arch/x86_64/kernel/mce_amd.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_extened_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                         if (high & MASK_OVERFLOW_HI) {
 224                                 rdmsrl(address, m.misc);
 225                                 rdmsrl(MSR_IA32_MC0_STATUS + bank * 4,
 226                                        m.status);
 227                                 m.bank = K8_MCE_THRESHOLD_BASE
 228                                        + bank * NR_BLOCKS
 229                                        + block;
 230                                 mce_log(&m);
 231                                 goto out;
 232                         }
 233                 }
 234         }
 235 out:
 236         irq_exit();
 237 }
 238
 239 /*
 240  * Sysfs Interface
 241  */
 242
 243 struct threshold_attr {
 244         struct attribute attr;
 245         ssize_t(*show) (struct threshold_block *, char *);
 246         ssize_t(*store) (struct threshold_block *, const char *, size_t count);
 247 };
 248
 249 static cpumask_t affinity_set(unsigned int cpu)
 250 {
 251         cpumask_t oldmask = current->cpus_allowed;
 252         cpumask_t newmask = CPU_MASK_NONE;
 253         cpu_set(cpu, newmask);
 254         set_cpus_allowed(current, newmask);
 255         return oldmask;
 256 }
 257
 258 static void affinity_restore(cpumask_t oldmask)
 259 {
 260         set_cpus_allowed(current, oldmask);
 261 }
 262
 263 #define SHOW_FIELDS(name)                                           \
 264 static ssize_t show_ ## name(struct threshold_block * b, char *buf) \
 265 {                                                                   \
 266         return sprintf(buf, "%lx\n", (unsigned long) b->name);      \
 267 }
 268 SHOW_FIELDS(interrupt_enable)
 269 SHOW_FIELDS(threshold_limit)
 270
 271 static ssize_t store_interrupt_enable(struct threshold_block *b,
 272                                       const char *buf, size_t count)
 273 {
 274         char *end;
 275         cpumask_t oldmask;
 276         unsigned long new = simple_strtoul(buf, &end, 0);
 277         if (end == buf)
 278                 return -EINVAL;
 279         b->interrupt_enable = !!new;
 280
 281         oldmask = affinity_set(b->cpu);
 282         threshold_restart_bank(b, 0, 0);
 283         affinity_restore(oldmask);
 284
 285         return end - buf;
 286 }
 287
 288 static ssize_t store_threshold_limit(struct threshold_block *b,
 289                                      const char *buf, size_t count)
 290 {
 291         char *end;
 292         cpumask_t oldmask;
 293         u16 old;
 294         unsigned long new = simple_strtoul(buf, &end, 0);
 295         if (end == buf)
 296                 return -EINVAL;
 297         if (new > THRESHOLD_MAX)
 298                 new = THRESHOLD_MAX;
 299         if (new < 1)
 300                 new = 1;
 301         old = b->threshold_limit;
 302         b->threshold_limit = new;
 303
 304         oldmask = affinity_set(b->cpu);
 305         threshold_restart_bank(b, 0, old);
 306         affinity_restore(oldmask);
 307
 308         return end - buf;
 309 }
 310
 311 static ssize_t show_error_count(struct threshold_block *b, char *buf)
 312 {
 313         u32 high, low;
 314         cpumask_t oldmask;
 315         oldmask = affinity_set(b->cpu);
 316         rdmsr(b->address, low, high);
 317         affinity_restore(oldmask);
 318         return sprintf(buf, "%x\n",
 319                        (high & 0xFFF) - (THRESHOLD_MAX - b->threshold_limit));
 320 }
 321
 322 static ssize_t store_error_count(struct threshold_block *b,
 323                                  const char *buf, size_t count)
 324 {
 325         cpumask_t oldmask;
 326         oldmask = affinity_set(b->cpu);
 327         threshold_restart_bank(b, 1, 0);
 328         affinity_restore(oldmask);
 329         return 1;
 330 }
 331
 332 #define THRESHOLD_ATTR(_name,_mode,_show,_store) {            \
 333         .attr = {.name = __stringify(_name), .mode = _mode }, \
 334         .show = _show,                                        \
 335         .store = _store,                                      \
 336 };
 337
 338 #define RW_ATTR(name)                                           \
 339 static struct threshold_attr name =                             \
 340         THRESHOLD_ATTR(name, 0644, show_## name, store_## name)
 341
 342 RW_ATTR(interrupt_enable);
 343 RW_ATTR(threshold_limit);
 344 RW_ATTR(error_count);
 345
 346 static struct attribute *default_attrs[] = {
 347         &interrupt_enable.attr,
 348         &threshold_limit.attr,
 349         &error_count.attr,
 350         NULL
 351 };
 352
 353 #define to_block(k) container_of(k, struct threshold_block, kobj)
 354 #define to_attr(a) container_of(a, struct threshold_attr, attr)
 355
 356 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
 357 {
 358         struct threshold_block *b = to_block(kobj);
 359         struct threshold_attr *a = to_attr(attr);
 360         ssize_t ret;
 361         ret = a->show ? a->show(b, buf) : -EIO;
 362         return ret;
 363 }
 364
 365 static ssize_t store(struct kobject *kobj, struct attribute *attr,
 366                      const char *buf, size_t count)
 367 {
 368         struct threshold_block *b = to_block(kobj);
 369         struct threshold_attr *a = to_attr(attr);
 370         ssize_t ret;
 371         ret = a->store ? a->store(b, buf, count) : -EIO;
 372         return ret;
 373 }
 374
 375 static struct sysfs_ops threshold_ops = {
 376         .show = show,
 377         .store = store,
 378 };
 379
 380 static struct kobj_type threshold_ktype = {
 381         .sysfs_ops = &threshold_ops,
 382         .default_attrs = default_attrs,
 383 };
 384
 385 static __cpuinit int allocate_threshold_blocks(unsigned int cpu,
 386                                                unsigned int bank,
 387                                                unsigned int block,
 388                                                u32 address)
 389 {
 390         int err;
 391         u32 low, high;
 392         struct threshold_block *b = NULL;
 393
 394         if ((bank >= NR_BANKS) || (block >= NR_BLOCKS))
 395                 return 0;
 396
 397         if (rdmsr_safe(address, &low, &high))
 398                 return 0;
 399
 400         if (!(high & MASK_VALID_HI)) {
 401                 if (block)
 402                         goto recurse;
 403                 else
 404                         return 0;
 405         }
 406
 407         if (!(high & MASK_CNTP_HI)  ||
 408              (high & MASK_LOCKED_HI))
 409                 goto recurse;
 410
 411         b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
 412         if (!b)
 413                 return -ENOMEM;
 414
 415         b->block = block;
 416         b->bank = bank;
 417         b->cpu = cpu;
 418         b->address = address;
 419         b->interrupt_enable = 0;
 420         b->threshold_limit = THRESHOLD_MAX;
 421
 422         INIT_LIST_HEAD(&b->miscj);
 423
 424         if (per_cpu(threshold_banks, cpu)[bank]->blocks)
 425                 list_add(&b->miscj,
 426                          &per_cpu(threshold_banks, cpu)[bank]->blocks->miscj);
 427         else
 428                 per_cpu(threshold_banks, cpu)[bank]->blocks = b;
 429
 430         kobject_set_name(&b->kobj, "misc%i", block);
 431         b->kobj.parent = &per_cpu(threshold_banks, cpu)[bank]->kobj;
 432         b->kobj.ktype = &threshold_ktype;
 433         err = kobject_register(&b->kobj);
 434         if (err)
 435                 goto out_free;
 436 recurse:
 437         if (!block) {
 438                 address = (low & MASK_BLKPTR_LO) >> 21;
 439                 if (!address)
 440                         return 0;
 441                 address += MCG_XBLK_ADDR;
 442         } else
 443                 ++address;
 444
 445         err = allocate_threshold_blocks(cpu, bank, ++block, address);
 446         if (err)
 447                 goto out_free;
 448
 449         return err;
 450
 451 out_free:
 452         if (b) {
 453                 kobject_unregister(&b->kobj);
 454                 kfree(b);
 455         }
 456         return err;
 457 }
 458
 459 /* symlinks sibling shared banks to first core.  first core owns dir/files. */
 460 static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank)
 461 {
 462         int i, err = 0;
 463         struct threshold_bank *b = NULL;
 464         cpumask_t oldmask = CPU_MASK_NONE;
 465         char name[32];
 466
 467         sprintf(name, "threshold_bank%i", bank);
 468
 469 #ifdef CONFIG_SMP
 470         if (cpu_data[cpu].cpu_core_id && shared_bank[bank]) {   /* symlink */
 471                 i = first_cpu(cpu_core_map[cpu]);
 472
 473                 /* first core not up yet */
 474                 if (cpu_data[i].cpu_core_id)
 475                         goto out;
 476
 477                 /* already linked */
 478                 if (per_cpu(threshold_banks, cpu)[bank])
 479                         goto out;
 480
 481                 b = per_cpu(threshold_banks, i)[bank];
 482
 483                 if (!b)
 484                         goto out;
 485
 486                 err = sysfs_create_link(&per_cpu(device_mce, cpu).kobj,
 487                                         &b->kobj, name);
 488                 if (err)
 489                         goto out;
 490
 491                 b->cpus = cpu_core_map[cpu];
 492                 per_cpu(threshold_banks, cpu)[bank] = b;
 493                 goto out;
 494         }
 495 #endif
 496
 497         b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
 498         if (!b) {
 499                 err = -ENOMEM;
 500                 goto out;
 501         }
 502
 503         kobject_set_name(&b->kobj, "threshold_bank%i", bank);
 504         b->kobj.parent = &per_cpu(device_mce, cpu).kobj;
 505 #ifndef CONFIG_SMP
 506         b->cpus = CPU_MASK_ALL;
 507 #else
 508         b->cpus = cpu_core_map[cpu];
 509 #endif
 510         err = kobject_register(&b->kobj);
 511         if (err)
 512                 goto out_free;
 513
 514         per_cpu(threshold_banks, cpu)[bank] = b;
 515
 516         oldmask = affinity_set(cpu);
 517         err = allocate_threshold_blocks(cpu, bank, 0,
 518                                         MSR_IA32_MC0_MISC + bank * 4);
 519         affinity_restore(oldmask);
 520
 521         if (err)
 522                 goto out_free;
 523
 524         for_each_cpu_mask(i, b->cpus) {
 525                 if (i == cpu)
 526                         continue;
 527
 528                 err = sysfs_create_link(&per_cpu(device_mce, i).kobj,
 529                                         &b->kobj, name);
 530                 if (err)
 531                         goto out;
 532
 533                 per_cpu(threshold_banks, i)[bank] = b;
 534         }
 535
 536         goto out;
 537
 538 out_free:
 539         per_cpu(threshold_banks, cpu)[bank] = NULL;
 540         kfree(b);
 541 out:
 542         return err;
 543 }
 544
 545 /* create dir/files for all valid threshold banks */
 546 static __cpuinit int threshold_create_device(unsigned int cpu)
 547 {
 548         unsigned int bank;
 549         int err = 0;
 550
 551         for (bank = 0; bank < NR_BANKS; ++bank) {
 552                 if (!(per_cpu(bank_map, cpu) & 1 << bank))
 553                         continue;
 554                 err = threshold_create_bank(cpu, bank);
 555                 if (err)
 556                         goto out;
 557         }
 558 out:
 559         return err;
 560 }
 561
 562 /*
 563  * let's be hotplug friendly.
 564  * in case of multiple core processors, the first core always takes ownership
 565  *   of shared sysfs dir/files, and rest of the cores will be symlinked to it.
 566  */
 567
 568 static void deallocate_threshold_block(unsigned int cpu,
 569                                                  unsigned int bank)
 570 {
 571         struct threshold_block *pos = NULL;
 572         struct threshold_block *tmp = NULL;
 573         struct threshold_bank *head = per_cpu(threshold_banks, cpu)[bank];
 574
 575         if (!head)
 576                 return;
 577
 578         list_for_each_entry_safe(pos, tmp, &head->blocks->miscj, miscj) {
 579                 kobject_unregister(&pos->kobj);
 580                 list_del(&pos->miscj);
 581                 kfree(pos);
 582         }
 583
 584         kfree(per_cpu(threshold_banks, cpu)[bank]->blocks);
 585         per_cpu(threshold_banks, cpu)[bank]->blocks = NULL;
 586 }
 587
 588 static void threshold_remove_bank(unsigned int cpu, int bank)
 589 {
 590         int i = 0;
 591         struct threshold_bank *b;
 592         char name[32];
 593
 594         b = per_cpu(threshold_banks, cpu)[bank];
 595
 596         if (!b)
 597                 return;
 598
 599         if (!b->blocks)
 600                 goto free_out;
 601
 602         sprintf(name, "threshold_bank%i", bank);
 603
 604 #ifdef CONFIG_SMP
 605         /* sibling symlink */
 606         if (shared_bank[bank] && b->blocks->cpu != cpu) {
 607                 sysfs_remove_link(&per_cpu(device_mce, cpu).kobj, name);
 608                 per_cpu(threshold_banks, cpu)[bank] = NULL;
 609                 return;
 610         }
 611 #endif
 612
 613         /* remove all sibling symlinks before unregistering */
 614         for_each_cpu_mask(i, b->cpus) {
 615                 if (i == cpu)
 616                         continue;
 617
 618                 sysfs_remove_link(&per_cpu(device_mce, i).kobj, name);
 619                 per_cpu(threshold_banks, i)[bank] = NULL;
 620         }
 621
 622         deallocate_threshold_block(cpu, bank);
 623
 624 free_out:
 625         kobject_unregister(&b->kobj);
 626         kfree(b);
 627         per_cpu(threshold_banks, cpu)[bank] = NULL;
 628 }
 629
 630 static void threshold_remove_device(unsigned int cpu)
 631 {
 632         unsigned int bank;
 633
 634         for (bank = 0; bank < NR_BANKS; ++bank) {
 635                 if (!(per_cpu(bank_map, cpu) & 1 << bank))
 636                         continue;
 637                 threshold_remove_bank(cpu, bank);
 638         }
 639 }
 640
 641 /* get notified when a cpu comes on/off */
 642 static int threshold_cpu_callback(struct notifier_block *nfb,
 643                                             unsigned long action, void *hcpu)
 644 {
 645         /* cpu was unsigned int to begin with */
 646         unsigned int cpu = (unsigned long)hcpu;
 647
 648         if (cpu >= NR_CPUS)
 649                 goto out;
 650
 651         switch (action) {
 652         case CPU_ONLINE:
 653                 threshold_create_device(cpu);
 654                 break;
 655         case CPU_DEAD:
 656                 threshold_remove_device(cpu);
 657                 break;
 658         default:
 659                 break;
 660         }
 661       out:
 662         return NOTIFY_OK;
 663 }
 664
 665 static struct notifier_block threshold_cpu_notifier = {
 666         .notifier_call = threshold_cpu_callback,
 667 };
 668
 669 static __init int threshold_init_device(void)
 670 {
 671         unsigned lcpu = 0;
 672
 673         /* to hit CPUs online before the notifier is up */
 674         for_each_online_cpu(lcpu) {
 675                 int err = threshold_create_device(lcpu);
 676                 if (err)
 677                         return err;
 678         }
 679         register_hotcpu_notifier(&threshold_cpu_notifier);
 680         return 0;
 681 }
 682
 683 device_initcall(threshold_init_device);