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_LVTOFF_HI    0x00F00000
  41 #define MASK_COUNT_EN_HI  0x00080000
  42 #define MASK_INT_TYPE_HI  0x00060000
  43 #define MASK_OVERFLOW_HI  0x00010000
  44 #define MASK_ERR_COUNT_HI 0x00000FFF
  45 #define MASK_BLKPTR_LO    0xFF000000
  46 #define MCG_XBLK_ADDR     0xC0000400
  47
  48 struct threshold_block {
  49         unsigned int block;
  50         unsigned int bank;
  51         unsigned int cpu;
  52         u32 address;
  53         u16 interrupt_enable;
  54         u16 threshold_limit;
  55         struct kobject kobj;
  56         struct list_head miscj;
  57 };
  58
  59 /* defaults used early on boot */
  60 static struct threshold_block threshold_defaults = {
  61         .interrupt_enable = 0,
  62         .threshold_limit = THRESHOLD_MAX,
  63 };
  64
  65 struct threshold_bank {
  66         struct kobject kobj;
  67         struct threshold_block *blocks;
  68         cpumask_t cpus;
  69 };
  70 static DEFINE_PER_CPU(struct threshold_bank *, threshold_banks[NR_BANKS]);
  71
  72 #ifdef CONFIG_SMP
  73 static unsigned char shared_bank[NR_BANKS] = {
  74         0, 0, 0, 0, 1
  75 };
  76 #endif
  77
  78 static DEFINE_PER_CPU(unsigned char, bank_map); /* see which banks are on */
  79
  80 /*
  81  * CPU Initialization
  82  */
  83
  84 /* must be called with correct cpu affinity */
  85 static void threshold_restart_bank(struct threshold_block *b,
  86                                    int reset, u16 old_limit)
  87 {
  88         u32 mci_misc_hi, mci_misc_lo;
  89
  90         rdmsr(b->address, mci_misc_lo, mci_misc_hi);
  91
  92         if (b->threshold_limit < (mci_misc_hi & THRESHOLD_MAX))
  93                 reset = 1;      /* limit cannot be lower than err count */
  94
  95         if (reset) {            /* reset err count and overflow bit */
  96                 mci_misc_hi =
  97                     (mci_misc_hi & ~(MASK_ERR_COUNT_HI | MASK_OVERFLOW_HI)) |
  98                     (THRESHOLD_MAX - b->threshold_limit);
  99         } else if (old_limit) { /* change limit w/o reset */
 100                 int new_count = (mci_misc_hi & THRESHOLD_MAX) +
 101                     (old_limit - b->threshold_limit);
 102                 mci_misc_hi = (mci_misc_hi & ~MASK_ERR_COUNT_HI) |
 103                     (new_count & THRESHOLD_MAX);
 104         }
 105
 106         b->interrupt_enable ?
 107             (mci_misc_hi = (mci_misc_hi & ~MASK_INT_TYPE_HI) | INT_TYPE_APIC) :
 108             (mci_misc_hi &= ~MASK_INT_TYPE_HI);
 109
 110         mci_misc_hi |= MASK_COUNT_EN_HI;
 111         wrmsr(b->address, mci_misc_lo, mci_misc_hi);
 112 }
 113
 114 /* cpu init entry point, called from mce.c with preempt off */
 115 void __cpuinit mce_amd_feature_init(struct cpuinfo_x86 *c)
 116 {
 117         unsigned int bank, block;
 118         unsigned int cpu = smp_processor_id();
 119         u32 low = 0, high = 0, address = 0;
 120
 121         for (bank = 0; bank < NR_BANKS; ++bank) {
 122                 for (block = 0; block < NR_BLOCKS; ++block) {
 123                         if (block == 0)
 124                                 address = MSR_IA32_MC0_MISC + bank * 4;
 125                         else if (block == 1)
 126                                 address = MCG_XBLK_ADDR
 127                                         + ((low & MASK_BLKPTR_LO) >> 21);
 128                         else
 129                                 ++address;
 130
 131                         if (rdmsr_safe(address, &low, &high))
 132                                 continue;
 133
 134                         if (!(high & MASK_VALID_HI)) {
 135                                 if (block)
 136                                         continue;
 137                                 else
 138                                         break;
 139                         }
 140
 141                         if (!(high & MASK_VALID_HI >> 1)  ||
 142                              (high & MASK_VALID_HI >> 2))
 143                                 continue;
 144
 145                         if (!block)
 146                                 per_cpu(bank_map, cpu) |= (1 << bank);
 147 #ifdef CONFIG_SMP
 148                         if (shared_bank[bank] && c->cpu_core_id)
 149                                 break;
 150 #endif
 151                         high &= ~MASK_LVTOFF_HI;
 152                         high |= K8_APIC_EXT_LVT_ENTRY_THRESHOLD << 20;
 153                         wrmsr(address, low, high);
 154
 155                         setup_APIC_extened_lvt(K8_APIC_EXT_LVT_ENTRY_THRESHOLD,
 156                                                THRESHOLD_APIC_VECTOR,
 157                                                K8_APIC_EXT_INT_MSG_FIX, 0);
 158
 159                         threshold_defaults.address = address;
 160                         threshold_restart_bank(&threshold_defaults, 0, 0);
 161                 }
 162         }
 163 }
 164
 165 /*
 166  * APIC Interrupt Handler
 167  */
 168
 169 /*
 170  * threshold interrupt handler will service THRESHOLD_APIC_VECTOR.
 171  * the interrupt goes off when error_count reaches threshold_limit.
 172  * the handler will simply log mcelog w/ software defined bank number.
 173  */
 174 asmlinkage void mce_threshold_interrupt(void)
 175 {
 176         unsigned int bank, block;
 177         struct mce m;
 178         u32 low = 0, high = 0, address = 0;
 179
 180         ack_APIC_irq();
 181         exit_idle();
 182         irq_enter();
 183
 184         memset(&m, 0, sizeof(m));
 185         rdtscll(m.tsc);
 186         m.cpu = smp_processor_id();
 187
 188         /* assume first bank caused it */
 189         for (bank = 0; bank < NR_BANKS; ++bank) {
 190                 for (block = 0; block < NR_BLOCKS; ++block) {
 191                         if (block == 0)
 192                                 address = MSR_IA32_MC0_MISC + bank * 4;
 193                         else if (block == 1)
 194                                 address = MCG_XBLK_ADDR
 195                                         + ((low & MASK_BLKPTR_LO) >> 21);
 196                         else
 197                                 ++address;
 198
 199                         if (rdmsr_safe(address, &low, &high))
 200                                 continue;
 201
 202                         if (!(high & MASK_VALID_HI)) {
 203                                 if (block)
 204                                         continue;
 205                                 else
 206                                         break;
 207                         }
 208
 209                         if (!(high & MASK_VALID_HI >> 1)  ||
 210                              (high & MASK_VALID_HI >> 2))
 211                                 continue;
 212
 213                         if (high & MASK_OVERFLOW_HI) {
 214                                 rdmsrl(address, m.misc);
 215                                 rdmsrl(MSR_IA32_MC0_STATUS + bank * 4,
 216                                        m.status);
 217                                 m.bank = K8_MCE_THRESHOLD_BASE
 218                                        + bank * NR_BLOCKS
 219                                        + block;
 220                                 mce_log(&m);
 221                                 goto out;
 222                         }
 223                 }
 224         }
 225 out:
 226         irq_exit();
 227 }
 228
 229 /*
 230  * Sysfs Interface
 231  */
 232
 233 struct threshold_attr {
 234         struct attribute attr;
 235         ssize_t(*show) (struct threshold_block *, char *);
 236         ssize_t(*store) (struct threshold_block *, const char *, size_t count);
 237 };
 238
 239 static cpumask_t affinity_set(unsigned int cpu)
 240 {
 241         cpumask_t oldmask = current->cpus_allowed;
 242         cpumask_t newmask = CPU_MASK_NONE;
 243         cpu_set(cpu, newmask);
 244         set_cpus_allowed(current, newmask);
 245         return oldmask;
 246 }
 247
 248 static void affinity_restore(cpumask_t oldmask)
 249 {
 250         set_cpus_allowed(current, oldmask);
 251 }
 252
 253 #define SHOW_FIELDS(name)                                           \
 254 static ssize_t show_ ## name(struct threshold_block * b, char *buf) \
 255 {                                                                   \
 256         return sprintf(buf, "%lx\n", (unsigned long) b->name);      \
 257 }
 258 SHOW_FIELDS(interrupt_enable)
 259 SHOW_FIELDS(threshold_limit)
 260
 261 static ssize_t store_interrupt_enable(struct threshold_block *b,
 262                                       const char *buf, size_t count)
 263 {
 264         char *end;
 265         cpumask_t oldmask;
 266         unsigned long new = simple_strtoul(buf, &end, 0);
 267         if (end == buf)
 268                 return -EINVAL;
 269         b->interrupt_enable = !!new;
 270
 271         oldmask = affinity_set(b->cpu);
 272         threshold_restart_bank(b, 0, 0);
 273         affinity_restore(oldmask);
 274
 275         return end - buf;
 276 }
 277
 278 static ssize_t store_threshold_limit(struct threshold_block *b,
 279                                      const char *buf, size_t count)
 280 {
 281         char *end;
 282         cpumask_t oldmask;
 283         u16 old;
 284         unsigned long new = simple_strtoul(buf, &end, 0);
 285         if (end == buf)
 286                 return -EINVAL;
 287         if (new > THRESHOLD_MAX)
 288                 new = THRESHOLD_MAX;
 289         if (new < 1)
 290                 new = 1;
 291         old = b->threshold_limit;
 292         b->threshold_limit = new;
 293
 294         oldmask = affinity_set(b->cpu);
 295         threshold_restart_bank(b, 0, old);
 296         affinity_restore(oldmask);
 297
 298         return end - buf;
 299 }
 300
 301 static ssize_t show_error_count(struct threshold_block *b, char *buf)
 302 {
 303         u32 high, low;
 304         cpumask_t oldmask;
 305         oldmask = affinity_set(b->cpu);
 306         rdmsr(b->address, low, high);
 307         affinity_restore(oldmask);
 308         return sprintf(buf, "%x\n",
 309                        (high & 0xFFF) - (THRESHOLD_MAX - b->threshold_limit));
 310 }
 311
 312 static ssize_t store_error_count(struct threshold_block *b,
 313                                  const char *buf, size_t count)
 314 {
 315         cpumask_t oldmask;
 316         oldmask = affinity_set(b->cpu);
 317         threshold_restart_bank(b, 1, 0);
 318         affinity_restore(oldmask);
 319         return 1;
 320 }
 321
 322 #define THRESHOLD_ATTR(_name,_mode,_show,_store) {            \
 323         .attr = {.name = __stringify(_name), .mode = _mode }, \
 324         .show = _show,                                        \
 325         .store = _store,                                      \
 326 };
 327
 328 #define RW_ATTR(name)                                           \
 329 static struct threshold_attr name =                             \
 330         THRESHOLD_ATTR(name, 0644, show_## name, store_## name)
 331
 332 RW_ATTR(interrupt_enable);
 333 RW_ATTR(threshold_limit);
 334 RW_ATTR(error_count);
 335
 336 static struct attribute *default_attrs[] = {
 337         &interrupt_enable.attr,
 338         &threshold_limit.attr,
 339         &error_count.attr,
 340         NULL
 341 };
 342
 343 #define to_block(k) container_of(k, struct threshold_block, kobj)
 344 #define to_attr(a) container_of(a, struct threshold_attr, attr)
 345
 346 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
 347 {
 348         struct threshold_block *b = to_block(kobj);
 349         struct threshold_attr *a = to_attr(attr);
 350         ssize_t ret;
 351         ret = a->show ? a->show(b, buf) : -EIO;
 352         return ret;
 353 }
 354
 355 static ssize_t store(struct kobject *kobj, struct attribute *attr,
 356                      const char *buf, size_t count)
 357 {
 358         struct threshold_block *b = to_block(kobj);
 359         struct threshold_attr *a = to_attr(attr);
 360         ssize_t ret;
 361         ret = a->store ? a->store(b, buf, count) : -EIO;
 362         return ret;
 363 }
 364
 365 static struct sysfs_ops threshold_ops = {
 366         .show = show,
 367         .store = store,
 368 };
 369
 370 static struct kobj_type threshold_ktype = {
 371         .sysfs_ops = &threshold_ops,
 372         .default_attrs = default_attrs,
 373 };
 374
 375 static __cpuinit int allocate_threshold_blocks(unsigned int cpu,
 376                                                unsigned int bank,
 377                                                unsigned int block,
 378                                                u32 address)
 379 {
 380         int err;
 381         u32 low, high;
 382         struct threshold_block *b = NULL;
 383
 384         if ((bank >= NR_BANKS) || (block >= NR_BLOCKS))
 385                 return 0;
 386
 387         if (rdmsr_safe(address, &low, &high))
 388                 goto recurse;
 389
 390         if (!(high & MASK_VALID_HI)) {
 391                 if (block)
 392                         goto recurse;
 393                 else
 394                         return 0;
 395         }
 396
 397         if (!(high & MASK_VALID_HI >> 1)  ||
 398              (high & MASK_VALID_HI >> 2))
 399                 goto recurse;
 400
 401         b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
 402         if (!b)
 403                 return -ENOMEM;
 404
 405         b->block = block;
 406         b->bank = bank;
 407         b->cpu = cpu;
 408         b->address = address;
 409         b->interrupt_enable = 0;
 410         b->threshold_limit = THRESHOLD_MAX;
 411
 412         INIT_LIST_HEAD(&b->miscj);
 413
 414         if (per_cpu(threshold_banks, cpu)[bank]->blocks)
 415                 list_add(&b->miscj,
 416                          &per_cpu(threshold_banks, cpu)[bank]->blocks->miscj);
 417         else
 418                 per_cpu(threshold_banks, cpu)[bank]->blocks = b;
 419
 420         kobject_set_name(&b->kobj, "misc%i", block);
 421         b->kobj.parent = &per_cpu(threshold_banks, cpu)[bank]->kobj;
 422         b->kobj.ktype = &threshold_ktype;
 423         err = kobject_register(&b->kobj);
 424         if (err)
 425                 goto out_free;
 426 recurse:
 427         if (!block) {
 428                 address = (low & MASK_BLKPTR_LO) >> 21;
 429                 if (!address)
 430                         return 0;
 431                 address += MCG_XBLK_ADDR;
 432         } else
 433                 ++address;
 434
 435         err = allocate_threshold_blocks(cpu, bank, ++block, address);
 436         if (err)
 437                 goto out_free;
 438
 439         return err;
 440
 441 out_free:
 442         if (b) {
 443                 kobject_unregister(&b->kobj);
 444                 kfree(b);
 445         }
 446         return err;
 447 }
 448
 449 /* symlinks sibling shared banks to first core.  first core owns dir/files. */
 450 static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank)
 451 {
 452         int i, err = 0;
 453         struct threshold_bank *b = NULL;
 454         cpumask_t oldmask = CPU_MASK_NONE;
 455         char name[32];
 456
 457         sprintf(name, "threshold_bank%i", bank);
 458
 459 #ifdef CONFIG_SMP
 460         if (cpu_data[cpu].cpu_core_id && shared_bank[bank]) {   /* symlink */
 461                 i = first_cpu(cpu_core_map[cpu]);
 462
 463                 /* first core not up yet */
 464                 if (cpu_data[i].cpu_core_id)
 465                         goto out;
 466
 467                 /* already linked */
 468                 if (per_cpu(threshold_banks, cpu)[bank])
 469                         goto out;
 470
 471                 b = per_cpu(threshold_banks, i)[bank];
 472
 473                 if (!b)
 474                         goto out;
 475
 476                 err = sysfs_create_link(&per_cpu(device_mce, cpu).kobj,
 477                                         &b->kobj, name);
 478                 if (err)
 479                         goto out;
 480
 481                 b->cpus = cpu_core_map[cpu];
 482                 per_cpu(threshold_banks, cpu)[bank] = b;
 483                 goto out;
 484         }
 485 #endif
 486
 487         b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
 488         if (!b) {
 489                 err = -ENOMEM;
 490                 goto out;
 491         }
 492
 493         kobject_set_name(&b->kobj, "threshold_bank%i", bank);
 494         b->kobj.parent = &per_cpu(device_mce, cpu).kobj;
 495 #ifndef CONFIG_SMP
 496         b->cpus = CPU_MASK_ALL;
 497 #else
 498         b->cpus = cpu_core_map[cpu];
 499 #endif
 500         err = kobject_register(&b->kobj);
 501         if (err)
 502                 goto out_free;
 503
 504         per_cpu(threshold_banks, cpu)[bank] = b;
 505
 506         oldmask = affinity_set(cpu);
 507         err = allocate_threshold_blocks(cpu, bank, 0,
 508                                         MSR_IA32_MC0_MISC + bank * 4);
 509         affinity_restore(oldmask);
 510
 511         if (err)
 512                 goto out_free;
 513
 514         for_each_cpu_mask(i, b->cpus) {
 515                 if (i == cpu)
 516                         continue;
 517
 518                 err = sysfs_create_link(&per_cpu(device_mce, i).kobj,
 519                                         &b->kobj, name);
 520                 if (err)
 521                         goto out;
 522
 523                 per_cpu(threshold_banks, i)[bank] = b;
 524         }
 525
 526         goto out;
 527
 528 out_free:
 529         per_cpu(threshold_banks, cpu)[bank] = NULL;
 530         kfree(b);
 531 out:
 532         return err;
 533 }
 534
 535 /* create dir/files for all valid threshold banks */
 536 static __cpuinit int threshold_create_device(unsigned int cpu)
 537 {
 538         unsigned int bank;
 539         int err = 0;
 540
 541         for (bank = 0; bank < NR_BANKS; ++bank) {
 542                 if (!(per_cpu(bank_map, cpu) & 1 << bank))
 543                         continue;
 544                 err = threshold_create_bank(cpu, bank);
 545                 if (err)
 546                         goto out;
 547         }
 548 out:
 549         return err;
 550 }
 551
 552 /*
 553  * let's be hotplug friendly.
 554  * in case of multiple core processors, the first core always takes ownership
 555  *   of shared sysfs dir/files, and rest of the cores will be symlinked to it.
 556  */
 557
 558 static void deallocate_threshold_block(unsigned int cpu,
 559                                                  unsigned int bank)
 560 {
 561         struct threshold_block *pos = NULL;
 562         struct threshold_block *tmp = NULL;
 563         struct threshold_bank *head = per_cpu(threshold_banks, cpu)[bank];
 564
 565         if (!head)
 566                 return;
 567
 568         list_for_each_entry_safe(pos, tmp, &head->blocks->miscj, miscj) {
 569                 kobject_unregister(&pos->kobj);
 570                 list_del(&pos->miscj);
 571                 kfree(pos);
 572         }
 573
 574         kfree(per_cpu(threshold_banks, cpu)[bank]->blocks);
 575         per_cpu(threshold_banks, cpu)[bank]->blocks = NULL;
 576 }
 577
 578 static void threshold_remove_bank(unsigned int cpu, int bank)
 579 {
 580         int i = 0;
 581         struct threshold_bank *b;
 582         char name[32];
 583
 584         b = per_cpu(threshold_banks, cpu)[bank];
 585
 586         if (!b)
 587                 return;
 588
 589         if (!b->blocks)
 590                 goto free_out;
 591
 592         sprintf(name, "threshold_bank%i", bank);
 593
 594 #ifdef CONFIG_SMP
 595         /* sibling symlink */
 596         if (shared_bank[bank] && b->blocks->cpu != cpu) {
 597                 sysfs_remove_link(&per_cpu(device_mce, cpu).kobj, name);
 598                 per_cpu(threshold_banks, cpu)[bank] = NULL;
 599                 return;
 600         }
 601 #endif
 602
 603         /* remove all sibling symlinks before unregistering */
 604         for_each_cpu_mask(i, b->cpus) {
 605                 if (i == cpu)
 606                         continue;
 607
 608                 sysfs_remove_link(&per_cpu(device_mce, i).kobj, name);
 609                 per_cpu(threshold_banks, i)[bank] = NULL;
 610         }
 611
 612         deallocate_threshold_block(cpu, bank);
 613
 614 free_out:
 615         kobject_unregister(&b->kobj);
 616         kfree(b);
 617         per_cpu(threshold_banks, cpu)[bank] = NULL;
 618 }
 619
 620 static void threshold_remove_device(unsigned int cpu)
 621 {
 622         unsigned int bank;
 623
 624         for (bank = 0; bank < NR_BANKS; ++bank) {
 625                 if (!(per_cpu(bank_map, cpu) & 1 << bank))
 626                         continue;
 627                 threshold_remove_bank(cpu, bank);
 628         }
 629 }
 630
 631 /* get notified when a cpu comes on/off */
 632 static int threshold_cpu_callback(struct notifier_block *nfb,
 633                                             unsigned long action, void *hcpu)
 634 {
 635         /* cpu was unsigned int to begin with */
 636         unsigned int cpu = (unsigned long)hcpu;
 637
 638         if (cpu >= NR_CPUS)
 639                 goto out;
 640
 641         switch (action) {
 642         case CPU_ONLINE:
 643                 threshold_create_device(cpu);
 644                 break;
 645         case CPU_DEAD:
 646                 threshold_remove_device(cpu);
 647                 break;
 648         default:
 649                 break;
 650         }
 651       out:
 652         return NOTIFY_OK;
 653 }
 654
 655 static struct notifier_block threshold_cpu_notifier = {
 656         .notifier_call = threshold_cpu_callback,
 657 };
 658
 659 static __init int threshold_init_device(void)
 660 {
 661         unsigned lcpu = 0;
 662
 663         /* to hit CPUs online before the notifier is up */
 664         for_each_online_cpu(lcpu) {
 665                 int err = threshold_create_device(lcpu);
 666                 if (err)
 667                         return err;
 668         }
 669         register_hotcpu_notifier(&threshold_cpu_notifier);
 670         return 0;
 671 }
 672
 673 device_initcall(threshold_init_device);