xref: /linux/arch/x86/kernel/smpboot.c (revision 606b2f490fb80e55d05cf0e6cec0b6c0ff0fc18f)
1 /*
2  *	x86 SMP booting functions
3  *
4  *	(c) 1995 Alan Cox, Building #3 <alan@lxorguk.ukuu.org.uk>
5  *	(c) 1998, 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
6  *	Copyright 2001 Andi Kleen, SuSE Labs.
7  *
8  *	Much of the core SMP work is based on previous work by Thomas Radke, to
9  *	whom a great many thanks are extended.
10  *
11  *	Thanks to Intel for making available several different Pentium,
12  *	Pentium Pro and Pentium-II/Xeon MP machines.
13  *	Original development of Linux SMP code supported by Caldera.
14  *
15  *	This code is released under the GNU General Public License version 2 or
16  *	later.
17  *
18  *	Fixes
19  *		Felix Koop	:	NR_CPUS used properly
20  *		Jose Renau	:	Handle single CPU case.
21  *		Alan Cox	:	By repeated request 8) - Total BogoMIPS report.
22  *		Greg Wright	:	Fix for kernel stacks panic.
23  *		Erich Boleyn	:	MP v1.4 and additional changes.
24  *	Matthias Sattler	:	Changes for 2.1 kernel map.
25  *	Michel Lespinasse	:	Changes for 2.1 kernel map.
26  *	Michael Chastain	:	Change trampoline.S to gnu as.
27  *		Alan Cox	:	Dumb bug: 'B' step PPro's are fine
28  *		Ingo Molnar	:	Added APIC timers, based on code
29  *					from Jose Renau
30  *		Ingo Molnar	:	various cleanups and rewrites
31  *		Tigran Aivazian	:	fixed "0.00 in /proc/uptime on SMP" bug.
32  *	Maciej W. Rozycki	:	Bits for genuine 82489DX APICs
33  *	Andi Kleen		:	Changed for SMP boot into long mode.
34  *		Martin J. Bligh	: 	Added support for multi-quad systems
35  *		Dave Jones	:	Report invalid combinations of Athlon CPUs.
36  *		Rusty Russell	:	Hacked into shape for new "hotplug" boot process.
37  *      Andi Kleen              :       Converted to new state machine.
38  *	Ashok Raj		: 	CPU hotplug support
39  *	Glauber Costa		:	i386 and x86_64 integration
40  */
41 
42 #include <linux/init.h>
43 #include <linux/smp.h>
44 #include <linux/module.h>
45 #include <linux/sched.h>
46 #include <linux/percpu.h>
47 #include <linux/bootmem.h>
48 #include <linux/err.h>
49 #include <linux/nmi.h>
50 #include <linux/tboot.h>
51 #include <linux/stackprotector.h>
52 #include <linux/gfp.h>
53 
54 #include <asm/acpi.h>
55 #include <asm/desc.h>
56 #include <asm/nmi.h>
57 #include <asm/irq.h>
58 #include <asm/idle.h>
59 #include <asm/trampoline.h>
60 #include <asm/cpu.h>
61 #include <asm/numa.h>
62 #include <asm/pgtable.h>
63 #include <asm/tlbflush.h>
64 #include <asm/mtrr.h>
65 #include <asm/vmi.h>
66 #include <asm/apic.h>
67 #include <asm/setup.h>
68 #include <asm/uv/uv.h>
69 #include <linux/mc146818rtc.h>
70 
71 #include <asm/smpboot_hooks.h>
72 #include <asm/i8259.h>
73 
74 #ifdef CONFIG_X86_32
75 u8 apicid_2_node[MAX_APICID];
76 #endif
77 
78 /* State of each CPU */
79 DEFINE_PER_CPU(int, cpu_state) = { 0 };
80 
81 /* Store all idle threads, this can be reused instead of creating
82 * a new thread. Also avoids complicated thread destroy functionality
83 * for idle threads.
84 */
85 #ifdef CONFIG_HOTPLUG_CPU
86 /*
87  * Needed only for CONFIG_HOTPLUG_CPU because __cpuinitdata is
88  * removed after init for !CONFIG_HOTPLUG_CPU.
89  */
90 static DEFINE_PER_CPU(struct task_struct *, idle_thread_array);
91 #define get_idle_for_cpu(x)      (per_cpu(idle_thread_array, x))
92 #define set_idle_for_cpu(x, p)   (per_cpu(idle_thread_array, x) = (p))
93 
94 /*
95  * We need this for trampoline_base protection from concurrent accesses when
96  * off- and onlining cores wildly.
97  */
98 static DEFINE_MUTEX(x86_cpu_hotplug_driver_mutex);
99 
100 void cpu_hotplug_driver_lock()
101 {
102         mutex_lock(&x86_cpu_hotplug_driver_mutex);
103 }
104 
105 void cpu_hotplug_driver_unlock()
106 {
107         mutex_unlock(&x86_cpu_hotplug_driver_mutex);
108 }
109 
110 ssize_t arch_cpu_probe(const char *buf, size_t count) { return -1; }
111 ssize_t arch_cpu_release(const char *buf, size_t count) { return -1; }
112 #else
113 static struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
114 #define get_idle_for_cpu(x)      (idle_thread_array[(x)])
115 #define set_idle_for_cpu(x, p)   (idle_thread_array[(x)] = (p))
116 #endif
117 
118 /* Number of siblings per CPU package */
119 int smp_num_siblings = 1;
120 EXPORT_SYMBOL(smp_num_siblings);
121 
122 /* Last level cache ID of each logical CPU */
123 DEFINE_PER_CPU(u16, cpu_llc_id) = BAD_APICID;
124 
125 /* representing HT siblings of each logical CPU */
126 DEFINE_PER_CPU(cpumask_var_t, cpu_sibling_map);
127 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
128 
129 /* representing HT and core siblings of each logical CPU */
130 DEFINE_PER_CPU(cpumask_var_t, cpu_core_map);
131 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
132 
133 /* Per CPU bogomips and other parameters */
134 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
135 EXPORT_PER_CPU_SYMBOL(cpu_info);
136 
137 atomic_t init_deasserted;
138 
139 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_32)
140 /* which node each logical CPU is on */
141 int cpu_to_node_map[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 };
142 EXPORT_SYMBOL(cpu_to_node_map);
143 
144 /* set up a mapping between cpu and node. */
145 static void map_cpu_to_node(int cpu, int node)
146 {
147 	printk(KERN_INFO "Mapping cpu %d to node %d\n", cpu, node);
148 	cpumask_set_cpu(cpu, node_to_cpumask_map[node]);
149 	cpu_to_node_map[cpu] = node;
150 }
151 
152 /* undo a mapping between cpu and node. */
153 static void unmap_cpu_to_node(int cpu)
154 {
155 	int node;
156 
157 	printk(KERN_INFO "Unmapping cpu %d from all nodes\n", cpu);
158 	for (node = 0; node < MAX_NUMNODES; node++)
159 		cpumask_clear_cpu(cpu, node_to_cpumask_map[node]);
160 	cpu_to_node_map[cpu] = 0;
161 }
162 #else /* !(CONFIG_NUMA && CONFIG_X86_32) */
163 #define map_cpu_to_node(cpu, node)	({})
164 #define unmap_cpu_to_node(cpu)	({})
165 #endif
166 
167 #ifdef CONFIG_X86_32
168 static int boot_cpu_logical_apicid;
169 
170 u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly =
171 					{ [0 ... NR_CPUS-1] = BAD_APICID };
172 
173 static void map_cpu_to_logical_apicid(void)
174 {
175 	int cpu = smp_processor_id();
176 	int apicid = logical_smp_processor_id();
177 	int node = apic->apicid_to_node(apicid);
178 
179 	if (!node_online(node))
180 		node = first_online_node;
181 
182 	cpu_2_logical_apicid[cpu] = apicid;
183 	map_cpu_to_node(cpu, node);
184 }
185 
186 void numa_remove_cpu(int cpu)
187 {
188 	cpu_2_logical_apicid[cpu] = BAD_APICID;
189 	unmap_cpu_to_node(cpu);
190 }
191 #else
192 #define map_cpu_to_logical_apicid()  do {} while (0)
193 #endif
194 
195 /*
196  * Report back to the Boot Processor.
197  * Running on AP.
198  */
199 static void __cpuinit smp_callin(void)
200 {
201 	int cpuid, phys_id;
202 	unsigned long timeout;
203 
204 	/*
205 	 * If waken up by an INIT in an 82489DX configuration
206 	 * we may get here before an INIT-deassert IPI reaches
207 	 * our local APIC.  We have to wait for the IPI or we'll
208 	 * lock up on an APIC access.
209 	 */
210 	if (apic->wait_for_init_deassert)
211 		apic->wait_for_init_deassert(&init_deasserted);
212 
213 	/*
214 	 * (This works even if the APIC is not enabled.)
215 	 */
216 	phys_id = read_apic_id();
217 	cpuid = smp_processor_id();
218 	if (cpumask_test_cpu(cpuid, cpu_callin_mask)) {
219 		panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
220 					phys_id, cpuid);
221 	}
222 	pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
223 
224 	/*
225 	 * STARTUP IPIs are fragile beasts as they might sometimes
226 	 * trigger some glue motherboard logic. Complete APIC bus
227 	 * silence for 1 second, this overestimates the time the
228 	 * boot CPU is spending to send the up to 2 STARTUP IPIs
229 	 * by a factor of two. This should be enough.
230 	 */
231 
232 	/*
233 	 * Waiting 2s total for startup (udelay is not yet working)
234 	 */
235 	timeout = jiffies + 2*HZ;
236 	while (time_before(jiffies, timeout)) {
237 		/*
238 		 * Has the boot CPU finished it's STARTUP sequence?
239 		 */
240 		if (cpumask_test_cpu(cpuid, cpu_callout_mask))
241 			break;
242 		cpu_relax();
243 	}
244 
245 	if (!time_before(jiffies, timeout)) {
246 		panic("%s: CPU%d started up but did not get a callout!\n",
247 		      __func__, cpuid);
248 	}
249 
250 	/*
251 	 * the boot CPU has finished the init stage and is spinning
252 	 * on callin_map until we finish. We are free to set up this
253 	 * CPU, first the APIC. (this is probably redundant on most
254 	 * boards)
255 	 */
256 
257 	pr_debug("CALLIN, before setup_local_APIC().\n");
258 	if (apic->smp_callin_clear_local_apic)
259 		apic->smp_callin_clear_local_apic();
260 	setup_local_APIC();
261 	end_local_APIC_setup();
262 	map_cpu_to_logical_apicid();
263 
264 	/*
265 	 * Need to setup vector mappings before we enable interrupts.
266 	 */
267 	setup_vector_irq(smp_processor_id());
268 	/*
269 	 * Get our bogomips.
270 	 *
271 	 * Need to enable IRQs because it can take longer and then
272 	 * the NMI watchdog might kill us.
273 	 */
274 	local_irq_enable();
275 	calibrate_delay();
276 	local_irq_disable();
277 	pr_debug("Stack at about %p\n", &cpuid);
278 
279 	/*
280 	 * Save our processor parameters
281 	 */
282 	smp_store_cpu_info(cpuid);
283 
284 	notify_cpu_starting(cpuid);
285 
286 	/*
287 	 * Allow the master to continue.
288 	 */
289 	cpumask_set_cpu(cpuid, cpu_callin_mask);
290 }
291 
292 /*
293  * Activate a secondary processor.
294  */
295 notrace static void __cpuinit start_secondary(void *unused)
296 {
297 	/*
298 	 * Don't put *anything* before cpu_init(), SMP booting is too
299 	 * fragile that we want to limit the things done here to the
300 	 * most necessary things.
301 	 */
302 
303 #ifdef CONFIG_X86_32
304 	/*
305 	 * Switch away from the trampoline page-table
306 	 *
307 	 * Do this before cpu_init() because it needs to access per-cpu
308 	 * data which may not be mapped in the trampoline page-table.
309 	 */
310 	load_cr3(swapper_pg_dir);
311 	__flush_tlb_all();
312 #endif
313 
314 	vmi_bringup();
315 	cpu_init();
316 	preempt_disable();
317 	smp_callin();
318 
319 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
320 	barrier();
321 	/*
322 	 * Check TSC synchronization with the BP:
323 	 */
324 	check_tsc_sync_target();
325 
326 	if (nmi_watchdog == NMI_IO_APIC) {
327 		legacy_pic->chip->mask(0);
328 		enable_NMI_through_LVT0();
329 		legacy_pic->chip->unmask(0);
330 	}
331 
332 	/* This must be done before setting cpu_online_mask */
333 	set_cpu_sibling_map(raw_smp_processor_id());
334 	wmb();
335 
336 	/*
337 	 * We need to hold call_lock, so there is no inconsistency
338 	 * between the time smp_call_function() determines number of
339 	 * IPI recipients, and the time when the determination is made
340 	 * for which cpus receive the IPI. Holding this
341 	 * lock helps us to not include this cpu in a currently in progress
342 	 * smp_call_function().
343 	 *
344 	 * We need to hold vector_lock so there the set of online cpus
345 	 * does not change while we are assigning vectors to cpus.  Holding
346 	 * this lock ensures we don't half assign or remove an irq from a cpu.
347 	 */
348 	ipi_call_lock();
349 	lock_vector_lock();
350 	set_cpu_online(smp_processor_id(), true);
351 	unlock_vector_lock();
352 	ipi_call_unlock();
353 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
354 	x86_platform.nmi_init();
355 
356 	/* enable local interrupts */
357 	local_irq_enable();
358 
359 	/* to prevent fake stack check failure in clock setup */
360 	boot_init_stack_canary();
361 
362 	x86_cpuinit.setup_percpu_clockev();
363 
364 	wmb();
365 	cpu_idle();
366 }
367 
368 #ifdef CONFIG_CPUMASK_OFFSTACK
369 /* In this case, llc_shared_map is a pointer to a cpumask. */
370 static inline void copy_cpuinfo_x86(struct cpuinfo_x86 *dst,
371 				    const struct cpuinfo_x86 *src)
372 {
373 	struct cpumask *llc = dst->llc_shared_map;
374 	*dst = *src;
375 	dst->llc_shared_map = llc;
376 }
377 #else
378 static inline void copy_cpuinfo_x86(struct cpuinfo_x86 *dst,
379 				    const struct cpuinfo_x86 *src)
380 {
381 	*dst = *src;
382 }
383 #endif /* CONFIG_CPUMASK_OFFSTACK */
384 
385 /*
386  * The bootstrap kernel entry code has set these up. Save them for
387  * a given CPU
388  */
389 
390 void __cpuinit smp_store_cpu_info(int id)
391 {
392 	struct cpuinfo_x86 *c = &cpu_data(id);
393 
394 	copy_cpuinfo_x86(c, &boot_cpu_data);
395 	c->cpu_index = id;
396 	if (id != 0)
397 		identify_secondary_cpu(c);
398 }
399 
400 
401 void __cpuinit set_cpu_sibling_map(int cpu)
402 {
403 	int i;
404 	struct cpuinfo_x86 *c = &cpu_data(cpu);
405 
406 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
407 
408 	if (smp_num_siblings > 1) {
409 		for_each_cpu(i, cpu_sibling_setup_mask) {
410 			struct cpuinfo_x86 *o = &cpu_data(i);
411 
412 			if (c->phys_proc_id == o->phys_proc_id &&
413 			    c->cpu_core_id == o->cpu_core_id) {
414 				cpumask_set_cpu(i, cpu_sibling_mask(cpu));
415 				cpumask_set_cpu(cpu, cpu_sibling_mask(i));
416 				cpumask_set_cpu(i, cpu_core_mask(cpu));
417 				cpumask_set_cpu(cpu, cpu_core_mask(i));
418 				cpumask_set_cpu(i, c->llc_shared_map);
419 				cpumask_set_cpu(cpu, o->llc_shared_map);
420 			}
421 		}
422 	} else {
423 		cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
424 	}
425 
426 	cpumask_set_cpu(cpu, c->llc_shared_map);
427 
428 	if (current_cpu_data.x86_max_cores == 1) {
429 		cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
430 		c->booted_cores = 1;
431 		return;
432 	}
433 
434 	for_each_cpu(i, cpu_sibling_setup_mask) {
435 		if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
436 		    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
437 			cpumask_set_cpu(i, c->llc_shared_map);
438 			cpumask_set_cpu(cpu, cpu_data(i).llc_shared_map);
439 		}
440 		if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
441 			cpumask_set_cpu(i, cpu_core_mask(cpu));
442 			cpumask_set_cpu(cpu, cpu_core_mask(i));
443 			/*
444 			 *  Does this new cpu bringup a new core?
445 			 */
446 			if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
447 				/*
448 				 * for each core in package, increment
449 				 * the booted_cores for this new cpu
450 				 */
451 				if (cpumask_first(cpu_sibling_mask(i)) == i)
452 					c->booted_cores++;
453 				/*
454 				 * increment the core count for all
455 				 * the other cpus in this package
456 				 */
457 				if (i != cpu)
458 					cpu_data(i).booted_cores++;
459 			} else if (i != cpu && !c->booted_cores)
460 				c->booted_cores = cpu_data(i).booted_cores;
461 		}
462 	}
463 }
464 
465 /* maps the cpu to the sched domain representing multi-core */
466 const struct cpumask *cpu_coregroup_mask(int cpu)
467 {
468 	struct cpuinfo_x86 *c = &cpu_data(cpu);
469 	/*
470 	 * For perf, we return last level cache shared map.
471 	 * And for power savings, we return cpu_core_map
472 	 */
473 	if ((sched_mc_power_savings || sched_smt_power_savings) &&
474 	    !(cpu_has(c, X86_FEATURE_AMD_DCM)))
475 		return cpu_core_mask(cpu);
476 	else
477 		return c->llc_shared_map;
478 }
479 
480 static void impress_friends(void)
481 {
482 	int cpu;
483 	unsigned long bogosum = 0;
484 	/*
485 	 * Allow the user to impress friends.
486 	 */
487 	pr_debug("Before bogomips.\n");
488 	for_each_possible_cpu(cpu)
489 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
490 			bogosum += cpu_data(cpu).loops_per_jiffy;
491 	printk(KERN_INFO
492 		"Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
493 		num_online_cpus(),
494 		bogosum/(500000/HZ),
495 		(bogosum/(5000/HZ))%100);
496 
497 	pr_debug("Before bogocount - setting activated=1.\n");
498 }
499 
500 void __inquire_remote_apic(int apicid)
501 {
502 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
503 	char *names[] = { "ID", "VERSION", "SPIV" };
504 	int timeout;
505 	u32 status;
506 
507 	printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
508 
509 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
510 		printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
511 
512 		/*
513 		 * Wait for idle.
514 		 */
515 		status = safe_apic_wait_icr_idle();
516 		if (status)
517 			printk(KERN_CONT
518 			       "a previous APIC delivery may have failed\n");
519 
520 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
521 
522 		timeout = 0;
523 		do {
524 			udelay(100);
525 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
526 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
527 
528 		switch (status) {
529 		case APIC_ICR_RR_VALID:
530 			status = apic_read(APIC_RRR);
531 			printk(KERN_CONT "%08x\n", status);
532 			break;
533 		default:
534 			printk(KERN_CONT "failed\n");
535 		}
536 	}
537 }
538 
539 /*
540  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
541  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
542  * won't ... remember to clear down the APIC, etc later.
543  */
544 int __cpuinit
545 wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
546 {
547 	unsigned long send_status, accept_status = 0;
548 	int maxlvt;
549 
550 	/* Target chip */
551 	/* Boot on the stack */
552 	/* Kick the second */
553 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, logical_apicid);
554 
555 	pr_debug("Waiting for send to finish...\n");
556 	send_status = safe_apic_wait_icr_idle();
557 
558 	/*
559 	 * Give the other CPU some time to accept the IPI.
560 	 */
561 	udelay(200);
562 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
563 		maxlvt = lapic_get_maxlvt();
564 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
565 			apic_write(APIC_ESR, 0);
566 		accept_status = (apic_read(APIC_ESR) & 0xEF);
567 	}
568 	pr_debug("NMI sent.\n");
569 
570 	if (send_status)
571 		printk(KERN_ERR "APIC never delivered???\n");
572 	if (accept_status)
573 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
574 
575 	return (send_status | accept_status);
576 }
577 
578 static int __cpuinit
579 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
580 {
581 	unsigned long send_status, accept_status = 0;
582 	int maxlvt, num_starts, j;
583 
584 	maxlvt = lapic_get_maxlvt();
585 
586 	/*
587 	 * Be paranoid about clearing APIC errors.
588 	 */
589 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
590 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
591 			apic_write(APIC_ESR, 0);
592 		apic_read(APIC_ESR);
593 	}
594 
595 	pr_debug("Asserting INIT.\n");
596 
597 	/*
598 	 * Turn INIT on target chip
599 	 */
600 	/*
601 	 * Send IPI
602 	 */
603 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
604 		       phys_apicid);
605 
606 	pr_debug("Waiting for send to finish...\n");
607 	send_status = safe_apic_wait_icr_idle();
608 
609 	mdelay(10);
610 
611 	pr_debug("Deasserting INIT.\n");
612 
613 	/* Target chip */
614 	/* Send IPI */
615 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
616 
617 	pr_debug("Waiting for send to finish...\n");
618 	send_status = safe_apic_wait_icr_idle();
619 
620 	mb();
621 	atomic_set(&init_deasserted, 1);
622 
623 	/*
624 	 * Should we send STARTUP IPIs ?
625 	 *
626 	 * Determine this based on the APIC version.
627 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
628 	 */
629 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
630 		num_starts = 2;
631 	else
632 		num_starts = 0;
633 
634 	/*
635 	 * Paravirt / VMI wants a startup IPI hook here to set up the
636 	 * target processor state.
637 	 */
638 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
639 			 (unsigned long)stack_start.sp);
640 
641 	/*
642 	 * Run STARTUP IPI loop.
643 	 */
644 	pr_debug("#startup loops: %d.\n", num_starts);
645 
646 	for (j = 1; j <= num_starts; j++) {
647 		pr_debug("Sending STARTUP #%d.\n", j);
648 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
649 			apic_write(APIC_ESR, 0);
650 		apic_read(APIC_ESR);
651 		pr_debug("After apic_write.\n");
652 
653 		/*
654 		 * STARTUP IPI
655 		 */
656 
657 		/* Target chip */
658 		/* Boot on the stack */
659 		/* Kick the second */
660 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
661 			       phys_apicid);
662 
663 		/*
664 		 * Give the other CPU some time to accept the IPI.
665 		 */
666 		udelay(300);
667 
668 		pr_debug("Startup point 1.\n");
669 
670 		pr_debug("Waiting for send to finish...\n");
671 		send_status = safe_apic_wait_icr_idle();
672 
673 		/*
674 		 * Give the other CPU some time to accept the IPI.
675 		 */
676 		udelay(200);
677 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
678 			apic_write(APIC_ESR, 0);
679 		accept_status = (apic_read(APIC_ESR) & 0xEF);
680 		if (send_status || accept_status)
681 			break;
682 	}
683 	pr_debug("After Startup.\n");
684 
685 	if (send_status)
686 		printk(KERN_ERR "APIC never delivered???\n");
687 	if (accept_status)
688 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
689 
690 	return (send_status | accept_status);
691 }
692 
693 struct create_idle {
694 	struct work_struct work;
695 	struct task_struct *idle;
696 	struct completion done;
697 	int cpu;
698 };
699 
700 static void __cpuinit do_fork_idle(struct work_struct *work)
701 {
702 	struct create_idle *c_idle =
703 		container_of(work, struct create_idle, work);
704 
705 	c_idle->idle = fork_idle(c_idle->cpu);
706 	complete(&c_idle->done);
707 }
708 
709 /* reduce the number of lines printed when booting a large cpu count system */
710 static void __cpuinit announce_cpu(int cpu, int apicid)
711 {
712 	static int current_node = -1;
713 	int node = early_cpu_to_node(cpu);
714 
715 	if (system_state == SYSTEM_BOOTING) {
716 		if (node != current_node) {
717 			if (current_node > (-1))
718 				pr_cont(" Ok.\n");
719 			current_node = node;
720 			pr_info("Booting Node %3d, Processors ", node);
721 		}
722 		pr_cont(" #%d%s", cpu, cpu == (nr_cpu_ids - 1) ? " Ok.\n" : "");
723 		return;
724 	} else
725 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
726 			node, cpu, apicid);
727 }
728 
729 /*
730  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
731  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
732  * Returns zero if CPU booted OK, else error code from
733  * ->wakeup_secondary_cpu.
734  */
735 static int __cpuinit do_boot_cpu(int apicid, int cpu)
736 {
737 	unsigned long boot_error = 0;
738 	unsigned long start_ip;
739 	int timeout;
740 	struct create_idle c_idle = {
741 		.cpu	= cpu,
742 		.done	= COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
743 	};
744 
745 	INIT_WORK_ON_STACK(&c_idle.work, do_fork_idle);
746 
747 	alternatives_smp_switch(1);
748 
749 	c_idle.idle = get_idle_for_cpu(cpu);
750 
751 	/*
752 	 * We can't use kernel_thread since we must avoid to
753 	 * reschedule the child.
754 	 */
755 	if (c_idle.idle) {
756 		c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
757 			(THREAD_SIZE +  task_stack_page(c_idle.idle))) - 1);
758 		init_idle(c_idle.idle, cpu);
759 		goto do_rest;
760 	}
761 
762 	schedule_work(&c_idle.work);
763 	wait_for_completion(&c_idle.done);
764 
765 	if (IS_ERR(c_idle.idle)) {
766 		printk("failed fork for CPU %d\n", cpu);
767 		destroy_work_on_stack(&c_idle.work);
768 		return PTR_ERR(c_idle.idle);
769 	}
770 
771 	set_idle_for_cpu(cpu, c_idle.idle);
772 do_rest:
773 	per_cpu(current_task, cpu) = c_idle.idle;
774 #ifdef CONFIG_X86_32
775 	/* Stack for startup_32 can be just as for start_secondary onwards */
776 	irq_ctx_init(cpu);
777 	initial_page_table = __pa(&trampoline_pg_dir);
778 #else
779 	clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
780 	initial_gs = per_cpu_offset(cpu);
781 	per_cpu(kernel_stack, cpu) =
782 		(unsigned long)task_stack_page(c_idle.idle) -
783 		KERNEL_STACK_OFFSET + THREAD_SIZE;
784 #endif
785 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
786 	initial_code = (unsigned long)start_secondary;
787 	stack_start.sp = (void *) c_idle.idle->thread.sp;
788 
789 	/* start_ip had better be page-aligned! */
790 	start_ip = setup_trampoline();
791 
792 	/* So we see what's up */
793 	announce_cpu(cpu, apicid);
794 
795 	/*
796 	 * This grunge runs the startup process for
797 	 * the targeted processor.
798 	 */
799 
800 	atomic_set(&init_deasserted, 0);
801 
802 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
803 
804 		pr_debug("Setting warm reset code and vector.\n");
805 
806 		smpboot_setup_warm_reset_vector(start_ip);
807 		/*
808 		 * Be paranoid about clearing APIC errors.
809 		*/
810 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
811 			apic_write(APIC_ESR, 0);
812 			apic_read(APIC_ESR);
813 		}
814 	}
815 
816 	/*
817 	 * Kick the secondary CPU. Use the method in the APIC driver
818 	 * if it's defined - or use an INIT boot APIC message otherwise:
819 	 */
820 	if (apic->wakeup_secondary_cpu)
821 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
822 	else
823 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
824 
825 	if (!boot_error) {
826 		/*
827 		 * allow APs to start initializing.
828 		 */
829 		pr_debug("Before Callout %d.\n", cpu);
830 		cpumask_set_cpu(cpu, cpu_callout_mask);
831 		pr_debug("After Callout %d.\n", cpu);
832 
833 		/*
834 		 * Wait 5s total for a response
835 		 */
836 		for (timeout = 0; timeout < 50000; timeout++) {
837 			if (cpumask_test_cpu(cpu, cpu_callin_mask))
838 				break;	/* It has booted */
839 			udelay(100);
840 			/*
841 			 * Allow other tasks to run while we wait for the
842 			 * AP to come online. This also gives a chance
843 			 * for the MTRR work(triggered by the AP coming online)
844 			 * to be completed in the stop machine context.
845 			 */
846 			schedule();
847 		}
848 
849 		if (cpumask_test_cpu(cpu, cpu_callin_mask))
850 			pr_debug("CPU%d: has booted.\n", cpu);
851 		else {
852 			boot_error = 1;
853 			if (*((volatile unsigned char *)trampoline_base)
854 					== 0xA5)
855 				/* trampoline started but...? */
856 				pr_err("CPU%d: Stuck ??\n", cpu);
857 			else
858 				/* trampoline code not run */
859 				pr_err("CPU%d: Not responding.\n", cpu);
860 			if (apic->inquire_remote_apic)
861 				apic->inquire_remote_apic(apicid);
862 		}
863 	}
864 
865 	if (boot_error) {
866 		/* Try to put things back the way they were before ... */
867 		numa_remove_cpu(cpu); /* was set by numa_add_cpu */
868 
869 		/* was set by do_boot_cpu() */
870 		cpumask_clear_cpu(cpu, cpu_callout_mask);
871 
872 		/* was set by cpu_init() */
873 		cpumask_clear_cpu(cpu, cpu_initialized_mask);
874 
875 		set_cpu_present(cpu, false);
876 		per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
877 	}
878 
879 	/* mark "stuck" area as not stuck */
880 	*((volatile unsigned long *)trampoline_base) = 0;
881 
882 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
883 		/*
884 		 * Cleanup possible dangling ends...
885 		 */
886 		smpboot_restore_warm_reset_vector();
887 	}
888 
889 	destroy_work_on_stack(&c_idle.work);
890 	return boot_error;
891 }
892 
893 int __cpuinit native_cpu_up(unsigned int cpu)
894 {
895 	int apicid = apic->cpu_present_to_apicid(cpu);
896 	unsigned long flags;
897 	int err;
898 
899 	WARN_ON(irqs_disabled());
900 
901 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
902 
903 	if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
904 	    !physid_isset(apicid, phys_cpu_present_map)) {
905 		printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
906 		return -EINVAL;
907 	}
908 
909 	/*
910 	 * Already booted CPU?
911 	 */
912 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
913 		pr_debug("do_boot_cpu %d Already started\n", cpu);
914 		return -ENOSYS;
915 	}
916 
917 	/*
918 	 * Save current MTRR state in case it was changed since early boot
919 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
920 	 */
921 	mtrr_save_state();
922 
923 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
924 
925 	err = do_boot_cpu(apicid, cpu);
926 
927 	if (err) {
928 		pr_debug("do_boot_cpu failed %d\n", err);
929 		return -EIO;
930 	}
931 
932 	/*
933 	 * Check TSC synchronization with the AP (keep irqs disabled
934 	 * while doing so):
935 	 */
936 	local_irq_save(flags);
937 	check_tsc_sync_source(cpu);
938 	local_irq_restore(flags);
939 
940 	while (!cpu_online(cpu)) {
941 		cpu_relax();
942 		touch_nmi_watchdog();
943 	}
944 
945 	return 0;
946 }
947 
948 /*
949  * Fall back to non SMP mode after errors.
950  *
951  * RED-PEN audit/test this more. I bet there is more state messed up here.
952  */
953 static __init void disable_smp(void)
954 {
955 	init_cpu_present(cpumask_of(0));
956 	init_cpu_possible(cpumask_of(0));
957 	smpboot_clear_io_apic_irqs();
958 
959 	if (smp_found_config)
960 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
961 	else
962 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
963 	map_cpu_to_logical_apicid();
964 	cpumask_set_cpu(0, cpu_sibling_mask(0));
965 	cpumask_set_cpu(0, cpu_core_mask(0));
966 }
967 
968 /*
969  * Various sanity checks.
970  */
971 static int __init smp_sanity_check(unsigned max_cpus)
972 {
973 	preempt_disable();
974 
975 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
976 	if (def_to_bigsmp && nr_cpu_ids > 8) {
977 		unsigned int cpu;
978 		unsigned nr;
979 
980 		printk(KERN_WARNING
981 		       "More than 8 CPUs detected - skipping them.\n"
982 		       "Use CONFIG_X86_BIGSMP.\n");
983 
984 		nr = 0;
985 		for_each_present_cpu(cpu) {
986 			if (nr >= 8)
987 				set_cpu_present(cpu, false);
988 			nr++;
989 		}
990 
991 		nr = 0;
992 		for_each_possible_cpu(cpu) {
993 			if (nr >= 8)
994 				set_cpu_possible(cpu, false);
995 			nr++;
996 		}
997 
998 		nr_cpu_ids = 8;
999 	}
1000 #endif
1001 
1002 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1003 		printk(KERN_WARNING
1004 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
1005 			hard_smp_processor_id());
1006 
1007 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1008 	}
1009 
1010 	/*
1011 	 * If we couldn't find an SMP configuration at boot time,
1012 	 * get out of here now!
1013 	 */
1014 	if (!smp_found_config && !acpi_lapic) {
1015 		preempt_enable();
1016 		printk(KERN_NOTICE "SMP motherboard not detected.\n");
1017 		disable_smp();
1018 		if (APIC_init_uniprocessor())
1019 			printk(KERN_NOTICE "Local APIC not detected."
1020 					   " Using dummy APIC emulation.\n");
1021 		return -1;
1022 	}
1023 
1024 	/*
1025 	 * Should not be necessary because the MP table should list the boot
1026 	 * CPU too, but we do it for the sake of robustness anyway.
1027 	 */
1028 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1029 		printk(KERN_NOTICE
1030 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
1031 			boot_cpu_physical_apicid);
1032 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1033 	}
1034 	preempt_enable();
1035 
1036 	/*
1037 	 * If we couldn't find a local APIC, then get out of here now!
1038 	 */
1039 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1040 	    !cpu_has_apic) {
1041 		if (!disable_apic) {
1042 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
1043 				boot_cpu_physical_apicid);
1044 			pr_err("... forcing use of dummy APIC emulation."
1045 				"(tell your hw vendor)\n");
1046 		}
1047 		smpboot_clear_io_apic();
1048 		arch_disable_smp_support();
1049 		return -1;
1050 	}
1051 
1052 	verify_local_APIC();
1053 
1054 	/*
1055 	 * If SMP should be disabled, then really disable it!
1056 	 */
1057 	if (!max_cpus) {
1058 		printk(KERN_INFO "SMP mode deactivated.\n");
1059 		smpboot_clear_io_apic();
1060 
1061 		localise_nmi_watchdog();
1062 
1063 		connect_bsp_APIC();
1064 		setup_local_APIC();
1065 		end_local_APIC_setup();
1066 		return -1;
1067 	}
1068 
1069 	return 0;
1070 }
1071 
1072 static void __init smp_cpu_index_default(void)
1073 {
1074 	int i;
1075 	struct cpuinfo_x86 *c;
1076 
1077 	for_each_possible_cpu(i) {
1078 		c = &cpu_data(i);
1079 		/* mark all to hotplug */
1080 		c->cpu_index = nr_cpu_ids;
1081 	}
1082 }
1083 
1084 /*
1085  * Prepare for SMP bootup.  The MP table or ACPI has been read
1086  * earlier.  Just do some sanity checking here and enable APIC mode.
1087  */
1088 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1089 {
1090 	unsigned int i;
1091 
1092 	preempt_disable();
1093 	smp_cpu_index_default();
1094 	current_cpu_data = boot_cpu_data;
1095 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1096 	mb();
1097 	/*
1098 	 * Setup boot CPU information
1099 	 */
1100 	smp_store_cpu_info(0); /* Final full version of the data */
1101 #ifdef CONFIG_X86_32
1102 	boot_cpu_logical_apicid = logical_smp_processor_id();
1103 #endif
1104 	current_thread_info()->cpu = 0;  /* needed? */
1105 	for_each_possible_cpu(i) {
1106 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1107 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1108 		zalloc_cpumask_var(&cpu_data(i).llc_shared_map, GFP_KERNEL);
1109 	}
1110 	set_cpu_sibling_map(0);
1111 
1112 	enable_IR_x2apic();
1113 	default_setup_apic_routing();
1114 
1115 	if (smp_sanity_check(max_cpus) < 0) {
1116 		printk(KERN_INFO "SMP disabled\n");
1117 		disable_smp();
1118 		goto out;
1119 	}
1120 
1121 	preempt_disable();
1122 	if (read_apic_id() != boot_cpu_physical_apicid) {
1123 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1124 		     read_apic_id(), boot_cpu_physical_apicid);
1125 		/* Or can we switch back to PIC here? */
1126 	}
1127 	preempt_enable();
1128 
1129 	connect_bsp_APIC();
1130 
1131 	/*
1132 	 * Switch from PIC to APIC mode.
1133 	 */
1134 	setup_local_APIC();
1135 
1136 	/*
1137 	 * Enable IO APIC before setting up error vector
1138 	 */
1139 	if (!skip_ioapic_setup && nr_ioapics)
1140 		enable_IO_APIC();
1141 
1142 	end_local_APIC_setup();
1143 
1144 	map_cpu_to_logical_apicid();
1145 
1146 	if (apic->setup_portio_remap)
1147 		apic->setup_portio_remap();
1148 
1149 	smpboot_setup_io_apic();
1150 	/*
1151 	 * Set up local APIC timer on boot CPU.
1152 	 */
1153 
1154 	printk(KERN_INFO "CPU%d: ", 0);
1155 	print_cpu_info(&cpu_data(0));
1156 	x86_init.timers.setup_percpu_clockev();
1157 
1158 	if (is_uv_system())
1159 		uv_system_init();
1160 
1161 	set_mtrr_aps_delayed_init();
1162 out:
1163 	preempt_enable();
1164 }
1165 
1166 void arch_enable_nonboot_cpus_begin(void)
1167 {
1168 	set_mtrr_aps_delayed_init();
1169 }
1170 
1171 void arch_enable_nonboot_cpus_end(void)
1172 {
1173 	mtrr_aps_init();
1174 }
1175 
1176 /*
1177  * Early setup to make printk work.
1178  */
1179 void __init native_smp_prepare_boot_cpu(void)
1180 {
1181 	int me = smp_processor_id();
1182 	switch_to_new_gdt(me);
1183 	/* already set me in cpu_online_mask in boot_cpu_init() */
1184 	cpumask_set_cpu(me, cpu_callout_mask);
1185 	per_cpu(cpu_state, me) = CPU_ONLINE;
1186 }
1187 
1188 void __init native_smp_cpus_done(unsigned int max_cpus)
1189 {
1190 	pr_debug("Boot done.\n");
1191 
1192 	impress_friends();
1193 #ifdef CONFIG_X86_IO_APIC
1194 	setup_ioapic_dest();
1195 #endif
1196 	check_nmi_watchdog();
1197 	mtrr_aps_init();
1198 }
1199 
1200 static int __initdata setup_possible_cpus = -1;
1201 static int __init _setup_possible_cpus(char *str)
1202 {
1203 	get_option(&str, &setup_possible_cpus);
1204 	return 0;
1205 }
1206 early_param("possible_cpus", _setup_possible_cpus);
1207 
1208 
1209 /*
1210  * cpu_possible_mask should be static, it cannot change as cpu's
1211  * are onlined, or offlined. The reason is per-cpu data-structures
1212  * are allocated by some modules at init time, and dont expect to
1213  * do this dynamically on cpu arrival/departure.
1214  * cpu_present_mask on the other hand can change dynamically.
1215  * In case when cpu_hotplug is not compiled, then we resort to current
1216  * behaviour, which is cpu_possible == cpu_present.
1217  * - Ashok Raj
1218  *
1219  * Three ways to find out the number of additional hotplug CPUs:
1220  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1221  * - The user can overwrite it with possible_cpus=NUM
1222  * - Otherwise don't reserve additional CPUs.
1223  * We do this because additional CPUs waste a lot of memory.
1224  * -AK
1225  */
1226 __init void prefill_possible_map(void)
1227 {
1228 	int i, possible;
1229 
1230 	/* no processor from mptable or madt */
1231 	if (!num_processors)
1232 		num_processors = 1;
1233 
1234 	i = setup_max_cpus ?: 1;
1235 	if (setup_possible_cpus == -1) {
1236 		possible = num_processors;
1237 #ifdef CONFIG_HOTPLUG_CPU
1238 		if (setup_max_cpus)
1239 			possible += disabled_cpus;
1240 #else
1241 		if (possible > i)
1242 			possible = i;
1243 #endif
1244 	} else
1245 		possible = setup_possible_cpus;
1246 
1247 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1248 
1249 	/* nr_cpu_ids could be reduced via nr_cpus= */
1250 	if (possible > nr_cpu_ids) {
1251 		printk(KERN_WARNING
1252 			"%d Processors exceeds NR_CPUS limit of %d\n",
1253 			possible, nr_cpu_ids);
1254 		possible = nr_cpu_ids;
1255 	}
1256 
1257 #ifdef CONFIG_HOTPLUG_CPU
1258 	if (!setup_max_cpus)
1259 #endif
1260 	if (possible > i) {
1261 		printk(KERN_WARNING
1262 			"%d Processors exceeds max_cpus limit of %u\n",
1263 			possible, setup_max_cpus);
1264 		possible = i;
1265 	}
1266 
1267 	printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1268 		possible, max_t(int, possible - num_processors, 0));
1269 
1270 	for (i = 0; i < possible; i++)
1271 		set_cpu_possible(i, true);
1272 	for (; i < NR_CPUS; i++)
1273 		set_cpu_possible(i, false);
1274 
1275 	nr_cpu_ids = possible;
1276 }
1277 
1278 #ifdef CONFIG_HOTPLUG_CPU
1279 
1280 static void remove_siblinginfo(int cpu)
1281 {
1282 	int sibling;
1283 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1284 
1285 	for_each_cpu(sibling, cpu_core_mask(cpu)) {
1286 		cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1287 		/*/
1288 		 * last thread sibling in this cpu core going down
1289 		 */
1290 		if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1291 			cpu_data(sibling).booted_cores--;
1292 	}
1293 
1294 	for_each_cpu(sibling, cpu_sibling_mask(cpu))
1295 		cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1296 	cpumask_clear(cpu_sibling_mask(cpu));
1297 	cpumask_clear(cpu_core_mask(cpu));
1298 	c->phys_proc_id = 0;
1299 	c->cpu_core_id = 0;
1300 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1301 }
1302 
1303 static void __ref remove_cpu_from_maps(int cpu)
1304 {
1305 	set_cpu_online(cpu, false);
1306 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1307 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1308 	/* was set by cpu_init() */
1309 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1310 	numa_remove_cpu(cpu);
1311 }
1312 
1313 void cpu_disable_common(void)
1314 {
1315 	int cpu = smp_processor_id();
1316 
1317 	remove_siblinginfo(cpu);
1318 
1319 	/* It's now safe to remove this processor from the online map */
1320 	lock_vector_lock();
1321 	remove_cpu_from_maps(cpu);
1322 	unlock_vector_lock();
1323 	fixup_irqs();
1324 }
1325 
1326 int native_cpu_disable(void)
1327 {
1328 	int cpu = smp_processor_id();
1329 
1330 	/*
1331 	 * Perhaps use cpufreq to drop frequency, but that could go
1332 	 * into generic code.
1333 	 *
1334 	 * We won't take down the boot processor on i386 due to some
1335 	 * interrupts only being able to be serviced by the BSP.
1336 	 * Especially so if we're not using an IOAPIC	-zwane
1337 	 */
1338 	if (cpu == 0)
1339 		return -EBUSY;
1340 
1341 	if (nmi_watchdog == NMI_LOCAL_APIC)
1342 		stop_apic_nmi_watchdog(NULL);
1343 	clear_local_APIC();
1344 
1345 	cpu_disable_common();
1346 	return 0;
1347 }
1348 
1349 void native_cpu_die(unsigned int cpu)
1350 {
1351 	/* We don't do anything here: idle task is faking death itself. */
1352 	unsigned int i;
1353 
1354 	for (i = 0; i < 10; i++) {
1355 		/* They ack this in play_dead by setting CPU_DEAD */
1356 		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1357 			if (system_state == SYSTEM_RUNNING)
1358 				pr_info("CPU %u is now offline\n", cpu);
1359 
1360 			if (1 == num_online_cpus())
1361 				alternatives_smp_switch(0);
1362 			return;
1363 		}
1364 		msleep(100);
1365 	}
1366 	pr_err("CPU %u didn't die...\n", cpu);
1367 }
1368 
1369 void play_dead_common(void)
1370 {
1371 	idle_task_exit();
1372 	reset_lazy_tlbstate();
1373 	irq_ctx_exit(raw_smp_processor_id());
1374 	c1e_remove_cpu(raw_smp_processor_id());
1375 
1376 	mb();
1377 	/* Ack it */
1378 	__get_cpu_var(cpu_state) = CPU_DEAD;
1379 
1380 	/*
1381 	 * With physical CPU hotplug, we should halt the cpu
1382 	 */
1383 	local_irq_disable();
1384 }
1385 
1386 void native_play_dead(void)
1387 {
1388 	play_dead_common();
1389 	tboot_shutdown(TB_SHUTDOWN_WFS);
1390 	wbinvd_halt();
1391 }
1392 
1393 #else /* ... !CONFIG_HOTPLUG_CPU */
1394 int native_cpu_disable(void)
1395 {
1396 	return -ENOSYS;
1397 }
1398 
1399 void native_cpu_die(unsigned int cpu)
1400 {
1401 	/* We said "no" in __cpu_disable */
1402 	BUG();
1403 }
1404 
1405 void native_play_dead(void)
1406 {
1407 	BUG();
1408 }
1409 
1410 #endif
1411