xref: /linux/arch/x86/kernel/smpboot.c (revision cb299ba8b5ef2239429484072fea394cd7581bd7)
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 	cpu_init();
303 	preempt_disable();
304 	smp_callin();
305 
306 #ifdef CONFIG_X86_32
307 	/* switch away from the initial page table */
308 	load_cr3(swapper_pg_dir);
309 	__flush_tlb_all();
310 #endif
311 
312 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
313 	barrier();
314 	/*
315 	 * Check TSC synchronization with the BP:
316 	 */
317 	check_tsc_sync_target();
318 
319 	if (nmi_watchdog == NMI_IO_APIC) {
320 		legacy_pic->mask(0);
321 		enable_NMI_through_LVT0();
322 		legacy_pic->unmask(0);
323 	}
324 
325 	/* This must be done before setting cpu_online_mask */
326 	set_cpu_sibling_map(raw_smp_processor_id());
327 	wmb();
328 
329 	/*
330 	 * We need to hold call_lock, so there is no inconsistency
331 	 * between the time smp_call_function() determines number of
332 	 * IPI recipients, and the time when the determination is made
333 	 * for which cpus receive the IPI. Holding this
334 	 * lock helps us to not include this cpu in a currently in progress
335 	 * smp_call_function().
336 	 *
337 	 * We need to hold vector_lock so there the set of online cpus
338 	 * does not change while we are assigning vectors to cpus.  Holding
339 	 * this lock ensures we don't half assign or remove an irq from a cpu.
340 	 */
341 	ipi_call_lock();
342 	lock_vector_lock();
343 	set_cpu_online(smp_processor_id(), true);
344 	unlock_vector_lock();
345 	ipi_call_unlock();
346 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
347 	x86_platform.nmi_init();
348 
349 	/* enable local interrupts */
350 	local_irq_enable();
351 
352 	/* to prevent fake stack check failure in clock setup */
353 	boot_init_stack_canary();
354 
355 	x86_cpuinit.setup_percpu_clockev();
356 
357 	wmb();
358 	cpu_idle();
359 }
360 
361 #ifdef CONFIG_CPUMASK_OFFSTACK
362 /* In this case, llc_shared_map is a pointer to a cpumask. */
363 static inline void copy_cpuinfo_x86(struct cpuinfo_x86 *dst,
364 				    const struct cpuinfo_x86 *src)
365 {
366 	struct cpumask *llc = dst->llc_shared_map;
367 	*dst = *src;
368 	dst->llc_shared_map = llc;
369 }
370 #else
371 static inline void copy_cpuinfo_x86(struct cpuinfo_x86 *dst,
372 				    const struct cpuinfo_x86 *src)
373 {
374 	*dst = *src;
375 }
376 #endif /* CONFIG_CPUMASK_OFFSTACK */
377 
378 /*
379  * The bootstrap kernel entry code has set these up. Save them for
380  * a given CPU
381  */
382 
383 void __cpuinit smp_store_cpu_info(int id)
384 {
385 	struct cpuinfo_x86 *c = &cpu_data(id);
386 
387 	copy_cpuinfo_x86(c, &boot_cpu_data);
388 	c->cpu_index = id;
389 	if (id != 0)
390 		identify_secondary_cpu(c);
391 }
392 
393 static void __cpuinit link_thread_siblings(int cpu1, int cpu2)
394 {
395 	struct cpuinfo_x86 *c1 = &cpu_data(cpu1);
396 	struct cpuinfo_x86 *c2 = &cpu_data(cpu2);
397 
398 	cpumask_set_cpu(cpu1, cpu_sibling_mask(cpu2));
399 	cpumask_set_cpu(cpu2, cpu_sibling_mask(cpu1));
400 	cpumask_set_cpu(cpu1, cpu_core_mask(cpu2));
401 	cpumask_set_cpu(cpu2, cpu_core_mask(cpu1));
402 	cpumask_set_cpu(cpu1, c2->llc_shared_map);
403 	cpumask_set_cpu(cpu2, c1->llc_shared_map);
404 }
405 
406 
407 void __cpuinit set_cpu_sibling_map(int cpu)
408 {
409 	int i;
410 	struct cpuinfo_x86 *c = &cpu_data(cpu);
411 
412 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
413 
414 	if (smp_num_siblings > 1) {
415 		for_each_cpu(i, cpu_sibling_setup_mask) {
416 			struct cpuinfo_x86 *o = &cpu_data(i);
417 
418 			if (cpu_has(c, X86_FEATURE_TOPOEXT)) {
419 				if (c->phys_proc_id == o->phys_proc_id &&
420 				    c->compute_unit_id == o->compute_unit_id)
421 					link_thread_siblings(cpu, i);
422 			} else if (c->phys_proc_id == o->phys_proc_id &&
423 				   c->cpu_core_id == o->cpu_core_id) {
424 				link_thread_siblings(cpu, i);
425 			}
426 		}
427 	} else {
428 		cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
429 	}
430 
431 	cpumask_set_cpu(cpu, c->llc_shared_map);
432 
433 	if (current_cpu_data.x86_max_cores == 1) {
434 		cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
435 		c->booted_cores = 1;
436 		return;
437 	}
438 
439 	for_each_cpu(i, cpu_sibling_setup_mask) {
440 		if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
441 		    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
442 			cpumask_set_cpu(i, c->llc_shared_map);
443 			cpumask_set_cpu(cpu, cpu_data(i).llc_shared_map);
444 		}
445 		if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
446 			cpumask_set_cpu(i, cpu_core_mask(cpu));
447 			cpumask_set_cpu(cpu, cpu_core_mask(i));
448 			/*
449 			 *  Does this new cpu bringup a new core?
450 			 */
451 			if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
452 				/*
453 				 * for each core in package, increment
454 				 * the booted_cores for this new cpu
455 				 */
456 				if (cpumask_first(cpu_sibling_mask(i)) == i)
457 					c->booted_cores++;
458 				/*
459 				 * increment the core count for all
460 				 * the other cpus in this package
461 				 */
462 				if (i != cpu)
463 					cpu_data(i).booted_cores++;
464 			} else if (i != cpu && !c->booted_cores)
465 				c->booted_cores = cpu_data(i).booted_cores;
466 		}
467 	}
468 }
469 
470 /* maps the cpu to the sched domain representing multi-core */
471 const struct cpumask *cpu_coregroup_mask(int cpu)
472 {
473 	struct cpuinfo_x86 *c = &cpu_data(cpu);
474 	/*
475 	 * For perf, we return last level cache shared map.
476 	 * And for power savings, we return cpu_core_map
477 	 */
478 	if ((sched_mc_power_savings || sched_smt_power_savings) &&
479 	    !(cpu_has(c, X86_FEATURE_AMD_DCM)))
480 		return cpu_core_mask(cpu);
481 	else
482 		return c->llc_shared_map;
483 }
484 
485 static void impress_friends(void)
486 {
487 	int cpu;
488 	unsigned long bogosum = 0;
489 	/*
490 	 * Allow the user to impress friends.
491 	 */
492 	pr_debug("Before bogomips.\n");
493 	for_each_possible_cpu(cpu)
494 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
495 			bogosum += cpu_data(cpu).loops_per_jiffy;
496 	printk(KERN_INFO
497 		"Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
498 		num_online_cpus(),
499 		bogosum/(500000/HZ),
500 		(bogosum/(5000/HZ))%100);
501 
502 	pr_debug("Before bogocount - setting activated=1.\n");
503 }
504 
505 void __inquire_remote_apic(int apicid)
506 {
507 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
508 	char *names[] = { "ID", "VERSION", "SPIV" };
509 	int timeout;
510 	u32 status;
511 
512 	printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
513 
514 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
515 		printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
516 
517 		/*
518 		 * Wait for idle.
519 		 */
520 		status = safe_apic_wait_icr_idle();
521 		if (status)
522 			printk(KERN_CONT
523 			       "a previous APIC delivery may have failed\n");
524 
525 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
526 
527 		timeout = 0;
528 		do {
529 			udelay(100);
530 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
531 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
532 
533 		switch (status) {
534 		case APIC_ICR_RR_VALID:
535 			status = apic_read(APIC_RRR);
536 			printk(KERN_CONT "%08x\n", status);
537 			break;
538 		default:
539 			printk(KERN_CONT "failed\n");
540 		}
541 	}
542 }
543 
544 /*
545  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
546  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
547  * won't ... remember to clear down the APIC, etc later.
548  */
549 int __cpuinit
550 wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
551 {
552 	unsigned long send_status, accept_status = 0;
553 	int maxlvt;
554 
555 	/* Target chip */
556 	/* Boot on the stack */
557 	/* Kick the second */
558 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, logical_apicid);
559 
560 	pr_debug("Waiting for send to finish...\n");
561 	send_status = safe_apic_wait_icr_idle();
562 
563 	/*
564 	 * Give the other CPU some time to accept the IPI.
565 	 */
566 	udelay(200);
567 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
568 		maxlvt = lapic_get_maxlvt();
569 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
570 			apic_write(APIC_ESR, 0);
571 		accept_status = (apic_read(APIC_ESR) & 0xEF);
572 	}
573 	pr_debug("NMI sent.\n");
574 
575 	if (send_status)
576 		printk(KERN_ERR "APIC never delivered???\n");
577 	if (accept_status)
578 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
579 
580 	return (send_status | accept_status);
581 }
582 
583 static int __cpuinit
584 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
585 {
586 	unsigned long send_status, accept_status = 0;
587 	int maxlvt, num_starts, j;
588 
589 	maxlvt = lapic_get_maxlvt();
590 
591 	/*
592 	 * Be paranoid about clearing APIC errors.
593 	 */
594 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
595 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
596 			apic_write(APIC_ESR, 0);
597 		apic_read(APIC_ESR);
598 	}
599 
600 	pr_debug("Asserting INIT.\n");
601 
602 	/*
603 	 * Turn INIT on target chip
604 	 */
605 	/*
606 	 * Send IPI
607 	 */
608 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
609 		       phys_apicid);
610 
611 	pr_debug("Waiting for send to finish...\n");
612 	send_status = safe_apic_wait_icr_idle();
613 
614 	mdelay(10);
615 
616 	pr_debug("Deasserting INIT.\n");
617 
618 	/* Target chip */
619 	/* Send IPI */
620 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
621 
622 	pr_debug("Waiting for send to finish...\n");
623 	send_status = safe_apic_wait_icr_idle();
624 
625 	mb();
626 	atomic_set(&init_deasserted, 1);
627 
628 	/*
629 	 * Should we send STARTUP IPIs ?
630 	 *
631 	 * Determine this based on the APIC version.
632 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
633 	 */
634 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
635 		num_starts = 2;
636 	else
637 		num_starts = 0;
638 
639 	/*
640 	 * Paravirt / VMI wants a startup IPI hook here to set up the
641 	 * target processor state.
642 	 */
643 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
644 			 (unsigned long)stack_start.sp);
645 
646 	/*
647 	 * Run STARTUP IPI loop.
648 	 */
649 	pr_debug("#startup loops: %d.\n", num_starts);
650 
651 	for (j = 1; j <= num_starts; j++) {
652 		pr_debug("Sending STARTUP #%d.\n", j);
653 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
654 			apic_write(APIC_ESR, 0);
655 		apic_read(APIC_ESR);
656 		pr_debug("After apic_write.\n");
657 
658 		/*
659 		 * STARTUP IPI
660 		 */
661 
662 		/* Target chip */
663 		/* Boot on the stack */
664 		/* Kick the second */
665 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
666 			       phys_apicid);
667 
668 		/*
669 		 * Give the other CPU some time to accept the IPI.
670 		 */
671 		udelay(300);
672 
673 		pr_debug("Startup point 1.\n");
674 
675 		pr_debug("Waiting for send to finish...\n");
676 		send_status = safe_apic_wait_icr_idle();
677 
678 		/*
679 		 * Give the other CPU some time to accept the IPI.
680 		 */
681 		udelay(200);
682 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
683 			apic_write(APIC_ESR, 0);
684 		accept_status = (apic_read(APIC_ESR) & 0xEF);
685 		if (send_status || accept_status)
686 			break;
687 	}
688 	pr_debug("After Startup.\n");
689 
690 	if (send_status)
691 		printk(KERN_ERR "APIC never delivered???\n");
692 	if (accept_status)
693 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
694 
695 	return (send_status | accept_status);
696 }
697 
698 struct create_idle {
699 	struct work_struct work;
700 	struct task_struct *idle;
701 	struct completion done;
702 	int cpu;
703 };
704 
705 static void __cpuinit do_fork_idle(struct work_struct *work)
706 {
707 	struct create_idle *c_idle =
708 		container_of(work, struct create_idle, work);
709 
710 	c_idle->idle = fork_idle(c_idle->cpu);
711 	complete(&c_idle->done);
712 }
713 
714 /* reduce the number of lines printed when booting a large cpu count system */
715 static void __cpuinit announce_cpu(int cpu, int apicid)
716 {
717 	static int current_node = -1;
718 	int node = early_cpu_to_node(cpu);
719 
720 	if (system_state == SYSTEM_BOOTING) {
721 		if (node != current_node) {
722 			if (current_node > (-1))
723 				pr_cont(" Ok.\n");
724 			current_node = node;
725 			pr_info("Booting Node %3d, Processors ", node);
726 		}
727 		pr_cont(" #%d%s", cpu, cpu == (nr_cpu_ids - 1) ? " Ok.\n" : "");
728 		return;
729 	} else
730 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
731 			node, cpu, apicid);
732 }
733 
734 /*
735  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
736  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
737  * Returns zero if CPU booted OK, else error code from
738  * ->wakeup_secondary_cpu.
739  */
740 static int __cpuinit do_boot_cpu(int apicid, int cpu)
741 {
742 	unsigned long boot_error = 0;
743 	unsigned long start_ip;
744 	int timeout;
745 	struct create_idle c_idle = {
746 		.cpu	= cpu,
747 		.done	= COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
748 	};
749 
750 	INIT_WORK_ONSTACK(&c_idle.work, do_fork_idle);
751 
752 	alternatives_smp_switch(1);
753 
754 	c_idle.idle = get_idle_for_cpu(cpu);
755 
756 	/*
757 	 * We can't use kernel_thread since we must avoid to
758 	 * reschedule the child.
759 	 */
760 	if (c_idle.idle) {
761 		c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
762 			(THREAD_SIZE +  task_stack_page(c_idle.idle))) - 1);
763 		init_idle(c_idle.idle, cpu);
764 		goto do_rest;
765 	}
766 
767 	schedule_work(&c_idle.work);
768 	wait_for_completion(&c_idle.done);
769 
770 	if (IS_ERR(c_idle.idle)) {
771 		printk("failed fork for CPU %d\n", cpu);
772 		destroy_work_on_stack(&c_idle.work);
773 		return PTR_ERR(c_idle.idle);
774 	}
775 
776 	set_idle_for_cpu(cpu, c_idle.idle);
777 do_rest:
778 	per_cpu(current_task, cpu) = c_idle.idle;
779 #ifdef CONFIG_X86_32
780 	/* Stack for startup_32 can be just as for start_secondary onwards */
781 	irq_ctx_init(cpu);
782 #else
783 	clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
784 	initial_gs = per_cpu_offset(cpu);
785 	per_cpu(kernel_stack, cpu) =
786 		(unsigned long)task_stack_page(c_idle.idle) -
787 		KERNEL_STACK_OFFSET + THREAD_SIZE;
788 #endif
789 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
790 	initial_code = (unsigned long)start_secondary;
791 	stack_start.sp = (void *) c_idle.idle->thread.sp;
792 
793 	/* start_ip had better be page-aligned! */
794 	start_ip = setup_trampoline();
795 
796 	/* So we see what's up */
797 	announce_cpu(cpu, apicid);
798 
799 	/*
800 	 * This grunge runs the startup process for
801 	 * the targeted processor.
802 	 */
803 
804 	atomic_set(&init_deasserted, 0);
805 
806 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
807 
808 		pr_debug("Setting warm reset code and vector.\n");
809 
810 		smpboot_setup_warm_reset_vector(start_ip);
811 		/*
812 		 * Be paranoid about clearing APIC errors.
813 		*/
814 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
815 			apic_write(APIC_ESR, 0);
816 			apic_read(APIC_ESR);
817 		}
818 	}
819 
820 	/*
821 	 * Kick the secondary CPU. Use the method in the APIC driver
822 	 * if it's defined - or use an INIT boot APIC message otherwise:
823 	 */
824 	if (apic->wakeup_secondary_cpu)
825 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
826 	else
827 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
828 
829 	if (!boot_error) {
830 		/*
831 		 * allow APs to start initializing.
832 		 */
833 		pr_debug("Before Callout %d.\n", cpu);
834 		cpumask_set_cpu(cpu, cpu_callout_mask);
835 		pr_debug("After Callout %d.\n", cpu);
836 
837 		/*
838 		 * Wait 5s total for a response
839 		 */
840 		for (timeout = 0; timeout < 50000; timeout++) {
841 			if (cpumask_test_cpu(cpu, cpu_callin_mask))
842 				break;	/* It has booted */
843 			udelay(100);
844 			/*
845 			 * Allow other tasks to run while we wait for the
846 			 * AP to come online. This also gives a chance
847 			 * for the MTRR work(triggered by the AP coming online)
848 			 * to be completed in the stop machine context.
849 			 */
850 			schedule();
851 		}
852 
853 		if (cpumask_test_cpu(cpu, cpu_callin_mask))
854 			pr_debug("CPU%d: has booted.\n", cpu);
855 		else {
856 			boot_error = 1;
857 			if (*((volatile unsigned char *)trampoline_base)
858 					== 0xA5)
859 				/* trampoline started but...? */
860 				pr_err("CPU%d: Stuck ??\n", cpu);
861 			else
862 				/* trampoline code not run */
863 				pr_err("CPU%d: Not responding.\n", cpu);
864 			if (apic->inquire_remote_apic)
865 				apic->inquire_remote_apic(apicid);
866 		}
867 	}
868 
869 	if (boot_error) {
870 		/* Try to put things back the way they were before ... */
871 		numa_remove_cpu(cpu); /* was set by numa_add_cpu */
872 
873 		/* was set by do_boot_cpu() */
874 		cpumask_clear_cpu(cpu, cpu_callout_mask);
875 
876 		/* was set by cpu_init() */
877 		cpumask_clear_cpu(cpu, cpu_initialized_mask);
878 
879 		set_cpu_present(cpu, false);
880 		per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
881 	}
882 
883 	/* mark "stuck" area as not stuck */
884 	*((volatile unsigned long *)trampoline_base) = 0;
885 
886 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
887 		/*
888 		 * Cleanup possible dangling ends...
889 		 */
890 		smpboot_restore_warm_reset_vector();
891 	}
892 
893 	destroy_work_on_stack(&c_idle.work);
894 	return boot_error;
895 }
896 
897 int __cpuinit native_cpu_up(unsigned int cpu)
898 {
899 	int apicid = apic->cpu_present_to_apicid(cpu);
900 	unsigned long flags;
901 	int err;
902 
903 	WARN_ON(irqs_disabled());
904 
905 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
906 
907 	if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
908 	    !physid_isset(apicid, phys_cpu_present_map)) {
909 		printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
910 		return -EINVAL;
911 	}
912 
913 	/*
914 	 * Already booted CPU?
915 	 */
916 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
917 		pr_debug("do_boot_cpu %d Already started\n", cpu);
918 		return -ENOSYS;
919 	}
920 
921 	/*
922 	 * Save current MTRR state in case it was changed since early boot
923 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
924 	 */
925 	mtrr_save_state();
926 
927 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
928 
929 	err = do_boot_cpu(apicid, cpu);
930 	if (err) {
931 		pr_debug("do_boot_cpu failed %d\n", err);
932 		return -EIO;
933 	}
934 
935 	/*
936 	 * Check TSC synchronization with the AP (keep irqs disabled
937 	 * while doing so):
938 	 */
939 	local_irq_save(flags);
940 	check_tsc_sync_source(cpu);
941 	local_irq_restore(flags);
942 
943 	while (!cpu_online(cpu)) {
944 		cpu_relax();
945 		touch_nmi_watchdog();
946 	}
947 
948 	return 0;
949 }
950 
951 /*
952  * Fall back to non SMP mode after errors.
953  *
954  * RED-PEN audit/test this more. I bet there is more state messed up here.
955  */
956 static __init void disable_smp(void)
957 {
958 	init_cpu_present(cpumask_of(0));
959 	init_cpu_possible(cpumask_of(0));
960 	smpboot_clear_io_apic_irqs();
961 
962 	if (smp_found_config)
963 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
964 	else
965 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
966 	map_cpu_to_logical_apicid();
967 	cpumask_set_cpu(0, cpu_sibling_mask(0));
968 	cpumask_set_cpu(0, cpu_core_mask(0));
969 }
970 
971 /*
972  * Various sanity checks.
973  */
974 static int __init smp_sanity_check(unsigned max_cpus)
975 {
976 	preempt_disable();
977 
978 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
979 	if (def_to_bigsmp && nr_cpu_ids > 8) {
980 		unsigned int cpu;
981 		unsigned nr;
982 
983 		printk(KERN_WARNING
984 		       "More than 8 CPUs detected - skipping them.\n"
985 		       "Use CONFIG_X86_BIGSMP.\n");
986 
987 		nr = 0;
988 		for_each_present_cpu(cpu) {
989 			if (nr >= 8)
990 				set_cpu_present(cpu, false);
991 			nr++;
992 		}
993 
994 		nr = 0;
995 		for_each_possible_cpu(cpu) {
996 			if (nr >= 8)
997 				set_cpu_possible(cpu, false);
998 			nr++;
999 		}
1000 
1001 		nr_cpu_ids = 8;
1002 	}
1003 #endif
1004 
1005 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1006 		printk(KERN_WARNING
1007 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
1008 			hard_smp_processor_id());
1009 
1010 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1011 	}
1012 
1013 	/*
1014 	 * If we couldn't find an SMP configuration at boot time,
1015 	 * get out of here now!
1016 	 */
1017 	if (!smp_found_config && !acpi_lapic) {
1018 		preempt_enable();
1019 		printk(KERN_NOTICE "SMP motherboard not detected.\n");
1020 		disable_smp();
1021 		if (APIC_init_uniprocessor())
1022 			printk(KERN_NOTICE "Local APIC not detected."
1023 					   " Using dummy APIC emulation.\n");
1024 		return -1;
1025 	}
1026 
1027 	/*
1028 	 * Should not be necessary because the MP table should list the boot
1029 	 * CPU too, but we do it for the sake of robustness anyway.
1030 	 */
1031 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1032 		printk(KERN_NOTICE
1033 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
1034 			boot_cpu_physical_apicid);
1035 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1036 	}
1037 	preempt_enable();
1038 
1039 	/*
1040 	 * If we couldn't find a local APIC, then get out of here now!
1041 	 */
1042 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1043 	    !cpu_has_apic) {
1044 		if (!disable_apic) {
1045 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
1046 				boot_cpu_physical_apicid);
1047 			pr_err("... forcing use of dummy APIC emulation."
1048 				"(tell your hw vendor)\n");
1049 		}
1050 		smpboot_clear_io_apic();
1051 		arch_disable_smp_support();
1052 		return -1;
1053 	}
1054 
1055 	verify_local_APIC();
1056 
1057 	/*
1058 	 * If SMP should be disabled, then really disable it!
1059 	 */
1060 	if (!max_cpus) {
1061 		printk(KERN_INFO "SMP mode deactivated.\n");
1062 		smpboot_clear_io_apic();
1063 
1064 		localise_nmi_watchdog();
1065 
1066 		connect_bsp_APIC();
1067 		setup_local_APIC();
1068 		end_local_APIC_setup();
1069 		return -1;
1070 	}
1071 
1072 	return 0;
1073 }
1074 
1075 static void __init smp_cpu_index_default(void)
1076 {
1077 	int i;
1078 	struct cpuinfo_x86 *c;
1079 
1080 	for_each_possible_cpu(i) {
1081 		c = &cpu_data(i);
1082 		/* mark all to hotplug */
1083 		c->cpu_index = nr_cpu_ids;
1084 	}
1085 }
1086 
1087 /*
1088  * Prepare for SMP bootup.  The MP table or ACPI has been read
1089  * earlier.  Just do some sanity checking here and enable APIC mode.
1090  */
1091 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1092 {
1093 	unsigned int i;
1094 
1095 	preempt_disable();
1096 	smp_cpu_index_default();
1097 	current_cpu_data = boot_cpu_data;
1098 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1099 	mb();
1100 	/*
1101 	 * Setup boot CPU information
1102 	 */
1103 	smp_store_cpu_info(0); /* Final full version of the data */
1104 #ifdef CONFIG_X86_32
1105 	boot_cpu_logical_apicid = logical_smp_processor_id();
1106 #endif
1107 	current_thread_info()->cpu = 0;  /* needed? */
1108 	for_each_possible_cpu(i) {
1109 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1110 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1111 		zalloc_cpumask_var(&cpu_data(i).llc_shared_map, GFP_KERNEL);
1112 	}
1113 	set_cpu_sibling_map(0);
1114 
1115 
1116 	if (smp_sanity_check(max_cpus) < 0) {
1117 		printk(KERN_INFO "SMP disabled\n");
1118 		disable_smp();
1119 		goto out;
1120 	}
1121 
1122 	default_setup_apic_routing();
1123 
1124 	preempt_disable();
1125 	if (read_apic_id() != boot_cpu_physical_apicid) {
1126 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1127 		     read_apic_id(), boot_cpu_physical_apicid);
1128 		/* Or can we switch back to PIC here? */
1129 	}
1130 	preempt_enable();
1131 
1132 	connect_bsp_APIC();
1133 
1134 	/*
1135 	 * Switch from PIC to APIC mode.
1136 	 */
1137 	setup_local_APIC();
1138 
1139 	/*
1140 	 * Enable IO APIC before setting up error vector
1141 	 */
1142 	if (!skip_ioapic_setup && nr_ioapics)
1143 		enable_IO_APIC();
1144 
1145 	end_local_APIC_setup();
1146 
1147 	map_cpu_to_logical_apicid();
1148 
1149 	if (apic->setup_portio_remap)
1150 		apic->setup_portio_remap();
1151 
1152 	smpboot_setup_io_apic();
1153 	/*
1154 	 * Set up local APIC timer on boot CPU.
1155 	 */
1156 
1157 	printk(KERN_INFO "CPU%d: ", 0);
1158 	print_cpu_info(&cpu_data(0));
1159 	x86_init.timers.setup_percpu_clockev();
1160 
1161 	if (is_uv_system())
1162 		uv_system_init();
1163 
1164 	set_mtrr_aps_delayed_init();
1165 out:
1166 	preempt_enable();
1167 }
1168 
1169 void arch_enable_nonboot_cpus_begin(void)
1170 {
1171 	set_mtrr_aps_delayed_init();
1172 }
1173 
1174 void arch_enable_nonboot_cpus_end(void)
1175 {
1176 	mtrr_aps_init();
1177 }
1178 
1179 /*
1180  * Early setup to make printk work.
1181  */
1182 void __init native_smp_prepare_boot_cpu(void)
1183 {
1184 	int me = smp_processor_id();
1185 	switch_to_new_gdt(me);
1186 	/* already set me in cpu_online_mask in boot_cpu_init() */
1187 	cpumask_set_cpu(me, cpu_callout_mask);
1188 	per_cpu(cpu_state, me) = CPU_ONLINE;
1189 }
1190 
1191 void __init native_smp_cpus_done(unsigned int max_cpus)
1192 {
1193 	pr_debug("Boot done.\n");
1194 
1195 	impress_friends();
1196 #ifdef CONFIG_X86_IO_APIC
1197 	setup_ioapic_dest();
1198 #endif
1199 	check_nmi_watchdog();
1200 	mtrr_aps_init();
1201 }
1202 
1203 static int __initdata setup_possible_cpus = -1;
1204 static int __init _setup_possible_cpus(char *str)
1205 {
1206 	get_option(&str, &setup_possible_cpus);
1207 	return 0;
1208 }
1209 early_param("possible_cpus", _setup_possible_cpus);
1210 
1211 
1212 /*
1213  * cpu_possible_mask should be static, it cannot change as cpu's
1214  * are onlined, or offlined. The reason is per-cpu data-structures
1215  * are allocated by some modules at init time, and dont expect to
1216  * do this dynamically on cpu arrival/departure.
1217  * cpu_present_mask on the other hand can change dynamically.
1218  * In case when cpu_hotplug is not compiled, then we resort to current
1219  * behaviour, which is cpu_possible == cpu_present.
1220  * - Ashok Raj
1221  *
1222  * Three ways to find out the number of additional hotplug CPUs:
1223  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1224  * - The user can overwrite it with possible_cpus=NUM
1225  * - Otherwise don't reserve additional CPUs.
1226  * We do this because additional CPUs waste a lot of memory.
1227  * -AK
1228  */
1229 __init void prefill_possible_map(void)
1230 {
1231 	int i, possible;
1232 
1233 	/* no processor from mptable or madt */
1234 	if (!num_processors)
1235 		num_processors = 1;
1236 
1237 	i = setup_max_cpus ?: 1;
1238 	if (setup_possible_cpus == -1) {
1239 		possible = num_processors;
1240 #ifdef CONFIG_HOTPLUG_CPU
1241 		if (setup_max_cpus)
1242 			possible += disabled_cpus;
1243 #else
1244 		if (possible > i)
1245 			possible = i;
1246 #endif
1247 	} else
1248 		possible = setup_possible_cpus;
1249 
1250 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1251 
1252 	/* nr_cpu_ids could be reduced via nr_cpus= */
1253 	if (possible > nr_cpu_ids) {
1254 		printk(KERN_WARNING
1255 			"%d Processors exceeds NR_CPUS limit of %d\n",
1256 			possible, nr_cpu_ids);
1257 		possible = nr_cpu_ids;
1258 	}
1259 
1260 #ifdef CONFIG_HOTPLUG_CPU
1261 	if (!setup_max_cpus)
1262 #endif
1263 	if (possible > i) {
1264 		printk(KERN_WARNING
1265 			"%d Processors exceeds max_cpus limit of %u\n",
1266 			possible, setup_max_cpus);
1267 		possible = i;
1268 	}
1269 
1270 	printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1271 		possible, max_t(int, possible - num_processors, 0));
1272 
1273 	for (i = 0; i < possible; i++)
1274 		set_cpu_possible(i, true);
1275 	for (; i < NR_CPUS; i++)
1276 		set_cpu_possible(i, false);
1277 
1278 	nr_cpu_ids = possible;
1279 }
1280 
1281 #ifdef CONFIG_HOTPLUG_CPU
1282 
1283 static void remove_siblinginfo(int cpu)
1284 {
1285 	int sibling;
1286 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1287 
1288 	for_each_cpu(sibling, cpu_core_mask(cpu)) {
1289 		cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1290 		/*/
1291 		 * last thread sibling in this cpu core going down
1292 		 */
1293 		if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1294 			cpu_data(sibling).booted_cores--;
1295 	}
1296 
1297 	for_each_cpu(sibling, cpu_sibling_mask(cpu))
1298 		cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1299 	cpumask_clear(cpu_sibling_mask(cpu));
1300 	cpumask_clear(cpu_core_mask(cpu));
1301 	c->phys_proc_id = 0;
1302 	c->cpu_core_id = 0;
1303 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1304 }
1305 
1306 static void __ref remove_cpu_from_maps(int cpu)
1307 {
1308 	set_cpu_online(cpu, false);
1309 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1310 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1311 	/* was set by cpu_init() */
1312 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1313 	numa_remove_cpu(cpu);
1314 }
1315 
1316 void cpu_disable_common(void)
1317 {
1318 	int cpu = smp_processor_id();
1319 
1320 	remove_siblinginfo(cpu);
1321 
1322 	/* It's now safe to remove this processor from the online map */
1323 	lock_vector_lock();
1324 	remove_cpu_from_maps(cpu);
1325 	unlock_vector_lock();
1326 	fixup_irqs();
1327 }
1328 
1329 int native_cpu_disable(void)
1330 {
1331 	int cpu = smp_processor_id();
1332 
1333 	/*
1334 	 * Perhaps use cpufreq to drop frequency, but that could go
1335 	 * into generic code.
1336 	 *
1337 	 * We won't take down the boot processor on i386 due to some
1338 	 * interrupts only being able to be serviced by the BSP.
1339 	 * Especially so if we're not using an IOAPIC	-zwane
1340 	 */
1341 	if (cpu == 0)
1342 		return -EBUSY;
1343 
1344 	if (nmi_watchdog == NMI_LOCAL_APIC)
1345 		stop_apic_nmi_watchdog(NULL);
1346 	clear_local_APIC();
1347 
1348 	cpu_disable_common();
1349 	return 0;
1350 }
1351 
1352 void native_cpu_die(unsigned int cpu)
1353 {
1354 	/* We don't do anything here: idle task is faking death itself. */
1355 	unsigned int i;
1356 
1357 	for (i = 0; i < 10; i++) {
1358 		/* They ack this in play_dead by setting CPU_DEAD */
1359 		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1360 			if (system_state == SYSTEM_RUNNING)
1361 				pr_info("CPU %u is now offline\n", cpu);
1362 
1363 			if (1 == num_online_cpus())
1364 				alternatives_smp_switch(0);
1365 			return;
1366 		}
1367 		msleep(100);
1368 	}
1369 	pr_err("CPU %u didn't die...\n", cpu);
1370 }
1371 
1372 void play_dead_common(void)
1373 {
1374 	idle_task_exit();
1375 	reset_lazy_tlbstate();
1376 	irq_ctx_exit(raw_smp_processor_id());
1377 	c1e_remove_cpu(raw_smp_processor_id());
1378 
1379 	mb();
1380 	/* Ack it */
1381 	__get_cpu_var(cpu_state) = CPU_DEAD;
1382 
1383 	/*
1384 	 * With physical CPU hotplug, we should halt the cpu
1385 	 */
1386 	local_irq_disable();
1387 }
1388 
1389 /*
1390  * We need to flush the caches before going to sleep, lest we have
1391  * dirty data in our caches when we come back up.
1392  */
1393 static inline void mwait_play_dead(void)
1394 {
1395 	unsigned int eax, ebx, ecx, edx;
1396 	unsigned int highest_cstate = 0;
1397 	unsigned int highest_subcstate = 0;
1398 	int i;
1399 	void *mwait_ptr;
1400 
1401 	if (!cpu_has(&current_cpu_data, X86_FEATURE_MWAIT))
1402 		return;
1403 	if (!cpu_has(&current_cpu_data, X86_FEATURE_CLFLSH))
1404 		return;
1405 	if (current_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
1406 		return;
1407 
1408 	eax = CPUID_MWAIT_LEAF;
1409 	ecx = 0;
1410 	native_cpuid(&eax, &ebx, &ecx, &edx);
1411 
1412 	/*
1413 	 * eax will be 0 if EDX enumeration is not valid.
1414 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1415 	 */
1416 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1417 		eax = 0;
1418 	} else {
1419 		edx >>= MWAIT_SUBSTATE_SIZE;
1420 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1421 			if (edx & MWAIT_SUBSTATE_MASK) {
1422 				highest_cstate = i;
1423 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1424 			}
1425 		}
1426 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1427 			(highest_subcstate - 1);
1428 	}
1429 
1430 	/*
1431 	 * This should be a memory location in a cache line which is
1432 	 * unlikely to be touched by other processors.  The actual
1433 	 * content is immaterial as it is not actually modified in any way.
1434 	 */
1435 	mwait_ptr = &current_thread_info()->flags;
1436 
1437 	wbinvd();
1438 
1439 	while (1) {
1440 		/*
1441 		 * The CLFLUSH is a workaround for erratum AAI65 for
1442 		 * the Xeon 7400 series.  It's not clear it is actually
1443 		 * needed, but it should be harmless in either case.
1444 		 * The WBINVD is insufficient due to the spurious-wakeup
1445 		 * case where we return around the loop.
1446 		 */
1447 		clflush(mwait_ptr);
1448 		__monitor(mwait_ptr, 0, 0);
1449 		mb();
1450 		__mwait(eax, 0);
1451 	}
1452 }
1453 
1454 static inline void hlt_play_dead(void)
1455 {
1456 	if (current_cpu_data.x86 >= 4)
1457 		wbinvd();
1458 
1459 	while (1) {
1460 		native_halt();
1461 	}
1462 }
1463 
1464 void native_play_dead(void)
1465 {
1466 	play_dead_common();
1467 	tboot_shutdown(TB_SHUTDOWN_WFS);
1468 
1469 	mwait_play_dead();	/* Only returns on failure */
1470 	hlt_play_dead();
1471 }
1472 
1473 #else /* ... !CONFIG_HOTPLUG_CPU */
1474 int native_cpu_disable(void)
1475 {
1476 	return -ENOSYS;
1477 }
1478 
1479 void native_cpu_die(unsigned int cpu)
1480 {
1481 	/* We said "no" in __cpu_disable */
1482 	BUG();
1483 }
1484 
1485 void native_play_dead(void)
1486 {
1487 	BUG();
1488 }
1489 
1490 #endif
1491