xref: /linux/arch/x86/kernel/smpboot.c (revision 9ffc93f203c18a70623f21950f1dd473c9ec48cd)
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 	/*
212 	 * Save our processor parameters. Note: this information
213 	 * is needed for clock calibration.
214 	 */
215 	smp_store_cpu_info(cpuid);
216 
217 	/*
218 	 * Get our bogomips.
219 	 * Update loops_per_jiffy in cpu_data. Previous call to
220 	 * smp_store_cpu_info() stored a value that is close but not as
221 	 * accurate as the value just calculated.
222 	 *
223 	 * Need to enable IRQs because it can take longer and then
224 	 * the NMI watchdog might kill us.
225 	 */
226 	local_irq_enable();
227 	calibrate_delay();
228 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
229 	local_irq_disable();
230 	pr_debug("Stack at about %p\n", &cpuid);
231 
232 	/*
233 	 * This must be done before setting cpu_online_mask
234 	 * or calling notify_cpu_starting.
235 	 */
236 	set_cpu_sibling_map(raw_smp_processor_id());
237 	wmb();
238 
239 	notify_cpu_starting(cpuid);
240 
241 	/*
242 	 * Allow the master to continue.
243 	 */
244 	cpumask_set_cpu(cpuid, cpu_callin_mask);
245 }
246 
247 /*
248  * Activate a secondary processor.
249  */
250 notrace static void __cpuinit start_secondary(void *unused)
251 {
252 	/*
253 	 * Don't put *anything* before cpu_init(), SMP booting is too
254 	 * fragile that we want to limit the things done here to the
255 	 * most necessary things.
256 	 */
257 	cpu_init();
258 	preempt_disable();
259 	smp_callin();
260 
261 #ifdef CONFIG_X86_32
262 	/* switch away from the initial page table */
263 	load_cr3(swapper_pg_dir);
264 	__flush_tlb_all();
265 #endif
266 
267 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
268 	barrier();
269 	/*
270 	 * Check TSC synchronization with the BP:
271 	 */
272 	check_tsc_sync_target();
273 
274 	/*
275 	 * We need to hold call_lock, so there is no inconsistency
276 	 * between the time smp_call_function() determines number of
277 	 * IPI recipients, and the time when the determination is made
278 	 * for which cpus receive the IPI. Holding this
279 	 * lock helps us to not include this cpu in a currently in progress
280 	 * smp_call_function().
281 	 *
282 	 * We need to hold vector_lock so there the set of online cpus
283 	 * does not change while we are assigning vectors to cpus.  Holding
284 	 * this lock ensures we don't half assign or remove an irq from a cpu.
285 	 */
286 	ipi_call_lock();
287 	lock_vector_lock();
288 	set_cpu_online(smp_processor_id(), true);
289 	unlock_vector_lock();
290 	ipi_call_unlock();
291 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
292 	x86_platform.nmi_init();
293 
294 	/* enable local interrupts */
295 	local_irq_enable();
296 
297 	/* to prevent fake stack check failure in clock setup */
298 	boot_init_stack_canary();
299 
300 	x86_cpuinit.setup_percpu_clockev();
301 
302 	wmb();
303 	cpu_idle();
304 }
305 
306 /*
307  * The bootstrap kernel entry code has set these up. Save them for
308  * a given CPU
309  */
310 
311 void __cpuinit smp_store_cpu_info(int id)
312 {
313 	struct cpuinfo_x86 *c = &cpu_data(id);
314 
315 	*c = boot_cpu_data;
316 	c->cpu_index = id;
317 	if (id != 0)
318 		identify_secondary_cpu(c);
319 }
320 
321 static void __cpuinit link_thread_siblings(int cpu1, int cpu2)
322 {
323 	cpumask_set_cpu(cpu1, cpu_sibling_mask(cpu2));
324 	cpumask_set_cpu(cpu2, cpu_sibling_mask(cpu1));
325 	cpumask_set_cpu(cpu1, cpu_core_mask(cpu2));
326 	cpumask_set_cpu(cpu2, cpu_core_mask(cpu1));
327 	cpumask_set_cpu(cpu1, cpu_llc_shared_mask(cpu2));
328 	cpumask_set_cpu(cpu2, cpu_llc_shared_mask(cpu1));
329 }
330 
331 
332 void __cpuinit set_cpu_sibling_map(int cpu)
333 {
334 	int i;
335 	struct cpuinfo_x86 *c = &cpu_data(cpu);
336 
337 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
338 
339 	if (smp_num_siblings > 1) {
340 		for_each_cpu(i, cpu_sibling_setup_mask) {
341 			struct cpuinfo_x86 *o = &cpu_data(i);
342 
343 			if (cpu_has(c, X86_FEATURE_TOPOEXT)) {
344 				if (c->phys_proc_id == o->phys_proc_id &&
345 				    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i) &&
346 				    c->compute_unit_id == o->compute_unit_id)
347 					link_thread_siblings(cpu, i);
348 			} else if (c->phys_proc_id == o->phys_proc_id &&
349 				   c->cpu_core_id == o->cpu_core_id) {
350 				link_thread_siblings(cpu, i);
351 			}
352 		}
353 	} else {
354 		cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
355 	}
356 
357 	cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
358 
359 	if (__this_cpu_read(cpu_info.x86_max_cores) == 1) {
360 		cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
361 		c->booted_cores = 1;
362 		return;
363 	}
364 
365 	for_each_cpu(i, cpu_sibling_setup_mask) {
366 		if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
367 		    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
368 			cpumask_set_cpu(i, cpu_llc_shared_mask(cpu));
369 			cpumask_set_cpu(cpu, cpu_llc_shared_mask(i));
370 		}
371 		if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
372 			cpumask_set_cpu(i, cpu_core_mask(cpu));
373 			cpumask_set_cpu(cpu, cpu_core_mask(i));
374 			/*
375 			 *  Does this new cpu bringup a new core?
376 			 */
377 			if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
378 				/*
379 				 * for each core in package, increment
380 				 * the booted_cores for this new cpu
381 				 */
382 				if (cpumask_first(cpu_sibling_mask(i)) == i)
383 					c->booted_cores++;
384 				/*
385 				 * increment the core count for all
386 				 * the other cpus in this package
387 				 */
388 				if (i != cpu)
389 					cpu_data(i).booted_cores++;
390 			} else if (i != cpu && !c->booted_cores)
391 				c->booted_cores = cpu_data(i).booted_cores;
392 		}
393 	}
394 }
395 
396 /* maps the cpu to the sched domain representing multi-core */
397 const struct cpumask *cpu_coregroup_mask(int cpu)
398 {
399 	struct cpuinfo_x86 *c = &cpu_data(cpu);
400 	/*
401 	 * For perf, we return last level cache shared map.
402 	 * And for power savings, we return cpu_core_map
403 	 */
404 	if ((sched_mc_power_savings || sched_smt_power_savings) &&
405 	    !(cpu_has(c, X86_FEATURE_AMD_DCM)))
406 		return cpu_core_mask(cpu);
407 	else
408 		return cpu_llc_shared_mask(cpu);
409 }
410 
411 static void impress_friends(void)
412 {
413 	int cpu;
414 	unsigned long bogosum = 0;
415 	/*
416 	 * Allow the user to impress friends.
417 	 */
418 	pr_debug("Before bogomips.\n");
419 	for_each_possible_cpu(cpu)
420 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
421 			bogosum += cpu_data(cpu).loops_per_jiffy;
422 	printk(KERN_INFO
423 		"Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
424 		num_online_cpus(),
425 		bogosum/(500000/HZ),
426 		(bogosum/(5000/HZ))%100);
427 
428 	pr_debug("Before bogocount - setting activated=1.\n");
429 }
430 
431 void __inquire_remote_apic(int apicid)
432 {
433 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
434 	const char * const names[] = { "ID", "VERSION", "SPIV" };
435 	int timeout;
436 	u32 status;
437 
438 	printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
439 
440 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
441 		printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
442 
443 		/*
444 		 * Wait for idle.
445 		 */
446 		status = safe_apic_wait_icr_idle();
447 		if (status)
448 			printk(KERN_CONT
449 			       "a previous APIC delivery may have failed\n");
450 
451 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
452 
453 		timeout = 0;
454 		do {
455 			udelay(100);
456 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
457 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
458 
459 		switch (status) {
460 		case APIC_ICR_RR_VALID:
461 			status = apic_read(APIC_RRR);
462 			printk(KERN_CONT "%08x\n", status);
463 			break;
464 		default:
465 			printk(KERN_CONT "failed\n");
466 		}
467 	}
468 }
469 
470 /*
471  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
472  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
473  * won't ... remember to clear down the APIC, etc later.
474  */
475 int __cpuinit
476 wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
477 {
478 	unsigned long send_status, accept_status = 0;
479 	int maxlvt;
480 
481 	/* Target chip */
482 	/* Boot on the stack */
483 	/* Kick the second */
484 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, logical_apicid);
485 
486 	pr_debug("Waiting for send to finish...\n");
487 	send_status = safe_apic_wait_icr_idle();
488 
489 	/*
490 	 * Give the other CPU some time to accept the IPI.
491 	 */
492 	udelay(200);
493 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
494 		maxlvt = lapic_get_maxlvt();
495 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
496 			apic_write(APIC_ESR, 0);
497 		accept_status = (apic_read(APIC_ESR) & 0xEF);
498 	}
499 	pr_debug("NMI sent.\n");
500 
501 	if (send_status)
502 		printk(KERN_ERR "APIC never delivered???\n");
503 	if (accept_status)
504 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
505 
506 	return (send_status | accept_status);
507 }
508 
509 static int __cpuinit
510 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
511 {
512 	unsigned long send_status, accept_status = 0;
513 	int maxlvt, num_starts, j;
514 
515 	maxlvt = lapic_get_maxlvt();
516 
517 	/*
518 	 * Be paranoid about clearing APIC errors.
519 	 */
520 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
521 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
522 			apic_write(APIC_ESR, 0);
523 		apic_read(APIC_ESR);
524 	}
525 
526 	pr_debug("Asserting INIT.\n");
527 
528 	/*
529 	 * Turn INIT on target chip
530 	 */
531 	/*
532 	 * Send IPI
533 	 */
534 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
535 		       phys_apicid);
536 
537 	pr_debug("Waiting for send to finish...\n");
538 	send_status = safe_apic_wait_icr_idle();
539 
540 	mdelay(10);
541 
542 	pr_debug("Deasserting INIT.\n");
543 
544 	/* Target chip */
545 	/* Send IPI */
546 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
547 
548 	pr_debug("Waiting for send to finish...\n");
549 	send_status = safe_apic_wait_icr_idle();
550 
551 	mb();
552 	atomic_set(&init_deasserted, 1);
553 
554 	/*
555 	 * Should we send STARTUP IPIs ?
556 	 *
557 	 * Determine this based on the APIC version.
558 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
559 	 */
560 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
561 		num_starts = 2;
562 	else
563 		num_starts = 0;
564 
565 	/*
566 	 * Paravirt / VMI wants a startup IPI hook here to set up the
567 	 * target processor state.
568 	 */
569 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
570 			 stack_start);
571 
572 	/*
573 	 * Run STARTUP IPI loop.
574 	 */
575 	pr_debug("#startup loops: %d.\n", num_starts);
576 
577 	for (j = 1; j <= num_starts; j++) {
578 		pr_debug("Sending STARTUP #%d.\n", j);
579 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
580 			apic_write(APIC_ESR, 0);
581 		apic_read(APIC_ESR);
582 		pr_debug("After apic_write.\n");
583 
584 		/*
585 		 * STARTUP IPI
586 		 */
587 
588 		/* Target chip */
589 		/* Boot on the stack */
590 		/* Kick the second */
591 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
592 			       phys_apicid);
593 
594 		/*
595 		 * Give the other CPU some time to accept the IPI.
596 		 */
597 		udelay(300);
598 
599 		pr_debug("Startup point 1.\n");
600 
601 		pr_debug("Waiting for send to finish...\n");
602 		send_status = safe_apic_wait_icr_idle();
603 
604 		/*
605 		 * Give the other CPU some time to accept the IPI.
606 		 */
607 		udelay(200);
608 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
609 			apic_write(APIC_ESR, 0);
610 		accept_status = (apic_read(APIC_ESR) & 0xEF);
611 		if (send_status || accept_status)
612 			break;
613 	}
614 	pr_debug("After Startup.\n");
615 
616 	if (send_status)
617 		printk(KERN_ERR "APIC never delivered???\n");
618 	if (accept_status)
619 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
620 
621 	return (send_status | accept_status);
622 }
623 
624 struct create_idle {
625 	struct work_struct work;
626 	struct task_struct *idle;
627 	struct completion done;
628 	int cpu;
629 };
630 
631 static void __cpuinit do_fork_idle(struct work_struct *work)
632 {
633 	struct create_idle *c_idle =
634 		container_of(work, struct create_idle, work);
635 
636 	c_idle->idle = fork_idle(c_idle->cpu);
637 	complete(&c_idle->done);
638 }
639 
640 /* reduce the number of lines printed when booting a large cpu count system */
641 static void __cpuinit announce_cpu(int cpu, int apicid)
642 {
643 	static int current_node = -1;
644 	int node = early_cpu_to_node(cpu);
645 
646 	if (system_state == SYSTEM_BOOTING) {
647 		if (node != current_node) {
648 			if (current_node > (-1))
649 				pr_cont(" Ok.\n");
650 			current_node = node;
651 			pr_info("Booting Node %3d, Processors ", node);
652 		}
653 		pr_cont(" #%d%s", cpu, cpu == (nr_cpu_ids - 1) ? " Ok.\n" : "");
654 		return;
655 	} else
656 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
657 			node, cpu, apicid);
658 }
659 
660 /*
661  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
662  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
663  * Returns zero if CPU booted OK, else error code from
664  * ->wakeup_secondary_cpu.
665  */
666 static int __cpuinit do_boot_cpu(int apicid, int cpu)
667 {
668 	unsigned long boot_error = 0;
669 	unsigned long start_ip;
670 	int timeout;
671 	struct create_idle c_idle = {
672 		.cpu	= cpu,
673 		.done	= COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
674 	};
675 
676 	INIT_WORK_ONSTACK(&c_idle.work, do_fork_idle);
677 
678 	alternatives_smp_switch(1);
679 
680 	c_idle.idle = get_idle_for_cpu(cpu);
681 
682 	/*
683 	 * We can't use kernel_thread since we must avoid to
684 	 * reschedule the child.
685 	 */
686 	if (c_idle.idle) {
687 		c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
688 			(THREAD_SIZE +  task_stack_page(c_idle.idle))) - 1);
689 		init_idle(c_idle.idle, cpu);
690 		goto do_rest;
691 	}
692 
693 	schedule_work(&c_idle.work);
694 	wait_for_completion(&c_idle.done);
695 
696 	if (IS_ERR(c_idle.idle)) {
697 		printk("failed fork for CPU %d\n", cpu);
698 		destroy_work_on_stack(&c_idle.work);
699 		return PTR_ERR(c_idle.idle);
700 	}
701 
702 	set_idle_for_cpu(cpu, c_idle.idle);
703 do_rest:
704 	per_cpu(current_task, cpu) = c_idle.idle;
705 #ifdef CONFIG_X86_32
706 	/* Stack for startup_32 can be just as for start_secondary onwards */
707 	irq_ctx_init(cpu);
708 #else
709 	clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
710 	initial_gs = per_cpu_offset(cpu);
711 	per_cpu(kernel_stack, cpu) =
712 		(unsigned long)task_stack_page(c_idle.idle) -
713 		KERNEL_STACK_OFFSET + THREAD_SIZE;
714 #endif
715 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
716 	initial_code = (unsigned long)start_secondary;
717 	stack_start  = c_idle.idle->thread.sp;
718 
719 	/* start_ip had better be page-aligned! */
720 	start_ip = trampoline_address();
721 
722 	/* So we see what's up */
723 	announce_cpu(cpu, apicid);
724 
725 	/*
726 	 * This grunge runs the startup process for
727 	 * the targeted processor.
728 	 */
729 
730 	atomic_set(&init_deasserted, 0);
731 
732 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
733 
734 		pr_debug("Setting warm reset code and vector.\n");
735 
736 		smpboot_setup_warm_reset_vector(start_ip);
737 		/*
738 		 * Be paranoid about clearing APIC errors.
739 		*/
740 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
741 			apic_write(APIC_ESR, 0);
742 			apic_read(APIC_ESR);
743 		}
744 	}
745 
746 	/*
747 	 * Kick the secondary CPU. Use the method in the APIC driver
748 	 * if it's defined - or use an INIT boot APIC message otherwise:
749 	 */
750 	if (apic->wakeup_secondary_cpu)
751 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
752 	else
753 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
754 
755 	if (!boot_error) {
756 		/*
757 		 * allow APs to start initializing.
758 		 */
759 		pr_debug("Before Callout %d.\n", cpu);
760 		cpumask_set_cpu(cpu, cpu_callout_mask);
761 		pr_debug("After Callout %d.\n", cpu);
762 
763 		/*
764 		 * Wait 5s total for a response
765 		 */
766 		for (timeout = 0; timeout < 50000; timeout++) {
767 			if (cpumask_test_cpu(cpu, cpu_callin_mask))
768 				break;	/* It has booted */
769 			udelay(100);
770 			/*
771 			 * Allow other tasks to run while we wait for the
772 			 * AP to come online. This also gives a chance
773 			 * for the MTRR work(triggered by the AP coming online)
774 			 * to be completed in the stop machine context.
775 			 */
776 			schedule();
777 		}
778 
779 		if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
780 			print_cpu_msr(&cpu_data(cpu));
781 			pr_debug("CPU%d: has booted.\n", cpu);
782 		} else {
783 			boot_error = 1;
784 			if (*(volatile u32 *)TRAMPOLINE_SYM(trampoline_status)
785 			    == 0xA5A5A5A5)
786 				/* trampoline started but...? */
787 				pr_err("CPU%d: Stuck ??\n", cpu);
788 			else
789 				/* trampoline code not run */
790 				pr_err("CPU%d: Not responding.\n", cpu);
791 			if (apic->inquire_remote_apic)
792 				apic->inquire_remote_apic(apicid);
793 		}
794 	}
795 
796 	if (boot_error) {
797 		/* Try to put things back the way they were before ... */
798 		numa_remove_cpu(cpu); /* was set by numa_add_cpu */
799 
800 		/* was set by do_boot_cpu() */
801 		cpumask_clear_cpu(cpu, cpu_callout_mask);
802 
803 		/* was set by cpu_init() */
804 		cpumask_clear_cpu(cpu, cpu_initialized_mask);
805 
806 		set_cpu_present(cpu, false);
807 		per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
808 	}
809 
810 	/* mark "stuck" area as not stuck */
811 	*(volatile u32 *)TRAMPOLINE_SYM(trampoline_status) = 0;
812 
813 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
814 		/*
815 		 * Cleanup possible dangling ends...
816 		 */
817 		smpboot_restore_warm_reset_vector();
818 	}
819 
820 	destroy_work_on_stack(&c_idle.work);
821 	return boot_error;
822 }
823 
824 int __cpuinit native_cpu_up(unsigned int cpu)
825 {
826 	int apicid = apic->cpu_present_to_apicid(cpu);
827 	unsigned long flags;
828 	int err;
829 
830 	WARN_ON(irqs_disabled());
831 
832 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
833 
834 	if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
835 	    !physid_isset(apicid, phys_cpu_present_map) ||
836 	    !apic->apic_id_valid(apicid)) {
837 		printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
838 		return -EINVAL;
839 	}
840 
841 	/*
842 	 * Already booted CPU?
843 	 */
844 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
845 		pr_debug("do_boot_cpu %d Already started\n", cpu);
846 		return -ENOSYS;
847 	}
848 
849 	/*
850 	 * Save current MTRR state in case it was changed since early boot
851 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
852 	 */
853 	mtrr_save_state();
854 
855 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
856 
857 	err = do_boot_cpu(apicid, cpu);
858 	if (err) {
859 		pr_debug("do_boot_cpu failed %d\n", err);
860 		return -EIO;
861 	}
862 
863 	/*
864 	 * Check TSC synchronization with the AP (keep irqs disabled
865 	 * while doing so):
866 	 */
867 	local_irq_save(flags);
868 	check_tsc_sync_source(cpu);
869 	local_irq_restore(flags);
870 
871 	while (!cpu_online(cpu)) {
872 		cpu_relax();
873 		touch_nmi_watchdog();
874 	}
875 
876 	return 0;
877 }
878 
879 /**
880  * arch_disable_smp_support() - disables SMP support for x86 at runtime
881  */
882 void arch_disable_smp_support(void)
883 {
884 	disable_ioapic_support();
885 }
886 
887 /*
888  * Fall back to non SMP mode after errors.
889  *
890  * RED-PEN audit/test this more. I bet there is more state messed up here.
891  */
892 static __init void disable_smp(void)
893 {
894 	init_cpu_present(cpumask_of(0));
895 	init_cpu_possible(cpumask_of(0));
896 	smpboot_clear_io_apic_irqs();
897 
898 	if (smp_found_config)
899 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
900 	else
901 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
902 	cpumask_set_cpu(0, cpu_sibling_mask(0));
903 	cpumask_set_cpu(0, cpu_core_mask(0));
904 }
905 
906 /*
907  * Various sanity checks.
908  */
909 static int __init smp_sanity_check(unsigned max_cpus)
910 {
911 	preempt_disable();
912 
913 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
914 	if (def_to_bigsmp && nr_cpu_ids > 8) {
915 		unsigned int cpu;
916 		unsigned nr;
917 
918 		printk(KERN_WARNING
919 		       "More than 8 CPUs detected - skipping them.\n"
920 		       "Use CONFIG_X86_BIGSMP.\n");
921 
922 		nr = 0;
923 		for_each_present_cpu(cpu) {
924 			if (nr >= 8)
925 				set_cpu_present(cpu, false);
926 			nr++;
927 		}
928 
929 		nr = 0;
930 		for_each_possible_cpu(cpu) {
931 			if (nr >= 8)
932 				set_cpu_possible(cpu, false);
933 			nr++;
934 		}
935 
936 		nr_cpu_ids = 8;
937 	}
938 #endif
939 
940 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
941 		printk(KERN_WARNING
942 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
943 			hard_smp_processor_id());
944 
945 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
946 	}
947 
948 	/*
949 	 * If we couldn't find an SMP configuration at boot time,
950 	 * get out of here now!
951 	 */
952 	if (!smp_found_config && !acpi_lapic) {
953 		preempt_enable();
954 		printk(KERN_NOTICE "SMP motherboard not detected.\n");
955 		disable_smp();
956 		if (APIC_init_uniprocessor())
957 			printk(KERN_NOTICE "Local APIC not detected."
958 					   " Using dummy APIC emulation.\n");
959 		return -1;
960 	}
961 
962 	/*
963 	 * Should not be necessary because the MP table should list the boot
964 	 * CPU too, but we do it for the sake of robustness anyway.
965 	 */
966 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
967 		printk(KERN_NOTICE
968 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
969 			boot_cpu_physical_apicid);
970 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
971 	}
972 	preempt_enable();
973 
974 	/*
975 	 * If we couldn't find a local APIC, then get out of here now!
976 	 */
977 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
978 	    !cpu_has_apic) {
979 		if (!disable_apic) {
980 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
981 				boot_cpu_physical_apicid);
982 			pr_err("... forcing use of dummy APIC emulation."
983 				"(tell your hw vendor)\n");
984 		}
985 		smpboot_clear_io_apic();
986 		disable_ioapic_support();
987 		return -1;
988 	}
989 
990 	verify_local_APIC();
991 
992 	/*
993 	 * If SMP should be disabled, then really disable it!
994 	 */
995 	if (!max_cpus) {
996 		printk(KERN_INFO "SMP mode deactivated.\n");
997 		smpboot_clear_io_apic();
998 
999 		connect_bsp_APIC();
1000 		setup_local_APIC();
1001 		bsp_end_local_APIC_setup();
1002 		return -1;
1003 	}
1004 
1005 	return 0;
1006 }
1007 
1008 static void __init smp_cpu_index_default(void)
1009 {
1010 	int i;
1011 	struct cpuinfo_x86 *c;
1012 
1013 	for_each_possible_cpu(i) {
1014 		c = &cpu_data(i);
1015 		/* mark all to hotplug */
1016 		c->cpu_index = nr_cpu_ids;
1017 	}
1018 }
1019 
1020 /*
1021  * Prepare for SMP bootup.  The MP table or ACPI has been read
1022  * earlier.  Just do some sanity checking here and enable APIC mode.
1023  */
1024 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1025 {
1026 	unsigned int i;
1027 
1028 	preempt_disable();
1029 	smp_cpu_index_default();
1030 
1031 	/*
1032 	 * Setup boot CPU information
1033 	 */
1034 	smp_store_cpu_info(0); /* Final full version of the data */
1035 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1036 	mb();
1037 
1038 	current_thread_info()->cpu = 0;  /* needed? */
1039 	for_each_possible_cpu(i) {
1040 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1041 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1042 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1043 	}
1044 	set_cpu_sibling_map(0);
1045 
1046 
1047 	if (smp_sanity_check(max_cpus) < 0) {
1048 		printk(KERN_INFO "SMP disabled\n");
1049 		disable_smp();
1050 		goto out;
1051 	}
1052 
1053 	default_setup_apic_routing();
1054 
1055 	preempt_disable();
1056 	if (read_apic_id() != boot_cpu_physical_apicid) {
1057 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1058 		     read_apic_id(), boot_cpu_physical_apicid);
1059 		/* Or can we switch back to PIC here? */
1060 	}
1061 	preempt_enable();
1062 
1063 	connect_bsp_APIC();
1064 
1065 	/*
1066 	 * Switch from PIC to APIC mode.
1067 	 */
1068 	setup_local_APIC();
1069 
1070 	/*
1071 	 * Enable IO APIC before setting up error vector
1072 	 */
1073 	if (!skip_ioapic_setup && nr_ioapics)
1074 		enable_IO_APIC();
1075 
1076 	bsp_end_local_APIC_setup();
1077 
1078 	if (apic->setup_portio_remap)
1079 		apic->setup_portio_remap();
1080 
1081 	smpboot_setup_io_apic();
1082 	/*
1083 	 * Set up local APIC timer on boot CPU.
1084 	 */
1085 
1086 	printk(KERN_INFO "CPU%d: ", 0);
1087 	print_cpu_info(&cpu_data(0));
1088 	x86_init.timers.setup_percpu_clockev();
1089 
1090 	if (is_uv_system())
1091 		uv_system_init();
1092 
1093 	set_mtrr_aps_delayed_init();
1094 out:
1095 	preempt_enable();
1096 }
1097 
1098 void arch_disable_nonboot_cpus_begin(void)
1099 {
1100 	/*
1101 	 * Avoid the smp alternatives switch during the disable_nonboot_cpus().
1102 	 * In the suspend path, we will be back in the SMP mode shortly anyways.
1103 	 */
1104 	skip_smp_alternatives = true;
1105 }
1106 
1107 void arch_disable_nonboot_cpus_end(void)
1108 {
1109 	skip_smp_alternatives = false;
1110 }
1111 
1112 void arch_enable_nonboot_cpus_begin(void)
1113 {
1114 	set_mtrr_aps_delayed_init();
1115 }
1116 
1117 void arch_enable_nonboot_cpus_end(void)
1118 {
1119 	mtrr_aps_init();
1120 }
1121 
1122 /*
1123  * Early setup to make printk work.
1124  */
1125 void __init native_smp_prepare_boot_cpu(void)
1126 {
1127 	int me = smp_processor_id();
1128 	switch_to_new_gdt(me);
1129 	/* already set me in cpu_online_mask in boot_cpu_init() */
1130 	cpumask_set_cpu(me, cpu_callout_mask);
1131 	per_cpu(cpu_state, me) = CPU_ONLINE;
1132 }
1133 
1134 void __init native_smp_cpus_done(unsigned int max_cpus)
1135 {
1136 	pr_debug("Boot done.\n");
1137 
1138 	nmi_selftest();
1139 	impress_friends();
1140 #ifdef CONFIG_X86_IO_APIC
1141 	setup_ioapic_dest();
1142 #endif
1143 	mtrr_aps_init();
1144 }
1145 
1146 static int __initdata setup_possible_cpus = -1;
1147 static int __init _setup_possible_cpus(char *str)
1148 {
1149 	get_option(&str, &setup_possible_cpus);
1150 	return 0;
1151 }
1152 early_param("possible_cpus", _setup_possible_cpus);
1153 
1154 
1155 /*
1156  * cpu_possible_mask should be static, it cannot change as cpu's
1157  * are onlined, or offlined. The reason is per-cpu data-structures
1158  * are allocated by some modules at init time, and dont expect to
1159  * do this dynamically on cpu arrival/departure.
1160  * cpu_present_mask on the other hand can change dynamically.
1161  * In case when cpu_hotplug is not compiled, then we resort to current
1162  * behaviour, which is cpu_possible == cpu_present.
1163  * - Ashok Raj
1164  *
1165  * Three ways to find out the number of additional hotplug CPUs:
1166  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1167  * - The user can overwrite it with possible_cpus=NUM
1168  * - Otherwise don't reserve additional CPUs.
1169  * We do this because additional CPUs waste a lot of memory.
1170  * -AK
1171  */
1172 __init void prefill_possible_map(void)
1173 {
1174 	int i, possible;
1175 
1176 	/* no processor from mptable or madt */
1177 	if (!num_processors)
1178 		num_processors = 1;
1179 
1180 	i = setup_max_cpus ?: 1;
1181 	if (setup_possible_cpus == -1) {
1182 		possible = num_processors;
1183 #ifdef CONFIG_HOTPLUG_CPU
1184 		if (setup_max_cpus)
1185 			possible += disabled_cpus;
1186 #else
1187 		if (possible > i)
1188 			possible = i;
1189 #endif
1190 	} else
1191 		possible = setup_possible_cpus;
1192 
1193 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1194 
1195 	/* nr_cpu_ids could be reduced via nr_cpus= */
1196 	if (possible > nr_cpu_ids) {
1197 		printk(KERN_WARNING
1198 			"%d Processors exceeds NR_CPUS limit of %d\n",
1199 			possible, nr_cpu_ids);
1200 		possible = nr_cpu_ids;
1201 	}
1202 
1203 #ifdef CONFIG_HOTPLUG_CPU
1204 	if (!setup_max_cpus)
1205 #endif
1206 	if (possible > i) {
1207 		printk(KERN_WARNING
1208 			"%d Processors exceeds max_cpus limit of %u\n",
1209 			possible, setup_max_cpus);
1210 		possible = i;
1211 	}
1212 
1213 	printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1214 		possible, max_t(int, possible - num_processors, 0));
1215 
1216 	for (i = 0; i < possible; i++)
1217 		set_cpu_possible(i, true);
1218 	for (; i < NR_CPUS; i++)
1219 		set_cpu_possible(i, false);
1220 
1221 	nr_cpu_ids = possible;
1222 }
1223 
1224 #ifdef CONFIG_HOTPLUG_CPU
1225 
1226 static void remove_siblinginfo(int cpu)
1227 {
1228 	int sibling;
1229 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1230 
1231 	for_each_cpu(sibling, cpu_core_mask(cpu)) {
1232 		cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1233 		/*/
1234 		 * last thread sibling in this cpu core going down
1235 		 */
1236 		if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1237 			cpu_data(sibling).booted_cores--;
1238 	}
1239 
1240 	for_each_cpu(sibling, cpu_sibling_mask(cpu))
1241 		cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1242 	cpumask_clear(cpu_sibling_mask(cpu));
1243 	cpumask_clear(cpu_core_mask(cpu));
1244 	c->phys_proc_id = 0;
1245 	c->cpu_core_id = 0;
1246 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1247 }
1248 
1249 static void __ref remove_cpu_from_maps(int cpu)
1250 {
1251 	set_cpu_online(cpu, false);
1252 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1253 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1254 	/* was set by cpu_init() */
1255 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1256 	numa_remove_cpu(cpu);
1257 }
1258 
1259 void cpu_disable_common(void)
1260 {
1261 	int cpu = smp_processor_id();
1262 
1263 	remove_siblinginfo(cpu);
1264 
1265 	/* It's now safe to remove this processor from the online map */
1266 	lock_vector_lock();
1267 	remove_cpu_from_maps(cpu);
1268 	unlock_vector_lock();
1269 	fixup_irqs();
1270 }
1271 
1272 int native_cpu_disable(void)
1273 {
1274 	int cpu = smp_processor_id();
1275 
1276 	/*
1277 	 * Perhaps use cpufreq to drop frequency, but that could go
1278 	 * into generic code.
1279 	 *
1280 	 * We won't take down the boot processor on i386 due to some
1281 	 * interrupts only being able to be serviced by the BSP.
1282 	 * Especially so if we're not using an IOAPIC	-zwane
1283 	 */
1284 	if (cpu == 0)
1285 		return -EBUSY;
1286 
1287 	clear_local_APIC();
1288 
1289 	cpu_disable_common();
1290 	return 0;
1291 }
1292 
1293 void native_cpu_die(unsigned int cpu)
1294 {
1295 	/* We don't do anything here: idle task is faking death itself. */
1296 	unsigned int i;
1297 
1298 	for (i = 0; i < 10; i++) {
1299 		/* They ack this in play_dead by setting CPU_DEAD */
1300 		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1301 			if (system_state == SYSTEM_RUNNING)
1302 				pr_info("CPU %u is now offline\n", cpu);
1303 
1304 			if (1 == num_online_cpus())
1305 				alternatives_smp_switch(0);
1306 			return;
1307 		}
1308 		msleep(100);
1309 	}
1310 	pr_err("CPU %u didn't die...\n", cpu);
1311 }
1312 
1313 void play_dead_common(void)
1314 {
1315 	idle_task_exit();
1316 	reset_lazy_tlbstate();
1317 	amd_e400_remove_cpu(raw_smp_processor_id());
1318 
1319 	mb();
1320 	/* Ack it */
1321 	__this_cpu_write(cpu_state, CPU_DEAD);
1322 
1323 	/*
1324 	 * With physical CPU hotplug, we should halt the cpu
1325 	 */
1326 	local_irq_disable();
1327 }
1328 
1329 /*
1330  * We need to flush the caches before going to sleep, lest we have
1331  * dirty data in our caches when we come back up.
1332  */
1333 static inline void mwait_play_dead(void)
1334 {
1335 	unsigned int eax, ebx, ecx, edx;
1336 	unsigned int highest_cstate = 0;
1337 	unsigned int highest_subcstate = 0;
1338 	int i;
1339 	void *mwait_ptr;
1340 	struct cpuinfo_x86 *c = __this_cpu_ptr(&cpu_info);
1341 
1342 	if (!(this_cpu_has(X86_FEATURE_MWAIT) && mwait_usable(c)))
1343 		return;
1344 	if (!this_cpu_has(X86_FEATURE_CLFLSH))
1345 		return;
1346 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1347 		return;
1348 
1349 	eax = CPUID_MWAIT_LEAF;
1350 	ecx = 0;
1351 	native_cpuid(&eax, &ebx, &ecx, &edx);
1352 
1353 	/*
1354 	 * eax will be 0 if EDX enumeration is not valid.
1355 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1356 	 */
1357 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1358 		eax = 0;
1359 	} else {
1360 		edx >>= MWAIT_SUBSTATE_SIZE;
1361 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1362 			if (edx & MWAIT_SUBSTATE_MASK) {
1363 				highest_cstate = i;
1364 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1365 			}
1366 		}
1367 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1368 			(highest_subcstate - 1);
1369 	}
1370 
1371 	/*
1372 	 * This should be a memory location in a cache line which is
1373 	 * unlikely to be touched by other processors.  The actual
1374 	 * content is immaterial as it is not actually modified in any way.
1375 	 */
1376 	mwait_ptr = &current_thread_info()->flags;
1377 
1378 	wbinvd();
1379 
1380 	while (1) {
1381 		/*
1382 		 * The CLFLUSH is a workaround for erratum AAI65 for
1383 		 * the Xeon 7400 series.  It's not clear it is actually
1384 		 * needed, but it should be harmless in either case.
1385 		 * The WBINVD is insufficient due to the spurious-wakeup
1386 		 * case where we return around the loop.
1387 		 */
1388 		clflush(mwait_ptr);
1389 		__monitor(mwait_ptr, 0, 0);
1390 		mb();
1391 		__mwait(eax, 0);
1392 	}
1393 }
1394 
1395 static inline void hlt_play_dead(void)
1396 {
1397 	if (__this_cpu_read(cpu_info.x86) >= 4)
1398 		wbinvd();
1399 
1400 	while (1) {
1401 		native_halt();
1402 	}
1403 }
1404 
1405 void native_play_dead(void)
1406 {
1407 	play_dead_common();
1408 	tboot_shutdown(TB_SHUTDOWN_WFS);
1409 
1410 	mwait_play_dead();	/* Only returns on failure */
1411 	hlt_play_dead();
1412 }
1413 
1414 #else /* ... !CONFIG_HOTPLUG_CPU */
1415 int native_cpu_disable(void)
1416 {
1417 	return -ENOSYS;
1418 }
1419 
1420 void native_cpu_die(unsigned int cpu)
1421 {
1422 	/* We said "no" in __cpu_disable */
1423 	BUG();
1424 }
1425 
1426 void native_play_dead(void)
1427 {
1428 	BUG();
1429 }
1430 
1431 #endif
1432