xref: /linux/arch/x86/kernel/smpboot.c (revision c54ea4918c2b7722d7242ea53271356501988a9b)
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 /* State of each CPU */
76 DEFINE_PER_CPU(int, cpu_state) = { 0 };
77 
78 /* Store all idle threads, this can be reused instead of creating
79 * a new thread. Also avoids complicated thread destroy functionality
80 * for idle threads.
81 */
82 #ifdef CONFIG_HOTPLUG_CPU
83 /*
84  * Needed only for CONFIG_HOTPLUG_CPU because __cpuinitdata is
85  * removed after init for !CONFIG_HOTPLUG_CPU.
86  */
87 static DEFINE_PER_CPU(struct task_struct *, idle_thread_array);
88 #define get_idle_for_cpu(x)      (per_cpu(idle_thread_array, x))
89 #define set_idle_for_cpu(x, p)   (per_cpu(idle_thread_array, x) = (p))
90 
91 /*
92  * We need this for trampoline_base protection from concurrent accesses when
93  * off- and onlining cores wildly.
94  */
95 static DEFINE_MUTEX(x86_cpu_hotplug_driver_mutex);
96 
97 void cpu_hotplug_driver_lock(void)
98 {
99         mutex_lock(&x86_cpu_hotplug_driver_mutex);
100 }
101 
102 void cpu_hotplug_driver_unlock(void)
103 {
104         mutex_unlock(&x86_cpu_hotplug_driver_mutex);
105 }
106 
107 ssize_t arch_cpu_probe(const char *buf, size_t count) { return -1; }
108 ssize_t arch_cpu_release(const char *buf, size_t count) { return -1; }
109 #else
110 static struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
111 #define get_idle_for_cpu(x)      (idle_thread_array[(x)])
112 #define set_idle_for_cpu(x, p)   (idle_thread_array[(x)] = (p))
113 #endif
114 
115 /* Number of siblings per CPU package */
116 int smp_num_siblings = 1;
117 EXPORT_SYMBOL(smp_num_siblings);
118 
119 /* Last level cache ID of each logical CPU */
120 DEFINE_PER_CPU(u16, cpu_llc_id) = BAD_APICID;
121 
122 /* representing HT siblings of each logical CPU */
123 DEFINE_PER_CPU(cpumask_var_t, cpu_sibling_map);
124 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
125 
126 /* representing HT and core siblings of each logical CPU */
127 DEFINE_PER_CPU(cpumask_var_t, cpu_core_map);
128 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
129 
130 DEFINE_PER_CPU(cpumask_var_t, cpu_llc_shared_map);
131 
132 /* Per CPU bogomips and other parameters */
133 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
134 EXPORT_PER_CPU_SYMBOL(cpu_info);
135 
136 atomic_t init_deasserted;
137 
138 /*
139  * Report back to the Boot Processor.
140  * Running on AP.
141  */
142 static void __cpuinit smp_callin(void)
143 {
144 	int cpuid, phys_id;
145 	unsigned long timeout;
146 
147 	/*
148 	 * If waken up by an INIT in an 82489DX configuration
149 	 * we may get here before an INIT-deassert IPI reaches
150 	 * our local APIC.  We have to wait for the IPI or we'll
151 	 * lock up on an APIC access.
152 	 */
153 	if (apic->wait_for_init_deassert)
154 		apic->wait_for_init_deassert(&init_deasserted);
155 
156 	/*
157 	 * (This works even if the APIC is not enabled.)
158 	 */
159 	phys_id = read_apic_id();
160 	cpuid = smp_processor_id();
161 	if (cpumask_test_cpu(cpuid, cpu_callin_mask)) {
162 		panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
163 					phys_id, cpuid);
164 	}
165 	pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
166 
167 	/*
168 	 * STARTUP IPIs are fragile beasts as they might sometimes
169 	 * trigger some glue motherboard logic. Complete APIC bus
170 	 * silence for 1 second, this overestimates the time the
171 	 * boot CPU is spending to send the up to 2 STARTUP IPIs
172 	 * by a factor of two. This should be enough.
173 	 */
174 
175 	/*
176 	 * Waiting 2s total for startup (udelay is not yet working)
177 	 */
178 	timeout = jiffies + 2*HZ;
179 	while (time_before(jiffies, timeout)) {
180 		/*
181 		 * Has the boot CPU finished it's STARTUP sequence?
182 		 */
183 		if (cpumask_test_cpu(cpuid, cpu_callout_mask))
184 			break;
185 		cpu_relax();
186 	}
187 
188 	if (!time_before(jiffies, timeout)) {
189 		panic("%s: CPU%d started up but did not get a callout!\n",
190 		      __func__, cpuid);
191 	}
192 
193 	/*
194 	 * the boot CPU has finished the init stage and is spinning
195 	 * on callin_map until we finish. We are free to set up this
196 	 * CPU, first the APIC. (this is probably redundant on most
197 	 * boards)
198 	 */
199 
200 	pr_debug("CALLIN, before setup_local_APIC().\n");
201 	if (apic->smp_callin_clear_local_apic)
202 		apic->smp_callin_clear_local_apic();
203 	setup_local_APIC();
204 	end_local_APIC_setup();
205 
206 	/*
207 	 * Need to setup vector mappings before we enable interrupts.
208 	 */
209 	setup_vector_irq(smp_processor_id());
210 	/*
211 	 * Get our bogomips.
212 	 *
213 	 * Need to enable IRQs because it can take longer and then
214 	 * the NMI watchdog might kill us.
215 	 */
216 	local_irq_enable();
217 	calibrate_delay();
218 	local_irq_disable();
219 	pr_debug("Stack at about %p\n", &cpuid);
220 
221 	/*
222 	 * Save our processor parameters
223 	 */
224 	smp_store_cpu_info(cpuid);
225 
226 	/*
227 	 * This must be done before setting cpu_online_mask
228 	 * or calling notify_cpu_starting.
229 	 */
230 	set_cpu_sibling_map(raw_smp_processor_id());
231 	wmb();
232 
233 	notify_cpu_starting(cpuid);
234 
235 	/*
236 	 * Allow the master to continue.
237 	 */
238 	cpumask_set_cpu(cpuid, cpu_callin_mask);
239 }
240 
241 /*
242  * Activate a secondary processor.
243  */
244 notrace static void __cpuinit start_secondary(void *unused)
245 {
246 	/*
247 	 * Don't put *anything* before cpu_init(), SMP booting is too
248 	 * fragile that we want to limit the things done here to the
249 	 * most necessary things.
250 	 */
251 	cpu_init();
252 	preempt_disable();
253 	smp_callin();
254 
255 #ifdef CONFIG_X86_32
256 	/* switch away from the initial page table */
257 	load_cr3(swapper_pg_dir);
258 	__flush_tlb_all();
259 #endif
260 
261 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
262 	barrier();
263 	/*
264 	 * Check TSC synchronization with the BP:
265 	 */
266 	check_tsc_sync_target();
267 
268 	/*
269 	 * We need to hold call_lock, so there is no inconsistency
270 	 * between the time smp_call_function() determines number of
271 	 * IPI recipients, and the time when the determination is made
272 	 * for which cpus receive the IPI. Holding this
273 	 * lock helps us to not include this cpu in a currently in progress
274 	 * smp_call_function().
275 	 *
276 	 * We need to hold vector_lock so there the set of online cpus
277 	 * does not change while we are assigning vectors to cpus.  Holding
278 	 * this lock ensures we don't half assign or remove an irq from a cpu.
279 	 */
280 	ipi_call_lock();
281 	lock_vector_lock();
282 	set_cpu_online(smp_processor_id(), true);
283 	unlock_vector_lock();
284 	ipi_call_unlock();
285 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
286 	x86_platform.nmi_init();
287 
288 	/* enable local interrupts */
289 	local_irq_enable();
290 
291 	/* to prevent fake stack check failure in clock setup */
292 	boot_init_stack_canary();
293 
294 	x86_cpuinit.setup_percpu_clockev();
295 
296 	wmb();
297 	cpu_idle();
298 }
299 
300 /*
301  * The bootstrap kernel entry code has set these up. Save them for
302  * a given CPU
303  */
304 
305 void __cpuinit smp_store_cpu_info(int id)
306 {
307 	struct cpuinfo_x86 *c = &cpu_data(id);
308 
309 	*c = boot_cpu_data;
310 	c->cpu_index = id;
311 	if (id != 0)
312 		identify_secondary_cpu(c);
313 }
314 
315 static void __cpuinit link_thread_siblings(int cpu1, int cpu2)
316 {
317 	cpumask_set_cpu(cpu1, cpu_sibling_mask(cpu2));
318 	cpumask_set_cpu(cpu2, cpu_sibling_mask(cpu1));
319 	cpumask_set_cpu(cpu1, cpu_core_mask(cpu2));
320 	cpumask_set_cpu(cpu2, cpu_core_mask(cpu1));
321 	cpumask_set_cpu(cpu1, cpu_llc_shared_mask(cpu2));
322 	cpumask_set_cpu(cpu2, cpu_llc_shared_mask(cpu1));
323 }
324 
325 
326 void __cpuinit set_cpu_sibling_map(int cpu)
327 {
328 	int i;
329 	struct cpuinfo_x86 *c = &cpu_data(cpu);
330 
331 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
332 
333 	if (smp_num_siblings > 1) {
334 		for_each_cpu(i, cpu_sibling_setup_mask) {
335 			struct cpuinfo_x86 *o = &cpu_data(i);
336 
337 			if (cpu_has(c, X86_FEATURE_TOPOEXT)) {
338 				if (c->phys_proc_id == o->phys_proc_id &&
339 				    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i) &&
340 				    c->compute_unit_id == o->compute_unit_id)
341 					link_thread_siblings(cpu, i);
342 			} else if (c->phys_proc_id == o->phys_proc_id &&
343 				   c->cpu_core_id == o->cpu_core_id) {
344 				link_thread_siblings(cpu, i);
345 			}
346 		}
347 	} else {
348 		cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
349 	}
350 
351 	cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
352 
353 	if (__this_cpu_read(cpu_info.x86_max_cores) == 1) {
354 		cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
355 		c->booted_cores = 1;
356 		return;
357 	}
358 
359 	for_each_cpu(i, cpu_sibling_setup_mask) {
360 		if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
361 		    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
362 			cpumask_set_cpu(i, cpu_llc_shared_mask(cpu));
363 			cpumask_set_cpu(cpu, cpu_llc_shared_mask(i));
364 		}
365 		if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
366 			cpumask_set_cpu(i, cpu_core_mask(cpu));
367 			cpumask_set_cpu(cpu, cpu_core_mask(i));
368 			/*
369 			 *  Does this new cpu bringup a new core?
370 			 */
371 			if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
372 				/*
373 				 * for each core in package, increment
374 				 * the booted_cores for this new cpu
375 				 */
376 				if (cpumask_first(cpu_sibling_mask(i)) == i)
377 					c->booted_cores++;
378 				/*
379 				 * increment the core count for all
380 				 * the other cpus in this package
381 				 */
382 				if (i != cpu)
383 					cpu_data(i).booted_cores++;
384 			} else if (i != cpu && !c->booted_cores)
385 				c->booted_cores = cpu_data(i).booted_cores;
386 		}
387 	}
388 }
389 
390 /* maps the cpu to the sched domain representing multi-core */
391 const struct cpumask *cpu_coregroup_mask(int cpu)
392 {
393 	struct cpuinfo_x86 *c = &cpu_data(cpu);
394 	/*
395 	 * For perf, we return last level cache shared map.
396 	 * And for power savings, we return cpu_core_map
397 	 */
398 	if ((sched_mc_power_savings || sched_smt_power_savings) &&
399 	    !(cpu_has(c, X86_FEATURE_AMD_DCM)))
400 		return cpu_core_mask(cpu);
401 	else
402 		return cpu_llc_shared_mask(cpu);
403 }
404 
405 static void impress_friends(void)
406 {
407 	int cpu;
408 	unsigned long bogosum = 0;
409 	/*
410 	 * Allow the user to impress friends.
411 	 */
412 	pr_debug("Before bogomips.\n");
413 	for_each_possible_cpu(cpu)
414 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
415 			bogosum += cpu_data(cpu).loops_per_jiffy;
416 	printk(KERN_INFO
417 		"Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
418 		num_online_cpus(),
419 		bogosum/(500000/HZ),
420 		(bogosum/(5000/HZ))%100);
421 
422 	pr_debug("Before bogocount - setting activated=1.\n");
423 }
424 
425 void __inquire_remote_apic(int apicid)
426 {
427 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
428 	char *names[] = { "ID", "VERSION", "SPIV" };
429 	int timeout;
430 	u32 status;
431 
432 	printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
433 
434 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
435 		printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
436 
437 		/*
438 		 * Wait for idle.
439 		 */
440 		status = safe_apic_wait_icr_idle();
441 		if (status)
442 			printk(KERN_CONT
443 			       "a previous APIC delivery may have failed\n");
444 
445 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
446 
447 		timeout = 0;
448 		do {
449 			udelay(100);
450 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
451 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
452 
453 		switch (status) {
454 		case APIC_ICR_RR_VALID:
455 			status = apic_read(APIC_RRR);
456 			printk(KERN_CONT "%08x\n", status);
457 			break;
458 		default:
459 			printk(KERN_CONT "failed\n");
460 		}
461 	}
462 }
463 
464 /*
465  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
466  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
467  * won't ... remember to clear down the APIC, etc later.
468  */
469 int __cpuinit
470 wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
471 {
472 	unsigned long send_status, accept_status = 0;
473 	int maxlvt;
474 
475 	/* Target chip */
476 	/* Boot on the stack */
477 	/* Kick the second */
478 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, logical_apicid);
479 
480 	pr_debug("Waiting for send to finish...\n");
481 	send_status = safe_apic_wait_icr_idle();
482 
483 	/*
484 	 * Give the other CPU some time to accept the IPI.
485 	 */
486 	udelay(200);
487 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
488 		maxlvt = lapic_get_maxlvt();
489 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
490 			apic_write(APIC_ESR, 0);
491 		accept_status = (apic_read(APIC_ESR) & 0xEF);
492 	}
493 	pr_debug("NMI sent.\n");
494 
495 	if (send_status)
496 		printk(KERN_ERR "APIC never delivered???\n");
497 	if (accept_status)
498 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
499 
500 	return (send_status | accept_status);
501 }
502 
503 static int __cpuinit
504 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
505 {
506 	unsigned long send_status, accept_status = 0;
507 	int maxlvt, num_starts, j;
508 
509 	maxlvt = lapic_get_maxlvt();
510 
511 	/*
512 	 * Be paranoid about clearing APIC errors.
513 	 */
514 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
515 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
516 			apic_write(APIC_ESR, 0);
517 		apic_read(APIC_ESR);
518 	}
519 
520 	pr_debug("Asserting INIT.\n");
521 
522 	/*
523 	 * Turn INIT on target chip
524 	 */
525 	/*
526 	 * Send IPI
527 	 */
528 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
529 		       phys_apicid);
530 
531 	pr_debug("Waiting for send to finish...\n");
532 	send_status = safe_apic_wait_icr_idle();
533 
534 	mdelay(10);
535 
536 	pr_debug("Deasserting INIT.\n");
537 
538 	/* Target chip */
539 	/* Send IPI */
540 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
541 
542 	pr_debug("Waiting for send to finish...\n");
543 	send_status = safe_apic_wait_icr_idle();
544 
545 	mb();
546 	atomic_set(&init_deasserted, 1);
547 
548 	/*
549 	 * Should we send STARTUP IPIs ?
550 	 *
551 	 * Determine this based on the APIC version.
552 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
553 	 */
554 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
555 		num_starts = 2;
556 	else
557 		num_starts = 0;
558 
559 	/*
560 	 * Paravirt / VMI wants a startup IPI hook here to set up the
561 	 * target processor state.
562 	 */
563 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
564 			 stack_start);
565 
566 	/*
567 	 * Run STARTUP IPI loop.
568 	 */
569 	pr_debug("#startup loops: %d.\n", num_starts);
570 
571 	for (j = 1; j <= num_starts; j++) {
572 		pr_debug("Sending STARTUP #%d.\n", j);
573 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
574 			apic_write(APIC_ESR, 0);
575 		apic_read(APIC_ESR);
576 		pr_debug("After apic_write.\n");
577 
578 		/*
579 		 * STARTUP IPI
580 		 */
581 
582 		/* Target chip */
583 		/* Boot on the stack */
584 		/* Kick the second */
585 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
586 			       phys_apicid);
587 
588 		/*
589 		 * Give the other CPU some time to accept the IPI.
590 		 */
591 		udelay(300);
592 
593 		pr_debug("Startup point 1.\n");
594 
595 		pr_debug("Waiting for send to finish...\n");
596 		send_status = safe_apic_wait_icr_idle();
597 
598 		/*
599 		 * Give the other CPU some time to accept the IPI.
600 		 */
601 		udelay(200);
602 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
603 			apic_write(APIC_ESR, 0);
604 		accept_status = (apic_read(APIC_ESR) & 0xEF);
605 		if (send_status || accept_status)
606 			break;
607 	}
608 	pr_debug("After Startup.\n");
609 
610 	if (send_status)
611 		printk(KERN_ERR "APIC never delivered???\n");
612 	if (accept_status)
613 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
614 
615 	return (send_status | accept_status);
616 }
617 
618 struct create_idle {
619 	struct work_struct work;
620 	struct task_struct *idle;
621 	struct completion done;
622 	int cpu;
623 };
624 
625 static void __cpuinit do_fork_idle(struct work_struct *work)
626 {
627 	struct create_idle *c_idle =
628 		container_of(work, struct create_idle, work);
629 
630 	c_idle->idle = fork_idle(c_idle->cpu);
631 	complete(&c_idle->done);
632 }
633 
634 /* reduce the number of lines printed when booting a large cpu count system */
635 static void __cpuinit announce_cpu(int cpu, int apicid)
636 {
637 	static int current_node = -1;
638 	int node = early_cpu_to_node(cpu);
639 
640 	if (system_state == SYSTEM_BOOTING) {
641 		if (node != current_node) {
642 			if (current_node > (-1))
643 				pr_cont(" Ok.\n");
644 			current_node = node;
645 			pr_info("Booting Node %3d, Processors ", node);
646 		}
647 		pr_cont(" #%d%s", cpu, cpu == (nr_cpu_ids - 1) ? " Ok.\n" : "");
648 		return;
649 	} else
650 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
651 			node, cpu, apicid);
652 }
653 
654 /*
655  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
656  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
657  * Returns zero if CPU booted OK, else error code from
658  * ->wakeup_secondary_cpu.
659  */
660 static int __cpuinit do_boot_cpu(int apicid, int cpu)
661 {
662 	unsigned long boot_error = 0;
663 	unsigned long start_ip;
664 	int timeout;
665 	struct create_idle c_idle = {
666 		.cpu	= cpu,
667 		.done	= COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
668 	};
669 
670 	INIT_WORK_ONSTACK(&c_idle.work, do_fork_idle);
671 
672 	alternatives_smp_switch(1);
673 
674 	c_idle.idle = get_idle_for_cpu(cpu);
675 
676 	/*
677 	 * We can't use kernel_thread since we must avoid to
678 	 * reschedule the child.
679 	 */
680 	if (c_idle.idle) {
681 		c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
682 			(THREAD_SIZE +  task_stack_page(c_idle.idle))) - 1);
683 		init_idle(c_idle.idle, cpu);
684 		goto do_rest;
685 	}
686 
687 	schedule_work(&c_idle.work);
688 	wait_for_completion(&c_idle.done);
689 
690 	if (IS_ERR(c_idle.idle)) {
691 		printk("failed fork for CPU %d\n", cpu);
692 		destroy_work_on_stack(&c_idle.work);
693 		return PTR_ERR(c_idle.idle);
694 	}
695 
696 	set_idle_for_cpu(cpu, c_idle.idle);
697 do_rest:
698 	per_cpu(current_task, cpu) = c_idle.idle;
699 #ifdef CONFIG_X86_32
700 	/* Stack for startup_32 can be just as for start_secondary onwards */
701 	irq_ctx_init(cpu);
702 #else
703 	clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
704 	initial_gs = per_cpu_offset(cpu);
705 	per_cpu(kernel_stack, cpu) =
706 		(unsigned long)task_stack_page(c_idle.idle) -
707 		KERNEL_STACK_OFFSET + THREAD_SIZE;
708 #endif
709 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
710 	initial_code = (unsigned long)start_secondary;
711 	stack_start  = c_idle.idle->thread.sp;
712 
713 	/* start_ip had better be page-aligned! */
714 	start_ip = trampoline_address();
715 
716 	/* So we see what's up */
717 	announce_cpu(cpu, apicid);
718 
719 	/*
720 	 * This grunge runs the startup process for
721 	 * the targeted processor.
722 	 */
723 
724 	printk(KERN_DEBUG "smpboot cpu %d: start_ip = %lx\n", cpu, start_ip);
725 
726 	atomic_set(&init_deasserted, 0);
727 
728 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
729 
730 		pr_debug("Setting warm reset code and vector.\n");
731 
732 		smpboot_setup_warm_reset_vector(start_ip);
733 		/*
734 		 * Be paranoid about clearing APIC errors.
735 		*/
736 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
737 			apic_write(APIC_ESR, 0);
738 			apic_read(APIC_ESR);
739 		}
740 	}
741 
742 	/*
743 	 * Kick the secondary CPU. Use the method in the APIC driver
744 	 * if it's defined - or use an INIT boot APIC message otherwise:
745 	 */
746 	if (apic->wakeup_secondary_cpu)
747 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
748 	else
749 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
750 
751 	if (!boot_error) {
752 		/*
753 		 * allow APs to start initializing.
754 		 */
755 		pr_debug("Before Callout %d.\n", cpu);
756 		cpumask_set_cpu(cpu, cpu_callout_mask);
757 		pr_debug("After Callout %d.\n", cpu);
758 
759 		/*
760 		 * Wait 5s total for a response
761 		 */
762 		for (timeout = 0; timeout < 50000; timeout++) {
763 			if (cpumask_test_cpu(cpu, cpu_callin_mask))
764 				break;	/* It has booted */
765 			udelay(100);
766 			/*
767 			 * Allow other tasks to run while we wait for the
768 			 * AP to come online. This also gives a chance
769 			 * for the MTRR work(triggered by the AP coming online)
770 			 * to be completed in the stop machine context.
771 			 */
772 			schedule();
773 		}
774 
775 		if (cpumask_test_cpu(cpu, cpu_callin_mask))
776 			pr_debug("CPU%d: has booted.\n", cpu);
777 		else {
778 			boot_error = 1;
779 			if (*(volatile u32 *)TRAMPOLINE_SYM(trampoline_status)
780 			    == 0xA5A5A5A5)
781 				/* trampoline started but...? */
782 				pr_err("CPU%d: Stuck ??\n", cpu);
783 			else
784 				/* trampoline code not run */
785 				pr_err("CPU%d: Not responding.\n", cpu);
786 			if (apic->inquire_remote_apic)
787 				apic->inquire_remote_apic(apicid);
788 		}
789 	}
790 
791 	if (boot_error) {
792 		/* Try to put things back the way they were before ... */
793 		numa_remove_cpu(cpu); /* was set by numa_add_cpu */
794 
795 		/* was set by do_boot_cpu() */
796 		cpumask_clear_cpu(cpu, cpu_callout_mask);
797 
798 		/* was set by cpu_init() */
799 		cpumask_clear_cpu(cpu, cpu_initialized_mask);
800 
801 		set_cpu_present(cpu, false);
802 		per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
803 	}
804 
805 	/* mark "stuck" area as not stuck */
806 	*(volatile u32 *)TRAMPOLINE_SYM(trampoline_status) = 0;
807 
808 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
809 		/*
810 		 * Cleanup possible dangling ends...
811 		 */
812 		smpboot_restore_warm_reset_vector();
813 	}
814 
815 	destroy_work_on_stack(&c_idle.work);
816 	return boot_error;
817 }
818 
819 int __cpuinit native_cpu_up(unsigned int cpu)
820 {
821 	int apicid = apic->cpu_present_to_apicid(cpu);
822 	unsigned long flags;
823 	int err;
824 
825 	WARN_ON(irqs_disabled());
826 
827 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
828 
829 	if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
830 	    !physid_isset(apicid, phys_cpu_present_map)) {
831 		printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
832 		return -EINVAL;
833 	}
834 
835 	/*
836 	 * Already booted CPU?
837 	 */
838 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
839 		pr_debug("do_boot_cpu %d Already started\n", cpu);
840 		return -ENOSYS;
841 	}
842 
843 	/*
844 	 * Save current MTRR state in case it was changed since early boot
845 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
846 	 */
847 	mtrr_save_state();
848 
849 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
850 
851 	err = do_boot_cpu(apicid, cpu);
852 	if (err) {
853 		pr_debug("do_boot_cpu failed %d\n", err);
854 		return -EIO;
855 	}
856 
857 	/*
858 	 * Check TSC synchronization with the AP (keep irqs disabled
859 	 * while doing so):
860 	 */
861 	local_irq_save(flags);
862 	check_tsc_sync_source(cpu);
863 	local_irq_restore(flags);
864 
865 	while (!cpu_online(cpu)) {
866 		cpu_relax();
867 		touch_nmi_watchdog();
868 	}
869 
870 	return 0;
871 }
872 
873 /**
874  * arch_disable_smp_support() - disables SMP support for x86 at runtime
875  */
876 void arch_disable_smp_support(void)
877 {
878 	disable_ioapic_support();
879 }
880 
881 /*
882  * Fall back to non SMP mode after errors.
883  *
884  * RED-PEN audit/test this more. I bet there is more state messed up here.
885  */
886 static __init void disable_smp(void)
887 {
888 	init_cpu_present(cpumask_of(0));
889 	init_cpu_possible(cpumask_of(0));
890 	smpboot_clear_io_apic_irqs();
891 
892 	if (smp_found_config)
893 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
894 	else
895 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
896 	cpumask_set_cpu(0, cpu_sibling_mask(0));
897 	cpumask_set_cpu(0, cpu_core_mask(0));
898 }
899 
900 /*
901  * Various sanity checks.
902  */
903 static int __init smp_sanity_check(unsigned max_cpus)
904 {
905 	preempt_disable();
906 
907 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
908 	if (def_to_bigsmp && nr_cpu_ids > 8) {
909 		unsigned int cpu;
910 		unsigned nr;
911 
912 		printk(KERN_WARNING
913 		       "More than 8 CPUs detected - skipping them.\n"
914 		       "Use CONFIG_X86_BIGSMP.\n");
915 
916 		nr = 0;
917 		for_each_present_cpu(cpu) {
918 			if (nr >= 8)
919 				set_cpu_present(cpu, false);
920 			nr++;
921 		}
922 
923 		nr = 0;
924 		for_each_possible_cpu(cpu) {
925 			if (nr >= 8)
926 				set_cpu_possible(cpu, false);
927 			nr++;
928 		}
929 
930 		nr_cpu_ids = 8;
931 	}
932 #endif
933 
934 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
935 		printk(KERN_WARNING
936 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
937 			hard_smp_processor_id());
938 
939 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
940 	}
941 
942 	/*
943 	 * If we couldn't find an SMP configuration at boot time,
944 	 * get out of here now!
945 	 */
946 	if (!smp_found_config && !acpi_lapic) {
947 		preempt_enable();
948 		printk(KERN_NOTICE "SMP motherboard not detected.\n");
949 		disable_smp();
950 		if (APIC_init_uniprocessor())
951 			printk(KERN_NOTICE "Local APIC not detected."
952 					   " Using dummy APIC emulation.\n");
953 		return -1;
954 	}
955 
956 	/*
957 	 * Should not be necessary because the MP table should list the boot
958 	 * CPU too, but we do it for the sake of robustness anyway.
959 	 */
960 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
961 		printk(KERN_NOTICE
962 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
963 			boot_cpu_physical_apicid);
964 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
965 	}
966 	preempt_enable();
967 
968 	/*
969 	 * If we couldn't find a local APIC, then get out of here now!
970 	 */
971 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
972 	    !cpu_has_apic) {
973 		if (!disable_apic) {
974 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
975 				boot_cpu_physical_apicid);
976 			pr_err("... forcing use of dummy APIC emulation."
977 				"(tell your hw vendor)\n");
978 		}
979 		smpboot_clear_io_apic();
980 		disable_ioapic_support();
981 		return -1;
982 	}
983 
984 	verify_local_APIC();
985 
986 	/*
987 	 * If SMP should be disabled, then really disable it!
988 	 */
989 	if (!max_cpus) {
990 		printk(KERN_INFO "SMP mode deactivated.\n");
991 		smpboot_clear_io_apic();
992 
993 		connect_bsp_APIC();
994 		setup_local_APIC();
995 		bsp_end_local_APIC_setup();
996 		return -1;
997 	}
998 
999 	return 0;
1000 }
1001 
1002 static void __init smp_cpu_index_default(void)
1003 {
1004 	int i;
1005 	struct cpuinfo_x86 *c;
1006 
1007 	for_each_possible_cpu(i) {
1008 		c = &cpu_data(i);
1009 		/* mark all to hotplug */
1010 		c->cpu_index = nr_cpu_ids;
1011 	}
1012 }
1013 
1014 /*
1015  * Prepare for SMP bootup.  The MP table or ACPI has been read
1016  * earlier.  Just do some sanity checking here and enable APIC mode.
1017  */
1018 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1019 {
1020 	unsigned int i;
1021 
1022 	preempt_disable();
1023 	smp_cpu_index_default();
1024 
1025 	/*
1026 	 * Setup boot CPU information
1027 	 */
1028 	smp_store_cpu_info(0); /* Final full version of the data */
1029 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1030 	mb();
1031 
1032 	current_thread_info()->cpu = 0;  /* needed? */
1033 	for_each_possible_cpu(i) {
1034 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1035 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1036 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1037 	}
1038 	set_cpu_sibling_map(0);
1039 
1040 
1041 	if (smp_sanity_check(max_cpus) < 0) {
1042 		printk(KERN_INFO "SMP disabled\n");
1043 		disable_smp();
1044 		goto out;
1045 	}
1046 
1047 	default_setup_apic_routing();
1048 
1049 	preempt_disable();
1050 	if (read_apic_id() != boot_cpu_physical_apicid) {
1051 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1052 		     read_apic_id(), boot_cpu_physical_apicid);
1053 		/* Or can we switch back to PIC here? */
1054 	}
1055 	preempt_enable();
1056 
1057 	connect_bsp_APIC();
1058 
1059 	/*
1060 	 * Switch from PIC to APIC mode.
1061 	 */
1062 	setup_local_APIC();
1063 
1064 	/*
1065 	 * Enable IO APIC before setting up error vector
1066 	 */
1067 	if (!skip_ioapic_setup && nr_ioapics)
1068 		enable_IO_APIC();
1069 
1070 	bsp_end_local_APIC_setup();
1071 
1072 	if (apic->setup_portio_remap)
1073 		apic->setup_portio_remap();
1074 
1075 	smpboot_setup_io_apic();
1076 	/*
1077 	 * Set up local APIC timer on boot CPU.
1078 	 */
1079 
1080 	printk(KERN_INFO "CPU%d: ", 0);
1081 	print_cpu_info(&cpu_data(0));
1082 	x86_init.timers.setup_percpu_clockev();
1083 
1084 	if (is_uv_system())
1085 		uv_system_init();
1086 
1087 	set_mtrr_aps_delayed_init();
1088 out:
1089 	preempt_enable();
1090 }
1091 
1092 void arch_disable_nonboot_cpus_begin(void)
1093 {
1094 	/*
1095 	 * Avoid the smp alternatives switch during the disable_nonboot_cpus().
1096 	 * In the suspend path, we will be back in the SMP mode shortly anyways.
1097 	 */
1098 	skip_smp_alternatives = true;
1099 }
1100 
1101 void arch_disable_nonboot_cpus_end(void)
1102 {
1103 	skip_smp_alternatives = false;
1104 }
1105 
1106 void arch_enable_nonboot_cpus_begin(void)
1107 {
1108 	set_mtrr_aps_delayed_init();
1109 }
1110 
1111 void arch_enable_nonboot_cpus_end(void)
1112 {
1113 	mtrr_aps_init();
1114 }
1115 
1116 /*
1117  * Early setup to make printk work.
1118  */
1119 void __init native_smp_prepare_boot_cpu(void)
1120 {
1121 	int me = smp_processor_id();
1122 	switch_to_new_gdt(me);
1123 	/* already set me in cpu_online_mask in boot_cpu_init() */
1124 	cpumask_set_cpu(me, cpu_callout_mask);
1125 	per_cpu(cpu_state, me) = CPU_ONLINE;
1126 }
1127 
1128 void __init native_smp_cpus_done(unsigned int max_cpus)
1129 {
1130 	pr_debug("Boot done.\n");
1131 
1132 	impress_friends();
1133 #ifdef CONFIG_X86_IO_APIC
1134 	setup_ioapic_dest();
1135 #endif
1136 	mtrr_aps_init();
1137 }
1138 
1139 static int __initdata setup_possible_cpus = -1;
1140 static int __init _setup_possible_cpus(char *str)
1141 {
1142 	get_option(&str, &setup_possible_cpus);
1143 	return 0;
1144 }
1145 early_param("possible_cpus", _setup_possible_cpus);
1146 
1147 
1148 /*
1149  * cpu_possible_mask should be static, it cannot change as cpu's
1150  * are onlined, or offlined. The reason is per-cpu data-structures
1151  * are allocated by some modules at init time, and dont expect to
1152  * do this dynamically on cpu arrival/departure.
1153  * cpu_present_mask on the other hand can change dynamically.
1154  * In case when cpu_hotplug is not compiled, then we resort to current
1155  * behaviour, which is cpu_possible == cpu_present.
1156  * - Ashok Raj
1157  *
1158  * Three ways to find out the number of additional hotplug CPUs:
1159  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1160  * - The user can overwrite it with possible_cpus=NUM
1161  * - Otherwise don't reserve additional CPUs.
1162  * We do this because additional CPUs waste a lot of memory.
1163  * -AK
1164  */
1165 __init void prefill_possible_map(void)
1166 {
1167 	int i, possible;
1168 
1169 	/* no processor from mptable or madt */
1170 	if (!num_processors)
1171 		num_processors = 1;
1172 
1173 	i = setup_max_cpus ?: 1;
1174 	if (setup_possible_cpus == -1) {
1175 		possible = num_processors;
1176 #ifdef CONFIG_HOTPLUG_CPU
1177 		if (setup_max_cpus)
1178 			possible += disabled_cpus;
1179 #else
1180 		if (possible > i)
1181 			possible = i;
1182 #endif
1183 	} else
1184 		possible = setup_possible_cpus;
1185 
1186 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1187 
1188 	/* nr_cpu_ids could be reduced via nr_cpus= */
1189 	if (possible > nr_cpu_ids) {
1190 		printk(KERN_WARNING
1191 			"%d Processors exceeds NR_CPUS limit of %d\n",
1192 			possible, nr_cpu_ids);
1193 		possible = nr_cpu_ids;
1194 	}
1195 
1196 #ifdef CONFIG_HOTPLUG_CPU
1197 	if (!setup_max_cpus)
1198 #endif
1199 	if (possible > i) {
1200 		printk(KERN_WARNING
1201 			"%d Processors exceeds max_cpus limit of %u\n",
1202 			possible, setup_max_cpus);
1203 		possible = i;
1204 	}
1205 
1206 	printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1207 		possible, max_t(int, possible - num_processors, 0));
1208 
1209 	for (i = 0; i < possible; i++)
1210 		set_cpu_possible(i, true);
1211 	for (; i < NR_CPUS; i++)
1212 		set_cpu_possible(i, false);
1213 
1214 	nr_cpu_ids = possible;
1215 }
1216 
1217 #ifdef CONFIG_HOTPLUG_CPU
1218 
1219 static void remove_siblinginfo(int cpu)
1220 {
1221 	int sibling;
1222 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1223 
1224 	for_each_cpu(sibling, cpu_core_mask(cpu)) {
1225 		cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1226 		/*/
1227 		 * last thread sibling in this cpu core going down
1228 		 */
1229 		if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1230 			cpu_data(sibling).booted_cores--;
1231 	}
1232 
1233 	for_each_cpu(sibling, cpu_sibling_mask(cpu))
1234 		cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1235 	cpumask_clear(cpu_sibling_mask(cpu));
1236 	cpumask_clear(cpu_core_mask(cpu));
1237 	c->phys_proc_id = 0;
1238 	c->cpu_core_id = 0;
1239 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1240 }
1241 
1242 static void __ref remove_cpu_from_maps(int cpu)
1243 {
1244 	set_cpu_online(cpu, false);
1245 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1246 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1247 	/* was set by cpu_init() */
1248 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1249 	numa_remove_cpu(cpu);
1250 }
1251 
1252 void cpu_disable_common(void)
1253 {
1254 	int cpu = smp_processor_id();
1255 
1256 	remove_siblinginfo(cpu);
1257 
1258 	/* It's now safe to remove this processor from the online map */
1259 	lock_vector_lock();
1260 	remove_cpu_from_maps(cpu);
1261 	unlock_vector_lock();
1262 	fixup_irqs();
1263 }
1264 
1265 int native_cpu_disable(void)
1266 {
1267 	int cpu = smp_processor_id();
1268 
1269 	/*
1270 	 * Perhaps use cpufreq to drop frequency, but that could go
1271 	 * into generic code.
1272 	 *
1273 	 * We won't take down the boot processor on i386 due to some
1274 	 * interrupts only being able to be serviced by the BSP.
1275 	 * Especially so if we're not using an IOAPIC	-zwane
1276 	 */
1277 	if (cpu == 0)
1278 		return -EBUSY;
1279 
1280 	clear_local_APIC();
1281 
1282 	cpu_disable_common();
1283 	return 0;
1284 }
1285 
1286 void native_cpu_die(unsigned int cpu)
1287 {
1288 	/* We don't do anything here: idle task is faking death itself. */
1289 	unsigned int i;
1290 
1291 	for (i = 0; i < 10; i++) {
1292 		/* They ack this in play_dead by setting CPU_DEAD */
1293 		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1294 			if (system_state == SYSTEM_RUNNING)
1295 				pr_info("CPU %u is now offline\n", cpu);
1296 
1297 			if (1 == num_online_cpus())
1298 				alternatives_smp_switch(0);
1299 			return;
1300 		}
1301 		msleep(100);
1302 	}
1303 	pr_err("CPU %u didn't die...\n", cpu);
1304 }
1305 
1306 void play_dead_common(void)
1307 {
1308 	idle_task_exit();
1309 	reset_lazy_tlbstate();
1310 	c1e_remove_cpu(raw_smp_processor_id());
1311 
1312 	mb();
1313 	/* Ack it */
1314 	__this_cpu_write(cpu_state, CPU_DEAD);
1315 
1316 	/*
1317 	 * With physical CPU hotplug, we should halt the cpu
1318 	 */
1319 	local_irq_disable();
1320 }
1321 
1322 /*
1323  * We need to flush the caches before going to sleep, lest we have
1324  * dirty data in our caches when we come back up.
1325  */
1326 static inline void mwait_play_dead(void)
1327 {
1328 	unsigned int eax, ebx, ecx, edx;
1329 	unsigned int highest_cstate = 0;
1330 	unsigned int highest_subcstate = 0;
1331 	int i;
1332 	void *mwait_ptr;
1333 	struct cpuinfo_x86 *c = __this_cpu_ptr(&cpu_info);
1334 
1335 	if (!(cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)))
1336 		return;
1337 	if (!cpu_has(__this_cpu_ptr(&cpu_info), X86_FEATURE_CLFLSH))
1338 		return;
1339 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1340 		return;
1341 
1342 	eax = CPUID_MWAIT_LEAF;
1343 	ecx = 0;
1344 	native_cpuid(&eax, &ebx, &ecx, &edx);
1345 
1346 	/*
1347 	 * eax will be 0 if EDX enumeration is not valid.
1348 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1349 	 */
1350 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1351 		eax = 0;
1352 	} else {
1353 		edx >>= MWAIT_SUBSTATE_SIZE;
1354 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1355 			if (edx & MWAIT_SUBSTATE_MASK) {
1356 				highest_cstate = i;
1357 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1358 			}
1359 		}
1360 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1361 			(highest_subcstate - 1);
1362 	}
1363 
1364 	/*
1365 	 * This should be a memory location in a cache line which is
1366 	 * unlikely to be touched by other processors.  The actual
1367 	 * content is immaterial as it is not actually modified in any way.
1368 	 */
1369 	mwait_ptr = &current_thread_info()->flags;
1370 
1371 	wbinvd();
1372 
1373 	while (1) {
1374 		/*
1375 		 * The CLFLUSH is a workaround for erratum AAI65 for
1376 		 * the Xeon 7400 series.  It's not clear it is actually
1377 		 * needed, but it should be harmless in either case.
1378 		 * The WBINVD is insufficient due to the spurious-wakeup
1379 		 * case where we return around the loop.
1380 		 */
1381 		clflush(mwait_ptr);
1382 		__monitor(mwait_ptr, 0, 0);
1383 		mb();
1384 		__mwait(eax, 0);
1385 	}
1386 }
1387 
1388 static inline void hlt_play_dead(void)
1389 {
1390 	if (__this_cpu_read(cpu_info.x86) >= 4)
1391 		wbinvd();
1392 
1393 	while (1) {
1394 		native_halt();
1395 	}
1396 }
1397 
1398 void native_play_dead(void)
1399 {
1400 	play_dead_common();
1401 	tboot_shutdown(TB_SHUTDOWN_WFS);
1402 
1403 	mwait_play_dead();	/* Only returns on failure */
1404 	hlt_play_dead();
1405 }
1406 
1407 #else /* ... !CONFIG_HOTPLUG_CPU */
1408 int native_cpu_disable(void)
1409 {
1410 	return -ENOSYS;
1411 }
1412 
1413 void native_cpu_die(unsigned int cpu)
1414 {
1415 	/* We said "no" in __cpu_disable */
1416 	BUG();
1417 }
1418 
1419 void native_play_dead(void)
1420 {
1421 	BUG();
1422 }
1423 
1424 #endif
1425