arch/x86_64/kernel/nmi.c

   1 /*
   2  *  linux/arch/x86_64/nmi.c
   3  *
   4  *  NMI watchdog support on APIC systems
   5  *
   6  *  Started by Ingo Molnar <mingo@redhat.com>
   7  *
   8  *  Fixes:
   9  *  Mikael Pettersson   : AMD K7 support for local APIC NMI watchdog.
  10  *  Mikael Pettersson   : Power Management for local APIC NMI watchdog.
  11  *  Pavel Machek and
  12  *  Mikael Pettersson   : PM converted to driver model. Disable/enable API.
  13  */
  14
  15 #include <linux/config.h>
  16 #include <linux/mm.h>
  17 #include <linux/irq.h>
  18 #include <linux/delay.h>
  19 #include <linux/bootmem.h>
  20 #include <linux/smp_lock.h>
  21 #include <linux/interrupt.h>
  22 #include <linux/mc146818rtc.h>
  23 #include <linux/kernel_stat.h>
  24 #include <linux/module.h>
  25 #include <linux/sysdev.h>
  26 #include <linux/nmi.h>
  27
  28 #include <asm/smp.h>
  29 #include <asm/mtrr.h>
  30 #include <asm/mpspec.h>
  31 #include <asm/nmi.h>
  32 #include <asm/msr.h>
  33 #include <asm/proto.h>
  34 #include <asm/kdebug.h>
  35
  36 /*
  37  * lapic_nmi_owner tracks the ownership of the lapic NMI hardware:
  38  * - it may be reserved by some other driver, or not
  39  * - when not reserved by some other driver, it may be used for
  40  *   the NMI watchdog, or not
  41  *
  42  * This is maintained separately from nmi_active because the NMI
  43  * watchdog may also be driven from the I/O APIC timer.
  44  */
  45 static spinlock_t lapic_nmi_owner_lock = SPIN_LOCK_UNLOCKED;
  46 static unsigned int lapic_nmi_owner;
  47 #define LAPIC_NMI_WATCHDOG      (1<<0)
  48 #define LAPIC_NMI_RESERVED      (1<<1)
  49
  50 /* nmi_active:
  51  * +1: the lapic NMI watchdog is active, but can be disabled
  52  *  0: the lapic NMI watchdog has not been set up, and cannot
  53  *     be enabled
  54  * -1: the lapic NMI watchdog is disabled, but can be enabled
  55  */
  56 int nmi_active;         /* oprofile uses this */
  57 static int panic_on_timeout;
  58
  59 unsigned int nmi_watchdog = NMI_DEFAULT;
  60 static unsigned int nmi_hz = HZ;
  61 unsigned int nmi_perfctr_msr;   /* the MSR to reset in NMI handler */
  62
  63 /* Note that these events don't tick when the CPU idles. This means
  64    the frequency varies with CPU load. */
  65
  66 #define K7_EVNTSEL_ENABLE       (1 << 22)
  67 #define K7_EVNTSEL_INT          (1 << 20)
  68 #define K7_EVNTSEL_OS           (1 << 17)
  69 #define K7_EVNTSEL_USR          (1 << 16)
  70 #define K7_EVENT_CYCLES_PROCESSOR_IS_RUNNING    0x76
  71 #define K7_NMI_EVENT            K7_EVENT_CYCLES_PROCESSOR_IS_RUNNING
  72
  73 #define P6_EVNTSEL0_ENABLE      (1 << 22)
  74 #define P6_EVNTSEL_INT          (1 << 20)
  75 #define P6_EVNTSEL_OS           (1 << 17)
  76 #define P6_EVNTSEL_USR          (1 << 16)
  77 #define P6_EVENT_CPU_CLOCKS_NOT_HALTED  0x79
  78 #define P6_NMI_EVENT            P6_EVENT_CPU_CLOCKS_NOT_HALTED
  79
  80 /* Run after command line and cpu_init init, but before all other checks */
  81 void __init nmi_watchdog_default(void)
  82 {
  83         if (nmi_watchdog != NMI_DEFAULT)
  84                 return;
  85
  86         /* For some reason the IO APIC watchdog doesn't work on the AMD
  87            8111 chipset. For now switch to local APIC mode using
  88            perfctr0 there.  On Intel CPUs we don't have code to handle
  89            the perfctr and the IO-APIC seems to work, so use that.  */
  90
  91         if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
  92                 nmi_watchdog = NMI_LOCAL_APIC;
  93                 printk(KERN_INFO
  94               "Using local APIC NMI watchdog using perfctr0\n");
  95         } else {
  96                 printk(KERN_INFO "Using IO APIC NMI watchdog\n");
  97                 nmi_watchdog = NMI_IO_APIC;
  98         }
  99 }
 100
 101 /* Why is there no CPUID flag for this? */
 102 static __init int cpu_has_lapic(void)
 103 {
 104         switch (boot_cpu_data.x86_vendor) {
 105         case X86_VENDOR_INTEL:
 106         case X86_VENDOR_AMD:
 107                 return boot_cpu_data.x86 >= 6;
 108         /* .... add more cpus here or find a different way to figure this out. */
 109         default:
 110                 return 0;
 111         }
 112 }
 113
 114 int __init check_nmi_watchdog (void)
 115 {
 116         int counts[NR_CPUS];
 117         int cpu;
 118
 119         if (nmi_watchdog == NMI_LOCAL_APIC && !cpu_has_lapic())  {
 120                 nmi_watchdog = NMI_NONE;
 121                 return -1;
 122         }
 123
 124         printk(KERN_INFO "testing NMI watchdog ... ");
 125
 126         for (cpu = 0; cpu < NR_CPUS; cpu++)
 127                 counts[cpu] = cpu_pda[cpu].__nmi_count;
 128         local_irq_enable();
 129         mdelay((10*1000)/nmi_hz); // wait 10 ticks
 130
 131         for (cpu = 0; cpu < NR_CPUS; cpu++) {
 132                 if (!cpu_online(cpu))
 133                         continue;
 134                 if (cpu_pda[cpu].__nmi_count - counts[cpu] <= 5) {
 135                         printk("CPU#%d: NMI appears to be stuck (%d)!\n",
 136                                cpu,
 137                                cpu_pda[cpu].__nmi_count);
 138                         nmi_active = 0;
 139                         lapic_nmi_owner &= ~LAPIC_NMI_WATCHDOG;
 140                         return -1;
 141                 }
 142         }
 143         printk("OK.\n");
 144
 145         /* now that we know it works we can reduce NMI frequency to
 146            something more reasonable; makes a difference in some configs */
 147         if (nmi_watchdog == NMI_LOCAL_APIC)
 148                 nmi_hz = 1;
 149
 150         return 0;
 151 }
 152
 153 int __init setup_nmi_watchdog(char *str)
 154 {
 155         int nmi;
 156
 157         if (!strncmp(str,"panic",5)) {
 158                 panic_on_timeout = 1;
 159                 str = strchr(str, ',');
 160                 if (!str)
 161                         return 1;
 162                 ++str;
 163         }
 164
 165         get_option(&str, &nmi);
 166
 167         if (nmi >= NMI_INVALID)
 168                 return 0;
 169                 nmi_watchdog = nmi;
 170         return 1;
 171 }
 172
 173 __setup("nmi_watchdog=", setup_nmi_watchdog);
 174
 175 static void disable_lapic_nmi_watchdog(void)
 176 {
 177         if (nmi_active <= 0)
 178                 return;
 179         switch (boot_cpu_data.x86_vendor) {
 180         case X86_VENDOR_AMD:
 181                 wrmsr(MSR_K7_EVNTSEL0, 0, 0);
 182                 break;
 183         case X86_VENDOR_INTEL:
 184                 wrmsr(MSR_IA32_EVNTSEL0, 0, 0);
 185                 break;
 186         }
 187         nmi_active = -1;
 188         /* tell do_nmi() and others that we're not active any more */
 189         nmi_watchdog = 0;
 190 }
 191
 192 static void enable_lapic_nmi_watchdog(void)
 193 {
 194         if (nmi_active < 0) {
 195                 nmi_watchdog = NMI_LOCAL_APIC;
 196                 setup_apic_nmi_watchdog();
 197         }
 198 }
 199
 200 int reserve_lapic_nmi(void)
 201 {
 202         unsigned int old_owner;
 203
 204         spin_lock(&lapic_nmi_owner_lock);
 205         old_owner = lapic_nmi_owner;
 206         lapic_nmi_owner |= LAPIC_NMI_RESERVED;
 207         spin_unlock(&lapic_nmi_owner_lock);
 208         if (old_owner & LAPIC_NMI_RESERVED)
 209                 return -EBUSY;
 210         if (old_owner & LAPIC_NMI_WATCHDOG)
 211                 disable_lapic_nmi_watchdog();
 212         return 0;
 213 }
 214
 215 void release_lapic_nmi(void)
 216 {
 217         unsigned int new_owner;
 218
 219         spin_lock(&lapic_nmi_owner_lock);
 220         new_owner = lapic_nmi_owner & ~LAPIC_NMI_RESERVED;
 221         lapic_nmi_owner = new_owner;
 222         spin_unlock(&lapic_nmi_owner_lock);
 223         if (new_owner & LAPIC_NMI_WATCHDOG)
 224                 enable_lapic_nmi_watchdog();
 225 }
 226
 227 void disable_timer_nmi_watchdog(void)
 228 {
 229         if ((nmi_watchdog != NMI_IO_APIC) || (nmi_active <= 0))
 230                 return;
 231
 232         disable_irq(0);
 233         unset_nmi_callback();
 234         nmi_active = -1;
 235         nmi_watchdog = NMI_NONE;
 236 }
 237
 238 void enable_timer_nmi_watchdog(void)
 239 {
 240         if (nmi_active < 0) {
 241                 nmi_watchdog = NMI_IO_APIC;
 242                 touch_nmi_watchdog();
 243                 nmi_active = 1;
 244                 enable_irq(0);
 245         }
 246 }
 247
 248 #ifdef CONFIG_PM
 249
 250 static int nmi_pm_active; /* nmi_active before suspend */
 251
 252 static int lapic_nmi_suspend(struct sys_device *dev, u32 state)
 253 {
 254         nmi_pm_active = nmi_active;
 255         disable_lapic_nmi_watchdog();
 256         return 0;
 257 }
 258
 259 static int lapic_nmi_resume(struct sys_device *dev)
 260 {
 261         if (nmi_pm_active > 0)
 262         enable_lapic_nmi_watchdog();
 263         return 0;
 264 }
 265
 266 static struct sysdev_class nmi_sysclass = {
 267         set_kset_name("lapic_nmi"),
 268         .resume         = lapic_nmi_resume,
 269         .suspend        = lapic_nmi_suspend,
 270 };
 271
 272 static struct sys_device device_lapic_nmi = {
 273         .id             = 0,
 274         .cls    = &nmi_sysclass,
 275 };
 276
 277 static int __init init_lapic_nmi_sysfs(void)
 278 {
 279         int error;
 280
 281         if (nmi_active == 0 || nmi_watchdog != NMI_LOCAL_APIC)
 282                 return 0;
 283
 284         error = sysdev_class_register(&nmi_sysclass);
 285         if (!error)
 286                 error = sysdev_register(&device_lapic_nmi);
 287         return error;
 288 }
 289 /* must come after the local APIC's device_initcall() */
 290 late_initcall(init_lapic_nmi_sysfs);
 291
 292 #endif  /* CONFIG_PM */
 293
 294 /*
 295  * Activate the NMI watchdog via the local APIC.
 296  * Original code written by Keith Owens.
 297  */
 298
 299 static void setup_k7_watchdog(void)
 300 {
 301         int i;
 302         unsigned int evntsel;
 303
 304         /* No check, so can start with slow frequency */
 305         nmi_hz = 1;
 306
 307         /* XXX should check these in EFER */
 308
 309         nmi_perfctr_msr = MSR_K7_PERFCTR0;
 310
 311         for(i = 0; i < 4; ++i) {
 312                 /* Simulator may not support it */
 313                 if (checking_wrmsrl(MSR_K7_EVNTSEL0+i, 0UL))
 314                         return;
 315                 wrmsrl(MSR_K7_PERFCTR0+i, 0UL);
 316         }
 317
 318         evntsel = K7_EVNTSEL_INT
 319                 | K7_EVNTSEL_OS
 320                 | K7_EVNTSEL_USR
 321                 | K7_NMI_EVENT;
 322
 323         wrmsr(MSR_K7_EVNTSEL0, evntsel, 0);
 324         wrmsrl(MSR_K7_PERFCTR0, -((u64)cpu_khz*1000) / nmi_hz);
 325         apic_write(APIC_LVTPC, APIC_DM_NMI);
 326         evntsel |= K7_EVNTSEL_ENABLE;
 327         wrmsr(MSR_K7_EVNTSEL0, evntsel, 0);
 328 }
 329
 330 void setup_apic_nmi_watchdog(void)
 331 {
 332         switch (boot_cpu_data.x86_vendor) {
 333         case X86_VENDOR_AMD:
 334                 if (boot_cpu_data.x86 < 6)
 335                         return;
 336                 if (strstr(boot_cpu_data.x86_model_id, "Screwdriver"))
 337                         return;
 338                 setup_k7_watchdog();
 339                 break;
 340         default:
 341                 return;
 342         }
 343         lapic_nmi_owner = LAPIC_NMI_WATCHDOG;
 344         nmi_active = 1;
 345 }
 346
 347 static spinlock_t nmi_print_lock = SPIN_LOCK_UNLOCKED;
 348
 349 /*
 350  * the best way to detect whether a CPU has a 'hard lockup' problem
 351  * is to check it's local APIC timer IRQ counts. If they are not
 352  * changing then that CPU has some problem.
 353  *
 354  * as these watchdog NMI IRQs are generated on every CPU, we only
 355  * have to check the current processor.
 356  *
 357  * since NMIs don't listen to _any_ locks, we have to be extremely
 358  * careful not to rely on unsafe variables. The printk might lock
 359  * up though, so we have to break up any console locks first ...
 360  * [when there will be more tty-related locks, break them up
 361  *  here too!]
 362  */
 363
 364 static unsigned int
 365         last_irq_sums [NR_CPUS],
 366         alert_counter [NR_CPUS];
 367
 368 void touch_nmi_watchdog (void)
 369 {
 370         int i;
 371
 372         /*
 373          * Just reset the alert counters, (other CPUs might be
 374          * spinning on locks we hold):
 375          */
 376         for (i = 0; i < NR_CPUS; i++)
 377                 alert_counter[i] = 0;
 378 }
 379
 380 void nmi_watchdog_tick (struct pt_regs * regs, unsigned reason)
 381 {
 382         int sum, cpu;
 383
 384         cpu = safe_smp_processor_id();
 385         sum = read_pda(apic_timer_irqs);
 386         if (last_irq_sums[cpu] == sum) {
 387                 /*
 388                  * Ayiee, looks like this CPU is stuck ...
 389                  * wait a few IRQs (5 seconds) before doing the oops ...
 390                  */
 391                 alert_counter[cpu]++;
 392                 if (alert_counter[cpu] == 5*nmi_hz) {
 393                         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT)
 394                                                         == NOTIFY_STOP) {
 395                                 alert_counter[cpu] = 0;
 396                                 return;
 397                         }
 398                         spin_lock(&nmi_print_lock);
 399                         /*
 400                          * We are in trouble anyway, lets at least try
 401                          * to get a message out.
 402                          */
 403                         bust_spinlocks(1);
 404                         printk("NMI Watchdog detected LOCKUP on CPU%d, registers:\n", cpu);
 405                         show_registers(regs);
 406                         if (panic_on_timeout || panic_on_oops)
 407                                 panic("nmi watchdog");
 408                         printk("console shuts up ...\n");
 409                         console_silent();
 410                         spin_unlock(&nmi_print_lock);
 411                         bust_spinlocks(0);
 412                         do_exit(SIGSEGV);
 413                 }
 414         } else {
 415                 last_irq_sums[cpu] = sum;
 416                 alert_counter[cpu] = 0;
 417         }
 418         if (nmi_perfctr_msr)
 419                 wrmsr(nmi_perfctr_msr, -(cpu_khz/nmi_hz*1000), -1);
 420 }
 421
 422 static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
 423 {
 424         return 0;
 425 }
 426
 427 static nmi_callback_t nmi_callback = dummy_nmi_callback;
 428
 429 asmlinkage void do_nmi(struct pt_regs * regs, long error_code)
 430 {
 431         int cpu = safe_smp_processor_id();
 432
 433         nmi_enter();
 434         add_pda(__nmi_count,1);
 435         if (!nmi_callback(regs, cpu))
 436                 default_do_nmi(regs);
 437         nmi_exit();
 438 }
 439
 440 void set_nmi_callback(nmi_callback_t callback)
 441 {
 442         nmi_callback = callback;
 443 }
 444
 445 void unset_nmi_callback(void)
 446 {
 447         nmi_callback = dummy_nmi_callback;
 448 }
 449
 450 EXPORT_SYMBOL(nmi_active);
 451 EXPORT_SYMBOL(nmi_watchdog);
 452 EXPORT_SYMBOL(reserve_lapic_nmi);
 453 EXPORT_SYMBOL(release_lapic_nmi);
 454 EXPORT_SYMBOL(disable_timer_nmi_watchdog);
 455 EXPORT_SYMBOL(enable_timer_nmi_watchdog);
 456 EXPORT_SYMBOL(touch_nmi_watchdog);