xref: /openbmc/linux/arch/x86/kernel/smpboot.c (revision fff74a93)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
43 
44 #include <linux/init.h>
45 #include <linux/smp.h>
46 #include <linux/module.h>
47 #include <linux/sched.h>
48 #include <linux/percpu.h>
49 #include <linux/bootmem.h>
50 #include <linux/err.h>
51 #include <linux/nmi.h>
52 #include <linux/tboot.h>
53 #include <linux/stackprotector.h>
54 #include <linux/gfp.h>
55 #include <linux/cpuidle.h>
56 
57 #include <asm/acpi.h>
58 #include <asm/desc.h>
59 #include <asm/nmi.h>
60 #include <asm/irq.h>
61 #include <asm/idle.h>
62 #include <asm/realmode.h>
63 #include <asm/cpu.h>
64 #include <asm/numa.h>
65 #include <asm/pgtable.h>
66 #include <asm/tlbflush.h>
67 #include <asm/mtrr.h>
68 #include <asm/mwait.h>
69 #include <asm/apic.h>
70 #include <asm/io_apic.h>
71 #include <asm/i387.h>
72 #include <asm/fpu-internal.h>
73 #include <asm/setup.h>
74 #include <asm/uv/uv.h>
75 #include <linux/mc146818rtc.h>
76 #include <asm/smpboot_hooks.h>
77 #include <asm/i8259.h>
78 #include <asm/realmode.h>
79 #include <asm/misc.h>
80 
81 /* State of each CPU */
82 DEFINE_PER_CPU(int, cpu_state) = { 0 };
83 
84 /* Number of siblings per CPU package */
85 int smp_num_siblings = 1;
86 EXPORT_SYMBOL(smp_num_siblings);
87 
88 /* Last level cache ID of each logical CPU */
89 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
90 
91 /* representing HT siblings of each logical CPU */
92 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
93 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
94 
95 /* representing HT and core siblings of each logical CPU */
96 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_map);
97 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
98 
99 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_llc_shared_map);
100 
101 /* Per CPU bogomips and other parameters */
102 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
103 EXPORT_PER_CPU_SYMBOL(cpu_info);
104 
105 atomic_t init_deasserted;
106 
107 /*
108  * Report back to the Boot Processor during boot time or to the caller processor
109  * during CPU online.
110  */
111 static void smp_callin(void)
112 {
113 	int cpuid, phys_id;
114 	unsigned long timeout;
115 
116 	/*
117 	 * If waken up by an INIT in an 82489DX configuration
118 	 * we may get here before an INIT-deassert IPI reaches
119 	 * our local APIC.  We have to wait for the IPI or we'll
120 	 * lock up on an APIC access.
121 	 *
122 	 * Since CPU0 is not wakened up by INIT, it doesn't wait for the IPI.
123 	 */
124 	cpuid = smp_processor_id();
125 	if (apic->wait_for_init_deassert && cpuid)
126 		while (!atomic_read(&init_deasserted))
127 			cpu_relax();
128 
129 	/*
130 	 * (This works even if the APIC is not enabled.)
131 	 */
132 	phys_id = read_apic_id();
133 	if (cpumask_test_cpu(cpuid, cpu_callin_mask)) {
134 		panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
135 					phys_id, cpuid);
136 	}
137 	pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
138 
139 	/*
140 	 * STARTUP IPIs are fragile beasts as they might sometimes
141 	 * trigger some glue motherboard logic. Complete APIC bus
142 	 * silence for 1 second, this overestimates the time the
143 	 * boot CPU is spending to send the up to 2 STARTUP IPIs
144 	 * by a factor of two. This should be enough.
145 	 */
146 
147 	/*
148 	 * Waiting 2s total for startup (udelay is not yet working)
149 	 */
150 	timeout = jiffies + 2*HZ;
151 	while (time_before(jiffies, timeout)) {
152 		/*
153 		 * Has the boot CPU finished it's STARTUP sequence?
154 		 */
155 		if (cpumask_test_cpu(cpuid, cpu_callout_mask))
156 			break;
157 		cpu_relax();
158 	}
159 
160 	if (!time_before(jiffies, timeout)) {
161 		panic("%s: CPU%d started up but did not get a callout!\n",
162 		      __func__, cpuid);
163 	}
164 
165 	/*
166 	 * the boot CPU has finished the init stage and is spinning
167 	 * on callin_map until we finish. We are free to set up this
168 	 * CPU, first the APIC. (this is probably redundant on most
169 	 * boards)
170 	 */
171 	setup_local_APIC();
172 	end_local_APIC_setup();
173 
174 	/*
175 	 * Need to setup vector mappings before we enable interrupts.
176 	 */
177 	setup_vector_irq(smp_processor_id());
178 
179 	/*
180 	 * Save our processor parameters. Note: this information
181 	 * is needed for clock calibration.
182 	 */
183 	smp_store_cpu_info(cpuid);
184 
185 	/*
186 	 * Get our bogomips.
187 	 * Update loops_per_jiffy in cpu_data. Previous call to
188 	 * smp_store_cpu_info() stored a value that is close but not as
189 	 * accurate as the value just calculated.
190 	 */
191 	calibrate_delay();
192 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
193 	pr_debug("Stack at about %p\n", &cpuid);
194 
195 	/*
196 	 * This must be done before setting cpu_online_mask
197 	 * or calling notify_cpu_starting.
198 	 */
199 	set_cpu_sibling_map(raw_smp_processor_id());
200 	wmb();
201 
202 	notify_cpu_starting(cpuid);
203 
204 	/*
205 	 * Allow the master to continue.
206 	 */
207 	cpumask_set_cpu(cpuid, cpu_callin_mask);
208 }
209 
210 static int cpu0_logical_apicid;
211 static int enable_start_cpu0;
212 /*
213  * Activate a secondary processor.
214  */
215 static void notrace start_secondary(void *unused)
216 {
217 	/*
218 	 * Don't put *anything* before cpu_init(), SMP booting is too
219 	 * fragile that we want to limit the things done here to the
220 	 * most necessary things.
221 	 */
222 	cpu_init();
223 	x86_cpuinit.early_percpu_clock_init();
224 	preempt_disable();
225 	smp_callin();
226 
227 	enable_start_cpu0 = 0;
228 
229 #ifdef CONFIG_X86_32
230 	/* switch away from the initial page table */
231 	load_cr3(swapper_pg_dir);
232 	__flush_tlb_all();
233 #endif
234 
235 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
236 	barrier();
237 	/*
238 	 * Check TSC synchronization with the BP:
239 	 */
240 	check_tsc_sync_target();
241 
242 	/*
243 	 * Enable the espfix hack for this CPU
244 	 */
245 #ifdef CONFIG_X86_ESPFIX64
246 	init_espfix_ap();
247 #endif
248 
249 	/*
250 	 * We need to hold vector_lock so there the set of online cpus
251 	 * does not change while we are assigning vectors to cpus.  Holding
252 	 * this lock ensures we don't half assign or remove an irq from a cpu.
253 	 */
254 	lock_vector_lock();
255 	set_cpu_online(smp_processor_id(), true);
256 	unlock_vector_lock();
257 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
258 	x86_platform.nmi_init();
259 
260 	/* enable local interrupts */
261 	local_irq_enable();
262 
263 	/* to prevent fake stack check failure in clock setup */
264 	boot_init_stack_canary();
265 
266 	x86_cpuinit.setup_percpu_clockev();
267 
268 	wmb();
269 	cpu_startup_entry(CPUHP_ONLINE);
270 }
271 
272 void __init smp_store_boot_cpu_info(void)
273 {
274 	int id = 0; /* CPU 0 */
275 	struct cpuinfo_x86 *c = &cpu_data(id);
276 
277 	*c = boot_cpu_data;
278 	c->cpu_index = id;
279 }
280 
281 /*
282  * The bootstrap kernel entry code has set these up. Save them for
283  * a given CPU
284  */
285 void smp_store_cpu_info(int id)
286 {
287 	struct cpuinfo_x86 *c = &cpu_data(id);
288 
289 	*c = boot_cpu_data;
290 	c->cpu_index = id;
291 	/*
292 	 * During boot time, CPU0 has this setup already. Save the info when
293 	 * bringing up AP or offlined CPU0.
294 	 */
295 	identify_secondary_cpu(c);
296 }
297 
298 static bool
299 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
300 {
301 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
302 
303 	return !WARN_ONCE(cpu_to_node(cpu1) != cpu_to_node(cpu2),
304 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
305 		"[node: %d != %d]. Ignoring dependency.\n",
306 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
307 }
308 
309 #define link_mask(_m, c1, c2)						\
310 do {									\
311 	cpumask_set_cpu((c1), cpu_##_m##_mask(c2));			\
312 	cpumask_set_cpu((c2), cpu_##_m##_mask(c1));			\
313 } while (0)
314 
315 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
316 {
317 	if (cpu_has_topoext) {
318 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
319 
320 		if (c->phys_proc_id == o->phys_proc_id &&
321 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2) &&
322 		    c->compute_unit_id == o->compute_unit_id)
323 			return topology_sane(c, o, "smt");
324 
325 	} else if (c->phys_proc_id == o->phys_proc_id &&
326 		   c->cpu_core_id == o->cpu_core_id) {
327 		return topology_sane(c, o, "smt");
328 	}
329 
330 	return false;
331 }
332 
333 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
334 {
335 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
336 
337 	if (per_cpu(cpu_llc_id, cpu1) != BAD_APICID &&
338 	    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2))
339 		return topology_sane(c, o, "llc");
340 
341 	return false;
342 }
343 
344 static bool match_mc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
345 {
346 	if (c->phys_proc_id == o->phys_proc_id) {
347 		if (cpu_has(c, X86_FEATURE_AMD_DCM))
348 			return true;
349 
350 		return topology_sane(c, o, "mc");
351 	}
352 	return false;
353 }
354 
355 void set_cpu_sibling_map(int cpu)
356 {
357 	bool has_smt = smp_num_siblings > 1;
358 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
359 	struct cpuinfo_x86 *c = &cpu_data(cpu);
360 	struct cpuinfo_x86 *o;
361 	int i;
362 
363 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
364 
365 	if (!has_mp) {
366 		cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
367 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
368 		cpumask_set_cpu(cpu, cpu_core_mask(cpu));
369 		c->booted_cores = 1;
370 		return;
371 	}
372 
373 	for_each_cpu(i, cpu_sibling_setup_mask) {
374 		o = &cpu_data(i);
375 
376 		if ((i == cpu) || (has_smt && match_smt(c, o)))
377 			link_mask(sibling, cpu, i);
378 
379 		if ((i == cpu) || (has_mp && match_llc(c, o)))
380 			link_mask(llc_shared, cpu, i);
381 
382 	}
383 
384 	/*
385 	 * This needs a separate iteration over the cpus because we rely on all
386 	 * cpu_sibling_mask links to be set-up.
387 	 */
388 	for_each_cpu(i, cpu_sibling_setup_mask) {
389 		o = &cpu_data(i);
390 
391 		if ((i == cpu) || (has_mp && match_mc(c, o))) {
392 			link_mask(core, cpu, i);
393 
394 			/*
395 			 *  Does this new cpu bringup a new core?
396 			 */
397 			if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
398 				/*
399 				 * for each core in package, increment
400 				 * the booted_cores for this new cpu
401 				 */
402 				if (cpumask_first(cpu_sibling_mask(i)) == i)
403 					c->booted_cores++;
404 				/*
405 				 * increment the core count for all
406 				 * the other cpus in this package
407 				 */
408 				if (i != cpu)
409 					cpu_data(i).booted_cores++;
410 			} else if (i != cpu && !c->booted_cores)
411 				c->booted_cores = cpu_data(i).booted_cores;
412 		}
413 	}
414 }
415 
416 /* maps the cpu to the sched domain representing multi-core */
417 const struct cpumask *cpu_coregroup_mask(int cpu)
418 {
419 	return cpu_llc_shared_mask(cpu);
420 }
421 
422 static void impress_friends(void)
423 {
424 	int cpu;
425 	unsigned long bogosum = 0;
426 	/*
427 	 * Allow the user to impress friends.
428 	 */
429 	pr_debug("Before bogomips\n");
430 	for_each_possible_cpu(cpu)
431 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
432 			bogosum += cpu_data(cpu).loops_per_jiffy;
433 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
434 		num_online_cpus(),
435 		bogosum/(500000/HZ),
436 		(bogosum/(5000/HZ))%100);
437 
438 	pr_debug("Before bogocount - setting activated=1\n");
439 }
440 
441 void __inquire_remote_apic(int apicid)
442 {
443 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
444 	const char * const names[] = { "ID", "VERSION", "SPIV" };
445 	int timeout;
446 	u32 status;
447 
448 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
449 
450 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
451 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
452 
453 		/*
454 		 * Wait for idle.
455 		 */
456 		status = safe_apic_wait_icr_idle();
457 		if (status)
458 			pr_cont("a previous APIC delivery may have failed\n");
459 
460 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
461 
462 		timeout = 0;
463 		do {
464 			udelay(100);
465 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
466 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
467 
468 		switch (status) {
469 		case APIC_ICR_RR_VALID:
470 			status = apic_read(APIC_RRR);
471 			pr_cont("%08x\n", status);
472 			break;
473 		default:
474 			pr_cont("failed\n");
475 		}
476 	}
477 }
478 
479 /*
480  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
481  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
482  * won't ... remember to clear down the APIC, etc later.
483  */
484 int
485 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
486 {
487 	unsigned long send_status, accept_status = 0;
488 	int maxlvt;
489 
490 	/* Target chip */
491 	/* Boot on the stack */
492 	/* Kick the second */
493 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, apicid);
494 
495 	pr_debug("Waiting for send to finish...\n");
496 	send_status = safe_apic_wait_icr_idle();
497 
498 	/*
499 	 * Give the other CPU some time to accept the IPI.
500 	 */
501 	udelay(200);
502 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
503 		maxlvt = lapic_get_maxlvt();
504 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
505 			apic_write(APIC_ESR, 0);
506 		accept_status = (apic_read(APIC_ESR) & 0xEF);
507 	}
508 	pr_debug("NMI sent\n");
509 
510 	if (send_status)
511 		pr_err("APIC never delivered???\n");
512 	if (accept_status)
513 		pr_err("APIC delivery error (%lx)\n", accept_status);
514 
515 	return (send_status | accept_status);
516 }
517 
518 static int
519 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
520 {
521 	unsigned long send_status, accept_status = 0;
522 	int maxlvt, num_starts, j;
523 
524 	maxlvt = lapic_get_maxlvt();
525 
526 	/*
527 	 * Be paranoid about clearing APIC errors.
528 	 */
529 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
530 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
531 			apic_write(APIC_ESR, 0);
532 		apic_read(APIC_ESR);
533 	}
534 
535 	pr_debug("Asserting INIT\n");
536 
537 	/*
538 	 * Turn INIT on target chip
539 	 */
540 	/*
541 	 * Send IPI
542 	 */
543 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
544 		       phys_apicid);
545 
546 	pr_debug("Waiting for send to finish...\n");
547 	send_status = safe_apic_wait_icr_idle();
548 
549 	mdelay(10);
550 
551 	pr_debug("Deasserting INIT\n");
552 
553 	/* Target chip */
554 	/* Send IPI */
555 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
556 
557 	pr_debug("Waiting for send to finish...\n");
558 	send_status = safe_apic_wait_icr_idle();
559 
560 	mb();
561 	atomic_set(&init_deasserted, 1);
562 
563 	/*
564 	 * Should we send STARTUP IPIs ?
565 	 *
566 	 * Determine this based on the APIC version.
567 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
568 	 */
569 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
570 		num_starts = 2;
571 	else
572 		num_starts = 0;
573 
574 	/*
575 	 * Paravirt / VMI wants a startup IPI hook here to set up the
576 	 * target processor state.
577 	 */
578 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
579 			 stack_start);
580 
581 	/*
582 	 * Run STARTUP IPI loop.
583 	 */
584 	pr_debug("#startup loops: %d\n", num_starts);
585 
586 	for (j = 1; j <= num_starts; j++) {
587 		pr_debug("Sending STARTUP #%d\n", j);
588 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
589 			apic_write(APIC_ESR, 0);
590 		apic_read(APIC_ESR);
591 		pr_debug("After apic_write\n");
592 
593 		/*
594 		 * STARTUP IPI
595 		 */
596 
597 		/* Target chip */
598 		/* Boot on the stack */
599 		/* Kick the second */
600 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
601 			       phys_apicid);
602 
603 		/*
604 		 * Give the other CPU some time to accept the IPI.
605 		 */
606 		udelay(300);
607 
608 		pr_debug("Startup point 1\n");
609 
610 		pr_debug("Waiting for send to finish...\n");
611 		send_status = safe_apic_wait_icr_idle();
612 
613 		/*
614 		 * Give the other CPU some time to accept the IPI.
615 		 */
616 		udelay(200);
617 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
618 			apic_write(APIC_ESR, 0);
619 		accept_status = (apic_read(APIC_ESR) & 0xEF);
620 		if (send_status || accept_status)
621 			break;
622 	}
623 	pr_debug("After Startup\n");
624 
625 	if (send_status)
626 		pr_err("APIC never delivered???\n");
627 	if (accept_status)
628 		pr_err("APIC delivery error (%lx)\n", accept_status);
629 
630 	return (send_status | accept_status);
631 }
632 
633 void smp_announce(void)
634 {
635 	int num_nodes = num_online_nodes();
636 
637 	printk(KERN_INFO "x86: Booted up %d node%s, %d CPUs\n",
638 	       num_nodes, (num_nodes > 1 ? "s" : ""), num_online_cpus());
639 }
640 
641 /* reduce the number of lines printed when booting a large cpu count system */
642 static void announce_cpu(int cpu, int apicid)
643 {
644 	static int current_node = -1;
645 	int node = early_cpu_to_node(cpu);
646 	static int width, node_width;
647 
648 	if (!width)
649 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
650 
651 	if (!node_width)
652 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
653 
654 	if (cpu == 1)
655 		printk(KERN_INFO "x86: Booting SMP configuration:\n");
656 
657 	if (system_state == SYSTEM_BOOTING) {
658 		if (node != current_node) {
659 			if (current_node > (-1))
660 				pr_cont("\n");
661 			current_node = node;
662 
663 			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
664 			       node_width - num_digits(node), " ", node);
665 		}
666 
667 		/* Add padding for the BSP */
668 		if (cpu == 1)
669 			pr_cont("%*s", width + 1, " ");
670 
671 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
672 
673 	} else
674 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
675 			node, cpu, apicid);
676 }
677 
678 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
679 {
680 	int cpu;
681 
682 	cpu = smp_processor_id();
683 	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
684 		return NMI_HANDLED;
685 
686 	return NMI_DONE;
687 }
688 
689 /*
690  * Wake up AP by INIT, INIT, STARTUP sequence.
691  *
692  * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
693  * boot-strap code which is not a desired behavior for waking up BSP. To
694  * void the boot-strap code, wake up CPU0 by NMI instead.
695  *
696  * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
697  * (i.e. physically hot removed and then hot added), NMI won't wake it up.
698  * We'll change this code in the future to wake up hard offlined CPU0 if
699  * real platform and request are available.
700  */
701 static int
702 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
703 	       int *cpu0_nmi_registered)
704 {
705 	int id;
706 	int boot_error;
707 
708 	preempt_disable();
709 
710 	/*
711 	 * Wake up AP by INIT, INIT, STARTUP sequence.
712 	 */
713 	if (cpu) {
714 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
715 		goto out;
716 	}
717 
718 	/*
719 	 * Wake up BSP by nmi.
720 	 *
721 	 * Register a NMI handler to help wake up CPU0.
722 	 */
723 	boot_error = register_nmi_handler(NMI_LOCAL,
724 					  wakeup_cpu0_nmi, 0, "wake_cpu0");
725 
726 	if (!boot_error) {
727 		enable_start_cpu0 = 1;
728 		*cpu0_nmi_registered = 1;
729 		if (apic->dest_logical == APIC_DEST_LOGICAL)
730 			id = cpu0_logical_apicid;
731 		else
732 			id = apicid;
733 		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
734 	}
735 
736 out:
737 	preempt_enable();
738 
739 	return boot_error;
740 }
741 
742 /*
743  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
744  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
745  * Returns zero if CPU booted OK, else error code from
746  * ->wakeup_secondary_cpu.
747  */
748 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
749 {
750 	volatile u32 *trampoline_status =
751 		(volatile u32 *) __va(real_mode_header->trampoline_status);
752 	/* start_ip had better be page-aligned! */
753 	unsigned long start_ip = real_mode_header->trampoline_start;
754 
755 	unsigned long boot_error = 0;
756 	int timeout;
757 	int cpu0_nmi_registered = 0;
758 
759 	/* Just in case we booted with a single CPU. */
760 	alternatives_enable_smp();
761 
762 	idle->thread.sp = (unsigned long) (((struct pt_regs *)
763 			  (THREAD_SIZE +  task_stack_page(idle))) - 1);
764 	per_cpu(current_task, cpu) = idle;
765 
766 #ifdef CONFIG_X86_32
767 	/* Stack for startup_32 can be just as for start_secondary onwards */
768 	irq_ctx_init(cpu);
769 #else
770 	clear_tsk_thread_flag(idle, TIF_FORK);
771 	initial_gs = per_cpu_offset(cpu);
772 #endif
773 	per_cpu(kernel_stack, cpu) =
774 		(unsigned long)task_stack_page(idle) -
775 		KERNEL_STACK_OFFSET + THREAD_SIZE;
776 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
777 	initial_code = (unsigned long)start_secondary;
778 	stack_start  = idle->thread.sp;
779 
780 	/* So we see what's up */
781 	announce_cpu(cpu, apicid);
782 
783 	/*
784 	 * This grunge runs the startup process for
785 	 * the targeted processor.
786 	 */
787 
788 	atomic_set(&init_deasserted, 0);
789 
790 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
791 
792 		pr_debug("Setting warm reset code and vector.\n");
793 
794 		smpboot_setup_warm_reset_vector(start_ip);
795 		/*
796 		 * Be paranoid about clearing APIC errors.
797 		*/
798 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
799 			apic_write(APIC_ESR, 0);
800 			apic_read(APIC_ESR);
801 		}
802 	}
803 
804 	/*
805 	 * Wake up a CPU in difference cases:
806 	 * - Use the method in the APIC driver if it's defined
807 	 * Otherwise,
808 	 * - Use an INIT boot APIC message for APs or NMI for BSP.
809 	 */
810 	if (apic->wakeup_secondary_cpu)
811 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
812 	else
813 		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
814 						     &cpu0_nmi_registered);
815 
816 	if (!boot_error) {
817 		/*
818 		 * allow APs to start initializing.
819 		 */
820 		pr_debug("Before Callout %d\n", cpu);
821 		cpumask_set_cpu(cpu, cpu_callout_mask);
822 		pr_debug("After Callout %d\n", cpu);
823 
824 		/*
825 		 * Wait 5s total for a response
826 		 */
827 		for (timeout = 0; timeout < 50000; timeout++) {
828 			if (cpumask_test_cpu(cpu, cpu_callin_mask))
829 				break;	/* It has booted */
830 			udelay(100);
831 			/*
832 			 * Allow other tasks to run while we wait for the
833 			 * AP to come online. This also gives a chance
834 			 * for the MTRR work(triggered by the AP coming online)
835 			 * to be completed in the stop machine context.
836 			 */
837 			schedule();
838 		}
839 
840 		if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
841 			print_cpu_msr(&cpu_data(cpu));
842 			pr_debug("CPU%d: has booted.\n", cpu);
843 		} else {
844 			boot_error = 1;
845 			if (*trampoline_status == 0xA5A5A5A5)
846 				/* trampoline started but...? */
847 				pr_err("CPU%d: Stuck ??\n", cpu);
848 			else
849 				/* trampoline code not run */
850 				pr_err("CPU%d: Not responding\n", cpu);
851 			if (apic->inquire_remote_apic)
852 				apic->inquire_remote_apic(apicid);
853 		}
854 	}
855 
856 	if (boot_error) {
857 		/* Try to put things back the way they were before ... */
858 		numa_remove_cpu(cpu); /* was set by numa_add_cpu */
859 
860 		/* was set by do_boot_cpu() */
861 		cpumask_clear_cpu(cpu, cpu_callout_mask);
862 
863 		/* was set by cpu_init() */
864 		cpumask_clear_cpu(cpu, cpu_initialized_mask);
865 	}
866 
867 	/* mark "stuck" area as not stuck */
868 	*trampoline_status = 0;
869 
870 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
871 		/*
872 		 * Cleanup possible dangling ends...
873 		 */
874 		smpboot_restore_warm_reset_vector();
875 	}
876 	/*
877 	 * Clean up the nmi handler. Do this after the callin and callout sync
878 	 * to avoid impact of possible long unregister time.
879 	 */
880 	if (cpu0_nmi_registered)
881 		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
882 
883 	return boot_error;
884 }
885 
886 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
887 {
888 	int apicid = apic->cpu_present_to_apicid(cpu);
889 	unsigned long flags;
890 	int err;
891 
892 	WARN_ON(irqs_disabled());
893 
894 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
895 
896 	if (apicid == BAD_APICID ||
897 	    !physid_isset(apicid, phys_cpu_present_map) ||
898 	    !apic->apic_id_valid(apicid)) {
899 		pr_err("%s: bad cpu %d\n", __func__, cpu);
900 		return -EINVAL;
901 	}
902 
903 	/*
904 	 * Already booted CPU?
905 	 */
906 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
907 		pr_debug("do_boot_cpu %d Already started\n", cpu);
908 		return -ENOSYS;
909 	}
910 
911 	/*
912 	 * Save current MTRR state in case it was changed since early boot
913 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
914 	 */
915 	mtrr_save_state();
916 
917 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
918 
919 	/* the FPU context is blank, nobody can own it */
920 	__cpu_disable_lazy_restore(cpu);
921 
922 	err = do_boot_cpu(apicid, cpu, tidle);
923 	if (err) {
924 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
925 		return -EIO;
926 	}
927 
928 	/*
929 	 * Check TSC synchronization with the AP (keep irqs disabled
930 	 * while doing so):
931 	 */
932 	local_irq_save(flags);
933 	check_tsc_sync_source(cpu);
934 	local_irq_restore(flags);
935 
936 	while (!cpu_online(cpu)) {
937 		cpu_relax();
938 		touch_nmi_watchdog();
939 	}
940 
941 	return 0;
942 }
943 
944 /**
945  * arch_disable_smp_support() - disables SMP support for x86 at runtime
946  */
947 void arch_disable_smp_support(void)
948 {
949 	disable_ioapic_support();
950 }
951 
952 /*
953  * Fall back to non SMP mode after errors.
954  *
955  * RED-PEN audit/test this more. I bet there is more state messed up here.
956  */
957 static __init void disable_smp(void)
958 {
959 	init_cpu_present(cpumask_of(0));
960 	init_cpu_possible(cpumask_of(0));
961 	smpboot_clear_io_apic_irqs();
962 
963 	if (smp_found_config)
964 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
965 	else
966 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
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 		pr_warn("More than 8 CPUs detected - skipping them\n"
984 			"Use CONFIG_X86_BIGSMP\n");
985 
986 		nr = 0;
987 		for_each_present_cpu(cpu) {
988 			if (nr >= 8)
989 				set_cpu_present(cpu, false);
990 			nr++;
991 		}
992 
993 		nr = 0;
994 		for_each_possible_cpu(cpu) {
995 			if (nr >= 8)
996 				set_cpu_possible(cpu, false);
997 			nr++;
998 		}
999 
1000 		nr_cpu_ids = 8;
1001 	}
1002 #endif
1003 
1004 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1005 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1006 			hard_smp_processor_id());
1007 
1008 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1009 	}
1010 
1011 	/*
1012 	 * If we couldn't find an SMP configuration at boot time,
1013 	 * get out of here now!
1014 	 */
1015 	if (!smp_found_config && !acpi_lapic) {
1016 		preempt_enable();
1017 		pr_notice("SMP motherboard not detected\n");
1018 		disable_smp();
1019 		if (APIC_init_uniprocessor())
1020 			pr_notice("Local APIC not detected. Using dummy APIC emulation.\n");
1021 		return -1;
1022 	}
1023 
1024 	/*
1025 	 * Should not be necessary because the MP table should list the boot
1026 	 * CPU too, but we do it for the sake of robustness anyway.
1027 	 */
1028 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1029 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1030 			  boot_cpu_physical_apicid);
1031 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1032 	}
1033 	preempt_enable();
1034 
1035 	/*
1036 	 * If we couldn't find a local APIC, then get out of here now!
1037 	 */
1038 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1039 	    !cpu_has_apic) {
1040 		if (!disable_apic) {
1041 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
1042 				boot_cpu_physical_apicid);
1043 			pr_err("... forcing use of dummy APIC emulation (tell your hw vendor)\n");
1044 		}
1045 		smpboot_clear_io_apic();
1046 		disable_ioapic_support();
1047 		return -1;
1048 	}
1049 
1050 	verify_local_APIC();
1051 
1052 	/*
1053 	 * If SMP should be disabled, then really disable it!
1054 	 */
1055 	if (!max_cpus) {
1056 		pr_info("SMP mode deactivated\n");
1057 		smpboot_clear_io_apic();
1058 
1059 		connect_bsp_APIC();
1060 		setup_local_APIC();
1061 		bsp_end_local_APIC_setup();
1062 		return -1;
1063 	}
1064 
1065 	return 0;
1066 }
1067 
1068 static void __init smp_cpu_index_default(void)
1069 {
1070 	int i;
1071 	struct cpuinfo_x86 *c;
1072 
1073 	for_each_possible_cpu(i) {
1074 		c = &cpu_data(i);
1075 		/* mark all to hotplug */
1076 		c->cpu_index = nr_cpu_ids;
1077 	}
1078 }
1079 
1080 /*
1081  * Prepare for SMP bootup.  The MP table or ACPI has been read
1082  * earlier.  Just do some sanity checking here and enable APIC mode.
1083  */
1084 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1085 {
1086 	unsigned int i;
1087 
1088 	preempt_disable();
1089 	smp_cpu_index_default();
1090 
1091 	/*
1092 	 * Setup boot CPU information
1093 	 */
1094 	smp_store_boot_cpu_info(); /* Final full version of the data */
1095 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1096 	mb();
1097 
1098 	current_thread_info()->cpu = 0;  /* needed? */
1099 	for_each_possible_cpu(i) {
1100 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1101 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1102 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1103 	}
1104 	set_cpu_sibling_map(0);
1105 
1106 
1107 	if (smp_sanity_check(max_cpus) < 0) {
1108 		pr_info("SMP disabled\n");
1109 		disable_smp();
1110 		goto out;
1111 	}
1112 
1113 	default_setup_apic_routing();
1114 
1115 	preempt_disable();
1116 	if (read_apic_id() != boot_cpu_physical_apicid) {
1117 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1118 		     read_apic_id(), boot_cpu_physical_apicid);
1119 		/* Or can we switch back to PIC here? */
1120 	}
1121 	preempt_enable();
1122 
1123 	connect_bsp_APIC();
1124 
1125 	/*
1126 	 * Switch from PIC to APIC mode.
1127 	 */
1128 	setup_local_APIC();
1129 
1130 	if (x2apic_mode)
1131 		cpu0_logical_apicid = apic_read(APIC_LDR);
1132 	else
1133 		cpu0_logical_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1134 
1135 	/*
1136 	 * Enable IO APIC before setting up error vector
1137 	 */
1138 	if (!skip_ioapic_setup && nr_ioapics)
1139 		enable_IO_APIC();
1140 
1141 	bsp_end_local_APIC_setup();
1142 	smpboot_setup_io_apic();
1143 	/*
1144 	 * Set up local APIC timer on boot CPU.
1145 	 */
1146 
1147 	pr_info("CPU%d: ", 0);
1148 	print_cpu_info(&cpu_data(0));
1149 	x86_init.timers.setup_percpu_clockev();
1150 
1151 	if (is_uv_system())
1152 		uv_system_init();
1153 
1154 	set_mtrr_aps_delayed_init();
1155 out:
1156 	preempt_enable();
1157 }
1158 
1159 void arch_enable_nonboot_cpus_begin(void)
1160 {
1161 	set_mtrr_aps_delayed_init();
1162 }
1163 
1164 void arch_enable_nonboot_cpus_end(void)
1165 {
1166 	mtrr_aps_init();
1167 }
1168 
1169 /*
1170  * Early setup to make printk work.
1171  */
1172 void __init native_smp_prepare_boot_cpu(void)
1173 {
1174 	int me = smp_processor_id();
1175 	switch_to_new_gdt(me);
1176 	/* already set me in cpu_online_mask in boot_cpu_init() */
1177 	cpumask_set_cpu(me, cpu_callout_mask);
1178 	per_cpu(cpu_state, me) = CPU_ONLINE;
1179 }
1180 
1181 void __init native_smp_cpus_done(unsigned int max_cpus)
1182 {
1183 	pr_debug("Boot done\n");
1184 
1185 	nmi_selftest();
1186 	impress_friends();
1187 #ifdef CONFIG_X86_IO_APIC
1188 	setup_ioapic_dest();
1189 #endif
1190 	mtrr_aps_init();
1191 }
1192 
1193 static int __initdata setup_possible_cpus = -1;
1194 static int __init _setup_possible_cpus(char *str)
1195 {
1196 	get_option(&str, &setup_possible_cpus);
1197 	return 0;
1198 }
1199 early_param("possible_cpus", _setup_possible_cpus);
1200 
1201 
1202 /*
1203  * cpu_possible_mask should be static, it cannot change as cpu's
1204  * are onlined, or offlined. The reason is per-cpu data-structures
1205  * are allocated by some modules at init time, and dont expect to
1206  * do this dynamically on cpu arrival/departure.
1207  * cpu_present_mask on the other hand can change dynamically.
1208  * In case when cpu_hotplug is not compiled, then we resort to current
1209  * behaviour, which is cpu_possible == cpu_present.
1210  * - Ashok Raj
1211  *
1212  * Three ways to find out the number of additional hotplug CPUs:
1213  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1214  * - The user can overwrite it with possible_cpus=NUM
1215  * - Otherwise don't reserve additional CPUs.
1216  * We do this because additional CPUs waste a lot of memory.
1217  * -AK
1218  */
1219 __init void prefill_possible_map(void)
1220 {
1221 	int i, possible;
1222 
1223 	/* no processor from mptable or madt */
1224 	if (!num_processors)
1225 		num_processors = 1;
1226 
1227 	i = setup_max_cpus ?: 1;
1228 	if (setup_possible_cpus == -1) {
1229 		possible = num_processors;
1230 #ifdef CONFIG_HOTPLUG_CPU
1231 		if (setup_max_cpus)
1232 			possible += disabled_cpus;
1233 #else
1234 		if (possible > i)
1235 			possible = i;
1236 #endif
1237 	} else
1238 		possible = setup_possible_cpus;
1239 
1240 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1241 
1242 	/* nr_cpu_ids could be reduced via nr_cpus= */
1243 	if (possible > nr_cpu_ids) {
1244 		pr_warn("%d Processors exceeds NR_CPUS limit of %d\n",
1245 			possible, nr_cpu_ids);
1246 		possible = nr_cpu_ids;
1247 	}
1248 
1249 #ifdef CONFIG_HOTPLUG_CPU
1250 	if (!setup_max_cpus)
1251 #endif
1252 	if (possible > i) {
1253 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1254 			possible, setup_max_cpus);
1255 		possible = i;
1256 	}
1257 
1258 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1259 		possible, max_t(int, possible - num_processors, 0));
1260 
1261 	for (i = 0; i < possible; i++)
1262 		set_cpu_possible(i, true);
1263 	for (; i < NR_CPUS; i++)
1264 		set_cpu_possible(i, false);
1265 
1266 	nr_cpu_ids = possible;
1267 }
1268 
1269 #ifdef CONFIG_HOTPLUG_CPU
1270 
1271 static void remove_siblinginfo(int cpu)
1272 {
1273 	int sibling;
1274 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1275 
1276 	for_each_cpu(sibling, cpu_core_mask(cpu)) {
1277 		cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1278 		/*/
1279 		 * last thread sibling in this cpu core going down
1280 		 */
1281 		if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1282 			cpu_data(sibling).booted_cores--;
1283 	}
1284 
1285 	for_each_cpu(sibling, cpu_sibling_mask(cpu))
1286 		cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1287 	cpumask_clear(cpu_sibling_mask(cpu));
1288 	cpumask_clear(cpu_core_mask(cpu));
1289 	c->phys_proc_id = 0;
1290 	c->cpu_core_id = 0;
1291 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1292 }
1293 
1294 static void __ref remove_cpu_from_maps(int cpu)
1295 {
1296 	set_cpu_online(cpu, false);
1297 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1298 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1299 	/* was set by cpu_init() */
1300 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1301 	numa_remove_cpu(cpu);
1302 }
1303 
1304 void cpu_disable_common(void)
1305 {
1306 	int cpu = smp_processor_id();
1307 
1308 	remove_siblinginfo(cpu);
1309 
1310 	/* It's now safe to remove this processor from the online map */
1311 	lock_vector_lock();
1312 	remove_cpu_from_maps(cpu);
1313 	unlock_vector_lock();
1314 	fixup_irqs();
1315 }
1316 
1317 int native_cpu_disable(void)
1318 {
1319 	int ret;
1320 
1321 	ret = check_irq_vectors_for_cpu_disable();
1322 	if (ret)
1323 		return ret;
1324 
1325 	clear_local_APIC();
1326 
1327 	cpu_disable_common();
1328 	return 0;
1329 }
1330 
1331 void native_cpu_die(unsigned int cpu)
1332 {
1333 	/* We don't do anything here: idle task is faking death itself. */
1334 	unsigned int i;
1335 
1336 	for (i = 0; i < 10; i++) {
1337 		/* They ack this in play_dead by setting CPU_DEAD */
1338 		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1339 			if (system_state == SYSTEM_RUNNING)
1340 				pr_info("CPU %u is now offline\n", cpu);
1341 			return;
1342 		}
1343 		msleep(100);
1344 	}
1345 	pr_err("CPU %u didn't die...\n", cpu);
1346 }
1347 
1348 void play_dead_common(void)
1349 {
1350 	idle_task_exit();
1351 	reset_lazy_tlbstate();
1352 	amd_e400_remove_cpu(raw_smp_processor_id());
1353 
1354 	mb();
1355 	/* Ack it */
1356 	__this_cpu_write(cpu_state, CPU_DEAD);
1357 
1358 	/*
1359 	 * With physical CPU hotplug, we should halt the cpu
1360 	 */
1361 	local_irq_disable();
1362 }
1363 
1364 static bool wakeup_cpu0(void)
1365 {
1366 	if (smp_processor_id() == 0 && enable_start_cpu0)
1367 		return true;
1368 
1369 	return false;
1370 }
1371 
1372 /*
1373  * We need to flush the caches before going to sleep, lest we have
1374  * dirty data in our caches when we come back up.
1375  */
1376 static inline void mwait_play_dead(void)
1377 {
1378 	unsigned int eax, ebx, ecx, edx;
1379 	unsigned int highest_cstate = 0;
1380 	unsigned int highest_subcstate = 0;
1381 	void *mwait_ptr;
1382 	int i;
1383 
1384 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1385 		return;
1386 	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1387 		return;
1388 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1389 		return;
1390 
1391 	eax = CPUID_MWAIT_LEAF;
1392 	ecx = 0;
1393 	native_cpuid(&eax, &ebx, &ecx, &edx);
1394 
1395 	/*
1396 	 * eax will be 0 if EDX enumeration is not valid.
1397 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1398 	 */
1399 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1400 		eax = 0;
1401 	} else {
1402 		edx >>= MWAIT_SUBSTATE_SIZE;
1403 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1404 			if (edx & MWAIT_SUBSTATE_MASK) {
1405 				highest_cstate = i;
1406 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1407 			}
1408 		}
1409 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1410 			(highest_subcstate - 1);
1411 	}
1412 
1413 	/*
1414 	 * This should be a memory location in a cache line which is
1415 	 * unlikely to be touched by other processors.  The actual
1416 	 * content is immaterial as it is not actually modified in any way.
1417 	 */
1418 	mwait_ptr = &current_thread_info()->flags;
1419 
1420 	wbinvd();
1421 
1422 	while (1) {
1423 		/*
1424 		 * The CLFLUSH is a workaround for erratum AAI65 for
1425 		 * the Xeon 7400 series.  It's not clear it is actually
1426 		 * needed, but it should be harmless in either case.
1427 		 * The WBINVD is insufficient due to the spurious-wakeup
1428 		 * case where we return around the loop.
1429 		 */
1430 		mb();
1431 		clflush(mwait_ptr);
1432 		mb();
1433 		__monitor(mwait_ptr, 0, 0);
1434 		mb();
1435 		__mwait(eax, 0);
1436 		/*
1437 		 * If NMI wants to wake up CPU0, start CPU0.
1438 		 */
1439 		if (wakeup_cpu0())
1440 			start_cpu0();
1441 	}
1442 }
1443 
1444 static inline void hlt_play_dead(void)
1445 {
1446 	if (__this_cpu_read(cpu_info.x86) >= 4)
1447 		wbinvd();
1448 
1449 	while (1) {
1450 		native_halt();
1451 		/*
1452 		 * If NMI wants to wake up CPU0, start CPU0.
1453 		 */
1454 		if (wakeup_cpu0())
1455 			start_cpu0();
1456 	}
1457 }
1458 
1459 void native_play_dead(void)
1460 {
1461 	play_dead_common();
1462 	tboot_shutdown(TB_SHUTDOWN_WFS);
1463 
1464 	mwait_play_dead();	/* Only returns on failure */
1465 	if (cpuidle_play_dead())
1466 		hlt_play_dead();
1467 }
1468 
1469 #else /* ... !CONFIG_HOTPLUG_CPU */
1470 int native_cpu_disable(void)
1471 {
1472 	return -ENOSYS;
1473 }
1474 
1475 void native_cpu_die(unsigned int cpu)
1476 {
1477 	/* We said "no" in __cpu_disable */
1478 	BUG();
1479 }
1480 
1481 void native_play_dead(void)
1482 {
1483 	BUG();
1484 }
1485 
1486 #endif
1487