xref: /linux/arch/x86/kernel/smpboot.c (revision 092e0e7e520a1fca03e13c9f2d157432a8657ff2)
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/mwait.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 	cpu_init();
315 	preempt_disable();
316 	smp_callin();
317 
318 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
319 	barrier();
320 	/*
321 	 * Check TSC synchronization with the BP:
322 	 */
323 	check_tsc_sync_target();
324 
325 	if (nmi_watchdog == NMI_IO_APIC) {
326 		legacy_pic->mask(0);
327 		enable_NMI_through_LVT0();
328 		legacy_pic->unmask(0);
329 	}
330 
331 	/* This must be done before setting cpu_online_mask */
332 	set_cpu_sibling_map(raw_smp_processor_id());
333 	wmb();
334 
335 	/*
336 	 * We need to hold call_lock, so there is no inconsistency
337 	 * between the time smp_call_function() determines number of
338 	 * IPI recipients, and the time when the determination is made
339 	 * for which cpus receive the IPI. Holding this
340 	 * lock helps us to not include this cpu in a currently in progress
341 	 * smp_call_function().
342 	 *
343 	 * We need to hold vector_lock so there the set of online cpus
344 	 * does not change while we are assigning vectors to cpus.  Holding
345 	 * this lock ensures we don't half assign or remove an irq from a cpu.
346 	 */
347 	ipi_call_lock();
348 	lock_vector_lock();
349 	set_cpu_online(smp_processor_id(), true);
350 	unlock_vector_lock();
351 	ipi_call_unlock();
352 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
353 	x86_platform.nmi_init();
354 
355 	/* enable local interrupts */
356 	local_irq_enable();
357 
358 	/* to prevent fake stack check failure in clock setup */
359 	boot_init_stack_canary();
360 
361 	x86_cpuinit.setup_percpu_clockev();
362 
363 	wmb();
364 	cpu_idle();
365 }
366 
367 #ifdef CONFIG_CPUMASK_OFFSTACK
368 /* In this case, llc_shared_map is a pointer to a cpumask. */
369 static inline void copy_cpuinfo_x86(struct cpuinfo_x86 *dst,
370 				    const struct cpuinfo_x86 *src)
371 {
372 	struct cpumask *llc = dst->llc_shared_map;
373 	*dst = *src;
374 	dst->llc_shared_map = llc;
375 }
376 #else
377 static inline void copy_cpuinfo_x86(struct cpuinfo_x86 *dst,
378 				    const struct cpuinfo_x86 *src)
379 {
380 	*dst = *src;
381 }
382 #endif /* CONFIG_CPUMASK_OFFSTACK */
383 
384 /*
385  * The bootstrap kernel entry code has set these up. Save them for
386  * a given CPU
387  */
388 
389 void __cpuinit smp_store_cpu_info(int id)
390 {
391 	struct cpuinfo_x86 *c = &cpu_data(id);
392 
393 	copy_cpuinfo_x86(c, &boot_cpu_data);
394 	c->cpu_index = id;
395 	if (id != 0)
396 		identify_secondary_cpu(c);
397 }
398 
399 static void __cpuinit link_thread_siblings(int cpu1, int cpu2)
400 {
401 	struct cpuinfo_x86 *c1 = &cpu_data(cpu1);
402 	struct cpuinfo_x86 *c2 = &cpu_data(cpu2);
403 
404 	cpumask_set_cpu(cpu1, cpu_sibling_mask(cpu2));
405 	cpumask_set_cpu(cpu2, cpu_sibling_mask(cpu1));
406 	cpumask_set_cpu(cpu1, cpu_core_mask(cpu2));
407 	cpumask_set_cpu(cpu2, cpu_core_mask(cpu1));
408 	cpumask_set_cpu(cpu1, c2->llc_shared_map);
409 	cpumask_set_cpu(cpu2, c1->llc_shared_map);
410 }
411 
412 
413 void __cpuinit set_cpu_sibling_map(int cpu)
414 {
415 	int i;
416 	struct cpuinfo_x86 *c = &cpu_data(cpu);
417 
418 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
419 
420 	if (smp_num_siblings > 1) {
421 		for_each_cpu(i, cpu_sibling_setup_mask) {
422 			struct cpuinfo_x86 *o = &cpu_data(i);
423 
424 			if (cpu_has(c, X86_FEATURE_TOPOEXT)) {
425 				if (c->phys_proc_id == o->phys_proc_id &&
426 				    c->compute_unit_id == o->compute_unit_id)
427 					link_thread_siblings(cpu, i);
428 			} else if (c->phys_proc_id == o->phys_proc_id &&
429 				   c->cpu_core_id == o->cpu_core_id) {
430 				link_thread_siblings(cpu, i);
431 			}
432 		}
433 	} else {
434 		cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
435 	}
436 
437 	cpumask_set_cpu(cpu, c->llc_shared_map);
438 
439 	if (current_cpu_data.x86_max_cores == 1) {
440 		cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
441 		c->booted_cores = 1;
442 		return;
443 	}
444 
445 	for_each_cpu(i, cpu_sibling_setup_mask) {
446 		if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
447 		    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
448 			cpumask_set_cpu(i, c->llc_shared_map);
449 			cpumask_set_cpu(cpu, cpu_data(i).llc_shared_map);
450 		}
451 		if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
452 			cpumask_set_cpu(i, cpu_core_mask(cpu));
453 			cpumask_set_cpu(cpu, cpu_core_mask(i));
454 			/*
455 			 *  Does this new cpu bringup a new core?
456 			 */
457 			if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
458 				/*
459 				 * for each core in package, increment
460 				 * the booted_cores for this new cpu
461 				 */
462 				if (cpumask_first(cpu_sibling_mask(i)) == i)
463 					c->booted_cores++;
464 				/*
465 				 * increment the core count for all
466 				 * the other cpus in this package
467 				 */
468 				if (i != cpu)
469 					cpu_data(i).booted_cores++;
470 			} else if (i != cpu && !c->booted_cores)
471 				c->booted_cores = cpu_data(i).booted_cores;
472 		}
473 	}
474 }
475 
476 /* maps the cpu to the sched domain representing multi-core */
477 const struct cpumask *cpu_coregroup_mask(int cpu)
478 {
479 	struct cpuinfo_x86 *c = &cpu_data(cpu);
480 	/*
481 	 * For perf, we return last level cache shared map.
482 	 * And for power savings, we return cpu_core_map
483 	 */
484 	if ((sched_mc_power_savings || sched_smt_power_savings) &&
485 	    !(cpu_has(c, X86_FEATURE_AMD_DCM)))
486 		return cpu_core_mask(cpu);
487 	else
488 		return c->llc_shared_map;
489 }
490 
491 static void impress_friends(void)
492 {
493 	int cpu;
494 	unsigned long bogosum = 0;
495 	/*
496 	 * Allow the user to impress friends.
497 	 */
498 	pr_debug("Before bogomips.\n");
499 	for_each_possible_cpu(cpu)
500 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
501 			bogosum += cpu_data(cpu).loops_per_jiffy;
502 	printk(KERN_INFO
503 		"Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
504 		num_online_cpus(),
505 		bogosum/(500000/HZ),
506 		(bogosum/(5000/HZ))%100);
507 
508 	pr_debug("Before bogocount - setting activated=1.\n");
509 }
510 
511 void __inquire_remote_apic(int apicid)
512 {
513 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
514 	char *names[] = { "ID", "VERSION", "SPIV" };
515 	int timeout;
516 	u32 status;
517 
518 	printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
519 
520 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
521 		printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
522 
523 		/*
524 		 * Wait for idle.
525 		 */
526 		status = safe_apic_wait_icr_idle();
527 		if (status)
528 			printk(KERN_CONT
529 			       "a previous APIC delivery may have failed\n");
530 
531 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
532 
533 		timeout = 0;
534 		do {
535 			udelay(100);
536 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
537 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
538 
539 		switch (status) {
540 		case APIC_ICR_RR_VALID:
541 			status = apic_read(APIC_RRR);
542 			printk(KERN_CONT "%08x\n", status);
543 			break;
544 		default:
545 			printk(KERN_CONT "failed\n");
546 		}
547 	}
548 }
549 
550 /*
551  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
552  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
553  * won't ... remember to clear down the APIC, etc later.
554  */
555 int __cpuinit
556 wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
557 {
558 	unsigned long send_status, accept_status = 0;
559 	int maxlvt;
560 
561 	/* Target chip */
562 	/* Boot on the stack */
563 	/* Kick the second */
564 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, logical_apicid);
565 
566 	pr_debug("Waiting for send to finish...\n");
567 	send_status = safe_apic_wait_icr_idle();
568 
569 	/*
570 	 * Give the other CPU some time to accept the IPI.
571 	 */
572 	udelay(200);
573 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
574 		maxlvt = lapic_get_maxlvt();
575 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
576 			apic_write(APIC_ESR, 0);
577 		accept_status = (apic_read(APIC_ESR) & 0xEF);
578 	}
579 	pr_debug("NMI sent.\n");
580 
581 	if (send_status)
582 		printk(KERN_ERR "APIC never delivered???\n");
583 	if (accept_status)
584 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
585 
586 	return (send_status | accept_status);
587 }
588 
589 static int __cpuinit
590 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
591 {
592 	unsigned long send_status, accept_status = 0;
593 	int maxlvt, num_starts, j;
594 
595 	maxlvt = lapic_get_maxlvt();
596 
597 	/*
598 	 * Be paranoid about clearing APIC errors.
599 	 */
600 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
601 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
602 			apic_write(APIC_ESR, 0);
603 		apic_read(APIC_ESR);
604 	}
605 
606 	pr_debug("Asserting INIT.\n");
607 
608 	/*
609 	 * Turn INIT on target chip
610 	 */
611 	/*
612 	 * Send IPI
613 	 */
614 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
615 		       phys_apicid);
616 
617 	pr_debug("Waiting for send to finish...\n");
618 	send_status = safe_apic_wait_icr_idle();
619 
620 	mdelay(10);
621 
622 	pr_debug("Deasserting INIT.\n");
623 
624 	/* Target chip */
625 	/* Send IPI */
626 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
627 
628 	pr_debug("Waiting for send to finish...\n");
629 	send_status = safe_apic_wait_icr_idle();
630 
631 	mb();
632 	atomic_set(&init_deasserted, 1);
633 
634 	/*
635 	 * Should we send STARTUP IPIs ?
636 	 *
637 	 * Determine this based on the APIC version.
638 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
639 	 */
640 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
641 		num_starts = 2;
642 	else
643 		num_starts = 0;
644 
645 	/*
646 	 * Paravirt / VMI wants a startup IPI hook here to set up the
647 	 * target processor state.
648 	 */
649 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
650 			 (unsigned long)stack_start.sp);
651 
652 	/*
653 	 * Run STARTUP IPI loop.
654 	 */
655 	pr_debug("#startup loops: %d.\n", num_starts);
656 
657 	for (j = 1; j <= num_starts; j++) {
658 		pr_debug("Sending STARTUP #%d.\n", j);
659 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
660 			apic_write(APIC_ESR, 0);
661 		apic_read(APIC_ESR);
662 		pr_debug("After apic_write.\n");
663 
664 		/*
665 		 * STARTUP IPI
666 		 */
667 
668 		/* Target chip */
669 		/* Boot on the stack */
670 		/* Kick the second */
671 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
672 			       phys_apicid);
673 
674 		/*
675 		 * Give the other CPU some time to accept the IPI.
676 		 */
677 		udelay(300);
678 
679 		pr_debug("Startup point 1.\n");
680 
681 		pr_debug("Waiting for send to finish...\n");
682 		send_status = safe_apic_wait_icr_idle();
683 
684 		/*
685 		 * Give the other CPU some time to accept the IPI.
686 		 */
687 		udelay(200);
688 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
689 			apic_write(APIC_ESR, 0);
690 		accept_status = (apic_read(APIC_ESR) & 0xEF);
691 		if (send_status || accept_status)
692 			break;
693 	}
694 	pr_debug("After Startup.\n");
695 
696 	if (send_status)
697 		printk(KERN_ERR "APIC never delivered???\n");
698 	if (accept_status)
699 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
700 
701 	return (send_status | accept_status);
702 }
703 
704 struct create_idle {
705 	struct work_struct work;
706 	struct task_struct *idle;
707 	struct completion done;
708 	int cpu;
709 };
710 
711 static void __cpuinit do_fork_idle(struct work_struct *work)
712 {
713 	struct create_idle *c_idle =
714 		container_of(work, struct create_idle, work);
715 
716 	c_idle->idle = fork_idle(c_idle->cpu);
717 	complete(&c_idle->done);
718 }
719 
720 /* reduce the number of lines printed when booting a large cpu count system */
721 static void __cpuinit announce_cpu(int cpu, int apicid)
722 {
723 	static int current_node = -1;
724 	int node = early_cpu_to_node(cpu);
725 
726 	if (system_state == SYSTEM_BOOTING) {
727 		if (node != current_node) {
728 			if (current_node > (-1))
729 				pr_cont(" Ok.\n");
730 			current_node = node;
731 			pr_info("Booting Node %3d, Processors ", node);
732 		}
733 		pr_cont(" #%d%s", cpu, cpu == (nr_cpu_ids - 1) ? " Ok.\n" : "");
734 		return;
735 	} else
736 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
737 			node, cpu, apicid);
738 }
739 
740 /*
741  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
742  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
743  * Returns zero if CPU booted OK, else error code from
744  * ->wakeup_secondary_cpu.
745  */
746 static int __cpuinit do_boot_cpu(int apicid, int cpu)
747 {
748 	unsigned long boot_error = 0;
749 	unsigned long start_ip;
750 	int timeout;
751 	struct create_idle c_idle = {
752 		.cpu	= cpu,
753 		.done	= COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
754 	};
755 
756 	INIT_WORK_ON_STACK(&c_idle.work, do_fork_idle);
757 
758 	alternatives_smp_switch(1);
759 
760 	c_idle.idle = get_idle_for_cpu(cpu);
761 
762 	/*
763 	 * We can't use kernel_thread since we must avoid to
764 	 * reschedule the child.
765 	 */
766 	if (c_idle.idle) {
767 		c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
768 			(THREAD_SIZE +  task_stack_page(c_idle.idle))) - 1);
769 		init_idle(c_idle.idle, cpu);
770 		goto do_rest;
771 	}
772 
773 	schedule_work(&c_idle.work);
774 	wait_for_completion(&c_idle.done);
775 
776 	if (IS_ERR(c_idle.idle)) {
777 		printk("failed fork for CPU %d\n", cpu);
778 		destroy_work_on_stack(&c_idle.work);
779 		return PTR_ERR(c_idle.idle);
780 	}
781 
782 	set_idle_for_cpu(cpu, c_idle.idle);
783 do_rest:
784 	per_cpu(current_task, cpu) = c_idle.idle;
785 #ifdef CONFIG_X86_32
786 	/* Stack for startup_32 can be just as for start_secondary onwards */
787 	irq_ctx_init(cpu);
788 	initial_page_table = __pa(&trampoline_pg_dir);
789 #else
790 	clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
791 	initial_gs = per_cpu_offset(cpu);
792 	per_cpu(kernel_stack, cpu) =
793 		(unsigned long)task_stack_page(c_idle.idle) -
794 		KERNEL_STACK_OFFSET + THREAD_SIZE;
795 #endif
796 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
797 	initial_code = (unsigned long)start_secondary;
798 	stack_start.sp = (void *) c_idle.idle->thread.sp;
799 
800 	/* start_ip had better be page-aligned! */
801 	start_ip = setup_trampoline();
802 
803 	/* So we see what's up */
804 	announce_cpu(cpu, apicid);
805 
806 	/*
807 	 * This grunge runs the startup process for
808 	 * the targeted processor.
809 	 */
810 
811 	atomic_set(&init_deasserted, 0);
812 
813 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
814 
815 		pr_debug("Setting warm reset code and vector.\n");
816 
817 		smpboot_setup_warm_reset_vector(start_ip);
818 		/*
819 		 * Be paranoid about clearing APIC errors.
820 		*/
821 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
822 			apic_write(APIC_ESR, 0);
823 			apic_read(APIC_ESR);
824 		}
825 	}
826 
827 	/*
828 	 * Kick the secondary CPU. Use the method in the APIC driver
829 	 * if it's defined - or use an INIT boot APIC message otherwise:
830 	 */
831 	if (apic->wakeup_secondary_cpu)
832 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
833 	else
834 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
835 
836 	if (!boot_error) {
837 		/*
838 		 * allow APs to start initializing.
839 		 */
840 		pr_debug("Before Callout %d.\n", cpu);
841 		cpumask_set_cpu(cpu, cpu_callout_mask);
842 		pr_debug("After Callout %d.\n", cpu);
843 
844 		/*
845 		 * Wait 5s total for a response
846 		 */
847 		for (timeout = 0; timeout < 50000; timeout++) {
848 			if (cpumask_test_cpu(cpu, cpu_callin_mask))
849 				break;	/* It has booted */
850 			udelay(100);
851 			/*
852 			 * Allow other tasks to run while we wait for the
853 			 * AP to come online. This also gives a chance
854 			 * for the MTRR work(triggered by the AP coming online)
855 			 * to be completed in the stop machine context.
856 			 */
857 			schedule();
858 		}
859 
860 		if (cpumask_test_cpu(cpu, cpu_callin_mask))
861 			pr_debug("CPU%d: has booted.\n", cpu);
862 		else {
863 			boot_error = 1;
864 			if (*((volatile unsigned char *)trampoline_base)
865 					== 0xA5)
866 				/* trampoline started but...? */
867 				pr_err("CPU%d: Stuck ??\n", cpu);
868 			else
869 				/* trampoline code not run */
870 				pr_err("CPU%d: Not responding.\n", cpu);
871 			if (apic->inquire_remote_apic)
872 				apic->inquire_remote_apic(apicid);
873 		}
874 	}
875 
876 	if (boot_error) {
877 		/* Try to put things back the way they were before ... */
878 		numa_remove_cpu(cpu); /* was set by numa_add_cpu */
879 
880 		/* was set by do_boot_cpu() */
881 		cpumask_clear_cpu(cpu, cpu_callout_mask);
882 
883 		/* was set by cpu_init() */
884 		cpumask_clear_cpu(cpu, cpu_initialized_mask);
885 
886 		set_cpu_present(cpu, false);
887 		per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
888 	}
889 
890 	/* mark "stuck" area as not stuck */
891 	*((volatile unsigned long *)trampoline_base) = 0;
892 
893 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
894 		/*
895 		 * Cleanup possible dangling ends...
896 		 */
897 		smpboot_restore_warm_reset_vector();
898 	}
899 
900 	destroy_work_on_stack(&c_idle.work);
901 	return boot_error;
902 }
903 
904 int __cpuinit native_cpu_up(unsigned int cpu)
905 {
906 	int apicid = apic->cpu_present_to_apicid(cpu);
907 	unsigned long flags;
908 	int err;
909 
910 	WARN_ON(irqs_disabled());
911 
912 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
913 
914 	if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
915 	    !physid_isset(apicid, phys_cpu_present_map)) {
916 		printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
917 		return -EINVAL;
918 	}
919 
920 	/*
921 	 * Already booted CPU?
922 	 */
923 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
924 		pr_debug("do_boot_cpu %d Already started\n", cpu);
925 		return -ENOSYS;
926 	}
927 
928 	/*
929 	 * Save current MTRR state in case it was changed since early boot
930 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
931 	 */
932 	mtrr_save_state();
933 
934 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
935 
936 	err = do_boot_cpu(apicid, cpu);
937 
938 	if (err) {
939 		pr_debug("do_boot_cpu failed %d\n", err);
940 		return -EIO;
941 	}
942 
943 	/*
944 	 * Check TSC synchronization with the AP (keep irqs disabled
945 	 * while doing so):
946 	 */
947 	local_irq_save(flags);
948 	check_tsc_sync_source(cpu);
949 	local_irq_restore(flags);
950 
951 	while (!cpu_online(cpu)) {
952 		cpu_relax();
953 		touch_nmi_watchdog();
954 	}
955 
956 	return 0;
957 }
958 
959 /*
960  * Fall back to non SMP mode after errors.
961  *
962  * RED-PEN audit/test this more. I bet there is more state messed up here.
963  */
964 static __init void disable_smp(void)
965 {
966 	init_cpu_present(cpumask_of(0));
967 	init_cpu_possible(cpumask_of(0));
968 	smpboot_clear_io_apic_irqs();
969 
970 	if (smp_found_config)
971 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
972 	else
973 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
974 	map_cpu_to_logical_apicid();
975 	cpumask_set_cpu(0, cpu_sibling_mask(0));
976 	cpumask_set_cpu(0, cpu_core_mask(0));
977 }
978 
979 /*
980  * Various sanity checks.
981  */
982 static int __init smp_sanity_check(unsigned max_cpus)
983 {
984 	preempt_disable();
985 
986 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
987 	if (def_to_bigsmp && nr_cpu_ids > 8) {
988 		unsigned int cpu;
989 		unsigned nr;
990 
991 		printk(KERN_WARNING
992 		       "More than 8 CPUs detected - skipping them.\n"
993 		       "Use CONFIG_X86_BIGSMP.\n");
994 
995 		nr = 0;
996 		for_each_present_cpu(cpu) {
997 			if (nr >= 8)
998 				set_cpu_present(cpu, false);
999 			nr++;
1000 		}
1001 
1002 		nr = 0;
1003 		for_each_possible_cpu(cpu) {
1004 			if (nr >= 8)
1005 				set_cpu_possible(cpu, false);
1006 			nr++;
1007 		}
1008 
1009 		nr_cpu_ids = 8;
1010 	}
1011 #endif
1012 
1013 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1014 		printk(KERN_WARNING
1015 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
1016 			hard_smp_processor_id());
1017 
1018 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1019 	}
1020 
1021 	/*
1022 	 * If we couldn't find an SMP configuration at boot time,
1023 	 * get out of here now!
1024 	 */
1025 	if (!smp_found_config && !acpi_lapic) {
1026 		preempt_enable();
1027 		printk(KERN_NOTICE "SMP motherboard not detected.\n");
1028 		disable_smp();
1029 		if (APIC_init_uniprocessor())
1030 			printk(KERN_NOTICE "Local APIC not detected."
1031 					   " Using dummy APIC emulation.\n");
1032 		return -1;
1033 	}
1034 
1035 	/*
1036 	 * Should not be necessary because the MP table should list the boot
1037 	 * CPU too, but we do it for the sake of robustness anyway.
1038 	 */
1039 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1040 		printk(KERN_NOTICE
1041 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
1042 			boot_cpu_physical_apicid);
1043 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1044 	}
1045 	preempt_enable();
1046 
1047 	/*
1048 	 * If we couldn't find a local APIC, then get out of here now!
1049 	 */
1050 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1051 	    !cpu_has_apic) {
1052 		if (!disable_apic) {
1053 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
1054 				boot_cpu_physical_apicid);
1055 			pr_err("... forcing use of dummy APIC emulation."
1056 				"(tell your hw vendor)\n");
1057 		}
1058 		smpboot_clear_io_apic();
1059 		arch_disable_smp_support();
1060 		return -1;
1061 	}
1062 
1063 	verify_local_APIC();
1064 
1065 	/*
1066 	 * If SMP should be disabled, then really disable it!
1067 	 */
1068 	if (!max_cpus) {
1069 		printk(KERN_INFO "SMP mode deactivated.\n");
1070 		smpboot_clear_io_apic();
1071 
1072 		localise_nmi_watchdog();
1073 
1074 		connect_bsp_APIC();
1075 		setup_local_APIC();
1076 		end_local_APIC_setup();
1077 		return -1;
1078 	}
1079 
1080 	return 0;
1081 }
1082 
1083 static void __init smp_cpu_index_default(void)
1084 {
1085 	int i;
1086 	struct cpuinfo_x86 *c;
1087 
1088 	for_each_possible_cpu(i) {
1089 		c = &cpu_data(i);
1090 		/* mark all to hotplug */
1091 		c->cpu_index = nr_cpu_ids;
1092 	}
1093 }
1094 
1095 /*
1096  * Prepare for SMP bootup.  The MP table or ACPI has been read
1097  * earlier.  Just do some sanity checking here and enable APIC mode.
1098  */
1099 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1100 {
1101 	unsigned int i;
1102 
1103 	preempt_disable();
1104 	smp_cpu_index_default();
1105 	current_cpu_data = boot_cpu_data;
1106 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1107 	mb();
1108 	/*
1109 	 * Setup boot CPU information
1110 	 */
1111 	smp_store_cpu_info(0); /* Final full version of the data */
1112 #ifdef CONFIG_X86_32
1113 	boot_cpu_logical_apicid = logical_smp_processor_id();
1114 #endif
1115 	current_thread_info()->cpu = 0;  /* needed? */
1116 	for_each_possible_cpu(i) {
1117 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1118 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1119 		zalloc_cpumask_var(&cpu_data(i).llc_shared_map, GFP_KERNEL);
1120 	}
1121 	set_cpu_sibling_map(0);
1122 
1123 
1124 	if (smp_sanity_check(max_cpus) < 0) {
1125 		printk(KERN_INFO "SMP disabled\n");
1126 		disable_smp();
1127 		goto out;
1128 	}
1129 
1130 	default_setup_apic_routing();
1131 
1132 	preempt_disable();
1133 	if (read_apic_id() != boot_cpu_physical_apicid) {
1134 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1135 		     read_apic_id(), boot_cpu_physical_apicid);
1136 		/* Or can we switch back to PIC here? */
1137 	}
1138 	preempt_enable();
1139 
1140 	connect_bsp_APIC();
1141 
1142 	/*
1143 	 * Switch from PIC to APIC mode.
1144 	 */
1145 	setup_local_APIC();
1146 
1147 	/*
1148 	 * Enable IO APIC before setting up error vector
1149 	 */
1150 	if (!skip_ioapic_setup && nr_ioapics)
1151 		enable_IO_APIC();
1152 
1153 	end_local_APIC_setup();
1154 
1155 	map_cpu_to_logical_apicid();
1156 
1157 	if (apic->setup_portio_remap)
1158 		apic->setup_portio_remap();
1159 
1160 	smpboot_setup_io_apic();
1161 	/*
1162 	 * Set up local APIC timer on boot CPU.
1163 	 */
1164 
1165 	printk(KERN_INFO "CPU%d: ", 0);
1166 	print_cpu_info(&cpu_data(0));
1167 	x86_init.timers.setup_percpu_clockev();
1168 
1169 	if (is_uv_system())
1170 		uv_system_init();
1171 
1172 	set_mtrr_aps_delayed_init();
1173 out:
1174 	preempt_enable();
1175 }
1176 
1177 void arch_enable_nonboot_cpus_begin(void)
1178 {
1179 	set_mtrr_aps_delayed_init();
1180 }
1181 
1182 void arch_enable_nonboot_cpus_end(void)
1183 {
1184 	mtrr_aps_init();
1185 }
1186 
1187 /*
1188  * Early setup to make printk work.
1189  */
1190 void __init native_smp_prepare_boot_cpu(void)
1191 {
1192 	int me = smp_processor_id();
1193 	switch_to_new_gdt(me);
1194 	/* already set me in cpu_online_mask in boot_cpu_init() */
1195 	cpumask_set_cpu(me, cpu_callout_mask);
1196 	per_cpu(cpu_state, me) = CPU_ONLINE;
1197 }
1198 
1199 void __init native_smp_cpus_done(unsigned int max_cpus)
1200 {
1201 	pr_debug("Boot done.\n");
1202 
1203 	impress_friends();
1204 #ifdef CONFIG_X86_IO_APIC
1205 	setup_ioapic_dest();
1206 #endif
1207 	check_nmi_watchdog();
1208 	mtrr_aps_init();
1209 }
1210 
1211 static int __initdata setup_possible_cpus = -1;
1212 static int __init _setup_possible_cpus(char *str)
1213 {
1214 	get_option(&str, &setup_possible_cpus);
1215 	return 0;
1216 }
1217 early_param("possible_cpus", _setup_possible_cpus);
1218 
1219 
1220 /*
1221  * cpu_possible_mask should be static, it cannot change as cpu's
1222  * are onlined, or offlined. The reason is per-cpu data-structures
1223  * are allocated by some modules at init time, and dont expect to
1224  * do this dynamically on cpu arrival/departure.
1225  * cpu_present_mask on the other hand can change dynamically.
1226  * In case when cpu_hotplug is not compiled, then we resort to current
1227  * behaviour, which is cpu_possible == cpu_present.
1228  * - Ashok Raj
1229  *
1230  * Three ways to find out the number of additional hotplug CPUs:
1231  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1232  * - The user can overwrite it with possible_cpus=NUM
1233  * - Otherwise don't reserve additional CPUs.
1234  * We do this because additional CPUs waste a lot of memory.
1235  * -AK
1236  */
1237 __init void prefill_possible_map(void)
1238 {
1239 	int i, possible;
1240 
1241 	/* no processor from mptable or madt */
1242 	if (!num_processors)
1243 		num_processors = 1;
1244 
1245 	i = setup_max_cpus ?: 1;
1246 	if (setup_possible_cpus == -1) {
1247 		possible = num_processors;
1248 #ifdef CONFIG_HOTPLUG_CPU
1249 		if (setup_max_cpus)
1250 			possible += disabled_cpus;
1251 #else
1252 		if (possible > i)
1253 			possible = i;
1254 #endif
1255 	} else
1256 		possible = setup_possible_cpus;
1257 
1258 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1259 
1260 	/* nr_cpu_ids could be reduced via nr_cpus= */
1261 	if (possible > nr_cpu_ids) {
1262 		printk(KERN_WARNING
1263 			"%d Processors exceeds NR_CPUS limit of %d\n",
1264 			possible, nr_cpu_ids);
1265 		possible = nr_cpu_ids;
1266 	}
1267 
1268 #ifdef CONFIG_HOTPLUG_CPU
1269 	if (!setup_max_cpus)
1270 #endif
1271 	if (possible > i) {
1272 		printk(KERN_WARNING
1273 			"%d Processors exceeds max_cpus limit of %u\n",
1274 			possible, setup_max_cpus);
1275 		possible = i;
1276 	}
1277 
1278 	printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1279 		possible, max_t(int, possible - num_processors, 0));
1280 
1281 	for (i = 0; i < possible; i++)
1282 		set_cpu_possible(i, true);
1283 	for (; i < NR_CPUS; i++)
1284 		set_cpu_possible(i, false);
1285 
1286 	nr_cpu_ids = possible;
1287 }
1288 
1289 #ifdef CONFIG_HOTPLUG_CPU
1290 
1291 static void remove_siblinginfo(int cpu)
1292 {
1293 	int sibling;
1294 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1295 
1296 	for_each_cpu(sibling, cpu_core_mask(cpu)) {
1297 		cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1298 		/*/
1299 		 * last thread sibling in this cpu core going down
1300 		 */
1301 		if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1302 			cpu_data(sibling).booted_cores--;
1303 	}
1304 
1305 	for_each_cpu(sibling, cpu_sibling_mask(cpu))
1306 		cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1307 	cpumask_clear(cpu_sibling_mask(cpu));
1308 	cpumask_clear(cpu_core_mask(cpu));
1309 	c->phys_proc_id = 0;
1310 	c->cpu_core_id = 0;
1311 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1312 }
1313 
1314 static void __ref remove_cpu_from_maps(int cpu)
1315 {
1316 	set_cpu_online(cpu, false);
1317 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1318 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1319 	/* was set by cpu_init() */
1320 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1321 	numa_remove_cpu(cpu);
1322 }
1323 
1324 void cpu_disable_common(void)
1325 {
1326 	int cpu = smp_processor_id();
1327 
1328 	remove_siblinginfo(cpu);
1329 
1330 	/* It's now safe to remove this processor from the online map */
1331 	lock_vector_lock();
1332 	remove_cpu_from_maps(cpu);
1333 	unlock_vector_lock();
1334 	fixup_irqs();
1335 }
1336 
1337 int native_cpu_disable(void)
1338 {
1339 	int cpu = smp_processor_id();
1340 
1341 	/*
1342 	 * Perhaps use cpufreq to drop frequency, but that could go
1343 	 * into generic code.
1344 	 *
1345 	 * We won't take down the boot processor on i386 due to some
1346 	 * interrupts only being able to be serviced by the BSP.
1347 	 * Especially so if we're not using an IOAPIC	-zwane
1348 	 */
1349 	if (cpu == 0)
1350 		return -EBUSY;
1351 
1352 	if (nmi_watchdog == NMI_LOCAL_APIC)
1353 		stop_apic_nmi_watchdog(NULL);
1354 	clear_local_APIC();
1355 
1356 	cpu_disable_common();
1357 	return 0;
1358 }
1359 
1360 void native_cpu_die(unsigned int cpu)
1361 {
1362 	/* We don't do anything here: idle task is faking death itself. */
1363 	unsigned int i;
1364 
1365 	for (i = 0; i < 10; i++) {
1366 		/* They ack this in play_dead by setting CPU_DEAD */
1367 		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1368 			if (system_state == SYSTEM_RUNNING)
1369 				pr_info("CPU %u is now offline\n", cpu);
1370 
1371 			if (1 == num_online_cpus())
1372 				alternatives_smp_switch(0);
1373 			return;
1374 		}
1375 		msleep(100);
1376 	}
1377 	pr_err("CPU %u didn't die...\n", cpu);
1378 }
1379 
1380 void play_dead_common(void)
1381 {
1382 	idle_task_exit();
1383 	reset_lazy_tlbstate();
1384 	irq_ctx_exit(raw_smp_processor_id());
1385 	c1e_remove_cpu(raw_smp_processor_id());
1386 
1387 	mb();
1388 	/* Ack it */
1389 	__get_cpu_var(cpu_state) = CPU_DEAD;
1390 
1391 	/*
1392 	 * With physical CPU hotplug, we should halt the cpu
1393 	 */
1394 	local_irq_disable();
1395 }
1396 
1397 /*
1398  * We need to flush the caches before going to sleep, lest we have
1399  * dirty data in our caches when we come back up.
1400  */
1401 static inline void mwait_play_dead(void)
1402 {
1403 	unsigned int eax, ebx, ecx, edx;
1404 	unsigned int highest_cstate = 0;
1405 	unsigned int highest_subcstate = 0;
1406 	int i;
1407 	void *mwait_ptr;
1408 
1409 	if (!cpu_has(&current_cpu_data, X86_FEATURE_MWAIT))
1410 		return;
1411 	if (!cpu_has(&current_cpu_data, X86_FEATURE_CLFLSH))
1412 		return;
1413 	if (current_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
1414 		return;
1415 
1416 	eax = CPUID_MWAIT_LEAF;
1417 	ecx = 0;
1418 	native_cpuid(&eax, &ebx, &ecx, &edx);
1419 
1420 	/*
1421 	 * eax will be 0 if EDX enumeration is not valid.
1422 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1423 	 */
1424 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1425 		eax = 0;
1426 	} else {
1427 		edx >>= MWAIT_SUBSTATE_SIZE;
1428 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1429 			if (edx & MWAIT_SUBSTATE_MASK) {
1430 				highest_cstate = i;
1431 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1432 			}
1433 		}
1434 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1435 			(highest_subcstate - 1);
1436 	}
1437 
1438 	/*
1439 	 * This should be a memory location in a cache line which is
1440 	 * unlikely to be touched by other processors.  The actual
1441 	 * content is immaterial as it is not actually modified in any way.
1442 	 */
1443 	mwait_ptr = &current_thread_info()->flags;
1444 
1445 	wbinvd();
1446 
1447 	while (1) {
1448 		/*
1449 		 * The CLFLUSH is a workaround for erratum AAI65 for
1450 		 * the Xeon 7400 series.  It's not clear it is actually
1451 		 * needed, but it should be harmless in either case.
1452 		 * The WBINVD is insufficient due to the spurious-wakeup
1453 		 * case where we return around the loop.
1454 		 */
1455 		clflush(mwait_ptr);
1456 		__monitor(mwait_ptr, 0, 0);
1457 		mb();
1458 		__mwait(eax, 0);
1459 	}
1460 }
1461 
1462 static inline void hlt_play_dead(void)
1463 {
1464 	if (current_cpu_data.x86 >= 4)
1465 		wbinvd();
1466 
1467 	while (1) {
1468 		native_halt();
1469 	}
1470 }
1471 
1472 void native_play_dead(void)
1473 {
1474 	play_dead_common();
1475 	tboot_shutdown(TB_SHUTDOWN_WFS);
1476 
1477 	mwait_play_dead();	/* Only returns on failure */
1478 	hlt_play_dead();
1479 }
1480 
1481 #else /* ... !CONFIG_HOTPLUG_CPU */
1482 int native_cpu_disable(void)
1483 {
1484 	return -ENOSYS;
1485 }
1486 
1487 void native_cpu_die(unsigned int cpu)
1488 {
1489 	/* We said "no" in __cpu_disable */
1490 	BUG();
1491 }
1492 
1493 void native_play_dead(void)
1494 {
1495 	BUG();
1496 }
1497 
1498 #endif
1499