\n",
	       m->cpu, m->mcgstatus, m->bank, m->status);
114
	if (m->ip) {
115
		printk(KERN_EMERG "RIP%s %02x:<%016Lx> ",
L
Linus Torvalds 已提交
116
		       !(m->mcgstatus & MCG_STATUS_EIPV) ? " !INEXACT!" : "",
117
		       m->cs, m->ip);
L
Linus Torvalds 已提交
118
		if (m->cs == __KERNEL_CS)
119
			print_symbol("{%s}", m->ip);
L
Linus Torvalds 已提交
120 121
		printk("\n");
	}
122
	printk(KERN_EMERG "TSC %Lx ", m->tsc);
L
Linus Torvalds 已提交
123 124 125
	if (m->addr)
		printk("ADDR %Lx ", m->addr);
	if (m->misc)
126
		printk("MISC %Lx ", m->misc);
L
Linus Torvalds 已提交
127
	printk("\n");
128
	printk(KERN_EMERG "This is not a software problem!\n");
129 130
	printk(KERN_EMERG "Run through mcelog --ascii to decode "
	       "and contact your hardware vendor\n");
L
Linus Torvalds 已提交
131 132 133
}

static void mce_panic(char *msg, struct mce *backup, unsigned long start)
134
{
L
Linus Torvalds 已提交
135
	int i;
136

L
Linus Torvalds 已提交
137 138 139
	oops_begin();
	for (i = 0; i < MCE_LOG_LEN; i++) {
		unsigned long tsc = mcelog.entry[i].tsc;
140

L
Linus Torvalds 已提交
141 142
		if (time_before(tsc, start))
			continue;
143
		print_mce(&mcelog.entry[i]);
L
Linus Torvalds 已提交
144 145 146 147 148
		if (backup && mcelog.entry[i].tsc == backup->tsc)
			backup = NULL;
	}
	if (backup)
		print_mce(backup);
149
	panic(msg);
150
}
L
Linus Torvalds 已提交
151 152 153

static int mce_available(struct cpuinfo_x86 *c)
{
154
	return cpu_has(c, X86_FEATURE_MCE) && cpu_has(c, X86_FEATURE_MCA);
L
Linus Torvalds 已提交
155 156
}

157 158 159
static inline void mce_get_rip(struct mce *m, struct pt_regs *regs)
{
	if (regs && (m->mcgstatus & MCG_STATUS_RIPV)) {
160
		m->ip = regs->ip;
161 162
		m->cs = regs->cs;
	} else {
163
		m->ip = 0;
164 165 166 167 168
		m->cs = 0;
	}
	if (rip_msr) {
		/* Assume the RIP in the MSR is exact. Is this true? */
		m->mcgstatus |= MCG_STATUS_EIPV;
169
		rdmsrl(rip_msr, m->ip);
170 171 172 173
		m->cs = 0;
	}
}

174
/*
L
Linus Torvalds 已提交
175 176 177 178 179 180 181 182
 * The actual machine check handler
 */
void do_machine_check(struct pt_regs * regs, long error_code)
{
	struct mce m, panicm;
	u64 mcestart = 0;
	int i;
	int panicm_found = 0;
183 184 185 186 187 188 189 190 191 192
	/*
	 * If no_way_out gets set, there is no safe way to recover from this
	 * MCE.  If tolerant is cranked up, we'll try anyway.
	 */
	int no_way_out = 0;
	/*
	 * If kill_it gets set, there might be a way to recover from this
	 * error.
	 */
	int kill_it = 0;
L
Linus Torvalds 已提交
193

194 195
	atomic_inc(&mce_entry);

196 197 198 199
	if ((regs
	     && notify_die(DIE_NMI, "machine check", regs, error_code,
			   18, SIGKILL) == NOTIFY_STOP)
	    || !banks)
200
		goto out2;
L
Linus Torvalds 已提交
201 202

	memset(&m, 0, sizeof(struct mce));
203
	m.cpu = smp_processor_id();
L
Linus Torvalds 已提交
204
	rdmsrl(MSR_IA32_MCG_STATUS, m.mcgstatus);
205
	/* if the restart IP is not valid, we're done for */
L
Linus Torvalds 已提交
206
	if (!(m.mcgstatus & MCG_STATUS_RIPV))
207
		no_way_out = 1;
208

L
Linus Torvalds 已提交
209 210 211 212
	rdtscll(mcestart);
	barrier();

	for (i = 0; i < banks; i++) {
213
		if (i < NR_SYSFS_BANKS && !bank[i])
L
Linus Torvalds 已提交
214
			continue;
215 216

		m.misc = 0;
L
Linus Torvalds 已提交
217 218 219 220 221 222 223 224 225
		m.addr = 0;
		m.bank = i;
		m.tsc = 0;

		rdmsrl(MSR_IA32_MC0_STATUS + i*4, m.status);
		if ((m.status & MCI_STATUS_VAL) == 0)
			continue;

		if (m.status & MCI_STATUS_EN) {
226 227 228 229 230 231 232 233 234 235 236 237
			/* if PCC was set, there's no way out */
			no_way_out |= !!(m.status & MCI_STATUS_PCC);
			/*
			 * If this error was uncorrectable and there was
			 * an overflow, we're in trouble.  If no overflow,
			 * we might get away with just killing a task.
			 */
			if (m.status & MCI_STATUS_UC) {
				if (tolerant < 1 || m.status & MCI_STATUS_OVER)
					no_way_out = 1;
				kill_it = 1;
			}
L
Linus Torvalds 已提交
238 239 240 241 242 243 244
		}

		if (m.status & MCI_STATUS_MISCV)
			rdmsrl(MSR_IA32_MC0_MISC + i*4, m.misc);
		if (m.status & MCI_STATUS_ADDRV)
			rdmsrl(MSR_IA32_MC0_ADDR + i*4, m.addr);

245
		mce_get_rip(&m, regs);
246
		if (error_code >= 0)
L
Linus Torvalds 已提交
247
			rdtscll(m.tsc);
248 249
		if (error_code != -2)
			mce_log(&m);
L
Linus Torvalds 已提交
250 251 252 253 254 255 256 257 258

		/* Did this bank cause the exception? */
		/* Assume that the bank with uncorrectable errors did it,
		   and that there is only a single one. */
		if ((m.status & MCI_STATUS_UC) && (m.status & MCI_STATUS_EN)) {
			panicm = m;
			panicm_found = 1;
		}

259
		add_taint(TAINT_MACHINE_CHECK);
L
Linus Torvalds 已提交
260 261 262
	}

	/* Never do anything final in the polling timer */
263
	if (!regs)
L
Linus Torvalds 已提交
264 265 266 267 268 269
		goto out;

	/* If we didn't find an uncorrectable error, pick
	   the last one (shouldn't happen, just being safe). */
	if (!panicm_found)
		panicm = m;
270 271 272 273 274 275

	/*
	 * If we have decided that we just CAN'T continue, and the user
	 *  has not set tolerant to an insane level, give up and die.
	 */
	if (no_way_out && tolerant < 3)
L
Linus Torvalds 已提交
276
		mce_panic("Machine check", &panicm, mcestart);
277 278 279 280 281 282 283 284

	/*
	 * If the error seems to be unrecoverable, something should be
	 * done.  Try to kill as little as possible.  If we can kill just
	 * one task, do that.  If the user has set the tolerance very
	 * high, don't try to do anything at all.
	 */
	if (kill_it && tolerant < 3) {
L
Linus Torvalds 已提交
285 286
		int user_space = 0;

287 288 289 290 291
		/*
		 * If the EIPV bit is set, it means the saved IP is the
		 * instruction which caused the MCE.
		 */
		if (m.mcgstatus & MCG_STATUS_EIPV)
292
			user_space = panicm.ip && (panicm.cs & 3);
293 294 295 296 297

		/*
		 * If we know that the error was in user space, send a
		 * SIGBUS.  Otherwise, panic if tolerance is low.
		 *
298
		 * force_sig() takes an awful lot of locks and has a slight
299 300 301
		 * risk of deadlocking.
		 */
		if (user_space) {
302
			force_sig(SIGBUS, current);
303 304 305 306
		} else if (panic_on_oops || tolerant < 2) {
			mce_panic("Uncorrected machine check",
				&panicm, mcestart);
		}
L
Linus Torvalds 已提交
307 308
	}

309 310 311
	/* notify userspace ASAP */
	set_thread_flag(TIF_MCE_NOTIFY);

L
Linus Torvalds 已提交
312
 out:
313 314 315
	/* the last thing we do is clear state */
	for (i = 0; i < banks; i++)
		wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0);
L
Linus Torvalds 已提交
316
	wrmsrl(MSR_IA32_MCG_STATUS, 0);
317 318
 out2:
	atomic_dec(&mce_entry);
L
Linus Torvalds 已提交
319 320
}

321 322 323
#ifdef CONFIG_X86_MCE_INTEL
/***
 * mce_log_therm_throt_event - Logs the thermal throttling event to mcelog
S
Simon Arlott 已提交
324
 * @cpu: The CPU on which the event occurred.
325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347
 * @status: Event status information
 *
 * This function should be called by the thermal interrupt after the
 * event has been processed and the decision was made to log the event
 * further.
 *
 * The status parameter will be saved to the 'status' field of 'struct mce'
 * and historically has been the register value of the
 * MSR_IA32_THERMAL_STATUS (Intel) msr.
 */
void mce_log_therm_throt_event(unsigned int cpu, __u64 status)
{
	struct mce m;

	memset(&m, 0, sizeof(m));
	m.cpu = cpu;
	m.bank = MCE_THERMAL_BANK;
	m.status = status;
	rdtscll(m.tsc);
	mce_log(&m);
}
#endif /* CONFIG_X86_MCE_INTEL */

L
Linus Torvalds 已提交
348
/*
349 350 351
 * Periodic polling timer for "silent" machine check errors.  If the
 * poller finds an MCE, poll 2x faster.  When the poller finds no more
 * errors, poll 2x slower (up to check_interval seconds).
L
Linus Torvalds 已提交
352 353 354
 */

static int check_interval = 5 * 60; /* 5 minutes */
355
static int next_interval; /* in jiffies */
356 357
static void mcheck_timer(unsigned long);
static DEFINE_PER_CPU(struct timer_list, mce_timer);
L
Linus Torvalds 已提交
358

359
static void mcheck_timer(unsigned long data)
L
Linus Torvalds 已提交
360
{
361 362 363 364
	struct timer_list *t = &per_cpu(mce_timer, data);

	WARN_ON(smp_processor_id() != data);

L
Linus Torvalds 已提交
365 366 367 368
	if (mce_available(&current_cpu_data))
		do_machine_check(NULL, 0);

	/*
369 370
	 * Alert userspace if needed.  If we logged an MCE, reduce the
	 * polling interval, otherwise increase the polling interval.
L
Linus Torvalds 已提交
371
	 */
372 373 374
	if (mce_notify_user()) {
		next_interval = max(next_interval/2, HZ/100);
	} else {
375
		next_interval = min(next_interval * 2,
376
				(int)round_jiffies_relative(check_interval*HZ));
377 378
	}

379 380
	t->expires = jiffies + next_interval;
	add_timer(t);
381 382
}

383 384 385 386 387 388 389
static void mce_do_trigger(struct work_struct *work)
{
	call_usermodehelper(trigger, trigger_argv, NULL, UMH_NO_WAIT);
}

static DECLARE_WORK(mce_trigger_work, mce_do_trigger);

390
/*
391 392 393
 * Notify the user(s) about new machine check events.
 * Can be called from interrupt context, but not from machine check/NMI
 * context.
394 395 396 397 398
 */
int mce_notify_user(void)
{
	clear_thread_flag(TIF_MCE_NOTIFY);
	if (test_and_clear_bit(0, &notify_user)) {
399 400 401
		static unsigned long last_print;
		unsigned long now = jiffies;

402
		wake_up_interruptible(&mce_wait);
403 404 405 406 407 408 409 410

		/*
		 * There is no risk of missing notifications because
		 * work_pending is always cleared before the function is
		 * executed.
		 */
		if (trigger[0] && !work_pending(&mce_trigger_work))
			schedule_work(&mce_trigger_work);
411

412 413 414 415
		if (time_after_eq(now, last_print + (check_interval*HZ))) {
			last_print = now;
			printk(KERN_INFO "Machine check events logged\n");
		}
416 417

		return 1;
L
Linus Torvalds 已提交
418
	}
419 420
	return 0;
}
421

422 423 424 425 426 427 428 429 430
/* see if the idle task needs to notify userspace */
static int
mce_idle_callback(struct notifier_block *nfb, unsigned long action, void *junk)
{
	/* IDLE_END should be safe - interrupts are back on */
	if (action == IDLE_END && test_thread_flag(TIF_MCE_NOTIFY))
		mce_notify_user();

	return NOTIFY_OK;
L
Linus Torvalds 已提交
431 432
}

433 434 435
static struct notifier_block mce_idle_notifier = {
	.notifier_call = mce_idle_callback,
};
L
Linus Torvalds 已提交
436 437

static __init int periodic_mcheck_init(void)
438
{
439 440
       idle_notifier_register(&mce_idle_notifier);
       return 0;
441
}
L
Linus Torvalds 已提交
442 443
__initcall(periodic_mcheck_init);

444
/*
L
Linus Torvalds 已提交
445 446 447 448 449 450 451 452 453
 * Initialize Machine Checks for a CPU.
 */
static void mce_init(void *dummy)
{
	u64 cap;
	int i;

	rdmsrl(MSR_IA32_MCG_CAP, cap);
	banks = cap & 0xff;
454
	if (banks > MCE_EXTENDED_BANK) {
455
		banks = MCE_EXTENDED_BANK;
456 457
		printk(KERN_INFO "MCE: warning: using only %d banks\n",
		       MCE_EXTENDED_BANK);
L
Linus Torvalds 已提交
458
	}
459 460 461
	/* Use accurate RIP reporting if available. */
	if ((cap & (1<<9)) && ((cap >> 16) & 0xff) >= 9)
		rip_msr = MSR_IA32_MCG_EIP;
L
Linus Torvalds 已提交
462 463 464

	/* Log the machine checks left over from the previous reset.
	   This also clears all registers */
465
	do_machine_check(NULL, mce_bootlog ? -1 : -2);
L
Linus Torvalds 已提交
466 467 468 469 470 471 472

	set_in_cr4(X86_CR4_MCE);

	if (cap & MCG_CTL_P)
		wrmsr(MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff);

	for (i = 0; i < banks; i++) {
473 474 475 476 477
		if (i < NR_SYSFS_BANKS)
			wrmsrl(MSR_IA32_MC0_CTL+4*i, bank[i]);
		else
			wrmsrl(MSR_IA32_MC0_CTL+4*i, ~0UL);

L
Linus Torvalds 已提交
478
		wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0);
479
	}
L
Linus Torvalds 已提交
480 481 482
}

/* Add per CPU specific workarounds here */
483
static void __cpuinit mce_cpu_quirks(struct cpuinfo_x86 *c)
484
{
L
Linus Torvalds 已提交
485
	/* This should be disabled by the BIOS, but isn't always */
486 487 488 489 490 491 492 493 494
	if (c->x86_vendor == X86_VENDOR_AMD) {
		if(c->x86 == 15)
			/* disable GART TBL walk error reporting, which trips off
			   incorrectly with the IOMMU & 3ware & Cerberus. */
			clear_bit(10, &bank[4]);
		if(c->x86 <= 17 && mce_bootlog < 0)
			/* Lots of broken BIOS around that don't clear them
			   by default and leave crap in there. Don't log. */
			mce_bootlog = 0;
L
Linus Torvalds 已提交
495
	}
496

497
}
L
Linus Torvalds 已提交
498

499
static void __cpuinit mce_cpu_features(struct cpuinfo_x86 *c)
L
Linus Torvalds 已提交
500 501 502 503 504
{
	switch (c->x86_vendor) {
	case X86_VENDOR_INTEL:
		mce_intel_feature_init(c);
		break;
505 506 507
	case X86_VENDOR_AMD:
		mce_amd_feature_init(c);
		break;
L
Linus Torvalds 已提交
508 509 510 511 512
	default:
		break;
	}
}

513 514 515 516 517 518 519 520 521 522 523 524 525 526
static void mce_init_timer(void)
{
	struct timer_list *t = &__get_cpu_var(mce_timer);

	/* data race harmless because everyone sets to the same value */
	if (!next_interval)
		next_interval = check_interval * HZ;
	if (!next_interval)
		return;
	setup_timer(t, mcheck_timer, smp_processor_id());
	t->expires = round_jiffies_relative(jiffies + next_interval);
	add_timer(t);
}

527
/*
L
Linus Torvalds 已提交
528
 * Called for each booted CPU to set up machine checks.
529
 * Must be called with preempt off.
L
Linus Torvalds 已提交
530
 */
531
void __cpuinit mcheck_init(struct cpuinfo_x86 *c)
L
Linus Torvalds 已提交
532
{
533
	mce_cpu_quirks(c);
L
Linus Torvalds 已提交
534 535 536 537 538 539 540

	if (mce_dont_init ||
	    !mce_available(c))
		return;

	mce_init(NULL);
	mce_cpu_features(c);
541
	mce_init_timer();
L
Linus Torvalds 已提交
542 543 544 545 546 547
}

/*
 * Character device to read and clear the MCE log.
 */

T
Tim Hockin 已提交
548 549 550 551 552 553
static DEFINE_SPINLOCK(mce_state_lock);
static int open_count;	/* #times opened */
static int open_exclu;	/* already open exclusive? */

static int mce_open(struct inode *inode, struct file *file)
{
A
Arnd Bergmann 已提交
554
	lock_kernel();
T
Tim Hockin 已提交
555 556 557 558
	spin_lock(&mce_state_lock);

	if (open_exclu || (open_count && (file->f_flags & O_EXCL))) {
		spin_unlock(&mce_state_lock);
A
Arnd Bergmann 已提交
559
		unlock_kernel();
T
Tim Hockin 已提交
560 561 562 563 564 565 566 567
		return -EBUSY;
	}

	if (file->f_flags & O_EXCL)
		open_exclu = 1;
	open_count++;

	spin_unlock(&mce_state_lock);
A
Arnd Bergmann 已提交
568
	unlock_kernel();
T
Tim Hockin 已提交
569

570
	return nonseekable_open(inode, file);
T
Tim Hockin 已提交
571 572 573 574 575 576 577 578 579 580 581 582 583 584
}

static int mce_release(struct inode *inode, struct file *file)
{
	spin_lock(&mce_state_lock);

	open_count--;
	open_exclu = 0;

	spin_unlock(&mce_state_lock);

	return 0;
}

585 586
static void collect_tscs(void *data)
{
L
Linus Torvalds 已提交
587
	unsigned long *cpu_tsc = (unsigned long *)data;
588

L
Linus Torvalds 已提交
589
	rdtscll(cpu_tsc[smp_processor_id()]);
590
}
L
Linus Torvalds 已提交
591

592 593
static ssize_t mce_read(struct file *filp, char __user *ubuf, size_t usize,
			loff_t *off)
L
Linus Torvalds 已提交
594
{
595
	unsigned long *cpu_tsc;
596
	static DEFINE_MUTEX(mce_read_mutex);
L
Linus Torvalds 已提交
597 598 599 600
	unsigned next;
	char __user *buf = ubuf;
	int i, err;

601
	cpu_tsc = kmalloc(nr_cpu_ids * sizeof(long), GFP_KERNEL);
602 603 604
	if (!cpu_tsc)
		return -ENOMEM;

605
	mutex_lock(&mce_read_mutex);
L
Linus Torvalds 已提交
606 607 608
	next = rcu_dereference(mcelog.next);

	/* Only supports full reads right now */
609
	if (*off != 0 || usize < MCE_LOG_LEN*sizeof(struct mce)) {
610
		mutex_unlock(&mce_read_mutex);
611
		kfree(cpu_tsc);
L
Linus Torvalds 已提交
612 613 614 615
		return -EINVAL;
	}

	err = 0;
616
	for (i = 0; i < next; i++) {
617
		unsigned long start = jiffies;
618

619
		while (!mcelog.entry[i].finished) {
620
			if (time_after_eq(jiffies, start + 2)) {
621
				memset(mcelog.entry + i,0, sizeof(struct mce));
622
				goto timeout;
623 624 625
			}
			cpu_relax();
		}
L
Linus Torvalds 已提交
626 627
		smp_rmb();
		err |= copy_to_user(buf, mcelog.entry + i, sizeof(struct mce));
628
		buf += sizeof(struct mce);
629 630
 timeout:
		;
631
	}
L
Linus Torvalds 已提交
632 633 634 635

	memset(mcelog.entry, 0, next * sizeof(struct mce));
	mcelog.next = 0;

636
	synchronize_sched();
L
Linus Torvalds 已提交
637

638 639 640 641
	/*
	 * Collect entries that were still getting written before the
	 * synchronize.
	 */
642
	on_each_cpu(collect_tscs, cpu_tsc, 1);
643 644 645 646 647
	for (i = next; i < MCE_LOG_LEN; i++) {
		if (mcelog.entry[i].finished &&
		    mcelog.entry[i].tsc < cpu_tsc[mcelog.entry[i].cpu]) {
			err |= copy_to_user(buf, mcelog.entry+i,
					    sizeof(struct mce));
L
Linus Torvalds 已提交
648 649 650 651
			smp_rmb();
			buf += sizeof(struct mce);
			memset(&mcelog.entry[i], 0, sizeof(struct mce));
		}
652
	}
653
	mutex_unlock(&mce_read_mutex);
654
	kfree(cpu_tsc);
655
	return err ? -EFAULT : buf - ubuf;
L
Linus Torvalds 已提交
656 657
}

658 659 660 661 662 663 664 665
static unsigned int mce_poll(struct file *file, poll_table *wait)
{
	poll_wait(file, &mce_wait, wait);
	if (rcu_dereference(mcelog.next))
		return POLLIN | POLLRDNORM;
	return 0;
}

666
static long mce_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
L
Linus Torvalds 已提交
667 668
{
	int __user *p = (int __user *)arg;
669

L
Linus Torvalds 已提交
670
	if (!capable(CAP_SYS_ADMIN))
671
		return -EPERM;
L
Linus Torvalds 已提交
672
	switch (cmd) {
673
	case MCE_GET_RECORD_LEN:
L
Linus Torvalds 已提交
674 675
		return put_user(sizeof(struct mce), p);
	case MCE_GET_LOG_LEN:
676
		return put_user(MCE_LOG_LEN, p);
L
Linus Torvalds 已提交
677 678
	case MCE_GETCLEAR_FLAGS: {
		unsigned flags;
679 680

		do {
L
Linus Torvalds 已提交
681
			flags = mcelog.flags;
682 683
		} while (cmpxchg(&mcelog.flags, flags, 0) != flags);
		return put_user(flags, p);
L
Linus Torvalds 已提交
684 685
	}
	default:
686 687
		return -ENOTTY;
	}
L
Linus Torvalds 已提交
688 689
}

690
static const struct file_operations mce_chrdev_ops = {
T
Tim Hockin 已提交
691 692
	.open = mce_open,
	.release = mce_release,
L
Linus Torvalds 已提交
693
	.read = mce_read,
694
	.poll = mce_poll,
695
	.unlocked_ioctl = mce_ioctl,
L
Linus Torvalds 已提交
696 697 698 699 700 701 702 703
};

static struct miscdevice mce_log_device = {
	MISC_MCELOG_MINOR,
	"mcelog",
	&mce_chrdev_ops,
};

704 705
/*
 * Old style boot options parsing. Only for compatibility.
L
Linus Torvalds 已提交
706 707 708 709
 */
static int __init mcheck_disable(char *str)
{
	mce_dont_init = 1;
710
	return 1;
L
Linus Torvalds 已提交
711 712
}

S
Simon Arlott 已提交
713
/* mce=off disables machine check. Note you can re-enable it later
714
   using sysfs.
715
   mce=TOLERANCELEVEL (number, see above)
716 717
   mce=bootlog Log MCEs from before booting. Disabled by default on AMD.
   mce=nobootlog Don't log MCEs from before booting. */
L
Linus Torvalds 已提交
718 719 720 721
static int __init mcheck_enable(char *str)
{
	if (!strcmp(str, "off"))
		mce_dont_init = 1;
722 723
	else if (!strcmp(str, "bootlog") || !strcmp(str,"nobootlog"))
		mce_bootlog = str[0] == 'b';
724 725
	else if (isdigit(str[0]))
		get_option(&str, &tolerant);
L
Linus Torvalds 已提交
726
	else
727
		printk("mce= argument %s ignored. Please use /sys", str);
728
	return 1;
L
Linus Torvalds 已提交
729 730 731
}

__setup("nomce", mcheck_disable);
732
__setup("mce=", mcheck_enable);
L
Linus Torvalds 已提交
733

734
/*
L
Linus Torvalds 已提交
735
 * Sysfs support
736
 */
L
Linus Torvalds 已提交
737

738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760
/*
 * Disable machine checks on suspend and shutdown. We can't really handle
 * them later.
 */
static int mce_disable(void)
{
	int i;

	for (i = 0; i < banks; i++)
		wrmsrl(MSR_IA32_MC0_CTL + i*4, 0);
	return 0;
}

static int mce_suspend(struct sys_device *dev, pm_message_t state)
{
	return mce_disable();
}

static int mce_shutdown(struct sys_device *dev)
{
	return mce_disable();
}

761 762 763
/* On resume clear all MCE state. Don't want to see leftovers from the BIOS.
   Only one CPU is active at this time, the others get readded later using
   CPU hotplug. */
L
Linus Torvalds 已提交
764 765
static int mce_resume(struct sys_device *dev)
{
766
	mce_init(NULL);
767
	mce_cpu_features(&current_cpu_data);
L
Linus Torvalds 已提交
768 769 770
	return 0;
}

771 772 773 774 775 776 777 778
static void mce_cpu_restart(void *data)
{
	del_timer_sync(&__get_cpu_var(mce_timer));
	if (mce_available(&current_cpu_data))
		mce_init(NULL);
	mce_init_timer();
}

L
Linus Torvalds 已提交
779
/* Reinit MCEs after user configuration changes */
780 781
static void mce_restart(void)
{
782
	next_interval = check_interval * HZ;
783
	on_each_cpu(mce_cpu_restart, NULL, 1);
L
Linus Torvalds 已提交
784 785 786
}

static struct sysdev_class mce_sysclass = {
787 788
	.suspend = mce_suspend,
	.shutdown = mce_shutdown,
L
Linus Torvalds 已提交
789
	.resume = mce_resume,
790
	.name = "machinecheck",
L
Linus Torvalds 已提交
791 792
};

793
DEFINE_PER_CPU(struct sys_device, device_mce);
794
void (*threshold_cpu_callback)(unsigned long action, unsigned int cpu) __cpuinitdata;
L
Linus Torvalds 已提交
795 796 797

/* Why are there no generic functions for this? */
#define ACCESSOR(name, var, start) \
798 799 800
	static ssize_t show_ ## name(struct sys_device *s,		\
				     struct sysdev_attribute *attr,	\
				     char *buf) {			\
801 802
		return sprintf(buf, "%lx\n", (unsigned long)var);	\
	}								\
803 804 805
	static ssize_t set_ ## name(struct sys_device *s,		\
				    struct sysdev_attribute *attr,	\
				    const char *buf, size_t siz) {	\
806 807 808 809 810 811 812
		char *end;						\
		unsigned long new = simple_strtoul(buf, &end, 0);	\
		if (end == buf) return -EINVAL;				\
		var = new;						\
		start;							\
		return end-buf;						\
	}								\
L
Linus Torvalds 已提交
813 814
	static SYSDEV_ATTR(name, 0644, show_ ## name, set_ ## name);

815 816 817 818
/*
 * TBD should generate these dynamically based on number of available banks.
 * Have only 6 contol banks in /sysfs until then.
 */
L
Linus Torvalds 已提交
819 820 821 822 823
ACCESSOR(bank0ctl,bank[0],mce_restart())
ACCESSOR(bank1ctl,bank[1],mce_restart())
ACCESSOR(bank2ctl,bank[2],mce_restart())
ACCESSOR(bank3ctl,bank[3],mce_restart())
ACCESSOR(bank4ctl,bank[4],mce_restart())
824
ACCESSOR(bank5ctl,bank[5],mce_restart())
825

826 827
static ssize_t show_trigger(struct sys_device *s, struct sysdev_attribute *attr,
				char *buf)
828 829 830 831 832 833
{
	strcpy(buf, trigger);
	strcat(buf, "\n");
	return strlen(trigger) + 1;
}

834 835
static ssize_t set_trigger(struct sys_device *s, struct sysdev_attribute *attr,
				const char *buf,size_t siz)
836 837 838 839 840 841 842 843 844 845 846 847
{
	char *p;
	int len;
	strncpy(trigger, buf, sizeof(trigger));
	trigger[sizeof(trigger)-1] = 0;
	len = strlen(trigger);
	p = strchr(trigger, '\n');
	if (*p) *p = 0;
	return len;
}

static SYSDEV_ATTR(trigger, 0644, show_trigger, set_trigger);
848
static SYSDEV_INT_ATTR(tolerant, 0644, tolerant);
L
Linus Torvalds 已提交
849
ACCESSOR(check_interval,check_interval,mce_restart())
850 851 852
static struct sysdev_attribute *mce_attributes[] = {
	&attr_bank0ctl, &attr_bank1ctl, &attr_bank2ctl,
	&attr_bank3ctl, &attr_bank4ctl, &attr_bank5ctl,
853
	&attr_tolerant.attr, &attr_check_interval, &attr_trigger,
854 855
	NULL
};
L
Linus Torvalds 已提交
856

857 858
static cpumask_t mce_device_initialized = CPU_MASK_NONE;

859 860
/* Per cpu sysdev init.  All of the cpus still share the same ctl bank */
static __cpuinit int mce_create_device(unsigned int cpu)
L
Linus Torvalds 已提交
861 862
{
	int err;
863
	int i;
864

A
Andreas Herrmann 已提交
865
	if (!mce_available(&boot_cpu_data))
866 867
		return -EIO;

868
	memset(&per_cpu(device_mce, cpu).kobj, 0, sizeof(struct kobject));
869 870 871 872
	per_cpu(device_mce,cpu).id = cpu;
	per_cpu(device_mce,cpu).cls = &mce_sysclass;

	err = sysdev_register(&per_cpu(device_mce,cpu));
873 874 875 876 877 878 879 880 881
	if (err)
		return err;

	for (i = 0; mce_attributes[i]; i++) {
		err = sysdev_create_file(&per_cpu(device_mce,cpu),
					 mce_attributes[i]);
		if (err)
			goto error;
	}
882
	cpu_set(cpu, mce_device_initialized);
883

884 885 886 887 888
	return 0;
error:
	while (i--) {
		sysdev_remove_file(&per_cpu(device_mce,cpu),
				   mce_attributes[i]);
889
	}
890 891
	sysdev_unregister(&per_cpu(device_mce,cpu));

892 893 894
	return err;
}

895
static __cpuinit void mce_remove_device(unsigned int cpu)
896
{
897 898
	int i;

899 900 901
	if (!cpu_isset(cpu, mce_device_initialized))
		return;

902
	for (i = 0; mce_attributes[i]; i++)
903
		sysdev_remove_file(&per_cpu(device_mce,cpu),
904
			mce_attributes[i]);
905
	sysdev_unregister(&per_cpu(device_mce,cpu));
906
	cpu_clear(cpu, mce_device_initialized);
907 908 909
}

/* Get notified when a cpu comes on/off. Be hotplug friendly. */
910 911
static int __cpuinit mce_cpu_callback(struct notifier_block *nfb,
				      unsigned long action, void *hcpu)
912 913
{
	unsigned int cpu = (unsigned long)hcpu;
914
	struct timer_list *t = &per_cpu(mce_timer, cpu);
915 916

	switch (action) {
917 918 919
	case CPU_ONLINE:
	case CPU_ONLINE_FROZEN:
		mce_create_device(cpu);
920 921
		if (threshold_cpu_callback)
			threshold_cpu_callback(action, cpu);
922 923
		break;
	case CPU_DEAD:
924
	case CPU_DEAD_FROZEN:
925 926
		if (threshold_cpu_callback)
			threshold_cpu_callback(action, cpu);
927 928
		mce_remove_device(cpu);
		break;
929 930 931 932 933 934 935 936 937
	case CPU_DOWN_PREPARE:
	case CPU_DOWN_PREPARE_FROZEN:
		del_timer_sync(t);
		break;
	case CPU_DOWN_FAILED:
	case CPU_DOWN_FAILED_FROZEN:
		t->expires = round_jiffies_relative(jiffies + next_interval);
		add_timer_on(t, cpu);
		break;
938
	}
939
	return NOTIFY_OK;
940 941
}

942
static struct notifier_block mce_cpu_notifier __cpuinitdata = {
943 944 945 946 947 948 949 950
	.notifier_call = mce_cpu_callback,
};

static __init int mce_init_device(void)
{
	int err;
	int i = 0;

L
Linus Torvalds 已提交
951 952 953
	if (!mce_available(&boot_cpu_data))
		return -EIO;
	err = sysdev_class_register(&mce_sysclass);
954 955
	if (err)
		return err;
956 957

	for_each_online_cpu(i) {
958 959 960
		err = mce_create_device(i);
		if (err)
			return err;
961 962
	}

963
	register_hotcpu_notifier(&mce_cpu_notifier);
L
Linus Torvalds 已提交
964 965 966
	misc_register(&mce_log_device);
	return err;
}
967

L
Linus Torvalds 已提交
968
device_initcall(mce_init_device);