arch/x86/kernel/cpu/mcheck/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         u8 lvt_off;
 122         u32 low = 0, high = 0, address = 0;
 123
 124         for (bank = 0; bank < NR_BANKS; ++bank) {
 125                 for (block = 0; block < NR_BLOCKS; ++block) {
 126                         if (block == 0)
 127                                 address = MSR_IA32_MC0_MISC + bank * 4;
 128                         else if (block == 1) {
 129                                 address = (low & MASK_BLKPTR_LO) >> 21;
 130                                 if (!address)
 131                                         break;
 132                                 address += MCG_XBLK_ADDR;
 133                         }
 134                         else
 135                                 ++address;
 136
 137                         if (rdmsr_safe(address, &low, &high))
 138                                 break;
 139
 140                         if (!(high & MASK_VALID_HI)) {
 141                                 if (block)
 142                                         continue;
 143                                 else
 144                                         break;
 145                         }
 146
 147                         if (!(high & MASK_CNTP_HI)  ||
 148                              (high & MASK_LOCKED_HI))
 149                                 continue;
 150
 151                         if (!block)
 152                                 per_cpu(bank_map, cpu) |= (1 << bank);
 153 #ifdef CONFIG_SMP
 154                         if (shared_bank[bank] && c->cpu_core_id)
 155                                 break;
 156 #endif
 157                         lvt_off = setup_APIC_eilvt_mce(THRESHOLD_APIC_VECTOR,
 158                                                        APIC_EILVT_MSG_FIX, 0);
 159
 160                         high &= ~MASK_LVTOFF_HI;
 161                         high |= lvt_off << 20;
 162                         wrmsr(address, low, high);
 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         add_pda(irq_threshold_count, 1);
 241         irq_exit();
 242 }
 243
 244 /*
 245  * Sysfs Interface
 246  */
 247
 248 struct threshold_attr {
 249         struct attribute attr;
 250         ssize_t(*show) (struct threshold_block *, char *);
 251         ssize_t(*store) (struct threshold_block *, const char *, size_t count);
 252 };
 253
 254 static cpumask_t affinity_set(unsigned int cpu)
 255 {
 256         cpumask_t oldmask = current->cpus_allowed;
 257         cpumask_t newmask = CPU_MASK_NONE;
 258         cpu_set(cpu, newmask);
 259         set_cpus_allowed(current, newmask);
 260         return oldmask;
 261 }
 262
 263 static void affinity_restore(cpumask_t oldmask)
 264 {
 265         set_cpus_allowed(current, oldmask);
 266 }
 267
 268 #define SHOW_FIELDS(name)                                           \
 269 static ssize_t show_ ## name(struct threshold_block * b, char *buf) \
 270 {                                                                   \
 271         return sprintf(buf, "%lx\n", (unsigned long) b->name);      \
 272 }
 273 SHOW_FIELDS(interrupt_enable)
 274 SHOW_FIELDS(threshold_limit)
 275
 276 static ssize_t store_interrupt_enable(struct threshold_block *b,
 277                                       const char *buf, size_t count)
 278 {
 279         char *end;
 280         cpumask_t oldmask;
 281         unsigned long new = simple_strtoul(buf, &end, 0);
 282         if (end == buf)
 283                 return -EINVAL;
 284         b->interrupt_enable = !!new;
 285
 286         oldmask = affinity_set(b->cpu);
 287         threshold_restart_bank(b, 0, 0);
 288         affinity_restore(oldmask);
 289
 290         return end - buf;
 291 }
 292
 293 static ssize_t store_threshold_limit(struct threshold_block *b,
 294                                      const char *buf, size_t count)
 295 {
 296         char *end;
 297         cpumask_t oldmask;
 298         u16 old;
 299         unsigned long new = simple_strtoul(buf, &end, 0);
 300         if (end == buf)
 301                 return -EINVAL;
 302         if (new > THRESHOLD_MAX)
 303                 new = THRESHOLD_MAX;
 304         if (new < 1)
 305                 new = 1;
 306         old = b->threshold_limit;
 307         b->threshold_limit = new;
 308
 309         oldmask = affinity_set(b->cpu);
 310         threshold_restart_bank(b, 0, old);
 311         affinity_restore(oldmask);
 312
 313         return end - buf;
 314 }
 315
 316 static ssize_t show_error_count(struct threshold_block *b, char *buf)
 317 {
 318         u32 high, low;
 319         cpumask_t oldmask;
 320         oldmask = affinity_set(b->cpu);
 321         rdmsr(b->address, low, high);
 322         affinity_restore(oldmask);
 323         return sprintf(buf, "%x\n",
 324                        (high & 0xFFF) - (THRESHOLD_MAX - b->threshold_limit));
 325 }
 326
 327 static ssize_t store_error_count(struct threshold_block *b,
 328                                  const char *buf, size_t count)
 329 {
 330         cpumask_t oldmask;
 331         oldmask = affinity_set(b->cpu);
 332         threshold_restart_bank(b, 1, 0);
 333         affinity_restore(oldmask);
 334         return 1;
 335 }
 336
 337 #define THRESHOLD_ATTR(_name,_mode,_show,_store) {            \
 338         .attr = {.name = __stringify(_name), .mode = _mode }, \
 339         .show = _show,                                        \
 340         .store = _store,                                      \
 341 };
 342
 343 #define RW_ATTR(name)                                           \
 344 static struct threshold_attr name =                             \
 345         THRESHOLD_ATTR(name, 0644, show_## name, store_## name)
 346
 347 RW_ATTR(interrupt_enable);
 348 RW_ATTR(threshold_limit);
 349 RW_ATTR(error_count);
 350
 351 static struct attribute *default_attrs[] = {
 352         &interrupt_enable.attr,
 353         &threshold_limit.attr,
 354         &error_count.attr,
 355         NULL
 356 };
 357
 358 #define to_block(k) container_of(k, struct threshold_block, kobj)
 359 #define to_attr(a) container_of(a, struct threshold_attr, attr)
 360
 361 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
 362 {
 363         struct threshold_block *b = to_block(kobj);
 364         struct threshold_attr *a = to_attr(attr);
 365         ssize_t ret;
 366         ret = a->show ? a->show(b, buf) : -EIO;
 367         return ret;
 368 }
 369
 370 static ssize_t store(struct kobject *kobj, struct attribute *attr,
 371                      const char *buf, size_t count)
 372 {
 373         struct threshold_block *b = to_block(kobj);
 374         struct threshold_attr *a = to_attr(attr);
 375         ssize_t ret;
 376         ret = a->store ? a->store(b, buf, count) : -EIO;
 377         return ret;
 378 }
 379
 380 static struct sysfs_ops threshold_ops = {
 381         .show = show,
 382         .store = store,
 383 };
 384
 385 static struct kobj_type threshold_ktype = {
 386         .sysfs_ops = &threshold_ops,
 387         .default_attrs = default_attrs,
 388 };
 389
 390 static __cpuinit int allocate_threshold_blocks(unsigned int cpu,
 391                                                unsigned int bank,
 392                                                unsigned int block,
 393                                                u32 address)
 394 {
 395         int err;
 396         u32 low, high;
 397         struct threshold_block *b = NULL;
 398
 399         if ((bank >= NR_BANKS) || (block >= NR_BLOCKS))
 400                 return 0;
 401
 402         if (rdmsr_safe(address, &low, &high))
 403                 return 0;
 404
 405         if (!(high & MASK_VALID_HI)) {
 406                 if (block)
 407                         goto recurse;
 408                 else
 409                         return 0;
 410         }
 411
 412         if (!(high & MASK_CNTP_HI)  ||
 413              (high & MASK_LOCKED_HI))
 414                 goto recurse;
 415
 416         b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
 417         if (!b)
 418                 return -ENOMEM;
 419
 420         b->block = block;
 421         b->bank = bank;
 422         b->cpu = cpu;
 423         b->address = address;
 424         b->interrupt_enable = 0;
 425         b->threshold_limit = THRESHOLD_MAX;
 426
 427         INIT_LIST_HEAD(&b->miscj);
 428
 429         if (per_cpu(threshold_banks, cpu)[bank]->blocks)
 430                 list_add(&b->miscj,
 431                          &per_cpu(threshold_banks, cpu)[bank]->blocks->miscj);
 432         else
 433                 per_cpu(threshold_banks, cpu)[bank]->blocks = b;
 434
 435         err = kobject_init_and_add(&b->kobj, &threshold_ktype,
 436                                    per_cpu(threshold_banks, cpu)[bank]->kobj,
 437                                    "misc%i", block);
 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         if (b)
 454                 kobject_uevent(&b->kobj, KOBJ_ADD);
 455
 456         return err;
 457
 458 out_free:
 459         if (b) {
 460                 kobject_put(&b->kobj);
 461                 kfree(b);
 462         }
 463         return err;
 464 }
 465
 466 /* symlinks sibling shared banks to first core.  first core owns dir/files. */
 467 static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank)
 468 {
 469         int i, err = 0;
 470         struct threshold_bank *b = NULL;
 471         cpumask_t oldmask = CPU_MASK_NONE;
 472         char name[32];
 473
 474         sprintf(name, "threshold_bank%i", bank);
 475
 476 #ifdef CONFIG_SMP
 477         if (cpu_data(cpu).cpu_core_id && shared_bank[bank]) {   /* symlink */
 478                 i = first_cpu(per_cpu(cpu_core_map, cpu));
 479
 480                 /* first core not up yet */
 481                 if (cpu_data(i).cpu_core_id)
 482                         goto out;
 483
 484                 /* already linked */
 485                 if (per_cpu(threshold_banks, cpu)[bank])
 486                         goto out;
 487
 488                 b = per_cpu(threshold_banks, i)[bank];
 489
 490                 if (!b)
 491                         goto out;
 492
 493                 err = sysfs_create_link(&per_cpu(device_mce, cpu).kobj,
 494                                         b->kobj, name);
 495                 if (err)
 496                         goto out;
 497
 498                 b->cpus = per_cpu(cpu_core_map, cpu);
 499                 per_cpu(threshold_banks, cpu)[bank] = b;
 500                 goto out;
 501         }
 502 #endif
 503
 504         b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
 505         if (!b) {
 506                 err = -ENOMEM;
 507                 goto out;
 508         }
 509
 510         b->kobj = kobject_create_and_add(name, &per_cpu(device_mce, cpu).kobj);
 511         if (!b->kobj)
 512                 goto out_free;
 513
 514 #ifndef CONFIG_SMP
 515         b->cpus = CPU_MASK_ALL;
 516 #else
 517         b->cpus = per_cpu(cpu_core_map, cpu);
 518 #endif
 519
 520         per_cpu(threshold_banks, cpu)[bank] = b;
 521
 522         oldmask = affinity_set(cpu);
 523         err = allocate_threshold_blocks(cpu, bank, 0,
 524                                         MSR_IA32_MC0_MISC + bank * 4);
 525         affinity_restore(oldmask);
 526
 527         if (err)
 528                 goto out_free;
 529
 530         for_each_cpu_mask(i, b->cpus) {
 531                 if (i == cpu)
 532                         continue;
 533
 534                 err = sysfs_create_link(&per_cpu(device_mce, i).kobj,
 535                                         b->kobj, name);
 536                 if (err)
 537                         goto out;
 538
 539                 per_cpu(threshold_banks, i)[bank] = b;
 540         }
 541
 542         goto out;
 543
 544 out_free:
 545         per_cpu(threshold_banks, cpu)[bank] = NULL;
 546         kfree(b);
 547 out:
 548         return err;
 549 }
 550
 551 /* create dir/files for all valid threshold banks */
 552 static __cpuinit int threshold_create_device(unsigned int cpu)
 553 {
 554         unsigned int bank;
 555         int err = 0;
 556
 557         for (bank = 0; bank < NR_BANKS; ++bank) {
 558                 if (!(per_cpu(bank_map, cpu) & (1 << bank)))
 559                         continue;
 560                 err = threshold_create_bank(cpu, bank);
 561                 if (err)
 562                         goto out;
 563         }
 564 out:
 565         return err;
 566 }
 567
 568 /*
 569  * let's be hotplug friendly.
 570  * in case of multiple core processors, the first core always takes ownership
 571  *   of shared sysfs dir/files, and rest of the cores will be symlinked to it.
 572  */
 573
 574 static void deallocate_threshold_block(unsigned int cpu,
 575                                                  unsigned int bank)
 576 {
 577         struct threshold_block *pos = NULL;
 578         struct threshold_block *tmp = NULL;
 579         struct threshold_bank *head = per_cpu(threshold_banks, cpu)[bank];
 580
 581         if (!head)
 582                 return;
 583
 584         list_for_each_entry_safe(pos, tmp, &head->blocks->miscj, miscj) {
 585                 kobject_put(&pos->kobj);
 586                 list_del(&pos->miscj);
 587                 kfree(pos);
 588         }
 589
 590         kfree(per_cpu(threshold_banks, cpu)[bank]->blocks);
 591         per_cpu(threshold_banks, cpu)[bank]->blocks = NULL;
 592 }
 593
 594 static void threshold_remove_bank(unsigned int cpu, int bank)
 595 {
 596         int i = 0;
 597         struct threshold_bank *b;
 598         char name[32];
 599
 600         b = per_cpu(threshold_banks, cpu)[bank];
 601
 602         if (!b)
 603                 return;
 604
 605         if (!b->blocks)
 606                 goto free_out;
 607
 608         sprintf(name, "threshold_bank%i", bank);
 609
 610 #ifdef CONFIG_SMP
 611         /* sibling symlink */
 612         if (shared_bank[bank] && b->blocks->cpu != cpu) {
 613                 sysfs_remove_link(&per_cpu(device_mce, cpu).kobj, name);
 614                 per_cpu(threshold_banks, cpu)[bank] = NULL;
 615                 return;
 616         }
 617 #endif
 618
 619         /* remove all sibling symlinks before unregistering */
 620         for_each_cpu_mask(i, b->cpus) {
 621                 if (i == cpu)
 622                         continue;
 623
 624                 sysfs_remove_link(&per_cpu(device_mce, i).kobj, name);
 625                 per_cpu(threshold_banks, i)[bank] = NULL;
 626         }
 627
 628         deallocate_threshold_block(cpu, bank);
 629
 630 free_out:
 631         kobject_put(b->kobj);
 632         kfree(b);
 633         per_cpu(threshold_banks, cpu)[bank] = NULL;
 634 }
 635
 636 static void threshold_remove_device(unsigned int cpu)
 637 {
 638         unsigned int bank;
 639
 640         for (bank = 0; bank < NR_BANKS; ++bank) {
 641                 if (!(per_cpu(bank_map, cpu) & (1 << bank)))
 642                         continue;
 643                 threshold_remove_bank(cpu, bank);
 644         }
 645 }
 646
 647 /* get notified when a cpu comes on/off */
 648 static int __cpuinit threshold_cpu_callback(struct notifier_block *nfb,
 649                                             unsigned long action, void *hcpu)
 650 {
 651         /* cpu was unsigned int to begin with */
 652         unsigned int cpu = (unsigned long)hcpu;
 653
 654         if (cpu >= NR_CPUS)
 655                 goto out;
 656
 657         switch (action) {
 658         case CPU_ONLINE:
 659         case CPU_ONLINE_FROZEN:
 660                 threshold_create_device(cpu);
 661                 break;
 662         case CPU_DEAD:
 663         case CPU_DEAD_FROZEN:
 664                 threshold_remove_device(cpu);
 665                 break;
 666         default:
 667                 break;
 668         }
 669       out:
 670         return NOTIFY_OK;
 671 }
 672
 673 static struct notifier_block threshold_cpu_notifier __cpuinitdata = {
 674         .notifier_call = threshold_cpu_callback,
 675 };
 676
 677 static __init int threshold_init_device(void)
 678 {
 679         unsigned lcpu = 0;
 680
 681         /* to hit CPUs online before the notifier is up */
 682         for_each_online_cpu(lcpu) {
 683                 int err = threshold_create_device(lcpu);
 684                 if (err)
 685                         return err;
 686         }
 687         register_hotcpu_notifier(&threshold_cpu_notifier);
 688         return 0;
 689 }
 690
 691 device_initcall(threshold_init_device);