xref: /linux/arch/x86/kernel/smpboot.c (revision 8ec3b8432e4fe8d452f88f1ed9a3450e715bb797)
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(void)
101 {
102         mutex_lock(&x86_cpu_hotplug_driver_mutex);
103 }
104 
105 void cpu_hotplug_driver_unlock(void)
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 	/*
285 	 * This must be done before setting cpu_online_mask
286 	 * or calling notify_cpu_starting.
287 	 */
288 	set_cpu_sibling_map(raw_smp_processor_id());
289 	wmb();
290 
291 	notify_cpu_starting(cpuid);
292 
293 	/*
294 	 * Allow the master to continue.
295 	 */
296 	cpumask_set_cpu(cpuid, cpu_callin_mask);
297 }
298 
299 /*
300  * Activate a secondary processor.
301  */
302 notrace static void __cpuinit start_secondary(void *unused)
303 {
304 	/*
305 	 * Don't put *anything* before cpu_init(), SMP booting is too
306 	 * fragile that we want to limit the things done here to the
307 	 * most necessary things.
308 	 */
309 	cpu_init();
310 	preempt_disable();
311 	smp_callin();
312 
313 #ifdef CONFIG_X86_32
314 	/* switch away from the initial page table */
315 	load_cr3(swapper_pg_dir);
316 	__flush_tlb_all();
317 #endif
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 	/*
327 	 * We need to hold call_lock, so there is no inconsistency
328 	 * between the time smp_call_function() determines number of
329 	 * IPI recipients, and the time when the determination is made
330 	 * for which cpus receive the IPI. Holding this
331 	 * lock helps us to not include this cpu in a currently in progress
332 	 * smp_call_function().
333 	 *
334 	 * We need to hold vector_lock so there the set of online cpus
335 	 * does not change while we are assigning vectors to cpus.  Holding
336 	 * this lock ensures we don't half assign or remove an irq from a cpu.
337 	 */
338 	ipi_call_lock();
339 	lock_vector_lock();
340 	set_cpu_online(smp_processor_id(), true);
341 	unlock_vector_lock();
342 	ipi_call_unlock();
343 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
344 	x86_platform.nmi_init();
345 
346 	/* enable local interrupts */
347 	local_irq_enable();
348 
349 	/* to prevent fake stack check failure in clock setup */
350 	boot_init_stack_canary();
351 
352 	x86_cpuinit.setup_percpu_clockev();
353 
354 	wmb();
355 	cpu_idle();
356 }
357 
358 #ifdef CONFIG_CPUMASK_OFFSTACK
359 /* In this case, llc_shared_map is a pointer to a cpumask. */
360 static inline void copy_cpuinfo_x86(struct cpuinfo_x86 *dst,
361 				    const struct cpuinfo_x86 *src)
362 {
363 	struct cpumask *llc = dst->llc_shared_map;
364 	*dst = *src;
365 	dst->llc_shared_map = llc;
366 }
367 #else
368 static inline void copy_cpuinfo_x86(struct cpuinfo_x86 *dst,
369 				    const struct cpuinfo_x86 *src)
370 {
371 	*dst = *src;
372 }
373 #endif /* CONFIG_CPUMASK_OFFSTACK */
374 
375 /*
376  * The bootstrap kernel entry code has set these up. Save them for
377  * a given CPU
378  */
379 
380 void __cpuinit smp_store_cpu_info(int id)
381 {
382 	struct cpuinfo_x86 *c = &cpu_data(id);
383 
384 	copy_cpuinfo_x86(c, &boot_cpu_data);
385 	c->cpu_index = id;
386 	if (id != 0)
387 		identify_secondary_cpu(c);
388 }
389 
390 static void __cpuinit link_thread_siblings(int cpu1, int cpu2)
391 {
392 	struct cpuinfo_x86 *c1 = &cpu_data(cpu1);
393 	struct cpuinfo_x86 *c2 = &cpu_data(cpu2);
394 
395 	cpumask_set_cpu(cpu1, cpu_sibling_mask(cpu2));
396 	cpumask_set_cpu(cpu2, cpu_sibling_mask(cpu1));
397 	cpumask_set_cpu(cpu1, cpu_core_mask(cpu2));
398 	cpumask_set_cpu(cpu2, cpu_core_mask(cpu1));
399 	cpumask_set_cpu(cpu1, c2->llc_shared_map);
400 	cpumask_set_cpu(cpu2, c1->llc_shared_map);
401 }
402 
403 
404 void __cpuinit set_cpu_sibling_map(int cpu)
405 {
406 	int i;
407 	struct cpuinfo_x86 *c = &cpu_data(cpu);
408 
409 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
410 
411 	if (smp_num_siblings > 1) {
412 		for_each_cpu(i, cpu_sibling_setup_mask) {
413 			struct cpuinfo_x86 *o = &cpu_data(i);
414 
415 			if (cpu_has(c, X86_FEATURE_TOPOEXT)) {
416 				if (c->phys_proc_id == o->phys_proc_id &&
417 				    c->compute_unit_id == o->compute_unit_id)
418 					link_thread_siblings(cpu, i);
419 			} else if (c->phys_proc_id == o->phys_proc_id &&
420 				   c->cpu_core_id == o->cpu_core_id) {
421 				link_thread_siblings(cpu, i);
422 			}
423 		}
424 	} else {
425 		cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
426 	}
427 
428 	cpumask_set_cpu(cpu, c->llc_shared_map);
429 
430 	if (__this_cpu_read(cpu_info.x86_max_cores) == 1) {
431 		cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
432 		c->booted_cores = 1;
433 		return;
434 	}
435 
436 	for_each_cpu(i, cpu_sibling_setup_mask) {
437 		if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
438 		    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
439 			cpumask_set_cpu(i, c->llc_shared_map);
440 			cpumask_set_cpu(cpu, cpu_data(i).llc_shared_map);
441 		}
442 		if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
443 			cpumask_set_cpu(i, cpu_core_mask(cpu));
444 			cpumask_set_cpu(cpu, cpu_core_mask(i));
445 			/*
446 			 *  Does this new cpu bringup a new core?
447 			 */
448 			if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
449 				/*
450 				 * for each core in package, increment
451 				 * the booted_cores for this new cpu
452 				 */
453 				if (cpumask_first(cpu_sibling_mask(i)) == i)
454 					c->booted_cores++;
455 				/*
456 				 * increment the core count for all
457 				 * the other cpus in this package
458 				 */
459 				if (i != cpu)
460 					cpu_data(i).booted_cores++;
461 			} else if (i != cpu && !c->booted_cores)
462 				c->booted_cores = cpu_data(i).booted_cores;
463 		}
464 	}
465 }
466 
467 /* maps the cpu to the sched domain representing multi-core */
468 const struct cpumask *cpu_coregroup_mask(int cpu)
469 {
470 	struct cpuinfo_x86 *c = &cpu_data(cpu);
471 	/*
472 	 * For perf, we return last level cache shared map.
473 	 * And for power savings, we return cpu_core_map
474 	 */
475 	if ((sched_mc_power_savings || sched_smt_power_savings) &&
476 	    !(cpu_has(c, X86_FEATURE_AMD_DCM)))
477 		return cpu_core_mask(cpu);
478 	else
479 		return c->llc_shared_map;
480 }
481 
482 static void impress_friends(void)
483 {
484 	int cpu;
485 	unsigned long bogosum = 0;
486 	/*
487 	 * Allow the user to impress friends.
488 	 */
489 	pr_debug("Before bogomips.\n");
490 	for_each_possible_cpu(cpu)
491 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
492 			bogosum += cpu_data(cpu).loops_per_jiffy;
493 	printk(KERN_INFO
494 		"Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
495 		num_online_cpus(),
496 		bogosum/(500000/HZ),
497 		(bogosum/(5000/HZ))%100);
498 
499 	pr_debug("Before bogocount - setting activated=1.\n");
500 }
501 
502 void __inquire_remote_apic(int apicid)
503 {
504 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
505 	char *names[] = { "ID", "VERSION", "SPIV" };
506 	int timeout;
507 	u32 status;
508 
509 	printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
510 
511 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
512 		printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
513 
514 		/*
515 		 * Wait for idle.
516 		 */
517 		status = safe_apic_wait_icr_idle();
518 		if (status)
519 			printk(KERN_CONT
520 			       "a previous APIC delivery may have failed\n");
521 
522 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
523 
524 		timeout = 0;
525 		do {
526 			udelay(100);
527 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
528 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
529 
530 		switch (status) {
531 		case APIC_ICR_RR_VALID:
532 			status = apic_read(APIC_RRR);
533 			printk(KERN_CONT "%08x\n", status);
534 			break;
535 		default:
536 			printk(KERN_CONT "failed\n");
537 		}
538 	}
539 }
540 
541 /*
542  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
543  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
544  * won't ... remember to clear down the APIC, etc later.
545  */
546 int __cpuinit
547 wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
548 {
549 	unsigned long send_status, accept_status = 0;
550 	int maxlvt;
551 
552 	/* Target chip */
553 	/* Boot on the stack */
554 	/* Kick the second */
555 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, logical_apicid);
556 
557 	pr_debug("Waiting for send to finish...\n");
558 	send_status = safe_apic_wait_icr_idle();
559 
560 	/*
561 	 * Give the other CPU some time to accept the IPI.
562 	 */
563 	udelay(200);
564 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
565 		maxlvt = lapic_get_maxlvt();
566 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
567 			apic_write(APIC_ESR, 0);
568 		accept_status = (apic_read(APIC_ESR) & 0xEF);
569 	}
570 	pr_debug("NMI sent.\n");
571 
572 	if (send_status)
573 		printk(KERN_ERR "APIC never delivered???\n");
574 	if (accept_status)
575 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
576 
577 	return (send_status | accept_status);
578 }
579 
580 static int __cpuinit
581 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
582 {
583 	unsigned long send_status, accept_status = 0;
584 	int maxlvt, num_starts, j;
585 
586 	maxlvt = lapic_get_maxlvt();
587 
588 	/*
589 	 * Be paranoid about clearing APIC errors.
590 	 */
591 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
592 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
593 			apic_write(APIC_ESR, 0);
594 		apic_read(APIC_ESR);
595 	}
596 
597 	pr_debug("Asserting INIT.\n");
598 
599 	/*
600 	 * Turn INIT on target chip
601 	 */
602 	/*
603 	 * Send IPI
604 	 */
605 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
606 		       phys_apicid);
607 
608 	pr_debug("Waiting for send to finish...\n");
609 	send_status = safe_apic_wait_icr_idle();
610 
611 	mdelay(10);
612 
613 	pr_debug("Deasserting INIT.\n");
614 
615 	/* Target chip */
616 	/* Send IPI */
617 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
618 
619 	pr_debug("Waiting for send to finish...\n");
620 	send_status = safe_apic_wait_icr_idle();
621 
622 	mb();
623 	atomic_set(&init_deasserted, 1);
624 
625 	/*
626 	 * Should we send STARTUP IPIs ?
627 	 *
628 	 * Determine this based on the APIC version.
629 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
630 	 */
631 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
632 		num_starts = 2;
633 	else
634 		num_starts = 0;
635 
636 	/*
637 	 * Paravirt / VMI wants a startup IPI hook here to set up the
638 	 * target processor state.
639 	 */
640 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
641 			 stack_start);
642 
643 	/*
644 	 * Run STARTUP IPI loop.
645 	 */
646 	pr_debug("#startup loops: %d.\n", num_starts);
647 
648 	for (j = 1; j <= num_starts; j++) {
649 		pr_debug("Sending STARTUP #%d.\n", j);
650 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
651 			apic_write(APIC_ESR, 0);
652 		apic_read(APIC_ESR);
653 		pr_debug("After apic_write.\n");
654 
655 		/*
656 		 * STARTUP IPI
657 		 */
658 
659 		/* Target chip */
660 		/* Boot on the stack */
661 		/* Kick the second */
662 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
663 			       phys_apicid);
664 
665 		/*
666 		 * Give the other CPU some time to accept the IPI.
667 		 */
668 		udelay(300);
669 
670 		pr_debug("Startup point 1.\n");
671 
672 		pr_debug("Waiting for send to finish...\n");
673 		send_status = safe_apic_wait_icr_idle();
674 
675 		/*
676 		 * Give the other CPU some time to accept the IPI.
677 		 */
678 		udelay(200);
679 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
680 			apic_write(APIC_ESR, 0);
681 		accept_status = (apic_read(APIC_ESR) & 0xEF);
682 		if (send_status || accept_status)
683 			break;
684 	}
685 	pr_debug("After Startup.\n");
686 
687 	if (send_status)
688 		printk(KERN_ERR "APIC never delivered???\n");
689 	if (accept_status)
690 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
691 
692 	return (send_status | accept_status);
693 }
694 
695 struct create_idle {
696 	struct work_struct work;
697 	struct task_struct *idle;
698 	struct completion done;
699 	int cpu;
700 };
701 
702 static void __cpuinit do_fork_idle(struct work_struct *work)
703 {
704 	struct create_idle *c_idle =
705 		container_of(work, struct create_idle, work);
706 
707 	c_idle->idle = fork_idle(c_idle->cpu);
708 	complete(&c_idle->done);
709 }
710 
711 /* reduce the number of lines printed when booting a large cpu count system */
712 static void __cpuinit announce_cpu(int cpu, int apicid)
713 {
714 	static int current_node = -1;
715 	int node = early_cpu_to_node(cpu);
716 
717 	if (system_state == SYSTEM_BOOTING) {
718 		if (node != current_node) {
719 			if (current_node > (-1))
720 				pr_cont(" Ok.\n");
721 			current_node = node;
722 			pr_info("Booting Node %3d, Processors ", node);
723 		}
724 		pr_cont(" #%d%s", cpu, cpu == (nr_cpu_ids - 1) ? " Ok.\n" : "");
725 		return;
726 	} else
727 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
728 			node, cpu, apicid);
729 }
730 
731 /*
732  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
733  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
734  * Returns zero if CPU booted OK, else error code from
735  * ->wakeup_secondary_cpu.
736  */
737 static int __cpuinit do_boot_cpu(int apicid, int cpu)
738 {
739 	unsigned long boot_error = 0;
740 	unsigned long start_ip;
741 	int timeout;
742 	struct create_idle c_idle = {
743 		.cpu	= cpu,
744 		.done	= COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
745 	};
746 
747 	INIT_WORK_ONSTACK(&c_idle.work, do_fork_idle);
748 
749 	alternatives_smp_switch(1);
750 
751 	c_idle.idle = get_idle_for_cpu(cpu);
752 
753 	/*
754 	 * We can't use kernel_thread since we must avoid to
755 	 * reschedule the child.
756 	 */
757 	if (c_idle.idle) {
758 		c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
759 			(THREAD_SIZE +  task_stack_page(c_idle.idle))) - 1);
760 		init_idle(c_idle.idle, cpu);
761 		goto do_rest;
762 	}
763 
764 	schedule_work(&c_idle.work);
765 	wait_for_completion(&c_idle.done);
766 
767 	if (IS_ERR(c_idle.idle)) {
768 		printk("failed fork for CPU %d\n", cpu);
769 		destroy_work_on_stack(&c_idle.work);
770 		return PTR_ERR(c_idle.idle);
771 	}
772 
773 	set_idle_for_cpu(cpu, c_idle.idle);
774 do_rest:
775 	per_cpu(current_task, cpu) = c_idle.idle;
776 #ifdef CONFIG_X86_32
777 	/* Stack for startup_32 can be just as for start_secondary onwards */
778 	irq_ctx_init(cpu);
779 #else
780 	clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
781 	initial_gs = per_cpu_offset(cpu);
782 	per_cpu(kernel_stack, cpu) =
783 		(unsigned long)task_stack_page(c_idle.idle) -
784 		KERNEL_STACK_OFFSET + THREAD_SIZE;
785 #endif
786 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
787 	initial_code = (unsigned long)start_secondary;
788 	stack_start  = c_idle.idle->thread.sp;
789 
790 	/* start_ip had better be page-aligned! */
791 	start_ip = setup_trampoline();
792 
793 	/* So we see what's up */
794 	announce_cpu(cpu, apicid);
795 
796 	/*
797 	 * This grunge runs the startup process for
798 	 * the targeted processor.
799 	 */
800 
801 	atomic_set(&init_deasserted, 0);
802 
803 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
804 
805 		pr_debug("Setting warm reset code and vector.\n");
806 
807 		smpboot_setup_warm_reset_vector(start_ip);
808 		/*
809 		 * Be paranoid about clearing APIC errors.
810 		*/
811 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
812 			apic_write(APIC_ESR, 0);
813 			apic_read(APIC_ESR);
814 		}
815 	}
816 
817 	/*
818 	 * Kick the secondary CPU. Use the method in the APIC driver
819 	 * if it's defined - or use an INIT boot APIC message otherwise:
820 	 */
821 	if (apic->wakeup_secondary_cpu)
822 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
823 	else
824 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
825 
826 	if (!boot_error) {
827 		/*
828 		 * allow APs to start initializing.
829 		 */
830 		pr_debug("Before Callout %d.\n", cpu);
831 		cpumask_set_cpu(cpu, cpu_callout_mask);
832 		pr_debug("After Callout %d.\n", cpu);
833 
834 		/*
835 		 * Wait 5s total for a response
836 		 */
837 		for (timeout = 0; timeout < 50000; timeout++) {
838 			if (cpumask_test_cpu(cpu, cpu_callin_mask))
839 				break;	/* It has booted */
840 			udelay(100);
841 			/*
842 			 * Allow other tasks to run while we wait for the
843 			 * AP to come online. This also gives a chance
844 			 * for the MTRR work(triggered by the AP coming online)
845 			 * to be completed in the stop machine context.
846 			 */
847 			schedule();
848 		}
849 
850 		if (cpumask_test_cpu(cpu, cpu_callin_mask))
851 			pr_debug("CPU%d: has booted.\n", cpu);
852 		else {
853 			boot_error = 1;
854 			if (*((volatile unsigned char *)trampoline_base)
855 					== 0xA5)
856 				/* trampoline started but...? */
857 				pr_err("CPU%d: Stuck ??\n", cpu);
858 			else
859 				/* trampoline code not run */
860 				pr_err("CPU%d: Not responding.\n", cpu);
861 			if (apic->inquire_remote_apic)
862 				apic->inquire_remote_apic(apicid);
863 		}
864 	}
865 
866 	if (boot_error) {
867 		/* Try to put things back the way they were before ... */
868 		numa_remove_cpu(cpu); /* was set by numa_add_cpu */
869 
870 		/* was set by do_boot_cpu() */
871 		cpumask_clear_cpu(cpu, cpu_callout_mask);
872 
873 		/* was set by cpu_init() */
874 		cpumask_clear_cpu(cpu, cpu_initialized_mask);
875 
876 		set_cpu_present(cpu, false);
877 		per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
878 	}
879 
880 	/* mark "stuck" area as not stuck */
881 	*((volatile unsigned long *)trampoline_base) = 0;
882 
883 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
884 		/*
885 		 * Cleanup possible dangling ends...
886 		 */
887 		smpboot_restore_warm_reset_vector();
888 	}
889 
890 	destroy_work_on_stack(&c_idle.work);
891 	return boot_error;
892 }
893 
894 int __cpuinit native_cpu_up(unsigned int cpu)
895 {
896 	int apicid = apic->cpu_present_to_apicid(cpu);
897 	unsigned long flags;
898 	int err;
899 
900 	WARN_ON(irqs_disabled());
901 
902 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
903 
904 	if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
905 	    !physid_isset(apicid, phys_cpu_present_map)) {
906 		printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
907 		return -EINVAL;
908 	}
909 
910 	/*
911 	 * Already booted CPU?
912 	 */
913 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
914 		pr_debug("do_boot_cpu %d Already started\n", cpu);
915 		return -ENOSYS;
916 	}
917 
918 	/*
919 	 * Save current MTRR state in case it was changed since early boot
920 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
921 	 */
922 	mtrr_save_state();
923 
924 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
925 
926 	err = do_boot_cpu(apicid, cpu);
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 		connect_bsp_APIC();
1062 		setup_local_APIC();
1063 		bsp_end_local_APIC_setup();
1064 		return -1;
1065 	}
1066 
1067 	return 0;
1068 }
1069 
1070 static void __init smp_cpu_index_default(void)
1071 {
1072 	int i;
1073 	struct cpuinfo_x86 *c;
1074 
1075 	for_each_possible_cpu(i) {
1076 		c = &cpu_data(i);
1077 		/* mark all to hotplug */
1078 		c->cpu_index = nr_cpu_ids;
1079 	}
1080 }
1081 
1082 /*
1083  * Prepare for SMP bootup.  The MP table or ACPI has been read
1084  * earlier.  Just do some sanity checking here and enable APIC mode.
1085  */
1086 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1087 {
1088 	unsigned int i;
1089 
1090 	preempt_disable();
1091 	smp_cpu_index_default();
1092 	memcpy(__this_cpu_ptr(&cpu_info), &boot_cpu_data, sizeof(cpu_info));
1093 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1094 	mb();
1095 	/*
1096 	 * Setup boot CPU information
1097 	 */
1098 	smp_store_cpu_info(0); /* Final full version of the data */
1099 #ifdef CONFIG_X86_32
1100 	boot_cpu_logical_apicid = logical_smp_processor_id();
1101 #endif
1102 	current_thread_info()->cpu = 0;  /* needed? */
1103 	for_each_possible_cpu(i) {
1104 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1105 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1106 		zalloc_cpumask_var(&cpu_data(i).llc_shared_map, GFP_KERNEL);
1107 	}
1108 	set_cpu_sibling_map(0);
1109 
1110 
1111 	if (smp_sanity_check(max_cpus) < 0) {
1112 		printk(KERN_INFO "SMP disabled\n");
1113 		disable_smp();
1114 		goto out;
1115 	}
1116 
1117 	default_setup_apic_routing();
1118 
1119 	preempt_disable();
1120 	if (read_apic_id() != boot_cpu_physical_apicid) {
1121 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1122 		     read_apic_id(), boot_cpu_physical_apicid);
1123 		/* Or can we switch back to PIC here? */
1124 	}
1125 	preempt_enable();
1126 
1127 	connect_bsp_APIC();
1128 
1129 	/*
1130 	 * Switch from PIC to APIC mode.
1131 	 */
1132 	setup_local_APIC();
1133 
1134 	/*
1135 	 * Enable IO APIC before setting up error vector
1136 	 */
1137 	if (!skip_ioapic_setup && nr_ioapics)
1138 		enable_IO_APIC();
1139 
1140 	bsp_end_local_APIC_setup();
1141 
1142 	map_cpu_to_logical_apicid();
1143 
1144 	if (apic->setup_portio_remap)
1145 		apic->setup_portio_remap();
1146 
1147 	smpboot_setup_io_apic();
1148 	/*
1149 	 * Set up local APIC timer on boot CPU.
1150 	 */
1151 
1152 	printk(KERN_INFO "CPU%d: ", 0);
1153 	print_cpu_info(&cpu_data(0));
1154 	x86_init.timers.setup_percpu_clockev();
1155 
1156 	if (is_uv_system())
1157 		uv_system_init();
1158 
1159 	set_mtrr_aps_delayed_init();
1160 out:
1161 	preempt_enable();
1162 }
1163 
1164 void arch_disable_nonboot_cpus_begin(void)
1165 {
1166 	/*
1167 	 * Avoid the smp alternatives switch during the disable_nonboot_cpus().
1168 	 * In the suspend path, we will be back in the SMP mode shortly anyways.
1169 	 */
1170 	skip_smp_alternatives = true;
1171 }
1172 
1173 void arch_disable_nonboot_cpus_end(void)
1174 {
1175 	skip_smp_alternatives = false;
1176 }
1177 
1178 void arch_enable_nonboot_cpus_begin(void)
1179 {
1180 	set_mtrr_aps_delayed_init();
1181 }
1182 
1183 void arch_enable_nonboot_cpus_end(void)
1184 {
1185 	mtrr_aps_init();
1186 }
1187 
1188 /*
1189  * Early setup to make printk work.
1190  */
1191 void __init native_smp_prepare_boot_cpu(void)
1192 {
1193 	int me = smp_processor_id();
1194 	switch_to_new_gdt(me);
1195 	/* already set me in cpu_online_mask in boot_cpu_init() */
1196 	cpumask_set_cpu(me, cpu_callout_mask);
1197 	per_cpu(cpu_state, me) = CPU_ONLINE;
1198 }
1199 
1200 void __init native_smp_cpus_done(unsigned int max_cpus)
1201 {
1202 	pr_debug("Boot done.\n");
1203 
1204 	impress_friends();
1205 #ifdef CONFIG_X86_IO_APIC
1206 	setup_ioapic_dest();
1207 #endif
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 	clear_local_APIC();
1353 
1354 	cpu_disable_common();
1355 	return 0;
1356 }
1357 
1358 void native_cpu_die(unsigned int cpu)
1359 {
1360 	/* We don't do anything here: idle task is faking death itself. */
1361 	unsigned int i;
1362 
1363 	for (i = 0; i < 10; i++) {
1364 		/* They ack this in play_dead by setting CPU_DEAD */
1365 		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1366 			if (system_state == SYSTEM_RUNNING)
1367 				pr_info("CPU %u is now offline\n", cpu);
1368 
1369 			if (1 == num_online_cpus())
1370 				alternatives_smp_switch(0);
1371 			return;
1372 		}
1373 		msleep(100);
1374 	}
1375 	pr_err("CPU %u didn't die...\n", cpu);
1376 }
1377 
1378 void play_dead_common(void)
1379 {
1380 	idle_task_exit();
1381 	reset_lazy_tlbstate();
1382 	c1e_remove_cpu(raw_smp_processor_id());
1383 
1384 	mb();
1385 	/* Ack it */
1386 	__this_cpu_write(cpu_state, CPU_DEAD);
1387 
1388 	/*
1389 	 * With physical CPU hotplug, we should halt the cpu
1390 	 */
1391 	local_irq_disable();
1392 }
1393 
1394 /*
1395  * We need to flush the caches before going to sleep, lest we have
1396  * dirty data in our caches when we come back up.
1397  */
1398 static inline void mwait_play_dead(void)
1399 {
1400 	unsigned int eax, ebx, ecx, edx;
1401 	unsigned int highest_cstate = 0;
1402 	unsigned int highest_subcstate = 0;
1403 	int i;
1404 	void *mwait_ptr;
1405 	struct cpuinfo_x86 *c = __this_cpu_ptr(&cpu_info);
1406 
1407 	if (!(cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)))
1408 		return;
1409 	if (!cpu_has(__this_cpu_ptr(&cpu_info), X86_FEATURE_CLFLSH))
1410 		return;
1411 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1412 		return;
1413 
1414 	eax = CPUID_MWAIT_LEAF;
1415 	ecx = 0;
1416 	native_cpuid(&eax, &ebx, &ecx, &edx);
1417 
1418 	/*
1419 	 * eax will be 0 if EDX enumeration is not valid.
1420 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1421 	 */
1422 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1423 		eax = 0;
1424 	} else {
1425 		edx >>= MWAIT_SUBSTATE_SIZE;
1426 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1427 			if (edx & MWAIT_SUBSTATE_MASK) {
1428 				highest_cstate = i;
1429 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1430 			}
1431 		}
1432 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1433 			(highest_subcstate - 1);
1434 	}
1435 
1436 	/*
1437 	 * This should be a memory location in a cache line which is
1438 	 * unlikely to be touched by other processors.  The actual
1439 	 * content is immaterial as it is not actually modified in any way.
1440 	 */
1441 	mwait_ptr = &current_thread_info()->flags;
1442 
1443 	wbinvd();
1444 
1445 	while (1) {
1446 		/*
1447 		 * The CLFLUSH is a workaround for erratum AAI65 for
1448 		 * the Xeon 7400 series.  It's not clear it is actually
1449 		 * needed, but it should be harmless in either case.
1450 		 * The WBINVD is insufficient due to the spurious-wakeup
1451 		 * case where we return around the loop.
1452 		 */
1453 		clflush(mwait_ptr);
1454 		__monitor(mwait_ptr, 0, 0);
1455 		mb();
1456 		__mwait(eax, 0);
1457 	}
1458 }
1459 
1460 static inline void hlt_play_dead(void)
1461 {
1462 	if (__this_cpu_read(cpu_info.x86) >= 4)
1463 		wbinvd();
1464 
1465 	while (1) {
1466 		native_halt();
1467 	}
1468 }
1469 
1470 void native_play_dead(void)
1471 {
1472 	play_dead_common();
1473 	tboot_shutdown(TB_SHUTDOWN_WFS);
1474 
1475 	mwait_play_dead();	/* Only returns on failure */
1476 	hlt_play_dead();
1477 }
1478 
1479 #else /* ... !CONFIG_HOTPLUG_CPU */
1480 int native_cpu_disable(void)
1481 {
1482 	return -ENOSYS;
1483 }
1484 
1485 void native_cpu_die(unsigned int cpu)
1486 {
1487 	/* We said "no" in __cpu_disable */
1488 	BUG();
1489 }
1490 
1491 void native_play_dead(void)
1492 {
1493 	BUG();
1494 }
1495 
1496 #endif
1497