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 struct thresh_restart {
  87         struct threshold_block *b;
  88         int reset;
  89         u16 old_limit;
  90 };
  91
  92 /* must be called with correct cpu affinity */
  93 static long threshold_restart_bank(void *_tr)
  94 {
  95         struct thresh_restart *tr = _tr;
  96         u32 mci_misc_hi, mci_misc_lo;
  97
  98         rdmsr(tr->b->address, mci_misc_lo, mci_misc_hi);
  99
 100         if (tr->b->threshold_limit < (mci_misc_hi & THRESHOLD_MAX))
 101                 tr->reset = 1;  /* limit cannot be lower than err count */
 102
 103         if (tr->reset) {                /* reset err count and overflow bit */
 104                 mci_misc_hi =
 105                     (mci_misc_hi & ~(MASK_ERR_COUNT_HI | MASK_OVERFLOW_HI)) |
 106                     (THRESHOLD_MAX - tr->b->threshold_limit);
 107         } else if (tr->old_limit) {     /* change limit w/o reset */
 108                 int new_count = (mci_misc_hi & THRESHOLD_MAX) +
 109                     (tr->old_limit - tr->b->threshold_limit);
 110                 mci_misc_hi = (mci_misc_hi & ~MASK_ERR_COUNT_HI) |
 111                     (new_count & THRESHOLD_MAX);
 112         }
 113
 114         tr->b->interrupt_enable ?
 115             (mci_misc_hi = (mci_misc_hi & ~MASK_INT_TYPE_HI) | INT_TYPE_APIC) :
 116             (mci_misc_hi &= ~MASK_INT_TYPE_HI);
 117
 118         mci_misc_hi |= MASK_COUNT_EN_HI;
 119         wrmsr(tr->b->address, mci_misc_lo, mci_misc_hi);
 120         return 0;
 121 }
 122
 123 /* cpu init entry point, called from mce.c with preempt off */
 124 void mce_amd_feature_init(struct cpuinfo_x86 *c)
 125 {
 126         unsigned int bank, block;
 127         unsigned int cpu = smp_processor_id();
 128         u8 lvt_off;
 129         u32 low = 0, high = 0, address = 0;
 130         struct thresh_restart tr;
 131
 132         for (bank = 0; bank < NR_BANKS; ++bank) {
 133                 for (block = 0; block < NR_BLOCKS; ++block) {
 134                         if (block == 0)
 135                                 address = MSR_IA32_MC0_MISC + bank * 4;
 136                         else if (block == 1) {
 137                                 address = (low & MASK_BLKPTR_LO) >> 21;
 138                                 if (!address)
 139                                         break;
 140                                 address += MCG_XBLK_ADDR;
 141                         }
 142                         else
 143                                 ++address;
 144
 145                         if (rdmsr_safe(address, &low, &high))
 146                                 break;
 147
 148                         if (!(high & MASK_VALID_HI)) {
 149                                 if (block)
 150                                         continue;
 151                                 else
 152                                         break;
 153                         }
 154
 155                         if (!(high & MASK_CNTP_HI)  ||
 156                              (high & MASK_LOCKED_HI))
 157                                 continue;
 158
 159                         if (!block)
 160                                 per_cpu(bank_map, cpu) |= (1 << bank);
 161 #ifdef CONFIG_SMP
 162                         if (shared_bank[bank] && c->cpu_core_id)
 163                                 break;
 164 #endif
 165                         lvt_off = setup_APIC_eilvt_mce(THRESHOLD_APIC_VECTOR,
 166                                                        APIC_EILVT_MSG_FIX, 0);
 167
 168                         high &= ~MASK_LVTOFF_HI;
 169                         high |= lvt_off << 20;
 170                         wrmsr(address, low, high);
 171
 172                         threshold_defaults.address = address;
 173                         tr.b = &threshold_defaults;
 174                         tr.reset = 0;
 175                         tr.old_limit = 0;
 176                         threshold_restart_bank(&tr);
 177                 }
 178         }
 179 }
 180
 181 /*
 182  * APIC Interrupt Handler
 183  */
 184
 185 /*
 186  * threshold interrupt handler will service THRESHOLD_APIC_VECTOR.
 187  * the interrupt goes off when error_count reaches threshold_limit.
 188  * the handler will simply log mcelog w/ software defined bank number.
 189  */
 190 asmlinkage void mce_threshold_interrupt(void)
 191 {
 192         unsigned int bank, block;
 193         struct mce m;
 194         u32 low = 0, high = 0, address = 0;
 195
 196         ack_APIC_irq();
 197         exit_idle();
 198         irq_enter();
 199
 200         mce_setup(&m);
 201
 202         /* assume first bank caused it */
 203         for (bank = 0; bank < NR_BANKS; ++bank) {
 204                 if (!(per_cpu(bank_map, m.cpu) & (1 << bank)))
 205                         continue;
 206                 for (block = 0; block < NR_BLOCKS; ++block) {
 207                         if (block == 0)
 208                                 address = MSR_IA32_MC0_MISC + bank * 4;
 209                         else if (block == 1) {
 210                                 address = (low & MASK_BLKPTR_LO) >> 21;
 211                                 if (!address)
 212                                         break;
 213                                 address += MCG_XBLK_ADDR;
 214                         }
 215                         else
 216                                 ++address;
 217
 218                         if (rdmsr_safe(address, &low, &high))
 219                                 break;
 220
 221                         if (!(high & MASK_VALID_HI)) {
 222                                 if (block)
 223                                         continue;
 224                                 else
 225                                         break;
 226                         }
 227
 228                         if (!(high & MASK_CNTP_HI)  ||
 229                              (high & MASK_LOCKED_HI))
 230                                 continue;
 231
 232                         /* Log the machine check that caused the threshold
 233                            event. */
 234                         machine_check_poll(MCP_TIMESTAMP);
 235
 236                         if (high & MASK_OVERFLOW_HI) {
 237                                 rdmsrl(address, m.misc);
 238                                 rdmsrl(MSR_IA32_MC0_STATUS + bank * 4,
 239                                        m.status);
 240                                 m.bank = K8_MCE_THRESHOLD_BASE
 241                                        + bank * NR_BLOCKS
 242                                        + block;
 243                                 mce_log(&m);
 244                                 goto out;
 245                         }
 246                 }
 247         }
 248 out:
 249         inc_irq_stat(irq_threshold_count);
 250         irq_exit();
 251 }
 252
 253 /*
 254  * Sysfs Interface
 255  */
 256
 257 struct threshold_attr {
 258         struct attribute attr;
 259         ssize_t(*show) (struct threshold_block *, char *);
 260         ssize_t(*store) (struct threshold_block *, const char *, size_t count);
 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         struct thresh_restart tr;
 276         unsigned long new = simple_strtoul(buf, &end, 0);
 277         if (end == buf)
 278                 return -EINVAL;
 279         b->interrupt_enable = !!new;
 280
 281         tr.b = b;
 282         tr.reset = 0;
 283         tr.old_limit = 0;
 284         work_on_cpu(b->cpu, threshold_restart_bank, &tr);
 285
 286         return end - buf;
 287 }
 288
 289 static ssize_t store_threshold_limit(struct threshold_block *b,
 290                                      const char *buf, size_t count)
 291 {
 292         char *end;
 293         struct thresh_restart tr;
 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         tr.old_limit = b->threshold_limit;
 302         b->threshold_limit = new;
 303         tr.b = b;
 304         tr.reset = 0;
 305
 306         work_on_cpu(b->cpu, threshold_restart_bank, &tr);
 307
 308         return end - buf;
 309 }
 310
 311 static long local_error_count(void *_b)
 312 {
 313         struct threshold_block *b = _b;
 314         u32 low, high;
 315
 316         rdmsr(b->address, low, high);
 317         return (high & 0xFFF) - (THRESHOLD_MAX - b->threshold_limit);
 318 }
 319
 320 static ssize_t show_error_count(struct threshold_block *b, char *buf)
 321 {
 322         return sprintf(buf, "%lx\n", work_on_cpu(b->cpu, local_error_count, b));
 323 }
 324
 325 static ssize_t store_error_count(struct threshold_block *b,
 326                                  const char *buf, size_t count)
 327 {
 328         struct thresh_restart tr = { .b = b, .reset = 1, .old_limit = 0 };
 329
 330         work_on_cpu(b->cpu, threshold_restart_bank, &tr);
 331         return 1;
 332 }
 333
 334 #define THRESHOLD_ATTR(_name,_mode,_show,_store) {            \
 335         .attr = {.name = __stringify(_name), .mode = _mode }, \
 336         .show = _show,                                        \
 337         .store = _store,                                      \
 338 };
 339
 340 #define RW_ATTR(name)                                           \
 341 static struct threshold_attr name =                             \
 342         THRESHOLD_ATTR(name, 0644, show_## name, store_## name)
 343
 344 RW_ATTR(interrupt_enable);
 345 RW_ATTR(threshold_limit);
 346 RW_ATTR(error_count);
 347
 348 static struct attribute *default_attrs[] = {
 349         &interrupt_enable.attr,
 350         &threshold_limit.attr,
 351         &error_count.attr,
 352         NULL
 353 };
 354
 355 #define to_block(k) container_of(k, struct threshold_block, kobj)
 356 #define to_attr(a) container_of(a, struct threshold_attr, attr)
 357
 358 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
 359 {
 360         struct threshold_block *b = to_block(kobj);
 361         struct threshold_attr *a = to_attr(attr);
 362         ssize_t ret;
 363         ret = a->show ? a->show(b, buf) : -EIO;
 364         return ret;
 365 }
 366
 367 static ssize_t store(struct kobject *kobj, struct attribute *attr,
 368                      const char *buf, size_t count)
 369 {
 370         struct threshold_block *b = to_block(kobj);
 371         struct threshold_attr *a = to_attr(attr);
 372         ssize_t ret;
 373         ret = a->store ? a->store(b, buf, count) : -EIO;
 374         return ret;
 375 }
 376
 377 static struct sysfs_ops threshold_ops = {
 378         .show = show,
 379         .store = store,
 380 };
 381
 382 static struct kobj_type threshold_ktype = {
 383         .sysfs_ops = &threshold_ops,
 384         .default_attrs = default_attrs,
 385 };
 386
 387 static __cpuinit int allocate_threshold_blocks(unsigned int cpu,
 388                                                unsigned int bank,
 389                                                unsigned int block,
 390                                                u32 address)
 391 {
 392         int err;
 393         u32 low, high;
 394         struct threshold_block *b = NULL;
 395
 396         if ((bank >= NR_BANKS) || (block >= NR_BLOCKS))
 397                 return 0;
 398
 399         if (rdmsr_safe(address, &low, &high))
 400                 return 0;
 401
 402         if (!(high & MASK_VALID_HI)) {
 403                 if (block)
 404                         goto recurse;
 405                 else
 406                         return 0;
 407         }
 408
 409         if (!(high & MASK_CNTP_HI)  ||
 410              (high & MASK_LOCKED_HI))
 411                 goto recurse;
 412
 413         b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
 414         if (!b)
 415                 return -ENOMEM;
 416
 417         b->block = block;
 418         b->bank = bank;
 419         b->cpu = cpu;
 420         b->address = address;
 421         b->interrupt_enable = 0;
 422         b->threshold_limit = THRESHOLD_MAX;
 423
 424         INIT_LIST_HEAD(&b->miscj);
 425
 426         if (per_cpu(threshold_banks, cpu)[bank]->blocks)
 427                 list_add(&b->miscj,
 428                          &per_cpu(threshold_banks, cpu)[bank]->blocks->miscj);
 429         else
 430                 per_cpu(threshold_banks, cpu)[bank]->blocks = b;
 431
 432         err = kobject_init_and_add(&b->kobj, &threshold_ktype,
 433                                    per_cpu(threshold_banks, cpu)[bank]->kobj,
 434                                    "misc%i", block);
 435         if (err)
 436                 goto out_free;
 437 recurse:
 438         if (!block) {
 439                 address = (low & MASK_BLKPTR_LO) >> 21;
 440                 if (!address)
 441                         return 0;
 442                 address += MCG_XBLK_ADDR;
 443         } else
 444                 ++address;
 445
 446         err = allocate_threshold_blocks(cpu, bank, ++block, address);
 447         if (err)
 448                 goto out_free;
 449
 450         if (b)
 451                 kobject_uevent(&b->kobj, KOBJ_ADD);
 452
 453         return err;
 454
 455 out_free:
 456         if (b) {
 457                 kobject_put(&b->kobj);
 458                 kfree(b);
 459         }
 460         return err;
 461 }
 462
 463 static __cpuinit long local_allocate_threshold_blocks(void *_bank)
 464 {
 465         unsigned int *bank = _bank;
 466
 467         return allocate_threshold_blocks(smp_processor_id(), *bank, 0,
 468                                          MSR_IA32_MC0_MISC + *bank * 4);
 469 }
 470
 471 /* symlinks sibling shared banks to first core.  first core owns dir/files. */
 472 static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank)
 473 {
 474         int i, err = 0;
 475         struct threshold_bank *b = NULL;
 476         char name[32];
 477
 478         sprintf(name, "threshold_bank%i", bank);
 479
 480 #ifdef CONFIG_SMP
 481         if (cpu_data(cpu).cpu_core_id && shared_bank[bank]) {   /* symlink */
 482                 i = first_cpu(per_cpu(cpu_core_map, cpu));
 483
 484                 /* first core not up yet */
 485                 if (cpu_data(i).cpu_core_id)
 486                         goto out;
 487
 488                 /* already linked */
 489                 if (per_cpu(threshold_banks, cpu)[bank])
 490                         goto out;
 491
 492                 b = per_cpu(threshold_banks, i)[bank];
 493
 494                 if (!b)
 495                         goto out;
 496
 497                 err = sysfs_create_link(&per_cpu(device_mce, cpu).kobj,
 498                                         b->kobj, name);
 499                 if (err)
 500                         goto out;
 501
 502                 b->cpus = per_cpu(cpu_core_map, cpu);
 503                 per_cpu(threshold_banks, cpu)[bank] = b;
 504                 goto out;
 505         }
 506 #endif
 507
 508         b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
 509         if (!b) {
 510                 err = -ENOMEM;
 511                 goto out;
 512         }
 513
 514         b->kobj = kobject_create_and_add(name, &per_cpu(device_mce, cpu).kobj);
 515         if (!b->kobj)
 516                 goto out_free;
 517
 518 #ifndef CONFIG_SMP
 519         b->cpus = CPU_MASK_ALL;
 520 #else
 521         b->cpus = per_cpu(cpu_core_map, cpu);
 522 #endif
 523
 524         per_cpu(threshold_banks, cpu)[bank] = b;
 525
 526         err = work_on_cpu(cpu, local_allocate_threshold_blocks, &bank);
 527         if (err)
 528                 goto out_free;
 529
 530         for_each_cpu_mask_nr(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_nr(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_del(b->kobj);
 632         kobject_put(b->kobj);
 633         kfree(b);
 634         per_cpu(threshold_banks, cpu)[bank] = NULL;
 635 }
 636
 637 static void threshold_remove_device(unsigned int cpu)
 638 {
 639         unsigned int bank;
 640
 641         for (bank = 0; bank < NR_BANKS; ++bank) {
 642                 if (!(per_cpu(bank_map, cpu) & (1 << bank)))
 643                         continue;
 644                 threshold_remove_bank(cpu, bank);
 645         }
 646 }
 647
 648 /* get notified when a cpu comes on/off */
 649 static void __cpuinit amd_64_threshold_cpu_callback(unsigned long action,
 650                                                      unsigned int cpu)
 651 {
 652         if (cpu >= NR_CPUS)
 653                 return;
 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 }
 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         threshold_cpu_callback = amd_64_threshold_cpu_callback;
 680         return 0;
 681 }
 682
 683 device_initcall(threshold_init_device);