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 2400's */
 196                 .callback = set_bios_reboot,
 197                 .ident = "Dell PowerEdge 2400",
 198                 .matches = {
 199                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Computer Corporation"),
 200                         DMI_MATCH(DMI_PRODUCT_NAME, "PowerEdge 2400"),
 201                 },
 202         },
 203         {       /* Handle problems with rebooting on Dell T5400's */
 204                 .callback = set_bios_reboot,
 205                 .ident = "Dell Precision T5400",
 206                 .matches = {
 207                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Inc."),
 208                         DMI_MATCH(DMI_PRODUCT_NAME, "Precision WorkStation T5400"),
 209                 },
 210         },
 211         {       /* Handle problems with rebooting on HP laptops */
 212                 .callback = set_bios_reboot,
 213                 .ident = "HP Compaq Laptop",
 214                 .matches = {
 215                         DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
 216                         DMI_MATCH(DMI_PRODUCT_NAME, "HP Compaq"),
 217                 },
 218         },
 219         {       /* Handle problems with rebooting on Dell XPS710 */
 220                 .callback = set_bios_reboot,
 221                 .ident = "Dell XPS710",
 222                 .matches = {
 223                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Inc."),
 224                         DMI_MATCH(DMI_PRODUCT_NAME, "Dell XPS710"),
 225                 },
 226         },
 227         {       /* Handle problems with rebooting on Dell DXP061 */
 228                 .callback = set_bios_reboot,
 229                 .ident = "Dell DXP061",
 230                 .matches = {
 231                         DMI_MATCH(DMI_SYS_VENDOR, "Dell Inc."),
 232                         DMI_MATCH(DMI_PRODUCT_NAME, "Dell DXP061"),
 233                 },
 234         },
 235         { }
 236 };
 237
 238 static int __init reboot_init(void)
 239 {
 240         dmi_check_system(reboot_dmi_table);
 241         return 0;
 242 }
 243 core_initcall(reboot_init);
 244
 245 /* The following code and data reboots the machine by switching to real
 246    mode and jumping to the BIOS reset entry point, as if the CPU has
 247    really been reset.  The previous version asked the keyboard
 248    controller to pulse the CPU reset line, which is more thorough, but
 249    doesn't work with at least one type of 486 motherboard.  It is easy
 250    to stop this code working; hence the copious comments. */
 251 static const unsigned long long
 252 real_mode_gdt_entries [3] =
 253 {
 254         0x0000000000000000ULL,  /* Null descriptor */
 255         0x00009b000000ffffULL,  /* 16-bit real-mode 64k code at 0x00000000 */
 256         0x000093000100ffffULL   /* 16-bit real-mode 64k data at 0x00000100 */
 257 };
 258
 259 static const struct desc_ptr
 260 real_mode_gdt = { sizeof (real_mode_gdt_entries) - 1, (long)real_mode_gdt_entries },
 261 real_mode_idt = { 0x3ff, 0 };
 262
 263 /* This is 16-bit protected mode code to disable paging and the cache,
 264    switch to real mode and jump to the BIOS reset code.
 265
 266    The instruction that switches to real mode by writing to CR0 must be
 267    followed immediately by a far jump instruction, which set CS to a
 268    valid value for real mode, and flushes the prefetch queue to avoid
 269    running instructions that have already been decoded in protected
 270    mode.
 271
 272    Clears all the flags except ET, especially PG (paging), PE
 273    (protected-mode enable) and TS (task switch for coprocessor state
 274    save).  Flushes the TLB after paging has been disabled.  Sets CD and
 275    NW, to disable the cache on a 486, and invalidates the cache.  This
 276    is more like the state of a 486 after reset.  I don't know if
 277    something else should be done for other chips.
 278
 279    More could be done here to set up the registers as if a CPU reset had
 280    occurred; hopefully real BIOSs don't assume much. */
 281 static const unsigned char real_mode_switch [] =
 282 {
 283         0x66, 0x0f, 0x20, 0xc0,                 /*    movl  %cr0,%eax        */
 284         0x66, 0x83, 0xe0, 0x11,                 /*    andl  $0x00000011,%eax */
 285         0x66, 0x0d, 0x00, 0x00, 0x00, 0x60,     /*    orl   $0x60000000,%eax */
 286         0x66, 0x0f, 0x22, 0xc0,                 /*    movl  %eax,%cr0        */
 287         0x66, 0x0f, 0x22, 0xd8,                 /*    movl  %eax,%cr3        */
 288         0x66, 0x0f, 0x20, 0xc3,                 /*    movl  %cr0,%ebx        */
 289         0x66, 0x81, 0xe3, 0x00, 0x00, 0x00, 0x60,       /*    andl  $0x60000000,%ebx */
 290         0x74, 0x02,                             /*    jz    f                */
 291         0x0f, 0x09,                             /*    wbinvd                 */
 292         0x24, 0x10,                             /* f: andb  $0x10,al         */
 293         0x66, 0x0f, 0x22, 0xc0                  /*    movl  %eax,%cr0        */
 294 };
 295 static const unsigned char jump_to_bios [] =
 296 {
 297         0xea, 0x00, 0x00, 0xff, 0xff            /*    ljmp  $0xffff,$0x0000  */
 298 };
 299
 300 /*
 301  * Switch to real mode and then execute the code
 302  * specified by the code and length parameters.
 303  * We assume that length will aways be less that 100!
 304  */
 305 void machine_real_restart(const unsigned char *code, int length)
 306 {
 307         local_irq_disable();
 308
 309         /* Write zero to CMOS register number 0x0f, which the BIOS POST
 310            routine will recognize as telling it to do a proper reboot.  (Well
 311            that's what this book in front of me says -- it may only apply to
 312            the Phoenix BIOS though, it's not clear).  At the same time,
 313            disable NMIs by setting the top bit in the CMOS address register,
 314            as we're about to do peculiar things to the CPU.  I'm not sure if
 315            `outb_p' is needed instead of just `outb'.  Use it to be on the
 316            safe side.  (Yes, CMOS_WRITE does outb_p's. -  Paul G.)
 317          */
 318         spin_lock(&rtc_lock);
 319         CMOS_WRITE(0x00, 0x8f);
 320         spin_unlock(&rtc_lock);
 321
 322         /* Remap the kernel at virtual address zero, as well as offset zero
 323            from the kernel segment.  This assumes the kernel segment starts at
 324            virtual address PAGE_OFFSET. */
 325         memcpy(swapper_pg_dir, swapper_pg_dir + KERNEL_PGD_BOUNDARY,
 326                 sizeof(swapper_pg_dir [0]) * KERNEL_PGD_PTRS);
 327
 328         /*
 329          * Use `swapper_pg_dir' as our page directory.
 330          */
 331         load_cr3(swapper_pg_dir);
 332
 333         /* Write 0x1234 to absolute memory location 0x472.  The BIOS reads
 334            this on booting to tell it to "Bypass memory test (also warm
 335            boot)".  This seems like a fairly standard thing that gets set by
 336            REBOOT.COM programs, and the previous reset routine did this
 337            too. */
 338         *((unsigned short *)0x472) = reboot_mode;
 339
 340         /* For the switch to real mode, copy some code to low memory.  It has
 341            to be in the first 64k because it is running in 16-bit mode, and it
 342            has to have the same physical and virtual address, because it turns
 343            off paging.  Copy it near the end of the first page, out of the way
 344            of BIOS variables. */
 345         memcpy((void *)(0x1000 - sizeof(real_mode_switch) - 100),
 346                 real_mode_switch, sizeof (real_mode_switch));
 347         memcpy((void *)(0x1000 - 100), code, length);
 348
 349         /* Set up the IDT for real mode. */
 350         load_idt(&real_mode_idt);
 351
 352         /* Set up a GDT from which we can load segment descriptors for real
 353            mode.  The GDT is not used in real mode; it is just needed here to
 354            prepare the descriptors. */
 355         load_gdt(&real_mode_gdt);
 356
 357         /* Load the data segment registers, and thus the descriptors ready for
 358            real mode.  The base address of each segment is 0x100, 16 times the
 359            selector value being loaded here.  This is so that the segment
 360            registers don't have to be reloaded after switching to real mode:
 361            the values are consistent for real mode operation already. */
 362         __asm__ __volatile__ ("movl $0x0010,%%eax\n"
 363                                 "\tmovl %%eax,%%ds\n"
 364                                 "\tmovl %%eax,%%es\n"
 365                                 "\tmovl %%eax,%%fs\n"
 366                                 "\tmovl %%eax,%%gs\n"
 367                                 "\tmovl %%eax,%%ss" : : : "eax");
 368
 369         /* Jump to the 16-bit code that we copied earlier.  It disables paging
 370            and the cache, switches to real mode, and jumps to the BIOS reset
 371            entry point. */
 372         __asm__ __volatile__ ("ljmp $0x0008,%0"
 373                                 :
 374                                 : "i" ((void *)(0x1000 - sizeof (real_mode_switch) - 100)));
 375 }
 376 #ifdef CONFIG_APM_MODULE
 377 EXPORT_SYMBOL(machine_real_restart);
 378 #endif
 379
 380 #endif /* CONFIG_X86_32 */
 381
 382 static inline void kb_wait(void)
 383 {
 384         int i;
 385
 386         for (i = 0; i < 0x10000; i++) {
 387                 if ((inb(0x64) & 0x02) == 0)
 388                         break;
 389                 udelay(2);
 390         }
 391 }
 392
 393 static void vmxoff_nmi(int cpu, struct die_args *args)
 394 {
 395         cpu_emergency_vmxoff();
 396 }
 397
 398 /* Use NMIs as IPIs to tell all CPUs to disable virtualization
 399  */
 400 static void emergency_vmx_disable_all(void)
 401 {
 402         /* Just make sure we won't change CPUs while doing this */
 403         local_irq_disable();
 404
 405         /* We need to disable VMX on all CPUs before rebooting, otherwise
 406          * we risk hanging up the machine, because the CPU ignore INIT
 407          * signals when VMX is enabled.
 408          *
 409          * We can't take any locks and we may be on an inconsistent
 410          * state, so we use NMIs as IPIs to tell the other CPUs to disable
 411          * VMX and halt.
 412          *
 413          * For safety, we will avoid running the nmi_shootdown_cpus()
 414          * stuff unnecessarily, but we don't have a way to check
 415          * if other CPUs have VMX enabled. So we will call it only if the
 416          * CPU we are running on has VMX enabled.
 417          *
 418          * We will miss cases where VMX is not enabled on all CPUs. This
 419          * shouldn't do much harm because KVM always enable VMX on all
 420          * CPUs anyway. But we can miss it on the small window where KVM
 421          * is still enabling VMX.
 422          */
 423         if (cpu_has_vmx() && cpu_vmx_enabled()) {
 424                 /* Disable VMX on this CPU.
 425                  */
 426                 cpu_vmxoff();
 427
 428                 /* Halt and disable VMX on the other CPUs */
 429                 nmi_shootdown_cpus(vmxoff_nmi);
 430
 431         }
 432 }
 433
 434
 435 void __attribute__((weak)) mach_reboot_fixups(void)
 436 {
 437 }
 438
 439 static void native_machine_emergency_restart(void)
 440 {
 441         int i;
 442
 443         if (reboot_emergency)
 444                 emergency_vmx_disable_all();
 445
 446         /* Tell the BIOS if we want cold or warm reboot */
 447         *((unsigned short *)__va(0x472)) = reboot_mode;
 448
 449         for (;;) {
 450                 /* Could also try the reset bit in the Hammer NB */
 451                 switch (reboot_type) {
 452                 case BOOT_KBD:
 453                         mach_reboot_fixups(); /* for board specific fixups */
 454
 455                         for (i = 0; i < 10; i++) {
 456                                 kb_wait();
 457                                 udelay(50);
 458                                 outb(0xfe, 0x64); /* pulse reset low */
 459                                 udelay(50);
 460                         }
 461
 462                 case BOOT_TRIPLE:
 463                         load_idt(&no_idt);
 464                         __asm__ __volatile__("int3");
 465
 466                         reboot_type = BOOT_KBD;
 467                         break;
 468
 469 #ifdef CONFIG_X86_32
 470                 case BOOT_BIOS:
 471                         machine_real_restart(jump_to_bios, sizeof(jump_to_bios));
 472
 473                         reboot_type = BOOT_KBD;
 474                         break;
 475 #endif
 476
 477                 case BOOT_ACPI:
 478                         acpi_reboot();
 479                         reboot_type = BOOT_KBD;
 480                         break;
 481
 482                 case BOOT_EFI:
 483                         if (efi_enabled)
 484                                 efi.reset_system(reboot_mode ?
 485                                                  EFI_RESET_WARM :
 486                                                  EFI_RESET_COLD,
 487                                                  EFI_SUCCESS, 0, NULL);
 488                         reboot_type = BOOT_KBD;
 489                         break;
 490
 491                 case BOOT_CF9:
 492                         port_cf9_safe = true;
 493                         /* fall through */
 494
 495                 case BOOT_CF9_COND:
 496                         if (port_cf9_safe) {
 497                                 u8 cf9 = inb(0xcf9) & ~6;
 498                                 outb(cf9|2, 0xcf9); /* Request hard reset */
 499                                 udelay(50);
 500                                 outb(cf9|6, 0xcf9); /* Actually do the reset */
 501                                 udelay(50);
 502                         }
 503                         reboot_type = BOOT_KBD;
 504                         break;
 505                 }
 506         }
 507 }
 508
 509 void native_machine_shutdown(void)
 510 {
 511         /* Stop the cpus and apics */
 512 #ifdef CONFIG_SMP
 513
 514         /* The boot cpu is always logical cpu 0 */
 515         int reboot_cpu_id = 0;
 516
 517 #ifdef CONFIG_X86_32
 518         /* See if there has been given a command line override */
 519         if ((reboot_cpu != -1) && (reboot_cpu < nr_cpu_ids) &&
 520                 cpu_online(reboot_cpu))
 521                 reboot_cpu_id = reboot_cpu;
 522 #endif
 523
 524         /* Make certain the cpu I'm about to reboot on is online */
 525         if (!cpu_online(reboot_cpu_id))
 526                 reboot_cpu_id = smp_processor_id();
 527
 528         /* Make certain I only run on the appropriate processor */
 529         set_cpus_allowed_ptr(current, cpumask_of(reboot_cpu_id));
 530
 531         /* O.K Now that I'm on the appropriate processor,
 532          * stop all of the others.
 533          */
 534         smp_send_stop();
 535 #endif
 536
 537         lapic_shutdown();
 538
 539 #ifdef CONFIG_X86_IO_APIC
 540         disable_IO_APIC();
 541 #endif
 542
 543 #ifdef CONFIG_HPET_TIMER
 544         hpet_disable();
 545 #endif
 546
 547 #ifdef CONFIG_X86_64
 548         pci_iommu_shutdown();
 549 #endif
 550 }
 551
 552 static void __machine_emergency_restart(int emergency)
 553 {
 554         reboot_emergency = emergency;
 555         machine_ops.emergency_restart();
 556 }
 557
 558 static void native_machine_restart(char *__unused)
 559 {
 560         printk("machine restart\n");
 561
 562         if (!reboot_force)
 563                 machine_shutdown();
 564         __machine_emergency_restart(0);
 565 }
 566
 567 static void native_machine_halt(void)
 568 {
 569         /* stop other cpus and apics */
 570         machine_shutdown();
 571
 572         /* stop this cpu */
 573         stop_this_cpu(NULL);
 574 }
 575
 576 static void native_machine_power_off(void)
 577 {
 578         if (pm_power_off) {
 579                 if (!reboot_force)
 580                         machine_shutdown();
 581                 pm_power_off();
 582         }
 583 }
 584
 585 struct machine_ops machine_ops = {
 586         .power_off = native_machine_power_off,
 587         .shutdown = native_machine_shutdown,
 588         .emergency_restart = native_machine_emergency_restart,
 589         .restart = native_machine_restart,
 590         .halt = native_machine_halt,
 591 #ifdef CONFIG_KEXEC
 592         .crash_shutdown = native_machine_crash_shutdown,
 593 #endif
 594 };
 595
 596 void machine_power_off(void)
 597 {
 598         machine_ops.power_off();
 599 }
 600
 601 void machine_shutdown(void)
 602 {
 603         machine_ops.shutdown();
 604 }
 605
 606 void machine_emergency_restart(void)
 607 {
 608         __machine_emergency_restart(1);
 609 }
 610
 611 void machine_restart(char *cmd)
 612 {
 613         machine_ops.restart(cmd);
 614 }
 615
 616 void machine_halt(void)
 617 {
 618         machine_ops.halt();
 619 }
 620
 621 #ifdef CONFIG_KEXEC
 622 void machine_crash_shutdown(struct pt_regs *regs)
 623 {
 624         machine_ops.crash_shutdown(regs);
 625 }
 626 #endif
 627
 628
 629 #if defined(CONFIG_SMP)
 630
 631 /* This keeps a track of which one is crashing cpu. */
 632 static int crashing_cpu;
 633 static nmi_shootdown_cb shootdown_callback;
 634
 635 static atomic_t waiting_for_crash_ipi;
 636
 637 static int crash_nmi_callback(struct notifier_block *self,
 638                         unsigned long val, void *data)
 639 {
 640         int cpu;
 641
 642         if (val != DIE_NMI_IPI)
 643                 return NOTIFY_OK;
 644
 645         cpu = raw_smp_processor_id();
 646
 647         /* Don't do anything if this handler is invoked on crashing cpu.
 648          * Otherwise, system will completely hang. Crashing cpu can get
 649          * an NMI if system was initially booted with nmi_watchdog parameter.
 650          */
 651         if (cpu == crashing_cpu)
 652                 return NOTIFY_STOP;
 653         local_irq_disable();
 654
 655         shootdown_callback(cpu, (struct die_args *)data);
 656
 657         atomic_dec(&waiting_for_crash_ipi);
 658         /* Assume hlt works */
 659         halt();
 660         for (;;)
 661                 cpu_relax();
 662
 663         return 1;
 664 }
 665
 666 static void smp_send_nmi_allbutself(void)
 667 {
 668         apic->send_IPI_allbutself(NMI_VECTOR);
 669 }
 670
 671 static struct notifier_block crash_nmi_nb = {
 672         .notifier_call = crash_nmi_callback,
 673 };
 674
 675 /* Halt all other CPUs, calling the specified function on each of them
 676  *
 677  * This function can be used to halt all other CPUs on crash
 678  * or emergency reboot time. The function passed as parameter
 679  * will be called inside a NMI handler on all CPUs.
 680  */
 681 void nmi_shootdown_cpus(nmi_shootdown_cb callback)
 682 {
 683         unsigned long msecs;
 684         local_irq_disable();
 685
 686         /* Make a note of crashing cpu. Will be used in NMI callback.*/
 687         crashing_cpu = safe_smp_processor_id();
 688
 689         shootdown_callback = callback;
 690
 691         atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
 692         /* Would it be better to replace the trap vector here? */
 693         if (register_die_notifier(&crash_nmi_nb))
 694                 return;         /* return what? */
 695         /* Ensure the new callback function is set before sending
 696          * out the NMI
 697          */
 698         wmb();
 699
 700         smp_send_nmi_allbutself();
 701
 702         msecs = 1000; /* Wait at most a second for the other cpus to stop */
 703         while ((atomic_read(&waiting_for_crash_ipi) > 0) && msecs) {
 704                 mdelay(1);
 705                 msecs--;
 706         }
 707
 708         /* Leave the nmi callback set */
 709 }
 710 #else /* !CONFIG_SMP */
 711 void nmi_shootdown_cpus(nmi_shootdown_cb callback)
 712 {
 713         /* No other CPUs to shoot down */
 714 }
 715 #endif