?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'akpm'
[linux-2.6/next.git] / arch / x86 / kernel / cpu / mcheck / mce_amd.c
blobf5474218cffede0c71793dfed39f1866cb3713c3
1 /*
2 * (c) 2005, 2006 Advanced Micro Devices, Inc.
3 * Your use of this code is subject to the terms and conditions of the
4 * GNU general public license version 2. See "COPYING" or
5 * http://www.gnu.org/licenses/gpl.html
6 *
7 * Written by Jacob Shin - AMD, Inc.
8 *
9 * Support : jacob.shin@amd.com
10 *
11 * April 2006
12 * - added support for AMD Family 0x10 processors
13 *
14 * All MC4_MISCi registers are shared between multi-cores
15 */
16 #include <linux/interrupt.h>
17 #include <linux/notifier.h>
18 #include <linux/kobject.h>
19 #include <linux/percpu.h>
20 #include <linux/sysdev.h>
21 #include <linux/errno.h>
22 #include <linux/sched.h>
23 #include <linux/sysfs.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/cpu.h>
27 #include <linux/smp.h>
28
29 #include <asm/apic.h>
30 #include <asm/idle.h>
31 #include <asm/mce.h>
32 #include <asm/msr.h>
33
34 #define NR_BANKS 6
35 #define NR_BLOCKS 9
36 #define THRESHOLD_MAX 0xFFF
37 #define INT_TYPE_APIC 0x00020000
38 #define MASK_VALID_HI 0x80000000
39 #define MASK_CNTP_HI 0x40000000
40 #define MASK_LOCKED_HI 0x20000000
41 #define MASK_LVTOFF_HI 0x00F00000
42 #define MASK_COUNT_EN_HI 0x00080000
43 #define MASK_INT_TYPE_HI 0x00060000
44 #define MASK_OVERFLOW_HI 0x00010000
45 #define MASK_ERR_COUNT_HI 0x00000FFF
46 #define MASK_BLKPTR_LO 0xFF000000
47 #define MCG_XBLK_ADDR 0xC0000400
48
49 struct threshold_block {
50 unsigned int block;
51 unsigned int bank;
52 unsigned int cpu;
53 u32 address;
54 u16 interrupt_enable;
55 u16 threshold_limit;
56 struct kobject kobj;
57 struct list_head miscj;
58 };
59
60 struct threshold_bank {
61 struct kobject *kobj;
62 struct threshold_block *blocks;
63 cpumask_var_t cpus;
64 };
65 static DEFINE_PER_CPU(struct threshold_bank * [NR_BANKS], threshold_banks);
66
67 #ifdef CONFIG_SMP
68 static unsigned char shared_bank[NR_BANKS] = {
69 0, 0, 0, 0, 1
70 };
71 #endif
72
73 static DEFINE_PER_CPU(unsigned char, bank_map); /* see which banks are on */
74
75 static void amd_threshold_interrupt(void);
76
77 /*
78 * CPU Initialization
79 */
80
81 struct thresh_restart {
82 struct threshold_block *b;
83 int reset;
84 int set_lvt_off;
85 int lvt_off;
86 u16 old_limit;
87 };
88
89 static int lvt_off_valid(struct threshold_block *b, int apic, u32 lo, u32 hi)
90 {
91 int msr = (hi & MASK_LVTOFF_HI) >> 20;
92
93 if (apic < 0) {
94 pr_err(FW_BUG "cpu %d, failed to setup threshold interrupt "
95 "for bank %d, block %d (MSR%08X=0x%x%08x)\n", b->cpu,
96 b->bank, b->block, b->address, hi, lo);
97 return 0;
98 }
99
100 if (apic != msr) {
101 pr_err(FW_BUG "cpu %d, invalid threshold interrupt offset %d "
102 "for bank %d, block %d (MSR%08X=0x%x%08x)\n",
103 b->cpu, apic, b->bank, b->block, b->address, hi, lo);
104 return 0;
105 }
106
107 return 1;
108 };
109
110 /* must be called with correct cpu affinity */
111 /* Called via smp_call_function_single() */
112 static void threshold_restart_bank(void *_tr)
113 {
114 struct thresh_restart *tr = _tr;
115 u32 hi, lo;
116
117 rdmsr(tr->b->address, lo, hi);
118
119 if (tr->b->threshold_limit < (hi & THRESHOLD_MAX))
120 tr->reset = 1; /* limit cannot be lower than err count */
121
122 if (tr->reset) { /* reset err count and overflow bit */
123 hi =
124 (hi & ~(MASK_ERR_COUNT_HI | MASK_OVERFLOW_HI)) |
125 (THRESHOLD_MAX - tr->b->threshold_limit);
126 } else if (tr->old_limit) { /* change limit w/o reset */
127 int new_count = (hi & THRESHOLD_MAX) +
128 (tr->old_limit - tr->b->threshold_limit);
129
130 hi = (hi & ~MASK_ERR_COUNT_HI) |
131 (new_count & THRESHOLD_MAX);
132 }
133
134 if (tr->set_lvt_off) {
135 if (lvt_off_valid(tr->b, tr->lvt_off, lo, hi)) {
136 /* set new lvt offset */
137 hi &= ~MASK_LVTOFF_HI;
138 hi |= tr->lvt_off << 20;
139 }
140 }
141
142 tr->b->interrupt_enable ?
143 (hi = (hi & ~MASK_INT_TYPE_HI) | INT_TYPE_APIC) :
144 (hi &= ~MASK_INT_TYPE_HI);
145
146 hi |= MASK_COUNT_EN_HI;
147 wrmsr(tr->b->address, lo, hi);
148 }
149
150 static void mce_threshold_block_init(struct threshold_block *b, int offset)
151 {
152 struct thresh_restart tr = {
153 .b = b,
154 .set_lvt_off = 1,
155 .lvt_off = offset,
156 };
157
158 b->threshold_limit = THRESHOLD_MAX;
159 threshold_restart_bank(&tr);
160 };
161
162 static int setup_APIC_mce(int reserved, int new)
163 {
164 if (reserved < 0 && !setup_APIC_eilvt(new, THRESHOLD_APIC_VECTOR,
165 APIC_EILVT_MSG_FIX, 0))
166 return new;
167
168 return reserved;
169 }
170
171 /* cpu init entry point, called from mce.c with preempt off */
172 void mce_amd_feature_init(struct cpuinfo_x86 *c)
173 {
174 struct threshold_block b;
175 unsigned int cpu = smp_processor_id();
176 u32 low = 0, high = 0, address = 0;
177 unsigned int bank, block;
178 int offset = -1;
179
180 for (bank = 0; bank < NR_BANKS; ++bank) {
181 for (block = 0; block < NR_BLOCKS; ++block) {
182 if (block == 0)
183 address = MSR_IA32_MC0_MISC + bank * 4;
184 else if (block == 1) {
185 address = (low & MASK_BLKPTR_LO) >> 21;
186 if (!address)
187 break;
188
189 address += MCG_XBLK_ADDR;
190 } else
191 ++address;
192
193 if (rdmsr_safe(address, &low, &high))
194 break;
195
196 if (!(high & MASK_VALID_HI))
197 continue;
198
199 if (!(high & MASK_CNTP_HI) ||
200 (high & MASK_LOCKED_HI))
201 continue;
202
203 if (!block)
204 per_cpu(bank_map, cpu) |= (1 << bank);
205 #ifdef CONFIG_SMP
206 if (shared_bank[bank] && c->cpu_core_id)
207 break;
208 #endif
209 offset = setup_APIC_mce(offset,
210 (high & MASK_LVTOFF_HI) >> 20);
211
212 memset(&b, 0, sizeof(b));
213 b.cpu = cpu;
214 b.bank = bank;
215 b.block = block;
216 b.address = address;
217
218 mce_threshold_block_init(&b, offset);
219 mce_threshold_vector = amd_threshold_interrupt;
220 }
221 }
222 }
223
224 /*
225 * APIC Interrupt Handler
226 */
227
228 /*
229 * threshold interrupt handler will service THRESHOLD_APIC_VECTOR.
230 * the interrupt goes off when error_count reaches threshold_limit.
231 * the handler will simply log mcelog w/ software defined bank number.
232 */
233 static void amd_threshold_interrupt(void)
234 {
235 u32 low = 0, high = 0, address = 0;
236 unsigned int bank, block;
237 struct mce m;
238
239 mce_setup(&m);
240
241 /* assume first bank caused it */
242 for (bank = 0; bank < NR_BANKS; ++bank) {
243 if (!(per_cpu(bank_map, m.cpu) & (1 << bank)))
244 continue;
245 for (block = 0; block < NR_BLOCKS; ++block) {
246 if (block == 0) {
247 address = MSR_IA32_MC0_MISC + bank * 4;
248 } else if (block == 1) {
249 address = (low & MASK_BLKPTR_LO) >> 21;
250 if (!address)
251 break;
252 address += MCG_XBLK_ADDR;
253 } else {
254 ++address;
255 }
256
257 if (rdmsr_safe(address, &low, &high))
258 break;
259
260 if (!(high & MASK_VALID_HI)) {
261 if (block)
262 continue;
263 else
264 break;
265 }
266
267 if (!(high & MASK_CNTP_HI) ||
268 (high & MASK_LOCKED_HI))
269 continue;
270
271 /*
272 * Log the machine check that caused the threshold
273 * event.
274 */
275 machine_check_poll(MCP_TIMESTAMP,
276 &__get_cpu_var(mce_poll_banks));
277
278 if (high & MASK_OVERFLOW_HI) {
279 rdmsrl(address, m.misc);
280 rdmsrl(MSR_IA32_MC0_STATUS + bank * 4,
281 m.status);
282 m.bank = K8_MCE_THRESHOLD_BASE
283 + bank * NR_BLOCKS
284 + block;
285 mce_log(&m);
286 return;
287 }
288 }
289 }
290 }
291
292 /*
293 * Sysfs Interface
294 */
295
296 struct threshold_attr {
297 struct attribute attr;
298 ssize_t (*show) (struct threshold_block *, char *);
299 ssize_t (*store) (struct threshold_block *, const char *, size_t count);
300 };
301
302 #define SHOW_FIELDS(name) \
303 static ssize_t show_ ## name(struct threshold_block *b, char *buf) \
304 { \
305 return sprintf(buf, "%lx\n", (unsigned long) b->name); \
306 }
307 SHOW_FIELDS(interrupt_enable)
308 SHOW_FIELDS(threshold_limit)
309
310 static ssize_t
311 store_interrupt_enable(struct threshold_block *b, const char *buf, size_t size)
312 {
313 struct thresh_restart tr;
314 unsigned long new;
315
316 if (strict_strtoul(buf, 0, &new) < 0)
317 return -EINVAL;
318
319 b->interrupt_enable = !!new;
320
321 memset(&tr, 0, sizeof(tr));
322 tr.b = b;
323
324 smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1);
325
326 return size;
327 }
328
329 static ssize_t
330 store_threshold_limit(struct threshold_block *b, const char *buf, size_t size)
331 {
332 struct thresh_restart tr;
333 unsigned long new;
334
335 if (strict_strtoul(buf, 0, &new) < 0)
336 return -EINVAL;
337
338 if (new > THRESHOLD_MAX)
339 new = THRESHOLD_MAX;
340 if (new < 1)
341 new = 1;
342
343 memset(&tr, 0, sizeof(tr));
344 tr.old_limit = b->threshold_limit;
345 b->threshold_limit = new;
346 tr.b = b;
347
348 smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1);
349
350 return size;
351 }
352
353 struct threshold_block_cross_cpu {
354 struct threshold_block *tb;
355 long retval;
356 };
357
358 static void local_error_count_handler(void *_tbcc)
359 {
360 struct threshold_block_cross_cpu *tbcc = _tbcc;
361 struct threshold_block *b = tbcc->tb;
362 u32 low, high;
363
364 rdmsr(b->address, low, high);
365 tbcc->retval = (high & 0xFFF) - (THRESHOLD_MAX - b->threshold_limit);
366 }
367
368 static ssize_t show_error_count(struct threshold_block *b, char *buf)
369 {
370 struct threshold_block_cross_cpu tbcc = { .tb = b, };
371
372 smp_call_function_single(b->cpu, local_error_count_handler, &tbcc, 1);
373 return sprintf(buf, "%lx\n", tbcc.retval);
374 }
375
376 static ssize_t store_error_count(struct threshold_block *b,
377 const char *buf, size_t count)
378 {
379 struct thresh_restart tr = { .b = b, .reset = 1, .old_limit = 0 };
380
381 smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1);
382 return 1;
383 }
384
385 #define RW_ATTR(val) \
386 static struct threshold_attr val = { \
387 .attr = {.name = __stringify(val), .mode = 0644 }, \
388 .show = show_## val, \
389 .store = store_## val, \
390 };
391
392 RW_ATTR(interrupt_enable);
393 RW_ATTR(threshold_limit);
394 RW_ATTR(error_count);
395
396 static struct attribute *default_attrs[] = {
397 &interrupt_enable.attr,
398 &threshold_limit.attr,
399 &error_count.attr,
400 NULL
401 };
402
403 #define to_block(k) container_of(k, struct threshold_block, kobj)
404 #define to_attr(a) container_of(a, struct threshold_attr, attr)
405
406 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
407 {
408 struct threshold_block *b = to_block(kobj);
409 struct threshold_attr *a = to_attr(attr);
410 ssize_t ret;
411
412 ret = a->show ? a->show(b, buf) : -EIO;
413
414 return ret;
415 }
416
417 static ssize_t store(struct kobject *kobj, struct attribute *attr,
418 const char *buf, size_t count)
419 {
420 struct threshold_block *b = to_block(kobj);
421 struct threshold_attr *a = to_attr(attr);
422 ssize_t ret;
423
424 ret = a->store ? a->store(b, buf, count) : -EIO;
425
426 return ret;
427 }
428
429 static const struct sysfs_ops threshold_ops = {
430 .show = show,
431 .store = store,
432 };
433
434 static struct kobj_type threshold_ktype = {
435 .sysfs_ops = &threshold_ops,
436 .default_attrs = default_attrs,
437 };
438
439 static __cpuinit int allocate_threshold_blocks(unsigned int cpu,
440 unsigned int bank,
441 unsigned int block,
442 u32 address)
443 {
444 struct threshold_block *b = NULL;
445 u32 low, high;
446 int err;
447
448 if ((bank >= NR_BANKS) || (block >= NR_BLOCKS))
449 return 0;
450
451 if (rdmsr_safe_on_cpu(cpu, address, &low, &high))
452 return 0;
453
454 if (!(high & MASK_VALID_HI)) {
455 if (block)
456 goto recurse;
457 else
458 return 0;
459 }
460
461 if (!(high & MASK_CNTP_HI) ||
462 (high & MASK_LOCKED_HI))
463 goto recurse;
464
465 b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
466 if (!b)
467 return -ENOMEM;
468
469 b->block = block;
470 b->bank = bank;
471 b->cpu = cpu;
472 b->address = address;
473 b->interrupt_enable = 0;
474 b->threshold_limit = THRESHOLD_MAX;
475
476 INIT_LIST_HEAD(&b->miscj);
477
478 if (per_cpu(threshold_banks, cpu)[bank]->blocks) {
479 list_add(&b->miscj,
480 &per_cpu(threshold_banks, cpu)[bank]->blocks->miscj);
481 } else {
482 per_cpu(threshold_banks, cpu)[bank]->blocks = b;
483 }
484
485 err = kobject_init_and_add(&b->kobj, &threshold_ktype,
486 per_cpu(threshold_banks, cpu)[bank]->kobj,
487 "misc%i", block);
488 if (err)
489 goto out_free;
490 recurse:
491 if (!block) {
492 address = (low & MASK_BLKPTR_LO) >> 21;
493 if (!address)
494 return 0;
495 address += MCG_XBLK_ADDR;
496 } else {
497 ++address;
498 }
499
500 err = allocate_threshold_blocks(cpu, bank, ++block, address);
501 if (err)
502 goto out_free;
503
504 if (b)
505 kobject_uevent(&b->kobj, KOBJ_ADD);
506
507 return err;
508
509 out_free:
510 if (b) {
511 kobject_put(&b->kobj);
512 list_del(&b->miscj);
513 kfree(b);
514 }
515 return err;
516 }
517
518 static __cpuinit long
519 local_allocate_threshold_blocks(int cpu, unsigned int bank)
520 {
521 return allocate_threshold_blocks(cpu, bank, 0,
522 MSR_IA32_MC0_MISC + bank * 4);
523 }
524
525 /* symlinks sibling shared banks to first core. first core owns dir/files. */
526 static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank)
527 {
528 int i, err = 0;
529 struct threshold_bank *b = NULL;
530 char name[32];
531
532 sprintf(name, "threshold_bank%i", bank);
533
534 #ifdef CONFIG_SMP
535 if (cpu_data(cpu).cpu_core_id && shared_bank[bank]) { /* symlink */
536 i = cpumask_first(cpu_llc_shared_mask(cpu));
537
538 /* first core not up yet */
539 if (cpu_data(i).cpu_core_id)
540 goto out;
541
542 /* already linked */
543 if (per_cpu(threshold_banks, cpu)[bank])
544 goto out;
545
546 b = per_cpu(threshold_banks, i)[bank];
547
548 if (!b)
549 goto out;
550
551 err = sysfs_create_link(&per_cpu(mce_sysdev, cpu).kobj,
552 b->kobj, name);
553 if (err)
554 goto out;
555
556 cpumask_copy(b->cpus, cpu_llc_shared_mask(cpu));
557 per_cpu(threshold_banks, cpu)[bank] = b;
558
559 goto out;
560 }
561 #endif
562
563 b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
564 if (!b) {
565 err = -ENOMEM;
566 goto out;
567 }
568 if (!zalloc_cpumask_var(&b->cpus, GFP_KERNEL)) {
569 kfree(b);
570 err = -ENOMEM;
571 goto out;
572 }
573
574 b->kobj = kobject_create_and_add(name, &per_cpu(mce_sysdev, cpu).kobj);
575 if (!b->kobj)
576 goto out_free;
577
578 #ifndef CONFIG_SMP
579 cpumask_setall(b->cpus);
580 #else
581 cpumask_set_cpu(cpu, b->cpus);
582 #endif
583
584 per_cpu(threshold_banks, cpu)[bank] = b;
585
586 err = local_allocate_threshold_blocks(cpu, bank);
587 if (err)
588 goto out_free;
589
590 for_each_cpu(i, b->cpus) {
591 if (i == cpu)
592 continue;
593
594 err = sysfs_create_link(&per_cpu(mce_sysdev, i).kobj,
595 b->kobj, name);
596 if (err)
597 goto out;
598
599 per_cpu(threshold_banks, i)[bank] = b;
600 }
601
602 goto out;
603
604 out_free:
605 per_cpu(threshold_banks, cpu)[bank] = NULL;
606 free_cpumask_var(b->cpus);
607 kfree(b);
608 out:
609 return err;
610 }
611
612 /* create dir/files for all valid threshold banks */
613 static __cpuinit int threshold_create_device(unsigned int cpu)
614 {
615 unsigned int bank;
616 int err = 0;
617
618 for (bank = 0; bank < NR_BANKS; ++bank) {
619 if (!(per_cpu(bank_map, cpu) & (1 << bank)))
620 continue;
621 err = threshold_create_bank(cpu, bank);
622 if (err)
623 return err;
624 }
625
626 return err;
627 }
628
629 /*
630 * let's be hotplug friendly.
631 * in case of multiple core processors, the first core always takes ownership
632 * of shared sysfs dir/files, and rest of the cores will be symlinked to it.
633 */
634
635 static void deallocate_threshold_block(unsigned int cpu,
636 unsigned int bank)
637 {
638 struct threshold_block *pos = NULL;
639 struct threshold_block *tmp = NULL;
640 struct threshold_bank *head = per_cpu(threshold_banks, cpu)[bank];
641
642 if (!head)
643 return;
644
645 list_for_each_entry_safe(pos, tmp, &head->blocks->miscj, miscj) {
646 kobject_put(&pos->kobj);
647 list_del(&pos->miscj);
648 kfree(pos);
649 }
650
651 kfree(per_cpu(threshold_banks, cpu)[bank]->blocks);
652 per_cpu(threshold_banks, cpu)[bank]->blocks = NULL;
653 }
654
655 static void threshold_remove_bank(unsigned int cpu, int bank)
656 {
657 struct threshold_bank *b;
658 char name[32];
659 int i = 0;
660
661 b = per_cpu(threshold_banks, cpu)[bank];
662 if (!b)
663 return;
664 if (!b->blocks)
665 goto free_out;
666
667 sprintf(name, "threshold_bank%i", bank);
668
669 #ifdef CONFIG_SMP
670 /* sibling symlink */
671 if (shared_bank[bank] && b->blocks->cpu != cpu) {
672 sysfs_remove_link(&per_cpu(mce_sysdev, cpu).kobj, name);
673 per_cpu(threshold_banks, cpu)[bank] = NULL;
674
675 return;
676 }
677 #endif
678
679 /* remove all sibling symlinks before unregistering */
680 for_each_cpu(i, b->cpus) {
681 if (i == cpu)
682 continue;
683
684 sysfs_remove_link(&per_cpu(mce_sysdev, i).kobj, name);
685 per_cpu(threshold_banks, i)[bank] = NULL;
686 }
687
688 deallocate_threshold_block(cpu, bank);
689
690 free_out:
691 kobject_del(b->kobj);
692 kobject_put(b->kobj);
693 free_cpumask_var(b->cpus);
694 kfree(b);
695 per_cpu(threshold_banks, cpu)[bank] = NULL;
696 }
697
698 static void threshold_remove_device(unsigned int cpu)
699 {
700 unsigned int bank;
701
702 for (bank = 0; bank < NR_BANKS; ++bank) {
703 if (!(per_cpu(bank_map, cpu) & (1 << bank)))
704 continue;
705 threshold_remove_bank(cpu, bank);
706 }
707 }
708
709 /* get notified when a cpu comes on/off */
710 static void __cpuinit
711 amd_64_threshold_cpu_callback(unsigned long action, unsigned int cpu)
712 {
713 switch (action) {
714 case CPU_ONLINE:
715 case CPU_ONLINE_FROZEN:
716 threshold_create_device(cpu);
717 break;
718 case CPU_DEAD:
719 case CPU_DEAD_FROZEN:
720 threshold_remove_device(cpu);
721 break;
722 default:
723 break;
724 }
725 }
726
727 static __init int threshold_init_device(void)
728 {
729 unsigned lcpu = 0;
730
731 /* to hit CPUs online before the notifier is up */
732 for_each_online_cpu(lcpu) {
733 int err = threshold_create_device(lcpu);
734
735 if (err)
736 return err;
737 }
738 threshold_cpu_callback = amd_64_threshold_cpu_callback;
739
740 return 0;
741 }
742 device_initcall(threshold_init_device);
linux-next