xref: /openbmc/linux/arch/x86/kernel/smpboot.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
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/io_apic.h>
68 #include <asm/setup.h>
69 #include <asm/uv/uv.h>
70 #include <linux/mc146818rtc.h>
71 
72 #include <asm/smpboot_hooks.h>
73 #include <asm/i8259.h>
74 
75 #ifdef CONFIG_X86_32
76 u8 apicid_2_node[MAX_APICID];
77 #endif
78 
79 /* State of each CPU */
80 DEFINE_PER_CPU(int, cpu_state) = { 0 };
81 
82 /* Store all idle threads, this can be reused instead of creating
83 * a new thread. Also avoids complicated thread destroy functionality
84 * for idle threads.
85 */
86 #ifdef CONFIG_HOTPLUG_CPU
87 /*
88  * Needed only for CONFIG_HOTPLUG_CPU because __cpuinitdata is
89  * removed after init for !CONFIG_HOTPLUG_CPU.
90  */
91 static DEFINE_PER_CPU(struct task_struct *, idle_thread_array);
92 #define get_idle_for_cpu(x)      (per_cpu(idle_thread_array, x))
93 #define set_idle_for_cpu(x, p)   (per_cpu(idle_thread_array, x) = (p))
94 
95 /*
96  * We need this for trampoline_base protection from concurrent accesses when
97  * off- and onlining cores wildly.
98  */
99 static DEFINE_MUTEX(x86_cpu_hotplug_driver_mutex);
100 
101 void cpu_hotplug_driver_lock(void)
102 {
103         mutex_lock(&x86_cpu_hotplug_driver_mutex);
104 }
105 
106 void cpu_hotplug_driver_unlock(void)
107 {
108         mutex_unlock(&x86_cpu_hotplug_driver_mutex);
109 }
110 
111 ssize_t arch_cpu_probe(const char *buf, size_t count) { return -1; }
112 ssize_t arch_cpu_release(const char *buf, size_t count) { return -1; }
113 #else
114 static struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
115 #define get_idle_for_cpu(x)      (idle_thread_array[(x)])
116 #define set_idle_for_cpu(x, p)   (idle_thread_array[(x)] = (p))
117 #endif
118 
119 /* Number of siblings per CPU package */
120 int smp_num_siblings = 1;
121 EXPORT_SYMBOL(smp_num_siblings);
122 
123 /* Last level cache ID of each logical CPU */
124 DEFINE_PER_CPU(u16, cpu_llc_id) = BAD_APICID;
125 
126 /* representing HT siblings of each logical CPU */
127 DEFINE_PER_CPU(cpumask_var_t, cpu_sibling_map);
128 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
129 
130 /* representing HT and core siblings of each logical CPU */
131 DEFINE_PER_CPU(cpumask_var_t, cpu_core_map);
132 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
133 
134 /* Per CPU bogomips and other parameters */
135 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
136 EXPORT_PER_CPU_SYMBOL(cpu_info);
137 
138 atomic_t init_deasserted;
139 
140 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_32)
141 /* which node each logical CPU is on */
142 int cpu_to_node_map[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 };
143 EXPORT_SYMBOL(cpu_to_node_map);
144 
145 /* set up a mapping between cpu and node. */
146 static void map_cpu_to_node(int cpu, int node)
147 {
148 	printk(KERN_INFO "Mapping cpu %d to node %d\n", cpu, node);
149 	cpumask_set_cpu(cpu, node_to_cpumask_map[node]);
150 	cpu_to_node_map[cpu] = node;
151 }
152 
153 /* undo a mapping between cpu and node. */
154 static void unmap_cpu_to_node(int cpu)
155 {
156 	int node;
157 
158 	printk(KERN_INFO "Unmapping cpu %d from all nodes\n", cpu);
159 	for (node = 0; node < MAX_NUMNODES; node++)
160 		cpumask_clear_cpu(cpu, node_to_cpumask_map[node]);
161 	cpu_to_node_map[cpu] = 0;
162 }
163 #else /* !(CONFIG_NUMA && CONFIG_X86_32) */
164 #define map_cpu_to_node(cpu, node)	({})
165 #define unmap_cpu_to_node(cpu)	({})
166 #endif
167 
168 #ifdef CONFIG_X86_32
169 static int boot_cpu_logical_apicid;
170 
171 u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly =
172 					{ [0 ... NR_CPUS-1] = BAD_APICID };
173 
174 static void map_cpu_to_logical_apicid(void)
175 {
176 	int cpu = smp_processor_id();
177 	int apicid = logical_smp_processor_id();
178 	int node = apic->apicid_to_node(apicid);
179 
180 	if (!node_online(node))
181 		node = first_online_node;
182 
183 	cpu_2_logical_apicid[cpu] = apicid;
184 	map_cpu_to_node(cpu, node);
185 }
186 
187 void numa_remove_cpu(int cpu)
188 {
189 	cpu_2_logical_apicid[cpu] = BAD_APICID;
190 	unmap_cpu_to_node(cpu);
191 }
192 #else
193 #define map_cpu_to_logical_apicid()  do {} while (0)
194 #endif
195 
196 /*
197  * Report back to the Boot Processor.
198  * Running on AP.
199  */
200 static void __cpuinit smp_callin(void)
201 {
202 	int cpuid, phys_id;
203 	unsigned long timeout;
204 
205 	/*
206 	 * If waken up by an INIT in an 82489DX configuration
207 	 * we may get here before an INIT-deassert IPI reaches
208 	 * our local APIC.  We have to wait for the IPI or we'll
209 	 * lock up on an APIC access.
210 	 */
211 	if (apic->wait_for_init_deassert)
212 		apic->wait_for_init_deassert(&init_deasserted);
213 
214 	/*
215 	 * (This works even if the APIC is not enabled.)
216 	 */
217 	phys_id = read_apic_id();
218 	cpuid = smp_processor_id();
219 	if (cpumask_test_cpu(cpuid, cpu_callin_mask)) {
220 		panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
221 					phys_id, cpuid);
222 	}
223 	pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
224 
225 	/*
226 	 * STARTUP IPIs are fragile beasts as they might sometimes
227 	 * trigger some glue motherboard logic. Complete APIC bus
228 	 * silence for 1 second, this overestimates the time the
229 	 * boot CPU is spending to send the up to 2 STARTUP IPIs
230 	 * by a factor of two. This should be enough.
231 	 */
232 
233 	/*
234 	 * Waiting 2s total for startup (udelay is not yet working)
235 	 */
236 	timeout = jiffies + 2*HZ;
237 	while (time_before(jiffies, timeout)) {
238 		/*
239 		 * Has the boot CPU finished it's STARTUP sequence?
240 		 */
241 		if (cpumask_test_cpu(cpuid, cpu_callout_mask))
242 			break;
243 		cpu_relax();
244 	}
245 
246 	if (!time_before(jiffies, timeout)) {
247 		panic("%s: CPU%d started up but did not get a callout!\n",
248 		      __func__, cpuid);
249 	}
250 
251 	/*
252 	 * the boot CPU has finished the init stage and is spinning
253 	 * on callin_map until we finish. We are free to set up this
254 	 * CPU, first the APIC. (this is probably redundant on most
255 	 * boards)
256 	 */
257 
258 	pr_debug("CALLIN, before setup_local_APIC().\n");
259 	if (apic->smp_callin_clear_local_apic)
260 		apic->smp_callin_clear_local_apic();
261 	setup_local_APIC();
262 	end_local_APIC_setup();
263 	map_cpu_to_logical_apicid();
264 
265 	/*
266 	 * Need to setup vector mappings before we enable interrupts.
267 	 */
268 	setup_vector_irq(smp_processor_id());
269 	/*
270 	 * Get our bogomips.
271 	 *
272 	 * Need to enable IRQs because it can take longer and then
273 	 * the NMI watchdog might kill us.
274 	 */
275 	local_irq_enable();
276 	calibrate_delay();
277 	local_irq_disable();
278 	pr_debug("Stack at about %p\n", &cpuid);
279 
280 	/*
281 	 * Save our processor parameters
282 	 */
283 	smp_store_cpu_info(cpuid);
284 
285 	/*
286 	 * This must be done before setting cpu_online_mask
287 	 * or calling notify_cpu_starting.
288 	 */
289 	set_cpu_sibling_map(raw_smp_processor_id());
290 	wmb();
291 
292 	notify_cpu_starting(cpuid);
293 
294 	/*
295 	 * Allow the master to continue.
296 	 */
297 	cpumask_set_cpu(cpuid, cpu_callin_mask);
298 }
299 
300 /*
301  * Activate a secondary processor.
302  */
303 notrace static void __cpuinit start_secondary(void *unused)
304 {
305 	/*
306 	 * Don't put *anything* before cpu_init(), SMP booting is too
307 	 * fragile that we want to limit the things done here to the
308 	 * most necessary things.
309 	 */
310 	cpu_init();
311 	preempt_disable();
312 	smp_callin();
313 
314 #ifdef CONFIG_X86_32
315 	/* switch away from the initial page table */
316 	load_cr3(swapper_pg_dir);
317 	__flush_tlb_all();
318 #endif
319 
320 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
321 	barrier();
322 	/*
323 	 * Check TSC synchronization with the BP:
324 	 */
325 	check_tsc_sync_target();
326 
327 	/*
328 	 * We need to hold call_lock, so there is no inconsistency
329 	 * between the time smp_call_function() determines number of
330 	 * IPI recipients, and the time when the determination is made
331 	 * for which cpus receive the IPI. Holding this
332 	 * lock helps us to not include this cpu in a currently in progress
333 	 * smp_call_function().
334 	 *
335 	 * We need to hold vector_lock so there the set of online cpus
336 	 * does not change while we are assigning vectors to cpus.  Holding
337 	 * this lock ensures we don't half assign or remove an irq from a cpu.
338 	 */
339 	ipi_call_lock();
340 	lock_vector_lock();
341 	set_cpu_online(smp_processor_id(), true);
342 	unlock_vector_lock();
343 	ipi_call_unlock();
344 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
345 	x86_platform.nmi_init();
346 
347 	/* enable local interrupts */
348 	local_irq_enable();
349 
350 	/* to prevent fake stack check failure in clock setup */
351 	boot_init_stack_canary();
352 
353 	x86_cpuinit.setup_percpu_clockev();
354 
355 	wmb();
356 	cpu_idle();
357 }
358 
359 #ifdef CONFIG_CPUMASK_OFFSTACK
360 /* In this case, llc_shared_map is a pointer to a cpumask. */
361 static inline void copy_cpuinfo_x86(struct cpuinfo_x86 *dst,
362 				    const struct cpuinfo_x86 *src)
363 {
364 	struct cpumask *llc = dst->llc_shared_map;
365 	*dst = *src;
366 	dst->llc_shared_map = llc;
367 }
368 #else
369 static inline void copy_cpuinfo_x86(struct cpuinfo_x86 *dst,
370 				    const struct cpuinfo_x86 *src)
371 {
372 	*dst = *src;
373 }
374 #endif /* CONFIG_CPUMASK_OFFSTACK */
375 
376 /*
377  * The bootstrap kernel entry code has set these up. Save them for
378  * a given CPU
379  */
380 
381 void __cpuinit smp_store_cpu_info(int id)
382 {
383 	struct cpuinfo_x86 *c = &cpu_data(id);
384 
385 	copy_cpuinfo_x86(c, &boot_cpu_data);
386 	c->cpu_index = id;
387 	if (id != 0)
388 		identify_secondary_cpu(c);
389 }
390 
391 static void __cpuinit link_thread_siblings(int cpu1, int cpu2)
392 {
393 	struct cpuinfo_x86 *c1 = &cpu_data(cpu1);
394 	struct cpuinfo_x86 *c2 = &cpu_data(cpu2);
395 
396 	cpumask_set_cpu(cpu1, cpu_sibling_mask(cpu2));
397 	cpumask_set_cpu(cpu2, cpu_sibling_mask(cpu1));
398 	cpumask_set_cpu(cpu1, cpu_core_mask(cpu2));
399 	cpumask_set_cpu(cpu2, cpu_core_mask(cpu1));
400 	cpumask_set_cpu(cpu1, c2->llc_shared_map);
401 	cpumask_set_cpu(cpu2, c1->llc_shared_map);
402 }
403 
404 
405 void __cpuinit set_cpu_sibling_map(int cpu)
406 {
407 	int i;
408 	struct cpuinfo_x86 *c = &cpu_data(cpu);
409 
410 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
411 
412 	if (smp_num_siblings > 1) {
413 		for_each_cpu(i, cpu_sibling_setup_mask) {
414 			struct cpuinfo_x86 *o = &cpu_data(i);
415 
416 			if (cpu_has(c, X86_FEATURE_TOPOEXT)) {
417 				if (c->phys_proc_id == o->phys_proc_id &&
418 				    c->compute_unit_id == o->compute_unit_id)
419 					link_thread_siblings(cpu, i);
420 			} else if (c->phys_proc_id == o->phys_proc_id &&
421 				   c->cpu_core_id == o->cpu_core_id) {
422 				link_thread_siblings(cpu, i);
423 			}
424 		}
425 	} else {
426 		cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
427 	}
428 
429 	cpumask_set_cpu(cpu, c->llc_shared_map);
430 
431 	if (__this_cpu_read(cpu_info.x86_max_cores) == 1) {
432 		cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
433 		c->booted_cores = 1;
434 		return;
435 	}
436 
437 	for_each_cpu(i, cpu_sibling_setup_mask) {
438 		if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
439 		    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
440 			cpumask_set_cpu(i, c->llc_shared_map);
441 			cpumask_set_cpu(cpu, cpu_data(i).llc_shared_map);
442 		}
443 		if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
444 			cpumask_set_cpu(i, cpu_core_mask(cpu));
445 			cpumask_set_cpu(cpu, cpu_core_mask(i));
446 			/*
447 			 *  Does this new cpu bringup a new core?
448 			 */
449 			if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
450 				/*
451 				 * for each core in package, increment
452 				 * the booted_cores for this new cpu
453 				 */
454 				if (cpumask_first(cpu_sibling_mask(i)) == i)
455 					c->booted_cores++;
456 				/*
457 				 * increment the core count for all
458 				 * the other cpus in this package
459 				 */
460 				if (i != cpu)
461 					cpu_data(i).booted_cores++;
462 			} else if (i != cpu && !c->booted_cores)
463 				c->booted_cores = cpu_data(i).booted_cores;
464 		}
465 	}
466 }
467 
468 /* maps the cpu to the sched domain representing multi-core */
469 const struct cpumask *cpu_coregroup_mask(int cpu)
470 {
471 	struct cpuinfo_x86 *c = &cpu_data(cpu);
472 	/*
473 	 * For perf, we return last level cache shared map.
474 	 * And for power savings, we return cpu_core_map
475 	 */
476 	if ((sched_mc_power_savings || sched_smt_power_savings) &&
477 	    !(cpu_has(c, X86_FEATURE_AMD_DCM)))
478 		return cpu_core_mask(cpu);
479 	else
480 		return c->llc_shared_map;
481 }
482 
483 static void impress_friends(void)
484 {
485 	int cpu;
486 	unsigned long bogosum = 0;
487 	/*
488 	 * Allow the user to impress friends.
489 	 */
490 	pr_debug("Before bogomips.\n");
491 	for_each_possible_cpu(cpu)
492 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
493 			bogosum += cpu_data(cpu).loops_per_jiffy;
494 	printk(KERN_INFO
495 		"Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
496 		num_online_cpus(),
497 		bogosum/(500000/HZ),
498 		(bogosum/(5000/HZ))%100);
499 
500 	pr_debug("Before bogocount - setting activated=1.\n");
501 }
502 
503 void __inquire_remote_apic(int apicid)
504 {
505 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
506 	char *names[] = { "ID", "VERSION", "SPIV" };
507 	int timeout;
508 	u32 status;
509 
510 	printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
511 
512 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
513 		printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
514 
515 		/*
516 		 * Wait for idle.
517 		 */
518 		status = safe_apic_wait_icr_idle();
519 		if (status)
520 			printk(KERN_CONT
521 			       "a previous APIC delivery may have failed\n");
522 
523 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
524 
525 		timeout = 0;
526 		do {
527 			udelay(100);
528 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
529 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
530 
531 		switch (status) {
532 		case APIC_ICR_RR_VALID:
533 			status = apic_read(APIC_RRR);
534 			printk(KERN_CONT "%08x\n", status);
535 			break;
536 		default:
537 			printk(KERN_CONT "failed\n");
538 		}
539 	}
540 }
541 
542 /*
543  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
544  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
545  * won't ... remember to clear down the APIC, etc later.
546  */
547 int __cpuinit
548 wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
549 {
550 	unsigned long send_status, accept_status = 0;
551 	int maxlvt;
552 
553 	/* Target chip */
554 	/* Boot on the stack */
555 	/* Kick the second */
556 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, logical_apicid);
557 
558 	pr_debug("Waiting for send to finish...\n");
559 	send_status = safe_apic_wait_icr_idle();
560 
561 	/*
562 	 * Give the other CPU some time to accept the IPI.
563 	 */
564 	udelay(200);
565 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
566 		maxlvt = lapic_get_maxlvt();
567 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
568 			apic_write(APIC_ESR, 0);
569 		accept_status = (apic_read(APIC_ESR) & 0xEF);
570 	}
571 	pr_debug("NMI sent.\n");
572 
573 	if (send_status)
574 		printk(KERN_ERR "APIC never delivered???\n");
575 	if (accept_status)
576 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
577 
578 	return (send_status | accept_status);
579 }
580 
581 static int __cpuinit
582 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
583 {
584 	unsigned long send_status, accept_status = 0;
585 	int maxlvt, num_starts, j;
586 
587 	maxlvt = lapic_get_maxlvt();
588 
589 	/*
590 	 * Be paranoid about clearing APIC errors.
591 	 */
592 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
593 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
594 			apic_write(APIC_ESR, 0);
595 		apic_read(APIC_ESR);
596 	}
597 
598 	pr_debug("Asserting INIT.\n");
599 
600 	/*
601 	 * Turn INIT on target chip
602 	 */
603 	/*
604 	 * Send IPI
605 	 */
606 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
607 		       phys_apicid);
608 
609 	pr_debug("Waiting for send to finish...\n");
610 	send_status = safe_apic_wait_icr_idle();
611 
612 	mdelay(10);
613 
614 	pr_debug("Deasserting INIT.\n");
615 
616 	/* Target chip */
617 	/* Send IPI */
618 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
619 
620 	pr_debug("Waiting for send to finish...\n");
621 	send_status = safe_apic_wait_icr_idle();
622 
623 	mb();
624 	atomic_set(&init_deasserted, 1);
625 
626 	/*
627 	 * Should we send STARTUP IPIs ?
628 	 *
629 	 * Determine this based on the APIC version.
630 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
631 	 */
632 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
633 		num_starts = 2;
634 	else
635 		num_starts = 0;
636 
637 	/*
638 	 * Paravirt / VMI wants a startup IPI hook here to set up the
639 	 * target processor state.
640 	 */
641 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
642 			 stack_start);
643 
644 	/*
645 	 * Run STARTUP IPI loop.
646 	 */
647 	pr_debug("#startup loops: %d.\n", num_starts);
648 
649 	for (j = 1; j <= num_starts; j++) {
650 		pr_debug("Sending STARTUP #%d.\n", j);
651 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
652 			apic_write(APIC_ESR, 0);
653 		apic_read(APIC_ESR);
654 		pr_debug("After apic_write.\n");
655 
656 		/*
657 		 * STARTUP IPI
658 		 */
659 
660 		/* Target chip */
661 		/* Boot on the stack */
662 		/* Kick the second */
663 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
664 			       phys_apicid);
665 
666 		/*
667 		 * Give the other CPU some time to accept the IPI.
668 		 */
669 		udelay(300);
670 
671 		pr_debug("Startup point 1.\n");
672 
673 		pr_debug("Waiting for send to finish...\n");
674 		send_status = safe_apic_wait_icr_idle();
675 
676 		/*
677 		 * Give the other CPU some time to accept the IPI.
678 		 */
679 		udelay(200);
680 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
681 			apic_write(APIC_ESR, 0);
682 		accept_status = (apic_read(APIC_ESR) & 0xEF);
683 		if (send_status || accept_status)
684 			break;
685 	}
686 	pr_debug("After Startup.\n");
687 
688 	if (send_status)
689 		printk(KERN_ERR "APIC never delivered???\n");
690 	if (accept_status)
691 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
692 
693 	return (send_status | accept_status);
694 }
695 
696 struct create_idle {
697 	struct work_struct work;
698 	struct task_struct *idle;
699 	struct completion done;
700 	int cpu;
701 };
702 
703 static void __cpuinit do_fork_idle(struct work_struct *work)
704 {
705 	struct create_idle *c_idle =
706 		container_of(work, struct create_idle, work);
707 
708 	c_idle->idle = fork_idle(c_idle->cpu);
709 	complete(&c_idle->done);
710 }
711 
712 /* reduce the number of lines printed when booting a large cpu count system */
713 static void __cpuinit announce_cpu(int cpu, int apicid)
714 {
715 	static int current_node = -1;
716 	int node = early_cpu_to_node(cpu);
717 
718 	if (system_state == SYSTEM_BOOTING) {
719 		if (node != current_node) {
720 			if (current_node > (-1))
721 				pr_cont(" Ok.\n");
722 			current_node = node;
723 			pr_info("Booting Node %3d, Processors ", node);
724 		}
725 		pr_cont(" #%d%s", cpu, cpu == (nr_cpu_ids - 1) ? " Ok.\n" : "");
726 		return;
727 	} else
728 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
729 			node, cpu, apicid);
730 }
731 
732 /*
733  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
734  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
735  * Returns zero if CPU booted OK, else error code from
736  * ->wakeup_secondary_cpu.
737  */
738 static int __cpuinit do_boot_cpu(int apicid, int cpu)
739 {
740 	unsigned long boot_error = 0;
741 	unsigned long start_ip;
742 	int timeout;
743 	struct create_idle c_idle = {
744 		.cpu	= cpu,
745 		.done	= COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
746 	};
747 
748 	INIT_WORK_ONSTACK(&c_idle.work, do_fork_idle);
749 
750 	alternatives_smp_switch(1);
751 
752 	c_idle.idle = get_idle_for_cpu(cpu);
753 
754 	/*
755 	 * We can't use kernel_thread since we must avoid to
756 	 * reschedule the child.
757 	 */
758 	if (c_idle.idle) {
759 		c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
760 			(THREAD_SIZE +  task_stack_page(c_idle.idle))) - 1);
761 		init_idle(c_idle.idle, cpu);
762 		goto do_rest;
763 	}
764 
765 	schedule_work(&c_idle.work);
766 	wait_for_completion(&c_idle.done);
767 
768 	if (IS_ERR(c_idle.idle)) {
769 		printk("failed fork for CPU %d\n", cpu);
770 		destroy_work_on_stack(&c_idle.work);
771 		return PTR_ERR(c_idle.idle);
772 	}
773 
774 	set_idle_for_cpu(cpu, c_idle.idle);
775 do_rest:
776 	per_cpu(current_task, cpu) = c_idle.idle;
777 #ifdef CONFIG_X86_32
778 	/* Stack for startup_32 can be just as for start_secondary onwards */
779 	irq_ctx_init(cpu);
780 #else
781 	clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
782 	initial_gs = per_cpu_offset(cpu);
783 	per_cpu(kernel_stack, cpu) =
784 		(unsigned long)task_stack_page(c_idle.idle) -
785 		KERNEL_STACK_OFFSET + THREAD_SIZE;
786 #endif
787 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
788 	initial_code = (unsigned long)start_secondary;
789 	stack_start  = c_idle.idle->thread.sp;
790 
791 	/* start_ip had better be page-aligned! */
792 	start_ip = setup_trampoline();
793 
794 	/* So we see what's up */
795 	announce_cpu(cpu, apicid);
796 
797 	/*
798 	 * This grunge runs the startup process for
799 	 * the targeted processor.
800 	 */
801 
802 	atomic_set(&init_deasserted, 0);
803 
804 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
805 
806 		pr_debug("Setting warm reset code and vector.\n");
807 
808 		smpboot_setup_warm_reset_vector(start_ip);
809 		/*
810 		 * Be paranoid about clearing APIC errors.
811 		*/
812 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
813 			apic_write(APIC_ESR, 0);
814 			apic_read(APIC_ESR);
815 		}
816 	}
817 
818 	/*
819 	 * Kick the secondary CPU. Use the method in the APIC driver
820 	 * if it's defined - or use an INIT boot APIC message otherwise:
821 	 */
822 	if (apic->wakeup_secondary_cpu)
823 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
824 	else
825 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
826 
827 	if (!boot_error) {
828 		/*
829 		 * allow APs to start initializing.
830 		 */
831 		pr_debug("Before Callout %d.\n", cpu);
832 		cpumask_set_cpu(cpu, cpu_callout_mask);
833 		pr_debug("After Callout %d.\n", cpu);
834 
835 		/*
836 		 * Wait 5s total for a response
837 		 */
838 		for (timeout = 0; timeout < 50000; timeout++) {
839 			if (cpumask_test_cpu(cpu, cpu_callin_mask))
840 				break;	/* It has booted */
841 			udelay(100);
842 			/*
843 			 * Allow other tasks to run while we wait for the
844 			 * AP to come online. This also gives a chance
845 			 * for the MTRR work(triggered by the AP coming online)
846 			 * to be completed in the stop machine context.
847 			 */
848 			schedule();
849 		}
850 
851 		if (cpumask_test_cpu(cpu, cpu_callin_mask))
852 			pr_debug("CPU%d: has booted.\n", cpu);
853 		else {
854 			boot_error = 1;
855 			if (*((volatile unsigned char *)trampoline_base)
856 					== 0xA5)
857 				/* trampoline started but...? */
858 				pr_err("CPU%d: Stuck ??\n", cpu);
859 			else
860 				/* trampoline code not run */
861 				pr_err("CPU%d: Not responding.\n", cpu);
862 			if (apic->inquire_remote_apic)
863 				apic->inquire_remote_apic(apicid);
864 		}
865 	}
866 
867 	if (boot_error) {
868 		/* Try to put things back the way they were before ... */
869 		numa_remove_cpu(cpu); /* was set by numa_add_cpu */
870 
871 		/* was set by do_boot_cpu() */
872 		cpumask_clear_cpu(cpu, cpu_callout_mask);
873 
874 		/* was set by cpu_init() */
875 		cpumask_clear_cpu(cpu, cpu_initialized_mask);
876 
877 		set_cpu_present(cpu, false);
878 		per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
879 	}
880 
881 	/* mark "stuck" area as not stuck */
882 	*((volatile unsigned long *)trampoline_base) = 0;
883 
884 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
885 		/*
886 		 * Cleanup possible dangling ends...
887 		 */
888 		smpboot_restore_warm_reset_vector();
889 	}
890 
891 	destroy_work_on_stack(&c_idle.work);
892 	return boot_error;
893 }
894 
895 int __cpuinit native_cpu_up(unsigned int cpu)
896 {
897 	int apicid = apic->cpu_present_to_apicid(cpu);
898 	unsigned long flags;
899 	int err;
900 
901 	WARN_ON(irqs_disabled());
902 
903 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
904 
905 	if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
906 	    !physid_isset(apicid, phys_cpu_present_map)) {
907 		printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
908 		return -EINVAL;
909 	}
910 
911 	/*
912 	 * Already booted CPU?
913 	 */
914 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
915 		pr_debug("do_boot_cpu %d Already started\n", cpu);
916 		return -ENOSYS;
917 	}
918 
919 	/*
920 	 * Save current MTRR state in case it was changed since early boot
921 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
922 	 */
923 	mtrr_save_state();
924 
925 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
926 
927 	err = do_boot_cpu(apicid, cpu);
928 	if (err) {
929 		pr_debug("do_boot_cpu failed %d\n", err);
930 		return -EIO;
931 	}
932 
933 	/*
934 	 * Check TSC synchronization with the AP (keep irqs disabled
935 	 * while doing so):
936 	 */
937 	local_irq_save(flags);
938 	check_tsc_sync_source(cpu);
939 	local_irq_restore(flags);
940 
941 	while (!cpu_online(cpu)) {
942 		cpu_relax();
943 		touch_nmi_watchdog();
944 	}
945 
946 	return 0;
947 }
948 
949 /**
950  * arch_disable_smp_support() - disables SMP support for x86 at runtime
951  */
952 void arch_disable_smp_support(void)
953 {
954 	disable_ioapic_support();
955 }
956 
957 /*
958  * Fall back to non SMP mode after errors.
959  *
960  * RED-PEN audit/test this more. I bet there is more state messed up here.
961  */
962 static __init void disable_smp(void)
963 {
964 	init_cpu_present(cpumask_of(0));
965 	init_cpu_possible(cpumask_of(0));
966 	smpboot_clear_io_apic_irqs();
967 
968 	if (smp_found_config)
969 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
970 	else
971 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
972 	map_cpu_to_logical_apicid();
973 	cpumask_set_cpu(0, cpu_sibling_mask(0));
974 	cpumask_set_cpu(0, cpu_core_mask(0));
975 }
976 
977 /*
978  * Various sanity checks.
979  */
980 static int __init smp_sanity_check(unsigned max_cpus)
981 {
982 	preempt_disable();
983 
984 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
985 	if (def_to_bigsmp && nr_cpu_ids > 8) {
986 		unsigned int cpu;
987 		unsigned nr;
988 
989 		printk(KERN_WARNING
990 		       "More than 8 CPUs detected - skipping them.\n"
991 		       "Use CONFIG_X86_BIGSMP.\n");
992 
993 		nr = 0;
994 		for_each_present_cpu(cpu) {
995 			if (nr >= 8)
996 				set_cpu_present(cpu, false);
997 			nr++;
998 		}
999 
1000 		nr = 0;
1001 		for_each_possible_cpu(cpu) {
1002 			if (nr >= 8)
1003 				set_cpu_possible(cpu, false);
1004 			nr++;
1005 		}
1006 
1007 		nr_cpu_ids = 8;
1008 	}
1009 #endif
1010 
1011 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1012 		printk(KERN_WARNING
1013 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
1014 			hard_smp_processor_id());
1015 
1016 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1017 	}
1018 
1019 	/*
1020 	 * If we couldn't find an SMP configuration at boot time,
1021 	 * get out of here now!
1022 	 */
1023 	if (!smp_found_config && !acpi_lapic) {
1024 		preempt_enable();
1025 		printk(KERN_NOTICE "SMP motherboard not detected.\n");
1026 		disable_smp();
1027 		if (APIC_init_uniprocessor())
1028 			printk(KERN_NOTICE "Local APIC not detected."
1029 					   " Using dummy APIC emulation.\n");
1030 		return -1;
1031 	}
1032 
1033 	/*
1034 	 * Should not be necessary because the MP table should list the boot
1035 	 * CPU too, but we do it for the sake of robustness anyway.
1036 	 */
1037 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1038 		printk(KERN_NOTICE
1039 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
1040 			boot_cpu_physical_apicid);
1041 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1042 	}
1043 	preempt_enable();
1044 
1045 	/*
1046 	 * If we couldn't find a local APIC, then get out of here now!
1047 	 */
1048 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1049 	    !cpu_has_apic) {
1050 		if (!disable_apic) {
1051 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
1052 				boot_cpu_physical_apicid);
1053 			pr_err("... forcing use of dummy APIC emulation."
1054 				"(tell your hw vendor)\n");
1055 		}
1056 		smpboot_clear_io_apic();
1057 		disable_ioapic_support();
1058 		return -1;
1059 	}
1060 
1061 	verify_local_APIC();
1062 
1063 	/*
1064 	 * If SMP should be disabled, then really disable it!
1065 	 */
1066 	if (!max_cpus) {
1067 		printk(KERN_INFO "SMP mode deactivated.\n");
1068 		smpboot_clear_io_apic();
1069 
1070 		connect_bsp_APIC();
1071 		setup_local_APIC();
1072 		bsp_end_local_APIC_setup();
1073 		return -1;
1074 	}
1075 
1076 	return 0;
1077 }
1078 
1079 static void __init smp_cpu_index_default(void)
1080 {
1081 	int i;
1082 	struct cpuinfo_x86 *c;
1083 
1084 	for_each_possible_cpu(i) {
1085 		c = &cpu_data(i);
1086 		/* mark all to hotplug */
1087 		c->cpu_index = nr_cpu_ids;
1088 	}
1089 }
1090 
1091 /*
1092  * Prepare for SMP bootup.  The MP table or ACPI has been read
1093  * earlier.  Just do some sanity checking here and enable APIC mode.
1094  */
1095 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1096 {
1097 	unsigned int i;
1098 
1099 	preempt_disable();
1100 	smp_cpu_index_default();
1101 	memcpy(__this_cpu_ptr(&cpu_info), &boot_cpu_data, sizeof(cpu_info));
1102 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1103 	mb();
1104 	/*
1105 	 * Setup boot CPU information
1106 	 */
1107 	smp_store_cpu_info(0); /* Final full version of the data */
1108 #ifdef CONFIG_X86_32
1109 	boot_cpu_logical_apicid = logical_smp_processor_id();
1110 #endif
1111 	current_thread_info()->cpu = 0;  /* needed? */
1112 	for_each_possible_cpu(i) {
1113 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1114 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1115 		zalloc_cpumask_var(&cpu_data(i).llc_shared_map, GFP_KERNEL);
1116 	}
1117 	set_cpu_sibling_map(0);
1118 
1119 
1120 	if (smp_sanity_check(max_cpus) < 0) {
1121 		printk(KERN_INFO "SMP disabled\n");
1122 		disable_smp();
1123 		goto out;
1124 	}
1125 
1126 	default_setup_apic_routing();
1127 
1128 	preempt_disable();
1129 	if (read_apic_id() != boot_cpu_physical_apicid) {
1130 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1131 		     read_apic_id(), boot_cpu_physical_apicid);
1132 		/* Or can we switch back to PIC here? */
1133 	}
1134 	preempt_enable();
1135 
1136 	connect_bsp_APIC();
1137 
1138 	/*
1139 	 * Switch from PIC to APIC mode.
1140 	 */
1141 	setup_local_APIC();
1142 
1143 	/*
1144 	 * Enable IO APIC before setting up error vector
1145 	 */
1146 	if (!skip_ioapic_setup && nr_ioapics)
1147 		enable_IO_APIC();
1148 
1149 	bsp_end_local_APIC_setup();
1150 
1151 	map_cpu_to_logical_apicid();
1152 
1153 	if (apic->setup_portio_remap)
1154 		apic->setup_portio_remap();
1155 
1156 	smpboot_setup_io_apic();
1157 	/*
1158 	 * Set up local APIC timer on boot CPU.
1159 	 */
1160 
1161 	printk(KERN_INFO "CPU%d: ", 0);
1162 	print_cpu_info(&cpu_data(0));
1163 	x86_init.timers.setup_percpu_clockev();
1164 
1165 	if (is_uv_system())
1166 		uv_system_init();
1167 
1168 	set_mtrr_aps_delayed_init();
1169 out:
1170 	preempt_enable();
1171 }
1172 
1173 void arch_disable_nonboot_cpus_begin(void)
1174 {
1175 	/*
1176 	 * Avoid the smp alternatives switch during the disable_nonboot_cpus().
1177 	 * In the suspend path, we will be back in the SMP mode shortly anyways.
1178 	 */
1179 	skip_smp_alternatives = true;
1180 }
1181 
1182 void arch_disable_nonboot_cpus_end(void)
1183 {
1184 	skip_smp_alternatives = false;
1185 }
1186 
1187 void arch_enable_nonboot_cpus_begin(void)
1188 {
1189 	set_mtrr_aps_delayed_init();
1190 }
1191 
1192 void arch_enable_nonboot_cpus_end(void)
1193 {
1194 	mtrr_aps_init();
1195 }
1196 
1197 /*
1198  * Early setup to make printk work.
1199  */
1200 void __init native_smp_prepare_boot_cpu(void)
1201 {
1202 	int me = smp_processor_id();
1203 	switch_to_new_gdt(me);
1204 	/* already set me in cpu_online_mask in boot_cpu_init() */
1205 	cpumask_set_cpu(me, cpu_callout_mask);
1206 	per_cpu(cpu_state, me) = CPU_ONLINE;
1207 }
1208 
1209 void __init native_smp_cpus_done(unsigned int max_cpus)
1210 {
1211 	pr_debug("Boot done.\n");
1212 
1213 	impress_friends();
1214 #ifdef CONFIG_X86_IO_APIC
1215 	setup_ioapic_dest();
1216 #endif
1217 	mtrr_aps_init();
1218 }
1219 
1220 static int __initdata setup_possible_cpus = -1;
1221 static int __init _setup_possible_cpus(char *str)
1222 {
1223 	get_option(&str, &setup_possible_cpus);
1224 	return 0;
1225 }
1226 early_param("possible_cpus", _setup_possible_cpus);
1227 
1228 
1229 /*
1230  * cpu_possible_mask should be static, it cannot change as cpu's
1231  * are onlined, or offlined. The reason is per-cpu data-structures
1232  * are allocated by some modules at init time, and dont expect to
1233  * do this dynamically on cpu arrival/departure.
1234  * cpu_present_mask on the other hand can change dynamically.
1235  * In case when cpu_hotplug is not compiled, then we resort to current
1236  * behaviour, which is cpu_possible == cpu_present.
1237  * - Ashok Raj
1238  *
1239  * Three ways to find out the number of additional hotplug CPUs:
1240  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1241  * - The user can overwrite it with possible_cpus=NUM
1242  * - Otherwise don't reserve additional CPUs.
1243  * We do this because additional CPUs waste a lot of memory.
1244  * -AK
1245  */
1246 __init void prefill_possible_map(void)
1247 {
1248 	int i, possible;
1249 
1250 	/* no processor from mptable or madt */
1251 	if (!num_processors)
1252 		num_processors = 1;
1253 
1254 	i = setup_max_cpus ?: 1;
1255 	if (setup_possible_cpus == -1) {
1256 		possible = num_processors;
1257 #ifdef CONFIG_HOTPLUG_CPU
1258 		if (setup_max_cpus)
1259 			possible += disabled_cpus;
1260 #else
1261 		if (possible > i)
1262 			possible = i;
1263 #endif
1264 	} else
1265 		possible = setup_possible_cpus;
1266 
1267 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1268 
1269 	/* nr_cpu_ids could be reduced via nr_cpus= */
1270 	if (possible > nr_cpu_ids) {
1271 		printk(KERN_WARNING
1272 			"%d Processors exceeds NR_CPUS limit of %d\n",
1273 			possible, nr_cpu_ids);
1274 		possible = nr_cpu_ids;
1275 	}
1276 
1277 #ifdef CONFIG_HOTPLUG_CPU
1278 	if (!setup_max_cpus)
1279 #endif
1280 	if (possible > i) {
1281 		printk(KERN_WARNING
1282 			"%d Processors exceeds max_cpus limit of %u\n",
1283 			possible, setup_max_cpus);
1284 		possible = i;
1285 	}
1286 
1287 	printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1288 		possible, max_t(int, possible - num_processors, 0));
1289 
1290 	for (i = 0; i < possible; i++)
1291 		set_cpu_possible(i, true);
1292 	for (; i < NR_CPUS; i++)
1293 		set_cpu_possible(i, false);
1294 
1295 	nr_cpu_ids = possible;
1296 }
1297 
1298 #ifdef CONFIG_HOTPLUG_CPU
1299 
1300 static void remove_siblinginfo(int cpu)
1301 {
1302 	int sibling;
1303 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1304 
1305 	for_each_cpu(sibling, cpu_core_mask(cpu)) {
1306 		cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1307 		/*/
1308 		 * last thread sibling in this cpu core going down
1309 		 */
1310 		if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1311 			cpu_data(sibling).booted_cores--;
1312 	}
1313 
1314 	for_each_cpu(sibling, cpu_sibling_mask(cpu))
1315 		cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1316 	cpumask_clear(cpu_sibling_mask(cpu));
1317 	cpumask_clear(cpu_core_mask(cpu));
1318 	c->phys_proc_id = 0;
1319 	c->cpu_core_id = 0;
1320 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1321 }
1322 
1323 static void __ref remove_cpu_from_maps(int cpu)
1324 {
1325 	set_cpu_online(cpu, false);
1326 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1327 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1328 	/* was set by cpu_init() */
1329 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1330 	numa_remove_cpu(cpu);
1331 }
1332 
1333 void cpu_disable_common(void)
1334 {
1335 	int cpu = smp_processor_id();
1336 
1337 	remove_siblinginfo(cpu);
1338 
1339 	/* It's now safe to remove this processor from the online map */
1340 	lock_vector_lock();
1341 	remove_cpu_from_maps(cpu);
1342 	unlock_vector_lock();
1343 	fixup_irqs();
1344 }
1345 
1346 int native_cpu_disable(void)
1347 {
1348 	int cpu = smp_processor_id();
1349 
1350 	/*
1351 	 * Perhaps use cpufreq to drop frequency, but that could go
1352 	 * into generic code.
1353 	 *
1354 	 * We won't take down the boot processor on i386 due to some
1355 	 * interrupts only being able to be serviced by the BSP.
1356 	 * Especially so if we're not using an IOAPIC	-zwane
1357 	 */
1358 	if (cpu == 0)
1359 		return -EBUSY;
1360 
1361 	clear_local_APIC();
1362 
1363 	cpu_disable_common();
1364 	return 0;
1365 }
1366 
1367 void native_cpu_die(unsigned int cpu)
1368 {
1369 	/* We don't do anything here: idle task is faking death itself. */
1370 	unsigned int i;
1371 
1372 	for (i = 0; i < 10; i++) {
1373 		/* They ack this in play_dead by setting CPU_DEAD */
1374 		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1375 			if (system_state == SYSTEM_RUNNING)
1376 				pr_info("CPU %u is now offline\n", cpu);
1377 
1378 			if (1 == num_online_cpus())
1379 				alternatives_smp_switch(0);
1380 			return;
1381 		}
1382 		msleep(100);
1383 	}
1384 	pr_err("CPU %u didn't die...\n", cpu);
1385 }
1386 
1387 void play_dead_common(void)
1388 {
1389 	idle_task_exit();
1390 	reset_lazy_tlbstate();
1391 	c1e_remove_cpu(raw_smp_processor_id());
1392 
1393 	mb();
1394 	/* Ack it */
1395 	__this_cpu_write(cpu_state, CPU_DEAD);
1396 
1397 	/*
1398 	 * With physical CPU hotplug, we should halt the cpu
1399 	 */
1400 	local_irq_disable();
1401 }
1402 
1403 /*
1404  * We need to flush the caches before going to sleep, lest we have
1405  * dirty data in our caches when we come back up.
1406  */
1407 static inline void mwait_play_dead(void)
1408 {
1409 	unsigned int eax, ebx, ecx, edx;
1410 	unsigned int highest_cstate = 0;
1411 	unsigned int highest_subcstate = 0;
1412 	int i;
1413 	void *mwait_ptr;
1414 	struct cpuinfo_x86 *c = __this_cpu_ptr(&cpu_info);
1415 
1416 	if (!(cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)))
1417 		return;
1418 	if (!cpu_has(__this_cpu_ptr(&cpu_info), X86_FEATURE_CLFLSH))
1419 		return;
1420 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1421 		return;
1422 
1423 	eax = CPUID_MWAIT_LEAF;
1424 	ecx = 0;
1425 	native_cpuid(&eax, &ebx, &ecx, &edx);
1426 
1427 	/*
1428 	 * eax will be 0 if EDX enumeration is not valid.
1429 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1430 	 */
1431 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1432 		eax = 0;
1433 	} else {
1434 		edx >>= MWAIT_SUBSTATE_SIZE;
1435 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1436 			if (edx & MWAIT_SUBSTATE_MASK) {
1437 				highest_cstate = i;
1438 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1439 			}
1440 		}
1441 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1442 			(highest_subcstate - 1);
1443 	}
1444 
1445 	/*
1446 	 * This should be a memory location in a cache line which is
1447 	 * unlikely to be touched by other processors.  The actual
1448 	 * content is immaterial as it is not actually modified in any way.
1449 	 */
1450 	mwait_ptr = &current_thread_info()->flags;
1451 
1452 	wbinvd();
1453 
1454 	while (1) {
1455 		/*
1456 		 * The CLFLUSH is a workaround for erratum AAI65 for
1457 		 * the Xeon 7400 series.  It's not clear it is actually
1458 		 * needed, but it should be harmless in either case.
1459 		 * The WBINVD is insufficient due to the spurious-wakeup
1460 		 * case where we return around the loop.
1461 		 */
1462 		clflush(mwait_ptr);
1463 		__monitor(mwait_ptr, 0, 0);
1464 		mb();
1465 		__mwait(eax, 0);
1466 	}
1467 }
1468 
1469 static inline void hlt_play_dead(void)
1470 {
1471 	if (__this_cpu_read(cpu_info.x86) >= 4)
1472 		wbinvd();
1473 
1474 	while (1) {
1475 		native_halt();
1476 	}
1477 }
1478 
1479 void native_play_dead(void)
1480 {
1481 	play_dead_common();
1482 	tboot_shutdown(TB_SHUTDOWN_WFS);
1483 
1484 	mwait_play_dead();	/* Only returns on failure */
1485 	hlt_play_dead();
1486 }
1487 
1488 #else /* ... !CONFIG_HOTPLUG_CPU */
1489 int native_cpu_disable(void)
1490 {
1491 	return -ENOSYS;
1492 }
1493 
1494 void native_cpu_die(unsigned int cpu)
1495 {
1496 	/* We said "no" in __cpu_disable */
1497 	BUG();
1498 }
1499 
1500 void native_play_dead(void)
1501 {
1502 	BUG();
1503 }
1504 
1505 #endif
1506