arch/x86/kernel/mce_amd_64.c

   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
  17 #include <linux/cpu.h>
  18 #include <linux/errno.h>
  19 #include <linux/init.h>
  20 #include <linux/interrupt.h>
  21 #include <linux/kobject.h>
  22 #include <linux/notifier.h>
  23 #include <linux/sched.h>
  24 #include <linux/smp.h>
  25 #include <linux/sysdev.h>
  26 #include <linux/sysfs.h>
  27 #include <asm/apic.h>
  28 #include <asm/mce.h>
  29 #include <asm/msr.h>
  30 #include <asm/percpu.h>
  31 #include <asm/idle.h>
  32
  33 #define PFX               "mce_threshold: "
  34 #define VERSION           "version 1.1.1"
  35 #define NR_BANKS          6
  36 #define NR_BLOCKS         9
  37 #define THRESHOLD_MAX     0xFFF
  38 #define INT_TYPE_APIC     0x00020000
  39 #define MASK_VALID_HI     0x80000000
  40 #define MASK_CNTP_HI      0x40000000
  41 #define MASK_LOCKED_HI    0x20000000
  42 #define MASK_LVTOFF_HI    0x00F00000
  43 #define MASK_COUNT_EN_HI  0x00080000
  44 #define MASK_INT_TYPE_HI  0x00060000
  45 #define MASK_OVERFLOW_HI  0x00010000
  46 #define MASK_ERR_COUNT_HI 0x00000FFF
  47 #define MASK_BLKPTR_LO    0xFF000000
  48 #define MCG_XBLK_ADDR     0xC0000400
  49
  50 struct threshold_block {
  51         unsigned int block;
  52         unsigned int bank;
  53         unsigned int cpu;
  54         u32 address;
  55         u16 interrupt_enable;
  56         u16 threshold_limit;
  57         struct kobject kobj;
  58         struct list_head miscj;
  59 };
  60
  61 /* defaults used early on boot */
  62 static struct threshold_block threshold_defaults = {
  63         .interrupt_enable = 0,
  64         .threshold_limit = THRESHOLD_MAX,
  65 };
  66
  67 struct threshold_bank {
  68         struct kobject kobj;
  69         struct threshold_block *blocks;
  70         cpumask_t cpus;
  71 };
  72 static DEFINE_PER_CPU(struct threshold_bank *, threshold_banks[NR_BANKS]);
  73
  74 #ifdef CONFIG_SMP
  75 static unsigned char shared_bank[NR_BANKS] = {
  76         0, 0, 0, 0, 1
  77 };
  78 #endif
  79
  80 static DEFINE_PER_CPU(unsigned char, bank_map); /* see which banks are on */
  81
  82 /*
  83  * CPU Initialization
  84  */
  85
  86 /* must be called with correct cpu affinity */
  87 static void threshold_restart_bank(struct threshold_block *b,
  88                                    int reset, u16 old_limit)
  89 {
  90         u32 mci_misc_hi, mci_misc_lo;
  91
  92         rdmsr(b->address, mci_misc_lo, mci_misc_hi);
  93
  94         if (b->threshold_limit < (mci_misc_hi & THRESHOLD_MAX))
  95                 reset = 1;      /* limit cannot be lower than err count */
  96
  97         if (reset) {            /* reset err count and overflow bit */
  98                 mci_misc_hi =
  99                     (mci_misc_hi & ~(MASK_ERR_COUNT_HI | MASK_OVERFLOW_HI)) |
 100                     (THRESHOLD_MAX - b->threshold_limit);
 101         } else if (old_limit) { /* change limit w/o reset */
 102                 int new_count = (mci_misc_hi & THRESHOLD_MAX) +
 103                     (old_limit - b->threshold_limit);
 104                 mci_misc_hi = (mci_misc_hi & ~MASK_ERR_COUNT_HI) |
 105                     (new_count & THRESHOLD_MAX);
 106         }
 107
 108         b->interrupt_enable ?
 109             (mci_misc_hi = (mci_misc_hi & ~MASK_INT_TYPE_HI) | INT_TYPE_APIC) :
 110             (mci_misc_hi &= ~MASK_INT_TYPE_HI);
 111
 112         mci_misc_hi |= MASK_COUNT_EN_HI;
 113         wrmsr(b->address, mci_misc_lo, mci_misc_hi);
 114 }
 115
 116 /* cpu init entry point, called from mce.c with preempt off */
 117 void __cpuinit mce_amd_feature_init(struct cpuinfo_x86 *c)
 118 {
 119         unsigned int bank, block;
 120         unsigned int cpu = smp_processor_id();
 121         u32 low = 0, high = 0, address = 0;
 122
 123         for (bank = 0; bank < NR_BANKS; ++bank) {
 124                 for (block = 0; block < NR_BLOCKS; ++block) {
 125                         if (block == 0)
 126                                 address = MSR_IA32_MC0_MISC + bank * 4;
 127                         else if (block == 1) {
 128                                 address = (low & MASK_BLKPTR_LO) >> 21;
 129                                 if (!address)
 130                                         break;
 131                                 address += MCG_XBLK_ADDR;
 132                         }
 133                         else
 134                                 ++address;
 135
 136                         if (rdmsr_safe(address, &low, &high))
 137                                 break;
 138
 139                         if (!(high & MASK_VALID_HI)) {
 140                                 if (block)
 141                                         continue;
 142                                 else
 143                                         break;
 144                         }
 145
 146                         if (!(high & MASK_CNTP_HI)  ||
 147                              (high & MASK_LOCKED_HI))
 148                                 continue;
 149
 150                         if (!block)
 151                                 per_cpu(bank_map, cpu) |= (1 << bank);
 152 #ifdef CONFIG_SMP
 153                         if (shared_bank[bank] && c->cpu_core_id)
 154                                 break;
 155 #endif
 156                         high &= ~MASK_LVTOFF_HI;
 157                         high |= K8_APIC_EXT_LVT_ENTRY_THRESHOLD << 20;
 158                         wrmsr(address, low, high);
 159
 160                         setup_APIC_extended_lvt(K8_APIC_EXT_LVT_ENTRY_THRESHOLD,
 161                                                 THRESHOLD_APIC_VECTOR,
 162                                                 K8_APIC_EXT_INT_MSG_FIX, 0);
 163
 164                         threshold_defaults.address = address;
 165                         threshold_restart_bank(&threshold_defaults, 0, 0);
 166                 }
 167         }
 168 }
 169
 170 /*
 171  * APIC Interrupt Handler
 172  */
 173
 174 /*
 175  * threshold interrupt handler will service THRESHOLD_APIC_VECTOR.
 176  * the interrupt goes off when error_count reaches threshold_limit.
 177  * the handler will simply log mcelog w/ software defined bank number.
 178  */
 179 asmlinkage void mce_threshold_interrupt(void)
 180 {
 181         unsigned int bank, block;
 182         struct mce m;
 183         u32 low = 0, high = 0, address = 0;
 184
 185         ack_APIC_irq();
 186         exit_idle();
 187         irq_enter();
 188
 189         memset(&m, 0, sizeof(m));
 190         rdtscll(m.tsc);
 191         m.cpu = smp_processor_id();
 192
 193         /* assume first bank caused it */
 194         for (bank = 0; bank < NR_BANKS; ++bank) {
 195                 if (!(per_cpu(bank_map, m.cpu) & (1 << bank)))
 196                         continue;
 197                 for (block = 0; block < NR_BLOCKS; ++block) {
 198                         if (block == 0)
 199                                 address = MSR_IA32_MC0_MISC + bank * 4;
 200                         else if (block == 1) {
 201                                 address = (low & MASK_BLKPTR_LO) >> 21;
 202                                 if (!address)
 203                                         break;
 204                                 address += MCG_XBLK_ADDR;
 205                         }
 206                         else
 207                                 ++address;
 208
 209                         if (rdmsr_safe(address, &low, &high))
 210                                 break;
 211
 212                         if (!(high & MASK_VALID_HI)) {
 213                                 if (block)
 214                                         continue;
 215                                 else
 216                                         break;
 217                         }
 218
 219                         if (!(high & MASK_CNTP_HI)  ||
 220                              (high & MASK_LOCKED_HI))
 221                                 continue;
 222
 223                         /* Log the machine check that caused the threshold
 224                            event. */
 225                         do_machine_check(NULL, 0);
 226
 227                         if (high & MASK_OVERFLOW_HI) {
 228                                 rdmsrl(address, m.misc);
 229                                 rdmsrl(MSR_IA32_MC0_STATUS + bank * 4,
 230                                        m.status);
 231                                 m.bank = K8_MCE_THRESHOLD_BASE
 232                                        + bank * NR_BLOCKS
 233                                        + block;
 234                                 mce_log(&m);
 235                                 goto out;
 236                         }
 237                 }
 238         }
 239 out:
 240         irq_exit();
 241 }
 242
 243 /*
 244  * Sysfs Interface
 245  */
 246
 247 struct threshold_attr {
 248         struct attribute attr;
 249         ssize_t(*show) (struct threshold_block *, char *);
 250         ssize_t(*store) (struct threshold_block *, const char *, size_t count);
 251 };
 252
 253 static cpumask_t affinity_set(unsigned int cpu)
 254 {
 255         cpumask_t oldmask = current->cpus_allowed;
 256         cpumask_t newmask = CPU_MASK_NONE;
 257         cpu_set(cpu, newmask);
 258         set_cpus_allowed(current, newmask);
 259         return oldmask;
 260 }
 261
 262 static void affinity_restore(cpumask_t oldmask)
 263 {
 264         set_cpus_allowed(current, oldmask);
 265 }
 266
 267 #define SHOW_FIELDS(name)                                           \
 268 static ssize_t show_ ## name(struct threshold_block * b, char *buf) \
 269 {                                                                   \
 270         return sprintf(buf, "%lx\n", (unsigned long) b->name);      \
 271 }
 272 SHOW_FIELDS(interrupt_enable)
 273 SHOW_FIELDS(threshold_limit)
 274
 275 static ssize_t store_interrupt_enable(struct threshold_block *b,
 276                                       const char *buf, size_t count)
 277 {
 278         char *end;
 279         cpumask_t oldmask;
 280         unsigned long new = simple_strtoul(buf, &end, 0);
 281         if (end == buf)
 282                 return -EINVAL;
 283         b->interrupt_enable = !!new;
 284
 285         oldmask = affinity_set(b->cpu);
 286         threshold_restart_bank(b, 0, 0);
 287         affinity_restore(oldmask);
 288
 289         return end - buf;
 290 }
 291
 292 static ssize_t store_threshold_limit(struct threshold_block *b,
 293                                      const char *buf, size_t count)
 294 {
 295         char *end;
 296         cpumask_t oldmask;
 297         u16 old;
 298         unsigned long new = simple_strtoul(buf, &end, 0);
 299         if (end == buf)
 300                 return -EINVAL;
 301         if (new > THRESHOLD_MAX)
 302                 new = THRESHOLD_MAX;
 303         if (new < 1)
 304                 new = 1;
 305         old = b->threshold_limit;
 306         b->threshold_limit = new;
 307
 308         oldmask = affinity_set(b->cpu);
 309         threshold_restart_bank(b, 0, old);
 310         affinity_restore(oldmask);
 311
 312         return end - buf;
 313 }
 314
 315 static ssize_t show_error_count(struct threshold_block *b, char *buf)
 316 {
 317         u32 high, low;
 318         cpumask_t oldmask;
 319         oldmask = affinity_set(b->cpu);
 320         rdmsr(b->address, low, high);
 321         affinity_restore(oldmask);
 322         return sprintf(buf, "%x\n",
 323                        (high & 0xFFF) - (THRESHOLD_MAX - b->threshold_limit));
 324 }
 325
 326 static ssize_t store_error_count(struct threshold_block *b,
 327                                  const char *buf, size_t count)
 328 {
 329         cpumask_t oldmask;
 330         oldmask = affinity_set(b->cpu);
 331         threshold_restart_bank(b, 1, 0);
 332         affinity_restore(oldmask);
 333         return 1;
 334 }
 335
 336 #define THRESHOLD_ATTR(_name,_mode,_show,_store) {            \
 337         .attr = {.name = __stringify(_name), .mode = _mode }, \
 338         .show = _show,                                        \
 339         .store = _store,                                      \
 340 };
 341
 342 #define RW_ATTR(name)                                           \
 343 static struct threshold_attr name =                             \
 344         THRESHOLD_ATTR(name, 0644, show_## name, store_## name)
 345
 346 RW_ATTR(interrupt_enable);
 347 RW_ATTR(threshold_limit);
 348 RW_ATTR(error_count);
 349
 350 static struct attribute *default_attrs[] = {
 351         &interrupt_enable.attr,
 352         &threshold_limit.attr,
 353         &error_count.attr,
 354         NULL
 355 };
 356
 357 #define to_block(k) container_of(k, struct threshold_block, kobj)
 358 #define to_attr(a) container_of(a, struct threshold_attr, attr)
 359
 360 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
 361 {
 362         struct threshold_block *b = to_block(kobj);
 363         struct threshold_attr *a = to_attr(attr);
 364         ssize_t ret;
 365         ret = a->show ? a->show(b, buf) : -EIO;
 366         return ret;
 367 }
 368
 369 static ssize_t store(struct kobject *kobj, struct attribute *attr,
 370                      const char *buf, size_t count)
 371 {
 372         struct threshold_block *b = to_block(kobj);
 373         struct threshold_attr *a = to_attr(attr);
 374         ssize_t ret;
 375         ret = a->store ? a->store(b, buf, count) : -EIO;
 376         return ret;
 377 }
 378
 379 static struct sysfs_ops threshold_ops = {
 380         .show = show,
 381         .store = store,
 382 };
 383
 384 static struct kobj_type threshold_ktype = {
 385         .sysfs_ops = &threshold_ops,
 386         .default_attrs = default_attrs,
 387 };
 388
 389 static __cpuinit int allocate_threshold_blocks(unsigned int cpu,
 390                                                unsigned int bank,
 391                                                unsigned int block,
 392                                                u32 address)
 393 {
 394         int err;
 395         u32 low, high;
 396         struct threshold_block *b = NULL;
 397
 398         if ((bank >= NR_BANKS) || (block >= NR_BLOCKS))
 399                 return 0;
 400
 401         if (rdmsr_safe(address, &low, &high))
 402                 return 0;
 403
 404         if (!(high & MASK_VALID_HI)) {
 405                 if (block)
 406                         goto recurse;
 407                 else
 408                         return 0;
 409         }
 410
 411         if (!(high & MASK_CNTP_HI)  ||
 412              (high & MASK_LOCKED_HI))
 413                 goto recurse;
 414
 415         b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
 416         if (!b)
 417                 return -ENOMEM;
 418
 419         b->block = block;
 420         b->bank = bank;
 421         b->cpu = cpu;
 422         b->address = address;
 423         b->interrupt_enable = 0;
 424         b->threshold_limit = THRESHOLD_MAX;
 425
 426         INIT_LIST_HEAD(&b->miscj);
 427
 428         if (per_cpu(threshold_banks, cpu)[bank]->blocks)
 429                 list_add(&b->miscj,
 430                          &per_cpu(threshold_banks, cpu)[bank]->blocks->miscj);
 431         else
 432                 per_cpu(threshold_banks, cpu)[bank]->blocks = b;
 433
 434         kobject_set_name(&b->kobj, "misc%i", block);
 435         b->kobj.parent = &per_cpu(threshold_banks, cpu)[bank]->kobj;
 436         b->kobj.ktype = &threshold_ktype;
 437         err = kobject_register(&b->kobj);
 438         if (err)
 439                 goto out_free;
 440 recurse:
 441         if (!block) {
 442                 address = (low & MASK_BLKPTR_LO) >> 21;
 443                 if (!address)
 444                         return 0;
 445                 address += MCG_XBLK_ADDR;
 446         } else
 447                 ++address;
 448
 449         err = allocate_threshold_blocks(cpu, bank, ++block, address);
 450         if (err)
 451                 goto out_free;
 452
 453         return err;
 454
 455 out_free:
 456         if (b) {
 457                 kobject_unregister(&b->kobj);
 458                 kfree(b);
 459         }
 460         return err;
 461 }
 462
 463 /* symlinks sibling shared banks to first core.  first core owns dir/files. */
 464 static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank)
 465 {
 466         int i, err = 0;
 467         struct threshold_bank *b = NULL;
 468         cpumask_t oldmask = CPU_MASK_NONE;
 469         char name[32];
 470
 471         sprintf(name, "threshold_bank%i", bank);
 472
 473 #ifdef CONFIG_SMP
 474         if (cpu_data[cpu].cpu_core_id && shared_bank[bank]) {   /* symlink */
 475                 i = first_cpu(cpu_core_map[cpu]);
 476
 477                 /* first core not up yet */
 478                 if (cpu_data[i].cpu_core_id)
 479                         goto out;
 480
 481                 /* already linked */
 482                 if (per_cpu(threshold_banks, cpu)[bank])
 483                         goto out;
 484
 485                 b = per_cpu(threshold_banks, i)[bank];
 486
 487                 if (!b)
 488                         goto out;
 489
 490                 err = sysfs_create_link(&per_cpu(device_mce, cpu).kobj,
 491                                         &b->kobj, name);
 492                 if (err)
 493                         goto out;
 494
 495                 b->cpus = cpu_core_map[cpu];
 496                 per_cpu(threshold_banks, cpu)[bank] = b;
 497                 goto out;
 498         }
 499 #endif
 500
 501         b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
 502         if (!b) {
 503                 err = -ENOMEM;
 504                 goto out;
 505         }
 506
 507         kobject_set_name(&b->kobj, "threshold_bank%i", bank);
 508         b->kobj.parent = &per_cpu(device_mce, cpu).kobj;
 509 #ifndef CONFIG_SMP
 510         b->cpus = CPU_MASK_ALL;
 511 #else
 512         b->cpus = cpu_core_map[cpu];
 513 #endif
 514         err = kobject_register(&b->kobj);
 515         if (err)
 516                 goto out_free;
 517
 518         per_cpu(threshold_banks, cpu)[bank] = b;
 519
 520         oldmask = affinity_set(cpu);
 521         err = allocate_threshold_blocks(cpu, bank, 0,
 522                                         MSR_IA32_MC0_MISC + bank * 4);
 523         affinity_restore(oldmask);
 524
 525         if (err)
 526                 goto out_free;
 527
 528         for_each_cpu_mask(i, b->cpus) {
 529                 if (i == cpu)
 530                         continue;
 531
 532                 err = sysfs_create_link(&per_cpu(device_mce, i).kobj,
 533                                         &b->kobj, name);
 534                 if (err)
 535                         goto out;
 536
 537                 per_cpu(threshold_banks, i)[bank] = b;
 538         }
 539
 540         goto out;
 541
 542 out_free:
 543         per_cpu(threshold_banks, cpu)[bank] = NULL;
 544         kfree(b);
 545 out:
 546         return err;
 547 }
 548
 549 /* create dir/files for all valid threshold banks */
 550 static __cpuinit int threshold_create_device(unsigned int cpu)
 551 {
 552         unsigned int bank;
 553         int err = 0;
 554
 555         for (bank = 0; bank < NR_BANKS; ++bank) {
 556                 if (!(per_cpu(bank_map, cpu) & 1 << bank))
 557                         continue;
 558                 err = threshold_create_bank(cpu, bank);
 559                 if (err)
 560                         goto out;
 561         }
 562 out:
 563         return err;
 564 }
 565
 566 /*
 567  * let's be hotplug friendly.
 568  * in case of multiple core processors, the first core always takes ownership
 569  *   of shared sysfs dir/files, and rest of the cores will be symlinked to it.
 570  */
 571
 572 static void deallocate_threshold_block(unsigned int cpu,
 573                                                  unsigned int bank)
 574 {
 575         struct threshold_block *pos = NULL;
 576         struct threshold_block *tmp = NULL;
 577         struct threshold_bank *head = per_cpu(threshold_banks, cpu)[bank];
 578
 579         if (!head)
 580                 return;
 581
 582         list_for_each_entry_safe(pos, tmp, &head->blocks->miscj, miscj) {
 583                 kobject_unregister(&pos->kobj);
 584                 list_del(&pos->miscj);
 585                 kfree(pos);
 586         }
 587
 588         kfree(per_cpu(threshold_banks, cpu)[bank]->blocks);
 589         per_cpu(threshold_banks, cpu)[bank]->blocks = NULL;
 590 }
 591
 592 static void threshold_remove_bank(unsigned int cpu, int bank)
 593 {
 594         int i = 0;
 595         struct threshold_bank *b;
 596         char name[32];
 597
 598         b = per_cpu(threshold_banks, cpu)[bank];
 599
 600         if (!b)
 601                 return;
 602
 603         if (!b->blocks)
 604                 goto free_out;
 605
 606         sprintf(name, "threshold_bank%i", bank);
 607
 608 #ifdef CONFIG_SMP
 609         /* sibling symlink */
 610         if (shared_bank[bank] && b->blocks->cpu != cpu) {
 611                 sysfs_remove_link(&per_cpu(device_mce, cpu).kobj, name);
 612                 per_cpu(threshold_banks, cpu)[bank] = NULL;
 613                 return;
 614         }
 615 #endif
 616
 617         /* remove all sibling symlinks before unregistering */
 618         for_each_cpu_mask(i, b->cpus) {
 619                 if (i == cpu)
 620                         continue;
 621
 622                 sysfs_remove_link(&per_cpu(device_mce, i).kobj, name);
 623                 per_cpu(threshold_banks, i)[bank] = NULL;
 624         }
 625
 626         deallocate_threshold_block(cpu, bank);
 627
 628 free_out:
 629         kobject_unregister(&b->kobj);
 630         kfree(b);
 631         per_cpu(threshold_banks, cpu)[bank] = NULL;
 632 }
 633
 634 static void threshold_remove_device(unsigned int cpu)
 635 {
 636         unsigned int bank;
 637
 638         for (bank = 0; bank < NR_BANKS; ++bank) {
 639                 if (!(per_cpu(bank_map, cpu) & 1 << bank))
 640                         continue;
 641                 threshold_remove_bank(cpu, bank);
 642         }
 643 }
 644
 645 /* get notified when a cpu comes on/off */
 646 static int threshold_cpu_callback(struct notifier_block *nfb,
 647                                             unsigned long action, void *hcpu)
 648 {
 649         /* cpu was unsigned int to begin with */
 650         unsigned int cpu = (unsigned long)hcpu;
 651
 652         if (cpu >= NR_CPUS)
 653                 goto out;
 654
 655         switch (action) {
 656         case CPU_ONLINE:
 657         case CPU_ONLINE_FROZEN:
 658                 threshold_create_device(cpu);
 659                 break;
 660         case CPU_DEAD:
 661         case CPU_DEAD_FROZEN:
 662                 threshold_remove_device(cpu);
 663                 break;
 664         default:
 665                 break;
 666         }
 667       out:
 668         return NOTIFY_OK;
 669 }
 670
 671 static struct notifier_block threshold_cpu_notifier = {
 672         .notifier_call = threshold_cpu_callback,
 673 };
 674
 675 static __init int threshold_init_device(void)
 676 {
 677         unsigned lcpu = 0;
 678
 679         /* to hit CPUs online before the notifier is up */
 680         for_each_online_cpu(lcpu) {
 681                 int err = threshold_create_device(lcpu);
 682                 if (err)
 683                         return err;
 684         }
 685         register_hotcpu_notifier(&threshold_cpu_notifier);
 686         return 0;
 687 }
 688
 689 device_initcall(threshold_init_device);