arch/x86_64/kernel/mce_amd.c

   1 /*
   2  *  (c) 2005 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  *  MC4_MISC0 DRAM ECC Error Threshold available under AMD K8 Rev F.
  12  *  MC4_MISC0 exists per physical processor.
  13  *
  14  */
  15
  16 #include <linux/cpu.h>
  17 #include <linux/errno.h>
  18 #include <linux/init.h>
  19 #include <linux/interrupt.h>
  20 #include <linux/kobject.h>
  21 #include <linux/notifier.h>
  22 #include <linux/sched.h>
  23 #include <linux/smp.h>
  24 #include <linux/sysdev.h>
  25 #include <linux/sysfs.h>
  26 #include <asm/apic.h>
  27 #include <asm/mce.h>
  28 #include <asm/msr.h>
  29 #include <asm/percpu.h>
  30
  31 #define PFX "mce_threshold: "
  32 #define VERSION "version 1.00.9"
  33 #define NR_BANKS 5
  34 #define THRESHOLD_MAX 0xFFF
  35 #define INT_TYPE_APIC 0x00020000
  36 #define MASK_VALID_HI 0x80000000
  37 #define MASK_LVTOFF_HI 0x00F00000
  38 #define MASK_COUNT_EN_HI 0x00080000
  39 #define MASK_INT_TYPE_HI 0x00060000
  40 #define MASK_OVERFLOW_HI 0x00010000
  41 #define MASK_ERR_COUNT_HI 0x00000FFF
  42 #define MASK_OVERFLOW 0x0001000000000000L
  43
  44 struct threshold_bank {
  45         unsigned int cpu;
  46         u8 bank;
  47         u8 interrupt_enable;
  48         u16 threshold_limit;
  49         struct kobject kobj;
  50 };
  51
  52 static struct threshold_bank threshold_defaults = {
  53         .interrupt_enable = 0,
  54         .threshold_limit = THRESHOLD_MAX,
  55 };
  56
  57 #ifdef CONFIG_SMP
  58 static unsigned char shared_bank[NR_BANKS] = {
  59         0, 0, 0, 0, 1
  60 };
  61 #endif
  62
  63 static DEFINE_PER_CPU(unsigned char, bank_map); /* see which banks are on */
  64
  65 /*
  66  * CPU Initialization
  67  */
  68
  69 /* must be called with correct cpu affinity */
  70 static void threshold_restart_bank(struct threshold_bank *b,
  71                                    int reset, u16 old_limit)
  72 {
  73         u32 mci_misc_hi, mci_misc_lo;
  74
  75         rdmsr(MSR_IA32_MC0_MISC + b->bank * 4, mci_misc_lo, mci_misc_hi);
  76
  77         if (b->threshold_limit < (mci_misc_hi & THRESHOLD_MAX))
  78                 reset = 1;      /* limit cannot be lower than err count */
  79
  80         if (reset) {            /* reset err count and overflow bit */
  81                 mci_misc_hi =
  82                     (mci_misc_hi & ~(MASK_ERR_COUNT_HI | MASK_OVERFLOW_HI)) |
  83                     (THRESHOLD_MAX - b->threshold_limit);
  84         } else if (old_limit) { /* change limit w/o reset */
  85                 int new_count = (mci_misc_hi & THRESHOLD_MAX) +
  86                     (old_limit - b->threshold_limit);
  87                 mci_misc_hi = (mci_misc_hi & ~MASK_ERR_COUNT_HI) |
  88                     (new_count & THRESHOLD_MAX);
  89         }
  90
  91         b->interrupt_enable ?
  92             (mci_misc_hi = (mci_misc_hi & ~MASK_INT_TYPE_HI) | INT_TYPE_APIC) :
  93             (mci_misc_hi &= ~MASK_INT_TYPE_HI);
  94
  95         mci_misc_hi |= MASK_COUNT_EN_HI;
  96         wrmsr(MSR_IA32_MC0_MISC + b->bank * 4, mci_misc_lo, mci_misc_hi);
  97 }
  98
  99 void __cpuinit mce_amd_feature_init(struct cpuinfo_x86 *c)
 100 {
 101         int bank;
 102         u32 mci_misc_lo, mci_misc_hi;
 103         unsigned int cpu = smp_processor_id();
 104
 105         for (bank = 0; bank < NR_BANKS; ++bank) {
 106                 rdmsr(MSR_IA32_MC0_MISC + bank * 4, mci_misc_lo, mci_misc_hi);
 107
 108                 /* !valid, !counter present, bios locked */
 109                 if (!(mci_misc_hi & MASK_VALID_HI) ||
 110                     !(mci_misc_hi & MASK_VALID_HI >> 1) ||
 111                     (mci_misc_hi & MASK_VALID_HI >> 2))
 112                         continue;
 113
 114                 per_cpu(bank_map, cpu) |= (1 << bank);
 115
 116 #ifdef CONFIG_SMP
 117                 if (shared_bank[bank] && cpu_core_id[cpu])
 118                         continue;
 119 #endif
 120
 121                 setup_threshold_lvt((mci_misc_hi & MASK_LVTOFF_HI) >> 20);
 122                 threshold_defaults.cpu = cpu;
 123                 threshold_defaults.bank = bank;
 124                 threshold_restart_bank(&threshold_defaults, 0, 0);
 125         }
 126 }
 127
 128 /*
 129  * APIC Interrupt Handler
 130  */
 131
 132 /*
 133  * threshold interrupt handler will service THRESHOLD_APIC_VECTOR.
 134  * the interrupt goes off when error_count reaches threshold_limit.
 135  * the handler will simply log mcelog w/ software defined bank number.
 136  */
 137 asmlinkage void mce_threshold_interrupt(void)
 138 {
 139         int bank;
 140         struct mce m;
 141
 142         ack_APIC_irq();
 143         irq_enter();
 144
 145         memset(&m, 0, sizeof(m));
 146         rdtscll(m.tsc);
 147         m.cpu = smp_processor_id();
 148
 149         /* assume first bank caused it */
 150         for (bank = 0; bank < NR_BANKS; ++bank) {
 151                 m.bank = MCE_THRESHOLD_BASE + bank;
 152                 rdmsrl(MSR_IA32_MC0_MISC + bank * 4, m.misc);
 153
 154                 if (m.misc & MASK_OVERFLOW) {
 155                         mce_log(&m);
 156                         goto out;
 157                 }
 158         }
 159       out:
 160         irq_exit();
 161 }
 162
 163 /*
 164  * Sysfs Interface
 165  */
 166
 167 static struct sysdev_class threshold_sysclass = {
 168         set_kset_name("threshold"),
 169 };
 170
 171 static DEFINE_PER_CPU(struct sys_device, device_threshold);
 172
 173 struct threshold_attr {
 174         struct attribute attr;
 175         ssize_t(*show) (struct threshold_bank *, char *);
 176         ssize_t(*store) (struct threshold_bank *, const char *, size_t count);
 177 };
 178
 179 static DEFINE_PER_CPU(struct threshold_bank *, threshold_banks[NR_BANKS]);
 180
 181 static cpumask_t affinity_set(unsigned int cpu)
 182 {
 183         cpumask_t oldmask = current->cpus_allowed;
 184         cpumask_t newmask = CPU_MASK_NONE;
 185         cpu_set(cpu, newmask);
 186         set_cpus_allowed(current, newmask);
 187         return oldmask;
 188 }
 189
 190 static void affinity_restore(cpumask_t oldmask)
 191 {
 192         set_cpus_allowed(current, oldmask);
 193 }
 194
 195 #define SHOW_FIELDS(name) \
 196         static ssize_t show_ ## name(struct threshold_bank * b, char *buf) \
 197         { \
 198                 return sprintf(buf, "%lx\n", (unsigned long) b->name); \
 199         }
 200 SHOW_FIELDS(interrupt_enable)
 201 SHOW_FIELDS(threshold_limit)
 202
 203 static ssize_t store_interrupt_enable(struct threshold_bank *b,
 204                                       const char *buf, size_t count)
 205 {
 206         char *end;
 207         cpumask_t oldmask;
 208         unsigned long new = simple_strtoul(buf, &end, 0);
 209         if (end == buf)
 210                 return -EINVAL;
 211         b->interrupt_enable = !!new;
 212
 213         oldmask = affinity_set(b->cpu);
 214         threshold_restart_bank(b, 0, 0);
 215         affinity_restore(oldmask);
 216
 217         return end - buf;
 218 }
 219
 220 static ssize_t store_threshold_limit(struct threshold_bank *b,
 221                                      const char *buf, size_t count)
 222 {
 223         char *end;
 224         cpumask_t oldmask;
 225         u16 old;
 226         unsigned long new = simple_strtoul(buf, &end, 0);
 227         if (end == buf)
 228                 return -EINVAL;
 229         if (new > THRESHOLD_MAX)
 230                 new = THRESHOLD_MAX;
 231         if (new < 1)
 232                 new = 1;
 233         old = b->threshold_limit;
 234         b->threshold_limit = new;
 235
 236         oldmask = affinity_set(b->cpu);
 237         threshold_restart_bank(b, 0, old);
 238         affinity_restore(oldmask);
 239
 240         return end - buf;
 241 }
 242
 243 static ssize_t show_error_count(struct threshold_bank *b, char *buf)
 244 {
 245         u32 high, low;
 246         cpumask_t oldmask;
 247         oldmask = affinity_set(b->cpu);
 248         rdmsr(MSR_IA32_MC0_MISC + b->bank * 4, low, high); /* ignore low 32 */
 249         affinity_restore(oldmask);
 250         return sprintf(buf, "%x\n",
 251                        (high & 0xFFF) - (THRESHOLD_MAX - b->threshold_limit));
 252 }
 253
 254 static ssize_t store_error_count(struct threshold_bank *b,
 255                                  const char *buf, size_t count)
 256 {
 257         cpumask_t oldmask;
 258         oldmask = affinity_set(b->cpu);
 259         threshold_restart_bank(b, 1, 0);
 260         affinity_restore(oldmask);
 261         return 1;
 262 }
 263
 264 #define THRESHOLD_ATTR(_name,_mode,_show,_store) {            \
 265         .attr = {.name = __stringify(_name), .mode = _mode }, \
 266         .show = _show,                                        \
 267         .store = _store,                                      \
 268 };
 269
 270 #define ATTR_FIELDS(name) \
 271         static struct threshold_attr name = \
 272         THRESHOLD_ATTR(name, 0644, show_## name, store_## name)
 273
 274 ATTR_FIELDS(interrupt_enable);
 275 ATTR_FIELDS(threshold_limit);
 276 ATTR_FIELDS(error_count);
 277
 278 static struct attribute *default_attrs[] = {
 279         &interrupt_enable.attr,
 280         &threshold_limit.attr,
 281         &error_count.attr,
 282         NULL
 283 };
 284
 285 #define to_bank(k) container_of(k,struct threshold_bank,kobj)
 286 #define to_attr(a) container_of(a,struct threshold_attr,attr)
 287
 288 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
 289 {
 290         struct threshold_bank *b = to_bank(kobj);
 291         struct threshold_attr *a = to_attr(attr);
 292         ssize_t ret;
 293         ret = a->show ? a->show(b, buf) : -EIO;
 294         return ret;
 295 }
 296
 297 static ssize_t store(struct kobject *kobj, struct attribute *attr,
 298                      const char *buf, size_t count)
 299 {
 300         struct threshold_bank *b = to_bank(kobj);
 301         struct threshold_attr *a = to_attr(attr);
 302         ssize_t ret;
 303         ret = a->store ? a->store(b, buf, count) : -EIO;
 304         return ret;
 305 }
 306
 307 static struct sysfs_ops threshold_ops = {
 308         .show = show,
 309         .store = store,
 310 };
 311
 312 static struct kobj_type threshold_ktype = {
 313         .sysfs_ops = &threshold_ops,
 314         .default_attrs = default_attrs,
 315 };
 316
 317 /* symlinks sibling shared banks to first core.  first core owns dir/files. */
 318 static __cpuinit int threshold_create_bank(unsigned int cpu, int bank)
 319 {
 320         int err = 0;
 321         struct threshold_bank *b = 0;
 322
 323 #ifdef CONFIG_SMP
 324         if (cpu_core_id[cpu] && shared_bank[bank]) {    /* symlink */
 325                 char name[16];
 326                 unsigned lcpu = first_cpu(cpu_core_map[cpu]);
 327                 if (cpu_core_id[lcpu])
 328                         goto out;       /* first core not up yet */
 329
 330                 b = per_cpu(threshold_banks, lcpu)[bank];
 331                 if (!b)
 332                         goto out;
 333                 sprintf(name, "bank%i", bank);
 334                 err = sysfs_create_link(&per_cpu(device_threshold, cpu).kobj,
 335                                         &b->kobj, name);
 336                 if (err)
 337                         goto out;
 338                 per_cpu(threshold_banks, cpu)[bank] = b;
 339                 goto out;
 340         }
 341 #endif
 342
 343         b = kmalloc(sizeof(struct threshold_bank), GFP_KERNEL);
 344         if (!b) {
 345                 err = -ENOMEM;
 346                 goto out;
 347         }
 348         memset(b, 0, sizeof(struct threshold_bank));
 349
 350         b->cpu = cpu;
 351         b->bank = bank;
 352         b->interrupt_enable = 0;
 353         b->threshold_limit = THRESHOLD_MAX;
 354         kobject_set_name(&b->kobj, "bank%i", bank);
 355         b->kobj.parent = &per_cpu(device_threshold, cpu).kobj;
 356         b->kobj.ktype = &threshold_ktype;
 357
 358         err = kobject_register(&b->kobj);
 359         if (err) {
 360                 kfree(b);
 361                 goto out;
 362         }
 363         per_cpu(threshold_banks, cpu)[bank] = b;
 364       out:
 365         return err;
 366 }
 367
 368 /* create dir/files for all valid threshold banks */
 369 static __cpuinit int threshold_create_device(unsigned int cpu)
 370 {
 371         int bank;
 372         int err = 0;
 373
 374         per_cpu(device_threshold, cpu).id = cpu;
 375         per_cpu(device_threshold, cpu).cls = &threshold_sysclass;
 376         err = sysdev_register(&per_cpu(device_threshold, cpu));
 377         if (err)
 378                 goto out;
 379
 380         for (bank = 0; bank < NR_BANKS; ++bank) {
 381                 if (!(per_cpu(bank_map, cpu) & 1 << bank))
 382                         continue;
 383                 err = threshold_create_bank(cpu, bank);
 384                 if (err)
 385                         goto out;
 386         }
 387       out:
 388         return err;
 389 }
 390
 391 #ifdef CONFIG_HOTPLUG_CPU
 392 /*
 393  * let's be hotplug friendly.
 394  * in case of multiple core processors, the first core always takes ownership
 395  *   of shared sysfs dir/files, and rest of the cores will be symlinked to it.
 396  */
 397
 398 /* cpu hotplug call removes all symlinks before first core dies */
 399 static __cpuinit void threshold_remove_bank(unsigned int cpu, int bank)
 400 {
 401         struct threshold_bank *b;
 402         char name[16];
 403
 404         b = per_cpu(threshold_banks, cpu)[bank];
 405         if (!b)
 406                 return;
 407         if (shared_bank[bank] && atomic_read(&b->kobj.kref.refcount) > 2) {
 408                 sprintf(name, "bank%i", bank);
 409                 sysfs_remove_link(&per_cpu(device_threshold, cpu).kobj, name);
 410                 per_cpu(threshold_banks, cpu)[bank] = 0;
 411         } else {
 412                 kobject_unregister(&b->kobj);
 413                 kfree(per_cpu(threshold_banks, cpu)[bank]);
 414         }
 415 }
 416
 417 static __cpuinit void threshold_remove_device(unsigned int cpu)
 418 {
 419         int bank;
 420
 421         for (bank = 0; bank < NR_BANKS; ++bank) {
 422                 if (!(per_cpu(bank_map, cpu) & 1 << bank))
 423                         continue;
 424                 threshold_remove_bank(cpu, bank);
 425         }
 426         sysdev_unregister(&per_cpu(device_threshold, cpu));
 427 }
 428
 429 /* link all existing siblings when first core comes up */
 430 static __cpuinit int threshold_create_symlinks(unsigned int cpu)
 431 {
 432         int bank, err = 0;
 433         unsigned int lcpu = 0;
 434
 435         if (cpu_core_id[cpu])
 436                 return 0;
 437         for_each_cpu_mask(lcpu, cpu_core_map[cpu]) {
 438                 if (lcpu == cpu)
 439                         continue;
 440                 for (bank = 0; bank < NR_BANKS; ++bank) {
 441                         if (!(per_cpu(bank_map, cpu) & 1 << bank))
 442                                 continue;
 443                         if (!shared_bank[bank])
 444                                 continue;
 445                         err = threshold_create_bank(lcpu, bank);
 446                 }
 447         }
 448         return err;
 449 }
 450
 451 /* remove all symlinks before first core dies. */
 452 static __cpuinit void threshold_remove_symlinks(unsigned int cpu)
 453 {
 454         int bank;
 455         unsigned int lcpu = 0;
 456         if (cpu_core_id[cpu])
 457                 return;
 458         for_each_cpu_mask(lcpu, cpu_core_map[cpu]) {
 459                 if (lcpu == cpu)
 460                         continue;
 461                 for (bank = 0; bank < NR_BANKS; ++bank) {
 462                         if (!(per_cpu(bank_map, cpu) & 1 << bank))
 463                                 continue;
 464                         if (!shared_bank[bank])
 465                                 continue;
 466                         threshold_remove_bank(lcpu, bank);
 467                 }
 468         }
 469 }
 470 #else /* !CONFIG_HOTPLUG_CPU */
 471 static __cpuinit void threshold_create_symlinks(unsigned int cpu)
 472 {
 473 }
 474 static __cpuinit void threshold_remove_symlinks(unsigned int cpu)
 475 {
 476 }
 477 static void threshold_remove_device(unsigned int cpu)
 478 {
 479 }
 480 #endif
 481
 482 /* get notified when a cpu comes on/off */
 483 static __cpuinit int threshold_cpu_callback(struct notifier_block *nfb,
 484                                             unsigned long action, void *hcpu)
 485 {
 486         /* cpu was unsigned int to begin with */
 487         unsigned int cpu = (unsigned long)hcpu;
 488
 489         if (cpu >= NR_CPUS)
 490                 goto out;
 491
 492         switch (action) {
 493         case CPU_ONLINE:
 494                 threshold_create_device(cpu);
 495                 threshold_create_symlinks(cpu);
 496                 break;
 497         case CPU_DOWN_PREPARE:
 498                 threshold_remove_symlinks(cpu);
 499                 break;
 500         case CPU_DOWN_FAILED:
 501                 threshold_create_symlinks(cpu);
 502                 break;
 503         case CPU_DEAD:
 504                 threshold_remove_device(cpu);
 505                 break;
 506         default:
 507                 break;
 508         }
 509       out:
 510         return NOTIFY_OK;
 511 }
 512
 513 static struct notifier_block threshold_cpu_notifier = {
 514         .notifier_call = threshold_cpu_callback,
 515 };
 516
 517 static __init int threshold_init_device(void)
 518 {
 519         int err;
 520         int lcpu = 0;
 521
 522         err = sysdev_class_register(&threshold_sysclass);
 523         if (err)
 524                 goto out;
 525
 526         /* to hit CPUs online before the notifier is up */
 527         for_each_online_cpu(lcpu) {
 528                 err = threshold_create_device(lcpu);
 529                 if (err)
 530                         goto out;
 531         }
 532         register_cpu_notifier(&threshold_cpu_notifier);
 533
 534       out:
 535         return err;
 536 }
 537
 538 device_initcall(threshold_init_device);