mce_amd.c 17.0 KB
Newer Older
1
/*
2
 *  (c) 2005-2012 Advanced Micro Devices, Inc.
3 4 5 6 7 8
 *  Your use of this code is subject to the terms and conditions of the
 *  GNU general public license version 2. See "COPYING" or
 *  http://www.gnu.org/licenses/gpl.html
 *
 *  Written by Jacob Shin - AMD, Inc.
 *
9
 *  Maintained by: Borislav Petkov <bp@alien8.de>
10
 *
11 12
 *  April 2006
 *     - added support for AMD Family 0x10 processors
13 14
 *  May 2012
 *     - major scrubbing
15
 *
16
 *  All MC4_MISCi registers are shared between multi-cores
17 18 19
 */
#include <linux/interrupt.h>
#include <linux/notifier.h>
I
Ingo Molnar 已提交
20
#include <linux/kobject.h>
21
#include <linux/percpu.h>
I
Ingo Molnar 已提交
22 23
#include <linux/errno.h>
#include <linux/sched.h>
24
#include <linux/sysfs.h>
25
#include <linux/slab.h>
I
Ingo Molnar 已提交
26 27 28 29
#include <linux/init.h>
#include <linux/cpu.h>
#include <linux/smp.h>

30
#include <asm/amd_nb.h>
31
#include <asm/apic.h>
I
Ingo Molnar 已提交
32
#include <asm/idle.h>
33 34 35
#include <asm/mce.h>
#include <asm/msr.h>

J
Jacob Shin 已提交
36 37 38 39
#define NR_BLOCKS         9
#define THRESHOLD_MAX     0xFFF
#define INT_TYPE_APIC     0x00020000
#define MASK_VALID_HI     0x80000000
40 41
#define MASK_CNTP_HI      0x40000000
#define MASK_LOCKED_HI    0x20000000
J
Jacob Shin 已提交
42 43 44 45
#define MASK_LVTOFF_HI    0x00F00000
#define MASK_COUNT_EN_HI  0x00080000
#define MASK_INT_TYPE_HI  0x00060000
#define MASK_OVERFLOW_HI  0x00010000
46
#define MASK_ERR_COUNT_HI 0x00000FFF
47 48
#define MASK_BLKPTR_LO    0xFF000000
#define MCG_XBLK_ADDR     0xC0000400
49

50 51 52 53 54 55 56 57 58
static const char * const th_names[] = {
	"load_store",
	"insn_fetch",
	"combined_unit",
	"",
	"northbridge",
	"execution_unit",
};

59
static DEFINE_PER_CPU(struct threshold_bank **, threshold_banks);
60 61
static DEFINE_PER_CPU(unsigned char, bank_map);	/* see which banks are on */

62 63
static void amd_threshold_interrupt(void);

64 65 66 67
/*
 * CPU Initialization
 */

68
struct thresh_restart {
I
Ingo Molnar 已提交
69 70
	struct threshold_block	*b;
	int			reset;
71 72
	int			set_lvt_off;
	int			lvt_off;
I
Ingo Molnar 已提交
73
	u16			old_limit;
74 75
};

76 77 78 79 80 81
static inline bool is_shared_bank(int bank)
{
	/* Bank 4 is for northbridge reporting and is thus shared */
	return (bank == 4);
}

82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101
static const char * const bank4_names(struct threshold_block *b)
{
	switch (b->address) {
	/* MSR4_MISC0 */
	case 0x00000413:
		return "dram";

	case 0xc0000408:
		return "ht_links";

	case 0xc0000409:
		return "l3_cache";

	default:
		WARN(1, "Funny MSR: 0x%08x\n", b->address);
		return "";
	}
};


102 103 104 105 106 107 108 109 110 111 112 113 114 115 116
static bool lvt_interrupt_supported(unsigned int bank, u32 msr_high_bits)
{
	/*
	 * bank 4 supports APIC LVT interrupts implicitly since forever.
	 */
	if (bank == 4)
		return true;

	/*
	 * IntP: interrupt present; if this bit is set, the thresholding
	 * bank can generate APIC LVT interrupts
	 */
	return msr_high_bits & BIT(28);
}

117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137
static int lvt_off_valid(struct threshold_block *b, int apic, u32 lo, u32 hi)
{
	int msr = (hi & MASK_LVTOFF_HI) >> 20;

	if (apic < 0) {
		pr_err(FW_BUG "cpu %d, failed to setup threshold interrupt "
		       "for bank %d, block %d (MSR%08X=0x%x%08x)\n", b->cpu,
		       b->bank, b->block, b->address, hi, lo);
		return 0;
	}

	if (apic != msr) {
		pr_err(FW_BUG "cpu %d, invalid threshold interrupt offset %d "
		       "for bank %d, block %d (MSR%08X=0x%x%08x)\n",
		       b->cpu, apic, b->bank, b->block, b->address, hi, lo);
		return 0;
	}

	return 1;
};

138 139 140 141
/*
 * Called via smp_call_function_single(), must be called with correct
 * cpu affinity.
 */
142
static void threshold_restart_bank(void *_tr)
143
{
144
	struct thresh_restart *tr = _tr;
145
	u32 hi, lo;
146

147
	rdmsr(tr->b->address, lo, hi);
148

149
	if (tr->b->threshold_limit < (hi & THRESHOLD_MAX))
150
		tr->reset = 1;	/* limit cannot be lower than err count */
151

152
	if (tr->reset) {		/* reset err count and overflow bit */
153 154
		hi =
		    (hi & ~(MASK_ERR_COUNT_HI | MASK_OVERFLOW_HI)) |
155 156
		    (THRESHOLD_MAX - tr->b->threshold_limit);
	} else if (tr->old_limit) {	/* change limit w/o reset */
157
		int new_count = (hi & THRESHOLD_MAX) +
158
		    (tr->old_limit - tr->b->threshold_limit);
I
Ingo Molnar 已提交
159

160
		hi = (hi & ~MASK_ERR_COUNT_HI) |
161 162 163
		    (new_count & THRESHOLD_MAX);
	}

164 165 166 167 168 169
	/* clear IntType */
	hi &= ~MASK_INT_TYPE_HI;

	if (!tr->b->interrupt_capable)
		goto done;

170
	if (tr->set_lvt_off) {
171 172 173 174 175
		if (lvt_off_valid(tr->b, tr->lvt_off, lo, hi)) {
			/* set new lvt offset */
			hi &= ~MASK_LVTOFF_HI;
			hi |= tr->lvt_off << 20;
		}
176 177
	}

178 179 180 181
	if (tr->b->interrupt_enable)
		hi |= INT_TYPE_APIC;

 done:
182

183 184
	hi |= MASK_COUNT_EN_HI;
	wrmsr(tr->b->address, lo, hi);
185 186
}

187 188 189 190 191 192 193 194 195 196 197 198
static void mce_threshold_block_init(struct threshold_block *b, int offset)
{
	struct thresh_restart tr = {
		.b			= b,
		.set_lvt_off		= 1,
		.lvt_off		= offset,
	};

	b->threshold_limit		= THRESHOLD_MAX;
	threshold_restart_bank(&tr);
};

199 200 201 202 203 204 205 206 207
static int setup_APIC_mce(int reserved, int new)
{
	if (reserved < 0 && !setup_APIC_eilvt(new, THRESHOLD_APIC_VECTOR,
					      APIC_EILVT_MSG_FIX, 0))
		return new;

	return reserved;
}

208
/* cpu init entry point, called from mce.c with preempt off */
209
void mce_amd_feature_init(struct cpuinfo_x86 *c)
210
{
211
	struct threshold_block b;
212
	unsigned int cpu = smp_processor_id();
213
	u32 low = 0, high = 0, address = 0;
I
Ingo Molnar 已提交
214
	unsigned int bank, block;
215
	int offset = -1;
216

217
	for (bank = 0; bank < mca_cfg.banks; ++bank) {
218 219 220
		for (block = 0; block < NR_BLOCKS; ++block) {
			if (block == 0)
				address = MSR_IA32_MC0_MISC + bank * 4;
221 222 223 224
			else if (block == 1) {
				address = (low & MASK_BLKPTR_LO) >> 21;
				if (!address)
					break;
225

226
				address += MCG_XBLK_ADDR;
I
Ingo Molnar 已提交
227
			} else
228 229 230
				++address;

			if (rdmsr_safe(address, &low, &high))
231
				break;
232

233 234
			if (!(high & MASK_VALID_HI))
				continue;
235

236 237
			if (!(high & MASK_CNTP_HI)  ||
			     (high & MASK_LOCKED_HI))
238 239 240 241
				continue;

			if (!block)
				per_cpu(bank_map, cpu) |= (1 << bank);
242

243
			memset(&b, 0, sizeof(b));
244 245 246 247 248 249 250 251 252 253
			b.cpu			= cpu;
			b.bank			= bank;
			b.block			= block;
			b.address		= address;
			b.interrupt_capable	= lvt_interrupt_supported(bank, high);

			if (b.interrupt_capable) {
				int new = (high & MASK_LVTOFF_HI) >> 20;
				offset  = setup_APIC_mce(offset, new);
			}
254

255
			mce_threshold_block_init(&b, offset);
256
			mce_threshold_vector = amd_threshold_interrupt;
257
		}
258 259 260 261 262 263 264 265 266 267 268 269
	}
}

/*
 * APIC Interrupt Handler
 */

/*
 * threshold interrupt handler will service THRESHOLD_APIC_VECTOR.
 * the interrupt goes off when error_count reaches threshold_limit.
 * the handler will simply log mcelog w/ software defined bank number.
 */
270
static void amd_threshold_interrupt(void)
271
{
I
Ingo Molnar 已提交
272
	u32 low = 0, high = 0, address = 0;
273
	unsigned int bank, block;
274 275
	struct mce m;

276
	mce_setup(&m);
277 278

	/* assume first bank caused it */
279
	for (bank = 0; bank < mca_cfg.banks; ++bank) {
280 281
		if (!(per_cpu(bank_map, m.cpu) & (1 << bank)))
			continue;
282
		for (block = 0; block < NR_BLOCKS; ++block) {
I
Ingo Molnar 已提交
283
			if (block == 0) {
284
				address = MSR_IA32_MC0_MISC + bank * 4;
I
Ingo Molnar 已提交
285
			} else if (block == 1) {
286 287 288 289
				address = (low & MASK_BLKPTR_LO) >> 21;
				if (!address)
					break;
				address += MCG_XBLK_ADDR;
I
Ingo Molnar 已提交
290
			} else {
291
				++address;
I
Ingo Molnar 已提交
292
			}
293 294

			if (rdmsr_safe(address, &low, &high))
295
				break;
296 297 298 299 300 301 302 303

			if (!(high & MASK_VALID_HI)) {
				if (block)
					continue;
				else
					break;
			}

304 305
			if (!(high & MASK_CNTP_HI)  ||
			     (high & MASK_LOCKED_HI))
306 307
				continue;

I
Ingo Molnar 已提交
308 309 310 311
			/*
			 * Log the machine check that caused the threshold
			 * event.
			 */
312
			machine_check_poll(MCP_TIMESTAMP,
313
					this_cpu_ptr(&mce_poll_banks));
314

315 316 317 318 319 320 321 322
			if (high & MASK_OVERFLOW_HI) {
				rdmsrl(address, m.misc);
				rdmsrl(MSR_IA32_MC0_STATUS + bank * 4,
				       m.status);
				m.bank = K8_MCE_THRESHOLD_BASE
				       + bank * NR_BLOCKS
				       + block;
				mce_log(&m);
323
				return;
324
			}
325 326 327 328 329 330 331 332 333
		}
	}
}

/*
 * Sysfs Interface
 */

struct threshold_attr {
J
Jacob Shin 已提交
334
	struct attribute attr;
I
Ingo Molnar 已提交
335 336
	ssize_t (*show) (struct threshold_block *, char *);
	ssize_t (*store) (struct threshold_block *, const char *, size_t count);
337 338
};

I
Ingo Molnar 已提交
339 340 341
#define SHOW_FIELDS(name)						\
static ssize_t show_ ## name(struct threshold_block *b, char *buf)	\
{									\
342
	return sprintf(buf, "%lu\n", (unsigned long) b->name);		\
J
Jacob Shin 已提交
343
}
344 345 346
SHOW_FIELDS(interrupt_enable)
SHOW_FIELDS(threshold_limit)

I
Ingo Molnar 已提交
347
static ssize_t
H
Hidetoshi Seto 已提交
348
store_interrupt_enable(struct threshold_block *b, const char *buf, size_t size)
349
{
350
	struct thresh_restart tr;
I
Ingo Molnar 已提交
351 352
	unsigned long new;

353 354 355
	if (!b->interrupt_capable)
		return -EINVAL;

356
	if (kstrtoul(buf, 0, &new) < 0)
357
		return -EINVAL;
I
Ingo Molnar 已提交
358

359 360
	b->interrupt_enable = !!new;

361
	memset(&tr, 0, sizeof(tr));
I
Ingo Molnar 已提交
362 363
	tr.b		= b;

364
	smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1);
365

H
Hidetoshi Seto 已提交
366
	return size;
367 368
}

I
Ingo Molnar 已提交
369
static ssize_t
H
Hidetoshi Seto 已提交
370
store_threshold_limit(struct threshold_block *b, const char *buf, size_t size)
371
{
372
	struct thresh_restart tr;
I
Ingo Molnar 已提交
373 374
	unsigned long new;

375
	if (kstrtoul(buf, 0, &new) < 0)
376
		return -EINVAL;
I
Ingo Molnar 已提交
377

378 379 380 381
	if (new > THRESHOLD_MAX)
		new = THRESHOLD_MAX;
	if (new < 1)
		new = 1;
I
Ingo Molnar 已提交
382

383
	memset(&tr, 0, sizeof(tr));
384
	tr.old_limit = b->threshold_limit;
385
	b->threshold_limit = new;
386
	tr.b = b;
387

388
	smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1);
389

H
Hidetoshi Seto 已提交
390
	return size;
391 392
}

393 394
static ssize_t show_error_count(struct threshold_block *b, char *buf)
{
395 396 397
	u32 lo, hi;

	rdmsr_on_cpu(b->cpu, b->address, &lo, &hi);
398

399 400
	return sprintf(buf, "%u\n", ((hi & THRESHOLD_MAX) -
				     (THRESHOLD_MAX - b->threshold_limit)));
401 402
}

403 404 405 406
static struct threshold_attr error_count = {
	.attr = {.name = __stringify(error_count), .mode = 0444 },
	.show = show_error_count,
};
407

408 409 410 411 412
#define RW_ATTR(val)							\
static struct threshold_attr val = {					\
	.attr	= {.name = __stringify(val), .mode = 0644 },		\
	.show	= show_## val,						\
	.store	= store_## val,						\
413 414
};

J
Jacob Shin 已提交
415 416
RW_ATTR(interrupt_enable);
RW_ATTR(threshold_limit);
417 418 419 420

static struct attribute *default_attrs[] = {
	&threshold_limit.attr,
	&error_count.attr,
421 422
	NULL,	/* possibly interrupt_enable if supported, see below */
	NULL,
423 424
};

I
Ingo Molnar 已提交
425 426
#define to_block(k)	container_of(k, struct threshold_block, kobj)
#define to_attr(a)	container_of(a, struct threshold_attr, attr)
427 428 429

static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
{
430
	struct threshold_block *b = to_block(kobj);
431 432
	struct threshold_attr *a = to_attr(attr);
	ssize_t ret;
I
Ingo Molnar 已提交
433

434
	ret = a->show ? a->show(b, buf) : -EIO;
I
Ingo Molnar 已提交
435

436 437 438 439 440 441
	return ret;
}

static ssize_t store(struct kobject *kobj, struct attribute *attr,
		     const char *buf, size_t count)
{
442
	struct threshold_block *b = to_block(kobj);
443 444
	struct threshold_attr *a = to_attr(attr);
	ssize_t ret;
I
Ingo Molnar 已提交
445

446
	ret = a->store ? a->store(b, buf, count) : -EIO;
I
Ingo Molnar 已提交
447

448 449 450
	return ret;
}

451
static const struct sysfs_ops threshold_ops = {
I
Ingo Molnar 已提交
452 453
	.show			= show,
	.store			= store,
454 455 456
};

static struct kobj_type threshold_ktype = {
I
Ingo Molnar 已提交
457 458
	.sysfs_ops		= &threshold_ops,
	.default_attrs		= default_attrs,
459 460
};

461 462
static int allocate_threshold_blocks(unsigned int cpu, unsigned int bank,
				     unsigned int block, u32 address)
463 464
{
	struct threshold_block *b = NULL;
I
Ingo Molnar 已提交
465 466
	u32 low, high;
	int err;
467

468
	if ((bank >= mca_cfg.banks) || (block >= NR_BLOCKS))
469 470
		return 0;

471
	if (rdmsr_safe_on_cpu(cpu, address, &low, &high))
472
		return 0;
473 474 475 476 477 478 479 480

	if (!(high & MASK_VALID_HI)) {
		if (block)
			goto recurse;
		else
			return 0;
	}

481 482
	if (!(high & MASK_CNTP_HI)  ||
	     (high & MASK_LOCKED_HI))
483 484 485 486 487 488
		goto recurse;

	b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
	if (!b)
		return -ENOMEM;

I
Ingo Molnar 已提交
489 490 491 492 493
	b->block		= block;
	b->bank			= bank;
	b->cpu			= cpu;
	b->address		= address;
	b->interrupt_enable	= 0;
494
	b->interrupt_capable	= lvt_interrupt_supported(bank, high);
I
Ingo Molnar 已提交
495
	b->threshold_limit	= THRESHOLD_MAX;
496

497 498 499 500 501
	if (b->interrupt_capable)
		threshold_ktype.default_attrs[2] = &interrupt_enable.attr;
	else
		threshold_ktype.default_attrs[2] = NULL;

502 503
	INIT_LIST_HEAD(&b->miscj);

I
Ingo Molnar 已提交
504
	if (per_cpu(threshold_banks, cpu)[bank]->blocks) {
505 506
		list_add(&b->miscj,
			 &per_cpu(threshold_banks, cpu)[bank]->blocks->miscj);
I
Ingo Molnar 已提交
507
	} else {
508
		per_cpu(threshold_banks, cpu)[bank]->blocks = b;
I
Ingo Molnar 已提交
509
	}
510

511 512
	err = kobject_init_and_add(&b->kobj, &threshold_ktype,
				   per_cpu(threshold_banks, cpu)[bank]->kobj,
513
				   (bank == 4 ? bank4_names(b) : th_names[bank]));
514 515 516 517 518 519 520 521
	if (err)
		goto out_free;
recurse:
	if (!block) {
		address = (low & MASK_BLKPTR_LO) >> 21;
		if (!address)
			return 0;
		address += MCG_XBLK_ADDR;
I
Ingo Molnar 已提交
522
	} else {
523
		++address;
I
Ingo Molnar 已提交
524
	}
525 526 527 528 529

	err = allocate_threshold_blocks(cpu, bank, ++block, address);
	if (err)
		goto out_free;

530 531
	if (b)
		kobject_uevent(&b->kobj, KOBJ_ADD);
532

533 534 535 536
	return err;

out_free:
	if (b) {
537
		kobject_put(&b->kobj);
538
		list_del(&b->miscj);
539 540 541 542 543
		kfree(b);
	}
	return err;
}

544
static int __threshold_add_blocks(struct threshold_bank *b)
545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567
{
	struct list_head *head = &b->blocks->miscj;
	struct threshold_block *pos = NULL;
	struct threshold_block *tmp = NULL;
	int err = 0;

	err = kobject_add(&b->blocks->kobj, b->kobj, b->blocks->kobj.name);
	if (err)
		return err;

	list_for_each_entry_safe(pos, tmp, head, miscj) {

		err = kobject_add(&pos->kobj, b->kobj, pos->kobj.name);
		if (err) {
			list_for_each_entry_safe_reverse(pos, tmp, head, miscj)
				kobject_del(&pos->kobj);

			return err;
		}
	}
	return err;
}

568
static int threshold_create_bank(unsigned int cpu, unsigned int bank)
569
{
570
	struct device *dev = per_cpu(mce_device, cpu);
571
	struct amd_northbridge *nb = NULL;
572
	struct threshold_bank *b = NULL;
573
	const char *name = th_names[bank];
574
	int err = 0;
575

576
	if (is_shared_bank(bank)) {
577 578 579
		nb = node_to_amd_nb(amd_get_nb_id(cpu));

		/* threshold descriptor already initialized on this node? */
580
		if (nb && nb->bank4) {
581 582 583 584 585 586 587 588 589 590 591 592 593 594 595
			/* yes, use it */
			b = nb->bank4;
			err = kobject_add(b->kobj, &dev->kobj, name);
			if (err)
				goto out;

			per_cpu(threshold_banks, cpu)[bank] = b;
			atomic_inc(&b->cpus);

			err = __threshold_add_blocks(b);

			goto out;
		}
	}

596
	b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
597 598 599 600 601
	if (!b) {
		err = -ENOMEM;
		goto out;
	}

602
	b->kobj = kobject_create_and_add(name, &dev->kobj);
603 604
	if (!b->kobj) {
		err = -EINVAL;
605
		goto out_free;
606
	}
607

608
	per_cpu(threshold_banks, cpu)[bank] = b;
609

610
	if (is_shared_bank(bank)) {
611 612 613
		atomic_set(&b->cpus, 1);

		/* nb is already initialized, see above */
614 615 616 617
		if (nb) {
			WARN_ON(nb->bank4);
			nb->bank4 = b;
		}
618 619
	}

620 621
	err = allocate_threshold_blocks(cpu, bank, 0,
					MSR_IA32_MC0_MISC + bank * 4);
622 623
	if (!err)
		goto out;
624

625
 out_free:
626
	kfree(b);
627 628

 out:
629 630 631 632
	return err;
}

/* create dir/files for all valid threshold banks */
633
static int threshold_create_device(unsigned int cpu)
634
{
J
Jacob Shin 已提交
635
	unsigned int bank;
636
	struct threshold_bank **bp;
637 638
	int err = 0;

639 640 641 642 643 644 645 646
	bp = kzalloc(sizeof(struct threshold_bank *) * mca_cfg.banks,
		     GFP_KERNEL);
	if (!bp)
		return -ENOMEM;

	per_cpu(threshold_banks, cpu) = bp;

	for (bank = 0; bank < mca_cfg.banks; ++bank) {
647
		if (!(per_cpu(bank_map, cpu) & (1 << bank)))
648 649 650
			continue;
		err = threshold_create_bank(cpu, bank);
		if (err)
651
			return err;
652
	}
653

654 655 656
	return err;
}

657
static void deallocate_threshold_block(unsigned int cpu,
658 659 660 661 662 663 664 665 666 667
						 unsigned int bank)
{
	struct threshold_block *pos = NULL;
	struct threshold_block *tmp = NULL;
	struct threshold_bank *head = per_cpu(threshold_banks, cpu)[bank];

	if (!head)
		return;

	list_for_each_entry_safe(pos, tmp, &head->blocks->miscj, miscj) {
668
		kobject_put(&pos->kobj);
669 670 671 672 673 674 675 676
		list_del(&pos->miscj);
		kfree(pos);
	}

	kfree(per_cpu(threshold_banks, cpu)[bank]->blocks);
	per_cpu(threshold_banks, cpu)[bank]->blocks = NULL;
}

677 678 679 680 681 682 683 684 685 686 687
static void __threshold_remove_blocks(struct threshold_bank *b)
{
	struct threshold_block *pos = NULL;
	struct threshold_block *tmp = NULL;

	kobject_del(b->kobj);

	list_for_each_entry_safe(pos, tmp, &b->blocks->miscj, miscj)
		kobject_del(&pos->kobj);
}

688
static void threshold_remove_bank(unsigned int cpu, int bank)
689
{
690
	struct amd_northbridge *nb;
691 692 693 694 695
	struct threshold_bank *b;

	b = per_cpu(threshold_banks, cpu)[bank];
	if (!b)
		return;
696

697 698 699
	if (!b->blocks)
		goto free_out;

700
	if (is_shared_bank(bank)) {
701 702 703 704 705 706 707 708 709 710 711 712 713 714
		if (!atomic_dec_and_test(&b->cpus)) {
			__threshold_remove_blocks(b);
			per_cpu(threshold_banks, cpu)[bank] = NULL;
			return;
		} else {
			/*
			 * the last CPU on this node using the shared bank is
			 * going away, remove that bank now.
			 */
			nb = node_to_amd_nb(amd_get_nb_id(cpu));
			nb->bank4 = NULL;
		}
	}

715 716 717
	deallocate_threshold_block(cpu, bank);

free_out:
718
	kobject_del(b->kobj);
719
	kobject_put(b->kobj);
720 721
	kfree(b);
	per_cpu(threshold_banks, cpu)[bank] = NULL;
722 723
}

724
static void threshold_remove_device(unsigned int cpu)
725
{
J
Jacob Shin 已提交
726
	unsigned int bank;
727

728
	for (bank = 0; bank < mca_cfg.banks; ++bank) {
729
		if (!(per_cpu(bank_map, cpu) & (1 << bank)))
730 731 732
			continue;
		threshold_remove_bank(cpu, bank);
	}
733
	kfree(per_cpu(threshold_banks, cpu));
734 735 736
}

/* get notified when a cpu comes on/off */
737
static void
I
Ingo Molnar 已提交
738
amd_64_threshold_cpu_callback(unsigned long action, unsigned int cpu)
739 740 741
{
	switch (action) {
	case CPU_ONLINE:
742
	case CPU_ONLINE_FROZEN:
743 744 745
		threshold_create_device(cpu);
		break;
	case CPU_DEAD:
746
	case CPU_DEAD_FROZEN:
747 748 749 750 751 752 753 754 755
		threshold_remove_device(cpu);
		break;
	default:
		break;
	}
}

static __init int threshold_init_device(void)
{
J
Jacob Shin 已提交
756
	unsigned lcpu = 0;
757 758 759

	/* to hit CPUs online before the notifier is up */
	for_each_online_cpu(lcpu) {
760
		int err = threshold_create_device(lcpu);
I
Ingo Molnar 已提交
761

762
		if (err)
763
			return err;
764
	}
765
	threshold_cpu_callback = amd_64_threshold_cpu_callback;
I
Ingo Molnar 已提交
766

767
	return 0;
768
}
769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789
/*
 * there are 3 funcs which need to be _initcalled in a logic sequence:
 * 1. xen_late_init_mcelog
 * 2. mcheck_init_device
 * 3. threshold_init_device
 *
 * xen_late_init_mcelog must register xen_mce_chrdev_device before
 * native mce_chrdev_device registration if running under xen platform;
 *
 * mcheck_init_device should be inited before threshold_init_device to
 * initialize mce_device, otherwise a NULL ptr dereference will cause panic.
 *
 * so we use following _initcalls
 * 1. device_initcall(xen_late_init_mcelog);
 * 2. device_initcall_sync(mcheck_init_device);
 * 3. late_initcall(threshold_init_device);
 *
 * when running under xen, the initcall order is 1,2,3;
 * on baremetal, we skip 1 and we do only 2 and 3.
 */
late_initcall(threshold_init_device);