arch/x86/kernel/reboot.c

   1 #include <linux/module.h>
   2 #include <linux/reboot.h>
   3 #include <linux/init.h>
   4 #include <linux/pm.h>
   5 #include <linux/efi.h>
   6 #include <acpi/reboot.h>
   7 #include <asm/io.h>
   8 #include <asm/apic.h>
   9 #include <asm/desc.h>
  10 #include <asm/hpet.h>
  11 #include <asm/pgtable.h>
  12 #include <asm/proto.h>
  13 #include <asm/reboot_fixups.h>
  14 #include <asm/reboot.h>
  15 #include <asm/pci_x86.h>
  16 #include <asm/virtext.h>
  17 #include <asm/cpu.h>
  18
  19 #ifdef CONFIG_X86_32
  20 # include <linux/dmi.h>
  21 # include <linux/ctype.h>
  22 # include <linux/mc146818rtc.h>
  23 #else
  24 # include <asm/iommu.h>
  25 #endif
  26
  27 /*
  28  * Power off function, if any
  29  */
  30 void (*pm_power_off)(void);
  31 EXPORT_SYMBOL(pm_power_off);
  32
  33 static const struct desc_ptr no_idt = {};
  34 static int reboot_mode;
  35 enum reboot_type reboot_type = BOOT_KBD;
  36 int reboot_force;
  37
  38 #if defined(CONFIG_X86_32) && defined(CONFIG_SMP)
  39 static int reboot_cpu = -1;
  40 #endif
  41
  42 /* This is set if we need to go through the 'emergency' path.
  43  * When machine_emergency_restart() is called, we may be on
  44  * an inconsistent state and won't be able to do a clean cleanup
  45  */
  46 static int reboot_emergency;
  47
  48 /* This is set by the PCI code if either type 1 or type 2 PCI is detected */
  49 bool port_cf9_safe = false;
  50
  51 /* reboot=b[ios] | s[mp] | t[riple] | k[bd] | e[fi] [, [w]arm | [c]old] | p[ci]
  52    warm   Don't set the cold reboot flag
  53    cold   Set the cold reboot flag
  54    bios   Reboot by jumping through the BIOS (only for X86_32)
  55    smp    Reboot by executing reset on BSP or other CPU (only for X86_32)
  56    triple Force a triple fault (init)
  57    kbd    Use the keyboard controller. cold reset (default)
  58    acpi   Use the RESET_REG in the FADT
  59    efi    Use efi reset_system runtime service
  60    pci    Use the so-called "PCI reset register", CF9
  61    force  Avoid anything that could hang.
  62  */
  63 static int __init reboot_setup(char *str)
  64 {
  65         for (;;) {
  66                 switch (*str) {
  67                 case 'w':
  68                         reboot_mode = 0x1234;
  69                         break;
  70
  71                 case 'c':
  72                         reboot_mode = 0;
  73                         break;
  74
  75 #ifdef CONFIG_X86_32
  76 #ifdef CONFIG_SMP
  77                 case 's':
  78                         if (isdigit(*(str+1))) {
  79                                 reboot_cpu = (int) (*(str+1) - '0');
  80                                 if (isdigit(*(str+2)))
  81                                         reboot_cpu = reboot_cpu*10 + (int)(*(str+2) - '0');
  82                         }
  83                                 /* we will leave sorting out the final value
  84                                    when we are ready to reboot, since we might not
  85                                    have set up boot_cpu_id or smp_num_cpu */
  86                         break;
  87 #endif /* CONFIG_SMP */
  88
  89                 case 'b':
  90 #endif
  91                 case 'a':
  92                 case 'k':
  93                 case 't':
  94                 case 'e':
  95                 case 'p':
  96                         reboot_type = *str;
  97                         break;
  98
  99                 case 'f':
 100                         reboot_force = 1;
 101                         break;
 102                 }
 103
 104                 str = strchr(str, ',');
 105                 if (str)
 106                         str++;
 107                 else
 108                         break;
 109         }
 110         return 1;
 111 }
 112
 113 __setup("reboot=", reboot_setup);
 114
 115
 116 #ifdef CONFIG_X86_32
 117 /*
 118  * Reboot options and system auto-detection code provided by
 119  * Dell Inc. so their systems "just work". :-)
 120  */
 121
 122 /*
 123  * Some machines require the "reboot=b"  commandline option,
 124  * this quirk makes that automatic.
 125  */
 126 static int __init set_bios_reboot(const struct dmi_system_id *d)
 127 {
 128         if (reboot_type != BOOT_BIOS) {
 129                 reboot_type = BOOT_BIOS;
 130                 printk(KERN_INFO "%s series board detected. Selecting BIOS-method for reboots.\n", d->ident);
 131         }
 132         return 0;
 133 }
 134
 135 static struct dmi_system_id __initdata reboot_dmi_table[] = {
 136         {       /* Handle problems with rebooting on Dell E520's */
 137                 .callback = set_bios_reboot,
 138                 .ident = "Dell E520",
 139                 .matches = {
 140                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Inc."),
 141                         DMI_MATCH(DMI_PRODUCT_NAME, "Dell DM061"),
 142                 },
 143         },
 144         {       /* Handle problems with rebooting on Dell 1300's */
 145                 .callback = set_bios_reboot,
 146                 .ident = "Dell PowerEdge 1300",
 147                 .matches = {
 148                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Computer Corporation"),
 149                         DMI_MATCH(DMI_PRODUCT_NAME, "PowerEdge 1300/"),
 150                 },
 151         },
 152         {       /* Handle problems with rebooting on Dell 300's */
 153                 .callback = set_bios_reboot,
 154                 .ident = "Dell PowerEdge 300",
 155                 .matches = {
 156                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Computer Corporation"),
 157                         DMI_MATCH(DMI_PRODUCT_NAME, "PowerEdge 300/"),
 158                 },
 159         },
 160         {       /* Handle problems with rebooting on Dell Optiplex 745's SFF*/
 161                 .callback = set_bios_reboot,
 162                 .ident = "Dell OptiPlex 745",
 163                 .matches = {
 164                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Inc."),
 165                         DMI_MATCH(DMI_PRODUCT_NAME, "OptiPlex 745"),
 166                 },
 167         },
 168         {       /* Handle problems with rebooting on Dell Optiplex 745's DFF*/
 169                 .callback = set_bios_reboot,
 170                 .ident = "Dell OptiPlex 745",
 171                 .matches = {
 172                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Inc."),
 173                         DMI_MATCH(DMI_PRODUCT_NAME, "OptiPlex 745"),
 174                         DMI_MATCH(DMI_BOARD_NAME, "0MM599"),
 175                 },
 176         },
 177         {       /* Handle problems with rebooting on Dell Optiplex 745 with 0KW626 */
 178                 .callback = set_bios_reboot,
 179                 .ident = "Dell OptiPlex 745",
 180                 .matches = {
 181                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Inc."),
 182                         DMI_MATCH(DMI_PRODUCT_NAME, "OptiPlex 745"),
 183                         DMI_MATCH(DMI_BOARD_NAME, "0KW626"),
 184                 },
 185         },
 186         {   /* Handle problems with rebooting on Dell Optiplex 330 with 0KP561 */
 187                 .callback = set_bios_reboot,
 188                 .ident = "Dell OptiPlex 330",
 189                 .matches = {
 190                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Inc."),
 191                         DMI_MATCH(DMI_PRODUCT_NAME, "OptiPlex 330"),
 192                         DMI_MATCH(DMI_BOARD_NAME, "0KP561"),
 193                 },
 194         },
 195         {   /* Handle problems with rebooting on Dell Optiplex 360 with 0T656F */
 196                 .callback = set_bios_reboot,
 197                 .ident = "Dell OptiPlex 360",
 198                 .matches = {
 199                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Inc."),
 200                         DMI_MATCH(DMI_PRODUCT_NAME, "OptiPlex 360"),
 201                         DMI_MATCH(DMI_BOARD_NAME, "0T656F"),
 202                 },
 203         },
 204         {       /* Handle problems with rebooting on Dell 2400's */
 205                 .callback = set_bios_reboot,
 206                 .ident = "Dell PowerEdge 2400",
 207                 .matches = {
 208                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Computer Corporation"),
 209                         DMI_MATCH(DMI_PRODUCT_NAME, "PowerEdge 2400"),
 210                 },
 211         },
 212         {       /* Handle problems with rebooting on Dell T5400's */
 213                 .callback = set_bios_reboot,
 214                 .ident = "Dell Precision T5400",
 215                 .matches = {
 216                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Inc."),
 217                         DMI_MATCH(DMI_PRODUCT_NAME, "Precision WorkStation T5400"),
 218                 },
 219         },
 220         {       /* Handle problems with rebooting on HP laptops */
 221                 .callback = set_bios_reboot,
 222                 .ident = "HP Compaq Laptop",
 223                 .matches = {
 224                         DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
 225                         DMI_MATCH(DMI_PRODUCT_NAME, "HP Compaq"),
 226                 },
 227         },
 228         {       /* Handle problems with rebooting on Dell XPS710 */
 229                 .callback = set_bios_reboot,
 230                 .ident = "Dell XPS710",
 231                 .matches = {
 232                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Inc."),
 233                         DMI_MATCH(DMI_PRODUCT_NAME, "Dell XPS710"),
 234                 },
 235         },
 236         { }
 237 };
 238
 239 static int __init reboot_init(void)
 240 {
 241         dmi_check_system(reboot_dmi_table);
 242         return 0;
 243 }
 244 core_initcall(reboot_init);
 245
 246 /* The following code and data reboots the machine by switching to real
 247    mode and jumping to the BIOS reset entry point, as if the CPU has
 248    really been reset.  The previous version asked the keyboard
 249    controller to pulse the CPU reset line, which is more thorough, but
 250    doesn't work with at least one type of 486 motherboard.  It is easy
 251    to stop this code working; hence the copious comments. */
 252 static const unsigned long long
 253 real_mode_gdt_entries [3] =
 254 {
 255         0x0000000000000000ULL,  /* Null descriptor */
 256         0x00009b000000ffffULL,  /* 16-bit real-mode 64k code at 0x00000000 */
 257         0x000093000100ffffULL   /* 16-bit real-mode 64k data at 0x00000100 */
 258 };
 259
 260 static const struct desc_ptr
 261 real_mode_gdt = { sizeof (real_mode_gdt_entries) - 1, (long)real_mode_gdt_entries },
 262 real_mode_idt = { 0x3ff, 0 };
 263
 264 /* This is 16-bit protected mode code to disable paging and the cache,
 265    switch to real mode and jump to the BIOS reset code.
 266
 267    The instruction that switches to real mode by writing to CR0 must be
 268    followed immediately by a far jump instruction, which set CS to a
 269    valid value for real mode, and flushes the prefetch queue to avoid
 270    running instructions that have already been decoded in protected
 271    mode.
 272
 273    Clears all the flags except ET, especially PG (paging), PE
 274    (protected-mode enable) and TS (task switch for coprocessor state
 275    save).  Flushes the TLB after paging has been disabled.  Sets CD and
 276    NW, to disable the cache on a 486, and invalidates the cache.  This
 277    is more like the state of a 486 after reset.  I don't know if
 278    something else should be done for other chips.
 279
 280    More could be done here to set up the registers as if a CPU reset had
 281    occurred; hopefully real BIOSs don't assume much. */
 282 static const unsigned char real_mode_switch [] =
 283 {
 284         0x66, 0x0f, 0x20, 0xc0,                 /*    movl  %cr0,%eax        */
 285         0x66, 0x83, 0xe0, 0x11,                 /*    andl  $0x00000011,%eax */
 286         0x66, 0x0d, 0x00, 0x00, 0x00, 0x60,     /*    orl   $0x60000000,%eax */
 287         0x66, 0x0f, 0x22, 0xc0,                 /*    movl  %eax,%cr0        */
 288         0x66, 0x0f, 0x22, 0xd8,                 /*    movl  %eax,%cr3        */
 289         0x66, 0x0f, 0x20, 0xc3,                 /*    movl  %cr0,%ebx        */
 290         0x66, 0x81, 0xe3, 0x00, 0x00, 0x00, 0x60,       /*    andl  $0x60000000,%ebx */
 291         0x74, 0x02,                             /*    jz    f                */
 292         0x0f, 0x09,                             /*    wbinvd                 */
 293         0x24, 0x10,                             /* f: andb  $0x10,al         */
 294         0x66, 0x0f, 0x22, 0xc0                  /*    movl  %eax,%cr0        */
 295 };
 296 static const unsigned char jump_to_bios [] =
 297 {
 298         0xea, 0x00, 0x00, 0xff, 0xff            /*    ljmp  $0xffff,$0x0000  */
 299 };
 300
 301 /*
 302  * Switch to real mode and then execute the code
 303  * specified by the code and length parameters.
 304  * We assume that length will aways be less that 100!
 305  */
 306 void machine_real_restart(const unsigned char *code, int length)
 307 {
 308         local_irq_disable();
 309
 310         /* Write zero to CMOS register number 0x0f, which the BIOS POST
 311            routine will recognize as telling it to do a proper reboot.  (Well
 312            that's what this book in front of me says -- it may only apply to
 313            the Phoenix BIOS though, it's not clear).  At the same time,
 314            disable NMIs by setting the top bit in the CMOS address register,
 315            as we're about to do peculiar things to the CPU.  I'm not sure if
 316            `outb_p' is needed instead of just `outb'.  Use it to be on the
 317            safe side.  (Yes, CMOS_WRITE does outb_p's. -  Paul G.)
 318          */
 319         spin_lock(&rtc_lock);
 320         CMOS_WRITE(0x00, 0x8f);
 321         spin_unlock(&rtc_lock);
 322
 323         /* Remap the kernel at virtual address zero, as well as offset zero
 324            from the kernel segment.  This assumes the kernel segment starts at
 325            virtual address PAGE_OFFSET. */
 326         memcpy(swapper_pg_dir, swapper_pg_dir + KERNEL_PGD_BOUNDARY,
 327                 sizeof(swapper_pg_dir [0]) * KERNEL_PGD_PTRS);
 328
 329         /*
 330          * Use `swapper_pg_dir' as our page directory.
 331          */
 332         load_cr3(swapper_pg_dir);
 333
 334         /* Write 0x1234 to absolute memory location 0x472.  The BIOS reads
 335            this on booting to tell it to "Bypass memory test (also warm
 336            boot)".  This seems like a fairly standard thing that gets set by
 337            REBOOT.COM programs, and the previous reset routine did this
 338            too. */
 339         *((unsigned short *)0x472) = reboot_mode;
 340
 341         /* For the switch to real mode, copy some code to low memory.  It has
 342            to be in the first 64k because it is running in 16-bit mode, and it
 343            has to have the same physical and virtual address, because it turns
 344            off paging.  Copy it near the end of the first page, out of the way
 345            of BIOS variables. */
 346         memcpy((void *)(0x1000 - sizeof(real_mode_switch) - 100),
 347                 real_mode_switch, sizeof (real_mode_switch));
 348         memcpy((void *)(0x1000 - 100), code, length);
 349
 350         /* Set up the IDT for real mode. */
 351         load_idt(&real_mode_idt);
 352
 353         /* Set up a GDT from which we can load segment descriptors for real
 354            mode.  The GDT is not used in real mode; it is just needed here to
 355            prepare the descriptors. */
 356         load_gdt(&real_mode_gdt);
 357
 358         /* Load the data segment registers, and thus the descriptors ready for
 359            real mode.  The base address of each segment is 0x100, 16 times the
 360            selector value being loaded here.  This is so that the segment
 361            registers don't have to be reloaded after switching to real mode:
 362            the values are consistent for real mode operation already. */
 363         __asm__ __volatile__ ("movl $0x0010,%%eax\n"
 364                                 "\tmovl %%eax,%%ds\n"
 365                                 "\tmovl %%eax,%%es\n"
 366                                 "\tmovl %%eax,%%fs\n"
 367                                 "\tmovl %%eax,%%gs\n"
 368                                 "\tmovl %%eax,%%ss" : : : "eax");
 369
 370         /* Jump to the 16-bit code that we copied earlier.  It disables paging
 371            and the cache, switches to real mode, and jumps to the BIOS reset
 372            entry point. */
 373         __asm__ __volatile__ ("ljmp $0x0008,%0"
 374                                 :
 375                                 : "i" ((void *)(0x1000 - sizeof (real_mode_switch) - 100)));
 376 }
 377 #ifdef CONFIG_APM_MODULE
 378 EXPORT_SYMBOL(machine_real_restart);
 379 #endif
 380
 381 #endif /* CONFIG_X86_32 */
 382
 383 static inline void kb_wait(void)
 384 {
 385         int i;
 386
 387         for (i = 0; i < 0x10000; i++) {
 388                 if ((inb(0x64) & 0x02) == 0)
 389                         break;
 390                 udelay(2);
 391         }
 392 }
 393
 394 static void vmxoff_nmi(int cpu, struct die_args *args)
 395 {
 396         cpu_emergency_vmxoff();
 397 }
 398
 399 /* Use NMIs as IPIs to tell all CPUs to disable virtualization
 400  */
 401 static void emergency_vmx_disable_all(void)
 402 {
 403         /* Just make sure we won't change CPUs while doing this */
 404         local_irq_disable();
 405
 406         /* We need to disable VMX on all CPUs before rebooting, otherwise
 407          * we risk hanging up the machine, because the CPU ignore INIT
 408          * signals when VMX is enabled.
 409          *
 410          * We can't take any locks and we may be on an inconsistent
 411          * state, so we use NMIs as IPIs to tell the other CPUs to disable
 412          * VMX and halt.
 413          *
 414          * For safety, we will avoid running the nmi_shootdown_cpus()
 415          * stuff unnecessarily, but we don't have a way to check
 416          * if other CPUs have VMX enabled. So we will call it only if the
 417          * CPU we are running on has VMX enabled.
 418          *
 419          * We will miss cases where VMX is not enabled on all CPUs. This
 420          * shouldn't do much harm because KVM always enable VMX on all
 421          * CPUs anyway. But we can miss it on the small window where KVM
 422          * is still enabling VMX.
 423          */
 424         if (cpu_has_vmx() && cpu_vmx_enabled()) {
 425                 /* Disable VMX on this CPU.
 426                  */
 427                 cpu_vmxoff();
 428
 429                 /* Halt and disable VMX on the other CPUs */
 430                 nmi_shootdown_cpus(vmxoff_nmi);
 431
 432         }
 433 }
 434
 435
 436 void __attribute__((weak)) mach_reboot_fixups(void)
 437 {
 438 }
 439
 440 static void native_machine_emergency_restart(void)
 441 {
 442         int i;
 443
 444         if (reboot_emergency)
 445                 emergency_vmx_disable_all();
 446
 447         /* Tell the BIOS if we want cold or warm reboot */
 448         *((unsigned short *)__va(0x472)) = reboot_mode;
 449
 450         for (;;) {
 451                 /* Could also try the reset bit in the Hammer NB */
 452                 switch (reboot_type) {
 453                 case BOOT_KBD:
 454                         mach_reboot_fixups(); /* for board specific fixups */
 455
 456                         for (i = 0; i < 10; i++) {
 457                                 kb_wait();
 458                                 udelay(50);
 459                                 outb(0xfe, 0x64); /* pulse reset low */
 460                                 udelay(50);
 461                         }
 462
 463                 case BOOT_TRIPLE:
 464                         load_idt(&no_idt);
 465                         __asm__ __volatile__("int3");
 466
 467                         reboot_type = BOOT_KBD;
 468                         break;
 469
 470 #ifdef CONFIG_X86_32
 471                 case BOOT_BIOS:
 472                         machine_real_restart(jump_to_bios, sizeof(jump_to_bios));
 473
 474                         reboot_type = BOOT_KBD;
 475                         break;
 476 #endif
 477
 478                 case BOOT_ACPI:
 479                         acpi_reboot();
 480                         reboot_type = BOOT_KBD;
 481                         break;
 482
 483                 case BOOT_EFI:
 484                         if (efi_enabled)
 485                                 efi.reset_system(reboot_mode ?
 486                                                  EFI_RESET_WARM :
 487                                                  EFI_RESET_COLD,
 488                                                  EFI_SUCCESS, 0, NULL);
 489                         reboot_type = BOOT_KBD;
 490                         break;
 491
 492                 case BOOT_CF9:
 493                         port_cf9_safe = true;
 494                         /* fall through */
 495
 496                 case BOOT_CF9_COND:
 497                         if (port_cf9_safe) {
 498                                 u8 cf9 = inb(0xcf9) & ~6;
 499                                 outb(cf9|2, 0xcf9); /* Request hard reset */
 500                                 udelay(50);
 501                                 outb(cf9|6, 0xcf9); /* Actually do the reset */
 502                                 udelay(50);
 503                         }
 504                         reboot_type = BOOT_KBD;
 505                         break;
 506                 }
 507         }
 508 }
 509
 510 void native_machine_shutdown(void)
 511 {
 512         /* Stop the cpus and apics */
 513 #ifdef CONFIG_SMP
 514
 515         /* The boot cpu is always logical cpu 0 */
 516         int reboot_cpu_id = 0;
 517
 518 #ifdef CONFIG_X86_32
 519         /* See if there has been given a command line override */
 520         if ((reboot_cpu != -1) && (reboot_cpu < nr_cpu_ids) &&
 521                 cpu_online(reboot_cpu))
 522                 reboot_cpu_id = reboot_cpu;
 523 #endif
 524
 525         /* Make certain the cpu I'm about to reboot on is online */
 526         if (!cpu_online(reboot_cpu_id))
 527                 reboot_cpu_id = smp_processor_id();
 528
 529         /* Make certain I only run on the appropriate processor */
 530         set_cpus_allowed_ptr(current, cpumask_of(reboot_cpu_id));
 531
 532         /* O.K Now that I'm on the appropriate processor,
 533          * stop all of the others.
 534          */
 535         smp_send_stop();
 536 #endif
 537
 538         lapic_shutdown();
 539
 540 #ifdef CONFIG_X86_IO_APIC
 541         disable_IO_APIC();
 542 #endif
 543
 544 #ifdef CONFIG_HPET_TIMER
 545         hpet_disable();
 546 #endif
 547
 548 #ifdef CONFIG_X86_64
 549         pci_iommu_shutdown();
 550 #endif
 551 }
 552
 553 static void __machine_emergency_restart(int emergency)
 554 {
 555         reboot_emergency = emergency;
 556         machine_ops.emergency_restart();
 557 }
 558
 559 static void native_machine_restart(char *__unused)
 560 {
 561         printk("machine restart\n");
 562
 563         if (!reboot_force)
 564                 machine_shutdown();
 565         __machine_emergency_restart(0);
 566 }
 567
 568 static void native_machine_halt(void)
 569 {
 570         /* stop other cpus and apics */
 571         machine_shutdown();
 572
 573         /* stop this cpu */
 574         stop_this_cpu(NULL);
 575 }
 576
 577 static void native_machine_power_off(void)
 578 {
 579         if (pm_power_off) {
 580                 if (!reboot_force)
 581                         machine_shutdown();
 582                 pm_power_off();
 583         }
 584 }
 585
 586 struct machine_ops machine_ops = {
 587         .power_off = native_machine_power_off,
 588         .shutdown = native_machine_shutdown,
 589         .emergency_restart = native_machine_emergency_restart,
 590         .restart = native_machine_restart,
 591         .halt = native_machine_halt,
 592 #ifdef CONFIG_KEXEC
 593         .crash_shutdown = native_machine_crash_shutdown,
 594 #endif
 595 };
 596
 597 void machine_power_off(void)
 598 {
 599         machine_ops.power_off();
 600 }
 601
 602 void machine_shutdown(void)
 603 {
 604         machine_ops.shutdown();
 605 }
 606
 607 void machine_emergency_restart(void)
 608 {
 609         __machine_emergency_restart(1);
 610 }
 611
 612 void machine_restart(char *cmd)
 613 {
 614         machine_ops.restart(cmd);
 615 }
 616
 617 void machine_halt(void)
 618 {
 619         machine_ops.halt();
 620 }
 621
 622 #ifdef CONFIG_KEXEC
 623 void machine_crash_shutdown(struct pt_regs *regs)
 624 {
 625         machine_ops.crash_shutdown(regs);
 626 }
 627 #endif
 628
 629
 630 #if defined(CONFIG_SMP)
 631
 632 /* This keeps a track of which one is crashing cpu. */
 633 static int crashing_cpu;
 634 static nmi_shootdown_cb shootdown_callback;
 635
 636 static atomic_t waiting_for_crash_ipi;
 637
 638 static int crash_nmi_callback(struct notifier_block *self,
 639                         unsigned long val, void *data)
 640 {
 641         int cpu;
 642
 643         if (val != DIE_NMI_IPI)
 644                 return NOTIFY_OK;
 645
 646         cpu = raw_smp_processor_id();
 647
 648         /* Don't do anything if this handler is invoked on crashing cpu.
 649          * Otherwise, system will completely hang. Crashing cpu can get
 650          * an NMI if system was initially booted with nmi_watchdog parameter.
 651          */
 652         if (cpu == crashing_cpu)
 653                 return NOTIFY_STOP;
 654         local_irq_disable();
 655
 656         shootdown_callback(cpu, (struct die_args *)data);
 657
 658         atomic_dec(&waiting_for_crash_ipi);
 659         /* Assume hlt works */
 660         halt();
 661         for (;;)
 662                 cpu_relax();
 663
 664         return 1;
 665 }
 666
 667 static void smp_send_nmi_allbutself(void)
 668 {
 669         apic->send_IPI_allbutself(NMI_VECTOR);
 670 }
 671
 672 static struct notifier_block crash_nmi_nb = {
 673         .notifier_call = crash_nmi_callback,
 674 };
 675
 676 /* Halt all other CPUs, calling the specified function on each of them
 677  *
 678  * This function can be used to halt all other CPUs on crash
 679  * or emergency reboot time. The function passed as parameter
 680  * will be called inside a NMI handler on all CPUs.
 681  */
 682 void nmi_shootdown_cpus(nmi_shootdown_cb callback)
 683 {
 684         unsigned long msecs;
 685         local_irq_disable();
 686
 687         /* Make a note of crashing cpu. Will be used in NMI callback.*/
 688         crashing_cpu = safe_smp_processor_id();
 689
 690         shootdown_callback = callback;
 691
 692         atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
 693         /* Would it be better to replace the trap vector here? */
 694         if (register_die_notifier(&crash_nmi_nb))
 695                 return;         /* return what? */
 696         /* Ensure the new callback function is set before sending
 697          * out the NMI
 698          */
 699         wmb();
 700
 701         smp_send_nmi_allbutself();
 702
 703         msecs = 1000; /* Wait at most a second for the other cpus to stop */
 704         while ((atomic_read(&waiting_for_crash_ipi) > 0) && msecs) {
 705                 mdelay(1);
 706                 msecs--;
 707         }
 708
 709         /* Leave the nmi callback set */
 710 }
 711 #else /* !CONFIG_SMP */
 712 void nmi_shootdown_cpus(nmi_shootdown_cb callback)
 713 {
 714         /* No other CPUs to shoot down */
 715 }
 716 #endif