xref: /openbmc/linux/arch/x86/kernel/reboot.c (revision 82ced6fd)
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 	{	/* Handle problems with rebooting on Sony VGN-Z540N */
236 		.callback = set_bios_reboot,
237 		.ident = "Sony VGN-Z540N",
238 		.matches = {
239 			DMI_MATCH(DMI_SYS_VENDOR, "Sony Corporation"),
240 			DMI_MATCH(DMI_PRODUCT_NAME, "VGN-Z540N"),
241 		},
242 	},
243 	{ }
244 };
245 
246 static int __init reboot_init(void)
247 {
248 	dmi_check_system(reboot_dmi_table);
249 	return 0;
250 }
251 core_initcall(reboot_init);
252 
253 /* The following code and data reboots the machine by switching to real
254    mode and jumping to the BIOS reset entry point, as if the CPU has
255    really been reset.  The previous version asked the keyboard
256    controller to pulse the CPU reset line, which is more thorough, but
257    doesn't work with at least one type of 486 motherboard.  It is easy
258    to stop this code working; hence the copious comments. */
259 static const unsigned long long
260 real_mode_gdt_entries [3] =
261 {
262 	0x0000000000000000ULL,	/* Null descriptor */
263 	0x00009b000000ffffULL,	/* 16-bit real-mode 64k code at 0x00000000 */
264 	0x000093000100ffffULL	/* 16-bit real-mode 64k data at 0x00000100 */
265 };
266 
267 static const struct desc_ptr
268 real_mode_gdt = { sizeof (real_mode_gdt_entries) - 1, (long)real_mode_gdt_entries },
269 real_mode_idt = { 0x3ff, 0 };
270 
271 /* This is 16-bit protected mode code to disable paging and the cache,
272    switch to real mode and jump to the BIOS reset code.
273 
274    The instruction that switches to real mode by writing to CR0 must be
275    followed immediately by a far jump instruction, which set CS to a
276    valid value for real mode, and flushes the prefetch queue to avoid
277    running instructions that have already been decoded in protected
278    mode.
279 
280    Clears all the flags except ET, especially PG (paging), PE
281    (protected-mode enable) and TS (task switch for coprocessor state
282    save).  Flushes the TLB after paging has been disabled.  Sets CD and
283    NW, to disable the cache on a 486, and invalidates the cache.  This
284    is more like the state of a 486 after reset.  I don't know if
285    something else should be done for other chips.
286 
287    More could be done here to set up the registers as if a CPU reset had
288    occurred; hopefully real BIOSs don't assume much. */
289 static const unsigned char real_mode_switch [] =
290 {
291 	0x66, 0x0f, 0x20, 0xc0,			/*    movl  %cr0,%eax        */
292 	0x66, 0x83, 0xe0, 0x11,			/*    andl  $0x00000011,%eax */
293 	0x66, 0x0d, 0x00, 0x00, 0x00, 0x60,	/*    orl   $0x60000000,%eax */
294 	0x66, 0x0f, 0x22, 0xc0,			/*    movl  %eax,%cr0        */
295 	0x66, 0x0f, 0x22, 0xd8,			/*    movl  %eax,%cr3        */
296 	0x66, 0x0f, 0x20, 0xc3,			/*    movl  %cr0,%ebx        */
297 	0x66, 0x81, 0xe3, 0x00, 0x00, 0x00, 0x60,	/*    andl  $0x60000000,%ebx */
298 	0x74, 0x02,				/*    jz    f                */
299 	0x0f, 0x09,				/*    wbinvd                 */
300 	0x24, 0x10,				/* f: andb  $0x10,al         */
301 	0x66, 0x0f, 0x22, 0xc0			/*    movl  %eax,%cr0        */
302 };
303 static const unsigned char jump_to_bios [] =
304 {
305 	0xea, 0x00, 0x00, 0xff, 0xff		/*    ljmp  $0xffff,$0x0000  */
306 };
307 
308 /*
309  * Switch to real mode and then execute the code
310  * specified by the code and length parameters.
311  * We assume that length will aways be less that 100!
312  */
313 void machine_real_restart(const unsigned char *code, int length)
314 {
315 	local_irq_disable();
316 
317 	/* Write zero to CMOS register number 0x0f, which the BIOS POST
318 	   routine will recognize as telling it to do a proper reboot.  (Well
319 	   that's what this book in front of me says -- it may only apply to
320 	   the Phoenix BIOS though, it's not clear).  At the same time,
321 	   disable NMIs by setting the top bit in the CMOS address register,
322 	   as we're about to do peculiar things to the CPU.  I'm not sure if
323 	   `outb_p' is needed instead of just `outb'.  Use it to be on the
324 	   safe side.  (Yes, CMOS_WRITE does outb_p's. -  Paul G.)
325 	 */
326 	spin_lock(&rtc_lock);
327 	CMOS_WRITE(0x00, 0x8f);
328 	spin_unlock(&rtc_lock);
329 
330 	/* Remap the kernel at virtual address zero, as well as offset zero
331 	   from the kernel segment.  This assumes the kernel segment starts at
332 	   virtual address PAGE_OFFSET. */
333 	memcpy(swapper_pg_dir, swapper_pg_dir + KERNEL_PGD_BOUNDARY,
334 		sizeof(swapper_pg_dir [0]) * KERNEL_PGD_PTRS);
335 
336 	/*
337 	 * Use `swapper_pg_dir' as our page directory.
338 	 */
339 	load_cr3(swapper_pg_dir);
340 
341 	/* Write 0x1234 to absolute memory location 0x472.  The BIOS reads
342 	   this on booting to tell it to "Bypass memory test (also warm
343 	   boot)".  This seems like a fairly standard thing that gets set by
344 	   REBOOT.COM programs, and the previous reset routine did this
345 	   too. */
346 	*((unsigned short *)0x472) = reboot_mode;
347 
348 	/* For the switch to real mode, copy some code to low memory.  It has
349 	   to be in the first 64k because it is running in 16-bit mode, and it
350 	   has to have the same physical and virtual address, because it turns
351 	   off paging.  Copy it near the end of the first page, out of the way
352 	   of BIOS variables. */
353 	memcpy((void *)(0x1000 - sizeof(real_mode_switch) - 100),
354 		real_mode_switch, sizeof (real_mode_switch));
355 	memcpy((void *)(0x1000 - 100), code, length);
356 
357 	/* Set up the IDT for real mode. */
358 	load_idt(&real_mode_idt);
359 
360 	/* Set up a GDT from which we can load segment descriptors for real
361 	   mode.  The GDT is not used in real mode; it is just needed here to
362 	   prepare the descriptors. */
363 	load_gdt(&real_mode_gdt);
364 
365 	/* Load the data segment registers, and thus the descriptors ready for
366 	   real mode.  The base address of each segment is 0x100, 16 times the
367 	   selector value being loaded here.  This is so that the segment
368 	   registers don't have to be reloaded after switching to real mode:
369 	   the values are consistent for real mode operation already. */
370 	__asm__ __volatile__ ("movl $0x0010,%%eax\n"
371 				"\tmovl %%eax,%%ds\n"
372 				"\tmovl %%eax,%%es\n"
373 				"\tmovl %%eax,%%fs\n"
374 				"\tmovl %%eax,%%gs\n"
375 				"\tmovl %%eax,%%ss" : : : "eax");
376 
377 	/* Jump to the 16-bit code that we copied earlier.  It disables paging
378 	   and the cache, switches to real mode, and jumps to the BIOS reset
379 	   entry point. */
380 	__asm__ __volatile__ ("ljmp $0x0008,%0"
381 				:
382 				: "i" ((void *)(0x1000 - sizeof (real_mode_switch) - 100)));
383 }
384 #ifdef CONFIG_APM_MODULE
385 EXPORT_SYMBOL(machine_real_restart);
386 #endif
387 
388 #endif /* CONFIG_X86_32 */
389 
390 static inline void kb_wait(void)
391 {
392 	int i;
393 
394 	for (i = 0; i < 0x10000; i++) {
395 		if ((inb(0x64) & 0x02) == 0)
396 			break;
397 		udelay(2);
398 	}
399 }
400 
401 static void vmxoff_nmi(int cpu, struct die_args *args)
402 {
403 	cpu_emergency_vmxoff();
404 }
405 
406 /* Use NMIs as IPIs to tell all CPUs to disable virtualization
407  */
408 static void emergency_vmx_disable_all(void)
409 {
410 	/* Just make sure we won't change CPUs while doing this */
411 	local_irq_disable();
412 
413 	/* We need to disable VMX on all CPUs before rebooting, otherwise
414 	 * we risk hanging up the machine, because the CPU ignore INIT
415 	 * signals when VMX is enabled.
416 	 *
417 	 * We can't take any locks and we may be on an inconsistent
418 	 * state, so we use NMIs as IPIs to tell the other CPUs to disable
419 	 * VMX and halt.
420 	 *
421 	 * For safety, we will avoid running the nmi_shootdown_cpus()
422 	 * stuff unnecessarily, but we don't have a way to check
423 	 * if other CPUs have VMX enabled. So we will call it only if the
424 	 * CPU we are running on has VMX enabled.
425 	 *
426 	 * We will miss cases where VMX is not enabled on all CPUs. This
427 	 * shouldn't do much harm because KVM always enable VMX on all
428 	 * CPUs anyway. But we can miss it on the small window where KVM
429 	 * is still enabling VMX.
430 	 */
431 	if (cpu_has_vmx() && cpu_vmx_enabled()) {
432 		/* Disable VMX on this CPU.
433 		 */
434 		cpu_vmxoff();
435 
436 		/* Halt and disable VMX on the other CPUs */
437 		nmi_shootdown_cpus(vmxoff_nmi);
438 
439 	}
440 }
441 
442 
443 void __attribute__((weak)) mach_reboot_fixups(void)
444 {
445 }
446 
447 static void native_machine_emergency_restart(void)
448 {
449 	int i;
450 
451 	if (reboot_emergency)
452 		emergency_vmx_disable_all();
453 
454 	/* Tell the BIOS if we want cold or warm reboot */
455 	*((unsigned short *)__va(0x472)) = reboot_mode;
456 
457 	for (;;) {
458 		/* Could also try the reset bit in the Hammer NB */
459 		switch (reboot_type) {
460 		case BOOT_KBD:
461 			mach_reboot_fixups(); /* for board specific fixups */
462 
463 			for (i = 0; i < 10; i++) {
464 				kb_wait();
465 				udelay(50);
466 				outb(0xfe, 0x64); /* pulse reset low */
467 				udelay(50);
468 			}
469 
470 		case BOOT_TRIPLE:
471 			load_idt(&no_idt);
472 			__asm__ __volatile__("int3");
473 
474 			reboot_type = BOOT_KBD;
475 			break;
476 
477 #ifdef CONFIG_X86_32
478 		case BOOT_BIOS:
479 			machine_real_restart(jump_to_bios, sizeof(jump_to_bios));
480 
481 			reboot_type = BOOT_KBD;
482 			break;
483 #endif
484 
485 		case BOOT_ACPI:
486 			acpi_reboot();
487 			reboot_type = BOOT_KBD;
488 			break;
489 
490 		case BOOT_EFI:
491 			if (efi_enabled)
492 				efi.reset_system(reboot_mode ?
493 						 EFI_RESET_WARM :
494 						 EFI_RESET_COLD,
495 						 EFI_SUCCESS, 0, NULL);
496 			reboot_type = BOOT_KBD;
497 			break;
498 
499 		case BOOT_CF9:
500 			port_cf9_safe = true;
501 			/* fall through */
502 
503 		case BOOT_CF9_COND:
504 			if (port_cf9_safe) {
505 				u8 cf9 = inb(0xcf9) & ~6;
506 				outb(cf9|2, 0xcf9); /* Request hard reset */
507 				udelay(50);
508 				outb(cf9|6, 0xcf9); /* Actually do the reset */
509 				udelay(50);
510 			}
511 			reboot_type = BOOT_KBD;
512 			break;
513 		}
514 	}
515 }
516 
517 void native_machine_shutdown(void)
518 {
519 	/* Stop the cpus and apics */
520 #ifdef CONFIG_SMP
521 
522 	/* The boot cpu is always logical cpu 0 */
523 	int reboot_cpu_id = 0;
524 
525 #ifdef CONFIG_X86_32
526 	/* See if there has been given a command line override */
527 	if ((reboot_cpu != -1) && (reboot_cpu < nr_cpu_ids) &&
528 		cpu_online(reboot_cpu))
529 		reboot_cpu_id = reboot_cpu;
530 #endif
531 
532 	/* Make certain the cpu I'm about to reboot on is online */
533 	if (!cpu_online(reboot_cpu_id))
534 		reboot_cpu_id = smp_processor_id();
535 
536 	/* Make certain I only run on the appropriate processor */
537 	set_cpus_allowed_ptr(current, cpumask_of(reboot_cpu_id));
538 
539 	/* O.K Now that I'm on the appropriate processor,
540 	 * stop all of the others.
541 	 */
542 	smp_send_stop();
543 #endif
544 
545 	lapic_shutdown();
546 
547 #ifdef CONFIG_X86_IO_APIC
548 	disable_IO_APIC();
549 #endif
550 
551 #ifdef CONFIG_HPET_TIMER
552 	hpet_disable();
553 #endif
554 
555 #ifdef CONFIG_X86_64
556 	pci_iommu_shutdown();
557 #endif
558 }
559 
560 static void __machine_emergency_restart(int emergency)
561 {
562 	reboot_emergency = emergency;
563 	machine_ops.emergency_restart();
564 }
565 
566 static void native_machine_restart(char *__unused)
567 {
568 	printk("machine restart\n");
569 
570 	if (!reboot_force)
571 		machine_shutdown();
572 	__machine_emergency_restart(0);
573 }
574 
575 static void native_machine_halt(void)
576 {
577 	/* stop other cpus and apics */
578 	machine_shutdown();
579 
580 	/* stop this cpu */
581 	stop_this_cpu(NULL);
582 }
583 
584 static void native_machine_power_off(void)
585 {
586 	if (pm_power_off) {
587 		if (!reboot_force)
588 			machine_shutdown();
589 		pm_power_off();
590 	}
591 }
592 
593 struct machine_ops machine_ops = {
594 	.power_off = native_machine_power_off,
595 	.shutdown = native_machine_shutdown,
596 	.emergency_restart = native_machine_emergency_restart,
597 	.restart = native_machine_restart,
598 	.halt = native_machine_halt,
599 #ifdef CONFIG_KEXEC
600 	.crash_shutdown = native_machine_crash_shutdown,
601 #endif
602 };
603 
604 void machine_power_off(void)
605 {
606 	machine_ops.power_off();
607 }
608 
609 void machine_shutdown(void)
610 {
611 	machine_ops.shutdown();
612 }
613 
614 void machine_emergency_restart(void)
615 {
616 	__machine_emergency_restart(1);
617 }
618 
619 void machine_restart(char *cmd)
620 {
621 	machine_ops.restart(cmd);
622 }
623 
624 void machine_halt(void)
625 {
626 	machine_ops.halt();
627 }
628 
629 #ifdef CONFIG_KEXEC
630 void machine_crash_shutdown(struct pt_regs *regs)
631 {
632 	machine_ops.crash_shutdown(regs);
633 }
634 #endif
635 
636 
637 #if defined(CONFIG_SMP)
638 
639 /* This keeps a track of which one is crashing cpu. */
640 static int crashing_cpu;
641 static nmi_shootdown_cb shootdown_callback;
642 
643 static atomic_t waiting_for_crash_ipi;
644 
645 static int crash_nmi_callback(struct notifier_block *self,
646 			unsigned long val, void *data)
647 {
648 	int cpu;
649 
650 	if (val != DIE_NMI_IPI)
651 		return NOTIFY_OK;
652 
653 	cpu = raw_smp_processor_id();
654 
655 	/* Don't do anything if this handler is invoked on crashing cpu.
656 	 * Otherwise, system will completely hang. Crashing cpu can get
657 	 * an NMI if system was initially booted with nmi_watchdog parameter.
658 	 */
659 	if (cpu == crashing_cpu)
660 		return NOTIFY_STOP;
661 	local_irq_disable();
662 
663 	shootdown_callback(cpu, (struct die_args *)data);
664 
665 	atomic_dec(&waiting_for_crash_ipi);
666 	/* Assume hlt works */
667 	halt();
668 	for (;;)
669 		cpu_relax();
670 
671 	return 1;
672 }
673 
674 static void smp_send_nmi_allbutself(void)
675 {
676 	apic->send_IPI_allbutself(NMI_VECTOR);
677 }
678 
679 static struct notifier_block crash_nmi_nb = {
680 	.notifier_call = crash_nmi_callback,
681 };
682 
683 /* Halt all other CPUs, calling the specified function on each of them
684  *
685  * This function can be used to halt all other CPUs on crash
686  * or emergency reboot time. The function passed as parameter
687  * will be called inside a NMI handler on all CPUs.
688  */
689 void nmi_shootdown_cpus(nmi_shootdown_cb callback)
690 {
691 	unsigned long msecs;
692 	local_irq_disable();
693 
694 	/* Make a note of crashing cpu. Will be used in NMI callback.*/
695 	crashing_cpu = safe_smp_processor_id();
696 
697 	shootdown_callback = callback;
698 
699 	atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
700 	/* Would it be better to replace the trap vector here? */
701 	if (register_die_notifier(&crash_nmi_nb))
702 		return;		/* return what? */
703 	/* Ensure the new callback function is set before sending
704 	 * out the NMI
705 	 */
706 	wmb();
707 
708 	smp_send_nmi_allbutself();
709 
710 	msecs = 1000; /* Wait at most a second for the other cpus to stop */
711 	while ((atomic_read(&waiting_for_crash_ipi) > 0) && msecs) {
712 		mdelay(1);
713 		msecs--;
714 	}
715 
716 	/* Leave the nmi callback set */
717 }
718 #else /* !CONFIG_SMP */
719 void nmi_shootdown_cpus(nmi_shootdown_cb callback)
720 {
721 	/* No other CPUs to shoot down */
722 }
723 #endif
724