xref: /openbmc/linux/arch/x86/kernel/smpboot.c (revision cd5d5810)
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 
77 #include <asm/smpboot_hooks.h>
78 #include <asm/i8259.h>
79 
80 #include <asm/realmode.h>
81 
82 /* State of each CPU */
83 DEFINE_PER_CPU(int, cpu_state) = { 0 };
84 
85 #ifdef CONFIG_HOTPLUG_CPU
86 /*
87  * We need this for trampoline_base protection from concurrent accesses when
88  * off- and onlining cores wildly.
89  */
90 static DEFINE_MUTEX(x86_cpu_hotplug_driver_mutex);
91 
92 void cpu_hotplug_driver_lock(void)
93 {
94 	mutex_lock(&x86_cpu_hotplug_driver_mutex);
95 }
96 
97 void cpu_hotplug_driver_unlock(void)
98 {
99 	mutex_unlock(&x86_cpu_hotplug_driver_mutex);
100 }
101 
102 ssize_t arch_cpu_probe(const char *buf, size_t count) { return -1; }
103 ssize_t arch_cpu_release(const char *buf, size_t count) { return -1; }
104 #endif
105 
106 /* Number of siblings per CPU package */
107 int smp_num_siblings = 1;
108 EXPORT_SYMBOL(smp_num_siblings);
109 
110 /* Last level cache ID of each logical CPU */
111 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
112 
113 /* representing HT siblings of each logical CPU */
114 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
115 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
116 
117 /* representing HT and core siblings of each logical CPU */
118 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_map);
119 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
120 
121 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_llc_shared_map);
122 
123 /* Per CPU bogomips and other parameters */
124 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
125 EXPORT_PER_CPU_SYMBOL(cpu_info);
126 
127 atomic_t init_deasserted;
128 
129 /*
130  * Report back to the Boot Processor during boot time or to the caller processor
131  * during CPU online.
132  */
133 static void smp_callin(void)
134 {
135 	int cpuid, phys_id;
136 	unsigned long timeout;
137 
138 	/*
139 	 * If waken up by an INIT in an 82489DX configuration
140 	 * we may get here before an INIT-deassert IPI reaches
141 	 * our local APIC.  We have to wait for the IPI or we'll
142 	 * lock up on an APIC access.
143 	 *
144 	 * Since CPU0 is not wakened up by INIT, it doesn't wait for the IPI.
145 	 */
146 	cpuid = smp_processor_id();
147 	if (apic->wait_for_init_deassert && cpuid != 0)
148 		apic->wait_for_init_deassert(&init_deasserted);
149 
150 	/*
151 	 * (This works even if the APIC is not enabled.)
152 	 */
153 	phys_id = read_apic_id();
154 	if (cpumask_test_cpu(cpuid, cpu_callin_mask)) {
155 		panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
156 					phys_id, cpuid);
157 	}
158 	pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
159 
160 	/*
161 	 * STARTUP IPIs are fragile beasts as they might sometimes
162 	 * trigger some glue motherboard logic. Complete APIC bus
163 	 * silence for 1 second, this overestimates the time the
164 	 * boot CPU is spending to send the up to 2 STARTUP IPIs
165 	 * by a factor of two. This should be enough.
166 	 */
167 
168 	/*
169 	 * Waiting 2s total for startup (udelay is not yet working)
170 	 */
171 	timeout = jiffies + 2*HZ;
172 	while (time_before(jiffies, timeout)) {
173 		/*
174 		 * Has the boot CPU finished it's STARTUP sequence?
175 		 */
176 		if (cpumask_test_cpu(cpuid, cpu_callout_mask))
177 			break;
178 		cpu_relax();
179 	}
180 
181 	if (!time_before(jiffies, timeout)) {
182 		panic("%s: CPU%d started up but did not get a callout!\n",
183 		      __func__, cpuid);
184 	}
185 
186 	/*
187 	 * the boot CPU has finished the init stage and is spinning
188 	 * on callin_map until we finish. We are free to set up this
189 	 * CPU, first the APIC. (this is probably redundant on most
190 	 * boards)
191 	 */
192 
193 	pr_debug("CALLIN, before setup_local_APIC()\n");
194 	if (apic->smp_callin_clear_local_apic)
195 		apic->smp_callin_clear_local_apic();
196 	setup_local_APIC();
197 	end_local_APIC_setup();
198 
199 	/*
200 	 * Need to setup vector mappings before we enable interrupts.
201 	 */
202 	setup_vector_irq(smp_processor_id());
203 
204 	/*
205 	 * Save our processor parameters. Note: this information
206 	 * is needed for clock calibration.
207 	 */
208 	smp_store_cpu_info(cpuid);
209 
210 	/*
211 	 * Get our bogomips.
212 	 * Update loops_per_jiffy in cpu_data. Previous call to
213 	 * smp_store_cpu_info() stored a value that is close but not as
214 	 * accurate as the value just calculated.
215 	 */
216 	calibrate_delay();
217 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
218 	pr_debug("Stack at about %p\n", &cpuid);
219 
220 	/*
221 	 * This must be done before setting cpu_online_mask
222 	 * or calling notify_cpu_starting.
223 	 */
224 	set_cpu_sibling_map(raw_smp_processor_id());
225 	wmb();
226 
227 	notify_cpu_starting(cpuid);
228 
229 	/*
230 	 * Allow the master to continue.
231 	 */
232 	cpumask_set_cpu(cpuid, cpu_callin_mask);
233 }
234 
235 static int cpu0_logical_apicid;
236 static int enable_start_cpu0;
237 /*
238  * Activate a secondary processor.
239  */
240 static void notrace start_secondary(void *unused)
241 {
242 	/*
243 	 * Don't put *anything* before cpu_init(), SMP booting is too
244 	 * fragile that we want to limit the things done here to the
245 	 * most necessary things.
246 	 */
247 	cpu_init();
248 	x86_cpuinit.early_percpu_clock_init();
249 	preempt_disable();
250 	smp_callin();
251 
252 	enable_start_cpu0 = 0;
253 
254 #ifdef CONFIG_X86_32
255 	/* switch away from the initial page table */
256 	load_cr3(swapper_pg_dir);
257 	__flush_tlb_all();
258 #endif
259 
260 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
261 	barrier();
262 	/*
263 	 * Check TSC synchronization with the BP:
264 	 */
265 	check_tsc_sync_target();
266 
267 	/*
268 	 * We need to hold vector_lock so there the set of online cpus
269 	 * does not change while we are assigning vectors to cpus.  Holding
270 	 * this lock ensures we don't half assign or remove an irq from a cpu.
271 	 */
272 	lock_vector_lock();
273 	set_cpu_online(smp_processor_id(), true);
274 	unlock_vector_lock();
275 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
276 	x86_platform.nmi_init();
277 
278 	/* enable local interrupts */
279 	local_irq_enable();
280 
281 	/* to prevent fake stack check failure in clock setup */
282 	boot_init_stack_canary();
283 
284 	x86_cpuinit.setup_percpu_clockev();
285 
286 	wmb();
287 	cpu_startup_entry(CPUHP_ONLINE);
288 }
289 
290 void __init smp_store_boot_cpu_info(void)
291 {
292 	int id = 0; /* CPU 0 */
293 	struct cpuinfo_x86 *c = &cpu_data(id);
294 
295 	*c = boot_cpu_data;
296 	c->cpu_index = id;
297 }
298 
299 /*
300  * The bootstrap kernel entry code has set these up. Save them for
301  * a given CPU
302  */
303 void smp_store_cpu_info(int id)
304 {
305 	struct cpuinfo_x86 *c = &cpu_data(id);
306 
307 	*c = boot_cpu_data;
308 	c->cpu_index = id;
309 	/*
310 	 * During boot time, CPU0 has this setup already. Save the info when
311 	 * bringing up AP or offlined CPU0.
312 	 */
313 	identify_secondary_cpu(c);
314 }
315 
316 static bool
317 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
318 {
319 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
320 
321 	return !WARN_ONCE(cpu_to_node(cpu1) != cpu_to_node(cpu2),
322 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
323 		"[node: %d != %d]. Ignoring dependency.\n",
324 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
325 }
326 
327 #define link_mask(_m, c1, c2)						\
328 do {									\
329 	cpumask_set_cpu((c1), cpu_##_m##_mask(c2));			\
330 	cpumask_set_cpu((c2), cpu_##_m##_mask(c1));			\
331 } while (0)
332 
333 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
334 {
335 	if (cpu_has_topoext) {
336 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
337 
338 		if (c->phys_proc_id == o->phys_proc_id &&
339 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2) &&
340 		    c->compute_unit_id == o->compute_unit_id)
341 			return topology_sane(c, o, "smt");
342 
343 	} else if (c->phys_proc_id == o->phys_proc_id &&
344 		   c->cpu_core_id == o->cpu_core_id) {
345 		return topology_sane(c, o, "smt");
346 	}
347 
348 	return false;
349 }
350 
351 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
352 {
353 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
354 
355 	if (per_cpu(cpu_llc_id, cpu1) != BAD_APICID &&
356 	    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2))
357 		return topology_sane(c, o, "llc");
358 
359 	return false;
360 }
361 
362 static bool match_mc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
363 {
364 	if (c->phys_proc_id == o->phys_proc_id) {
365 		if (cpu_has(c, X86_FEATURE_AMD_DCM))
366 			return true;
367 
368 		return topology_sane(c, o, "mc");
369 	}
370 	return false;
371 }
372 
373 void set_cpu_sibling_map(int cpu)
374 {
375 	bool has_smt = smp_num_siblings > 1;
376 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
377 	struct cpuinfo_x86 *c = &cpu_data(cpu);
378 	struct cpuinfo_x86 *o;
379 	int i;
380 
381 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
382 
383 	if (!has_mp) {
384 		cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
385 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
386 		cpumask_set_cpu(cpu, cpu_core_mask(cpu));
387 		c->booted_cores = 1;
388 		return;
389 	}
390 
391 	for_each_cpu(i, cpu_sibling_setup_mask) {
392 		o = &cpu_data(i);
393 
394 		if ((i == cpu) || (has_smt && match_smt(c, o)))
395 			link_mask(sibling, cpu, i);
396 
397 		if ((i == cpu) || (has_mp && match_llc(c, o)))
398 			link_mask(llc_shared, cpu, i);
399 
400 	}
401 
402 	/*
403 	 * This needs a separate iteration over the cpus because we rely on all
404 	 * cpu_sibling_mask links to be set-up.
405 	 */
406 	for_each_cpu(i, cpu_sibling_setup_mask) {
407 		o = &cpu_data(i);
408 
409 		if ((i == cpu) || (has_mp && match_mc(c, o))) {
410 			link_mask(core, cpu, i);
411 
412 			/*
413 			 *  Does this new cpu bringup a new core?
414 			 */
415 			if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
416 				/*
417 				 * for each core in package, increment
418 				 * the booted_cores for this new cpu
419 				 */
420 				if (cpumask_first(cpu_sibling_mask(i)) == i)
421 					c->booted_cores++;
422 				/*
423 				 * increment the core count for all
424 				 * the other cpus in this package
425 				 */
426 				if (i != cpu)
427 					cpu_data(i).booted_cores++;
428 			} else if (i != cpu && !c->booted_cores)
429 				c->booted_cores = cpu_data(i).booted_cores;
430 		}
431 	}
432 }
433 
434 /* maps the cpu to the sched domain representing multi-core */
435 const struct cpumask *cpu_coregroup_mask(int cpu)
436 {
437 	return cpu_llc_shared_mask(cpu);
438 }
439 
440 static void impress_friends(void)
441 {
442 	int cpu;
443 	unsigned long bogosum = 0;
444 	/*
445 	 * Allow the user to impress friends.
446 	 */
447 	pr_debug("Before bogomips\n");
448 	for_each_possible_cpu(cpu)
449 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
450 			bogosum += cpu_data(cpu).loops_per_jiffy;
451 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
452 		num_online_cpus(),
453 		bogosum/(500000/HZ),
454 		(bogosum/(5000/HZ))%100);
455 
456 	pr_debug("Before bogocount - setting activated=1\n");
457 }
458 
459 void __inquire_remote_apic(int apicid)
460 {
461 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
462 	const char * const names[] = { "ID", "VERSION", "SPIV" };
463 	int timeout;
464 	u32 status;
465 
466 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
467 
468 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
469 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
470 
471 		/*
472 		 * Wait for idle.
473 		 */
474 		status = safe_apic_wait_icr_idle();
475 		if (status)
476 			pr_cont("a previous APIC delivery may have failed\n");
477 
478 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
479 
480 		timeout = 0;
481 		do {
482 			udelay(100);
483 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
484 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
485 
486 		switch (status) {
487 		case APIC_ICR_RR_VALID:
488 			status = apic_read(APIC_RRR);
489 			pr_cont("%08x\n", status);
490 			break;
491 		default:
492 			pr_cont("failed\n");
493 		}
494 	}
495 }
496 
497 /*
498  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
499  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
500  * won't ... remember to clear down the APIC, etc later.
501  */
502 int
503 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
504 {
505 	unsigned long send_status, accept_status = 0;
506 	int maxlvt;
507 
508 	/* Target chip */
509 	/* Boot on the stack */
510 	/* Kick the second */
511 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, apicid);
512 
513 	pr_debug("Waiting for send to finish...\n");
514 	send_status = safe_apic_wait_icr_idle();
515 
516 	/*
517 	 * Give the other CPU some time to accept the IPI.
518 	 */
519 	udelay(200);
520 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
521 		maxlvt = lapic_get_maxlvt();
522 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
523 			apic_write(APIC_ESR, 0);
524 		accept_status = (apic_read(APIC_ESR) & 0xEF);
525 	}
526 	pr_debug("NMI sent\n");
527 
528 	if (send_status)
529 		pr_err("APIC never delivered???\n");
530 	if (accept_status)
531 		pr_err("APIC delivery error (%lx)\n", accept_status);
532 
533 	return (send_status | accept_status);
534 }
535 
536 static int
537 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
538 {
539 	unsigned long send_status, accept_status = 0;
540 	int maxlvt, num_starts, j;
541 
542 	maxlvt = lapic_get_maxlvt();
543 
544 	/*
545 	 * Be paranoid about clearing APIC errors.
546 	 */
547 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
548 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
549 			apic_write(APIC_ESR, 0);
550 		apic_read(APIC_ESR);
551 	}
552 
553 	pr_debug("Asserting INIT\n");
554 
555 	/*
556 	 * Turn INIT on target chip
557 	 */
558 	/*
559 	 * Send IPI
560 	 */
561 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
562 		       phys_apicid);
563 
564 	pr_debug("Waiting for send to finish...\n");
565 	send_status = safe_apic_wait_icr_idle();
566 
567 	mdelay(10);
568 
569 	pr_debug("Deasserting INIT\n");
570 
571 	/* Target chip */
572 	/* Send IPI */
573 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
574 
575 	pr_debug("Waiting for send to finish...\n");
576 	send_status = safe_apic_wait_icr_idle();
577 
578 	mb();
579 	atomic_set(&init_deasserted, 1);
580 
581 	/*
582 	 * Should we send STARTUP IPIs ?
583 	 *
584 	 * Determine this based on the APIC version.
585 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
586 	 */
587 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
588 		num_starts = 2;
589 	else
590 		num_starts = 0;
591 
592 	/*
593 	 * Paravirt / VMI wants a startup IPI hook here to set up the
594 	 * target processor state.
595 	 */
596 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
597 			 stack_start);
598 
599 	/*
600 	 * Run STARTUP IPI loop.
601 	 */
602 	pr_debug("#startup loops: %d\n", num_starts);
603 
604 	for (j = 1; j <= num_starts; j++) {
605 		pr_debug("Sending STARTUP #%d\n", j);
606 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
607 			apic_write(APIC_ESR, 0);
608 		apic_read(APIC_ESR);
609 		pr_debug("After apic_write\n");
610 
611 		/*
612 		 * STARTUP IPI
613 		 */
614 
615 		/* Target chip */
616 		/* Boot on the stack */
617 		/* Kick the second */
618 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
619 			       phys_apicid);
620 
621 		/*
622 		 * Give the other CPU some time to accept the IPI.
623 		 */
624 		udelay(300);
625 
626 		pr_debug("Startup point 1\n");
627 
628 		pr_debug("Waiting for send to finish...\n");
629 		send_status = safe_apic_wait_icr_idle();
630 
631 		/*
632 		 * Give the other CPU some time to accept the IPI.
633 		 */
634 		udelay(200);
635 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
636 			apic_write(APIC_ESR, 0);
637 		accept_status = (apic_read(APIC_ESR) & 0xEF);
638 		if (send_status || accept_status)
639 			break;
640 	}
641 	pr_debug("After Startup\n");
642 
643 	if (send_status)
644 		pr_err("APIC never delivered???\n");
645 	if (accept_status)
646 		pr_err("APIC delivery error (%lx)\n", accept_status);
647 
648 	return (send_status | accept_status);
649 }
650 
651 /* reduce the number of lines printed when booting a large cpu count system */
652 static void announce_cpu(int cpu, int apicid)
653 {
654 	static int current_node = -1;
655 	int node = early_cpu_to_node(cpu);
656 	int max_cpu_present = find_last_bit(cpumask_bits(cpu_present_mask), NR_CPUS);
657 
658 	if (system_state == SYSTEM_BOOTING) {
659 		if (node != current_node) {
660 			if (current_node > (-1))
661 				pr_cont(" OK\n");
662 			current_node = node;
663 			pr_info("Booting Node %3d, Processors ", node);
664 		}
665 		pr_cont(" #%4d%s", cpu, cpu == max_cpu_present ? " OK\n" : "");
666 		return;
667 	} else
668 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
669 			node, cpu, apicid);
670 }
671 
672 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
673 {
674 	int cpu;
675 
676 	cpu = smp_processor_id();
677 	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
678 		return NMI_HANDLED;
679 
680 	return NMI_DONE;
681 }
682 
683 /*
684  * Wake up AP by INIT, INIT, STARTUP sequence.
685  *
686  * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
687  * boot-strap code which is not a desired behavior for waking up BSP. To
688  * void the boot-strap code, wake up CPU0 by NMI instead.
689  *
690  * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
691  * (i.e. physically hot removed and then hot added), NMI won't wake it up.
692  * We'll change this code in the future to wake up hard offlined CPU0 if
693  * real platform and request are available.
694  */
695 static int
696 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
697 	       int *cpu0_nmi_registered)
698 {
699 	int id;
700 	int boot_error;
701 
702 	/*
703 	 * Wake up AP by INIT, INIT, STARTUP sequence.
704 	 */
705 	if (cpu)
706 		return wakeup_secondary_cpu_via_init(apicid, start_ip);
707 
708 	/*
709 	 * Wake up BSP by nmi.
710 	 *
711 	 * Register a NMI handler to help wake up CPU0.
712 	 */
713 	boot_error = register_nmi_handler(NMI_LOCAL,
714 					  wakeup_cpu0_nmi, 0, "wake_cpu0");
715 
716 	if (!boot_error) {
717 		enable_start_cpu0 = 1;
718 		*cpu0_nmi_registered = 1;
719 		if (apic->dest_logical == APIC_DEST_LOGICAL)
720 			id = cpu0_logical_apicid;
721 		else
722 			id = apicid;
723 		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
724 	}
725 
726 	return boot_error;
727 }
728 
729 /*
730  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
731  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
732  * Returns zero if CPU booted OK, else error code from
733  * ->wakeup_secondary_cpu.
734  */
735 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
736 {
737 	volatile u32 *trampoline_status =
738 		(volatile u32 *) __va(real_mode_header->trampoline_status);
739 	/* start_ip had better be page-aligned! */
740 	unsigned long start_ip = real_mode_header->trampoline_start;
741 
742 	unsigned long boot_error = 0;
743 	int timeout;
744 	int cpu0_nmi_registered = 0;
745 
746 	/* Just in case we booted with a single CPU. */
747 	alternatives_enable_smp();
748 
749 	idle->thread.sp = (unsigned long) (((struct pt_regs *)
750 			  (THREAD_SIZE +  task_stack_page(idle))) - 1);
751 	per_cpu(current_task, cpu) = idle;
752 
753 #ifdef CONFIG_X86_32
754 	/* Stack for startup_32 can be just as for start_secondary onwards */
755 	irq_ctx_init(cpu);
756 #else
757 	clear_tsk_thread_flag(idle, TIF_FORK);
758 	initial_gs = per_cpu_offset(cpu);
759 	per_cpu(kernel_stack, cpu) =
760 		(unsigned long)task_stack_page(idle) -
761 		KERNEL_STACK_OFFSET + THREAD_SIZE;
762 #endif
763 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
764 	initial_code = (unsigned long)start_secondary;
765 	stack_start  = idle->thread.sp;
766 
767 	/* So we see what's up */
768 	announce_cpu(cpu, apicid);
769 
770 	/*
771 	 * This grunge runs the startup process for
772 	 * the targeted processor.
773 	 */
774 
775 	atomic_set(&init_deasserted, 0);
776 
777 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
778 
779 		pr_debug("Setting warm reset code and vector.\n");
780 
781 		smpboot_setup_warm_reset_vector(start_ip);
782 		/*
783 		 * Be paranoid about clearing APIC errors.
784 		*/
785 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
786 			apic_write(APIC_ESR, 0);
787 			apic_read(APIC_ESR);
788 		}
789 	}
790 
791 	/*
792 	 * Wake up a CPU in difference cases:
793 	 * - Use the method in the APIC driver if it's defined
794 	 * Otherwise,
795 	 * - Use an INIT boot APIC message for APs or NMI for BSP.
796 	 */
797 	if (apic->wakeup_secondary_cpu)
798 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
799 	else
800 		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
801 						     &cpu0_nmi_registered);
802 
803 	if (!boot_error) {
804 		/*
805 		 * allow APs to start initializing.
806 		 */
807 		pr_debug("Before Callout %d\n", cpu);
808 		cpumask_set_cpu(cpu, cpu_callout_mask);
809 		pr_debug("After Callout %d\n", cpu);
810 
811 		/*
812 		 * Wait 5s total for a response
813 		 */
814 		for (timeout = 0; timeout < 50000; timeout++) {
815 			if (cpumask_test_cpu(cpu, cpu_callin_mask))
816 				break;	/* It has booted */
817 			udelay(100);
818 			/*
819 			 * Allow other tasks to run while we wait for the
820 			 * AP to come online. This also gives a chance
821 			 * for the MTRR work(triggered by the AP coming online)
822 			 * to be completed in the stop machine context.
823 			 */
824 			schedule();
825 		}
826 
827 		if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
828 			print_cpu_msr(&cpu_data(cpu));
829 			pr_debug("CPU%d: has booted.\n", cpu);
830 		} else {
831 			boot_error = 1;
832 			if (*trampoline_status == 0xA5A5A5A5)
833 				/* trampoline started but...? */
834 				pr_err("CPU%d: Stuck ??\n", cpu);
835 			else
836 				/* trampoline code not run */
837 				pr_err("CPU%d: Not responding\n", cpu);
838 			if (apic->inquire_remote_apic)
839 				apic->inquire_remote_apic(apicid);
840 		}
841 	}
842 
843 	if (boot_error) {
844 		/* Try to put things back the way they were before ... */
845 		numa_remove_cpu(cpu); /* was set by numa_add_cpu */
846 
847 		/* was set by do_boot_cpu() */
848 		cpumask_clear_cpu(cpu, cpu_callout_mask);
849 
850 		/* was set by cpu_init() */
851 		cpumask_clear_cpu(cpu, cpu_initialized_mask);
852 
853 		set_cpu_present(cpu, false);
854 		per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
855 	}
856 
857 	/* mark "stuck" area as not stuck */
858 	*trampoline_status = 0;
859 
860 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
861 		/*
862 		 * Cleanup possible dangling ends...
863 		 */
864 		smpboot_restore_warm_reset_vector();
865 	}
866 	/*
867 	 * Clean up the nmi handler. Do this after the callin and callout sync
868 	 * to avoid impact of possible long unregister time.
869 	 */
870 	if (cpu0_nmi_registered)
871 		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
872 
873 	return boot_error;
874 }
875 
876 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
877 {
878 	int apicid = apic->cpu_present_to_apicid(cpu);
879 	unsigned long flags;
880 	int err;
881 
882 	WARN_ON(irqs_disabled());
883 
884 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
885 
886 	if (apicid == BAD_APICID ||
887 	    !physid_isset(apicid, phys_cpu_present_map) ||
888 	    !apic->apic_id_valid(apicid)) {
889 		pr_err("%s: bad cpu %d\n", __func__, cpu);
890 		return -EINVAL;
891 	}
892 
893 	/*
894 	 * Already booted CPU?
895 	 */
896 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
897 		pr_debug("do_boot_cpu %d Already started\n", cpu);
898 		return -ENOSYS;
899 	}
900 
901 	/*
902 	 * Save current MTRR state in case it was changed since early boot
903 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
904 	 */
905 	mtrr_save_state();
906 
907 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
908 
909 	/* the FPU context is blank, nobody can own it */
910 	__cpu_disable_lazy_restore(cpu);
911 
912 	err = do_boot_cpu(apicid, cpu, tidle);
913 	if (err) {
914 		pr_debug("do_boot_cpu failed %d\n", err);
915 		return -EIO;
916 	}
917 
918 	/*
919 	 * Check TSC synchronization with the AP (keep irqs disabled
920 	 * while doing so):
921 	 */
922 	local_irq_save(flags);
923 	check_tsc_sync_source(cpu);
924 	local_irq_restore(flags);
925 
926 	while (!cpu_online(cpu)) {
927 		cpu_relax();
928 		touch_nmi_watchdog();
929 	}
930 
931 	return 0;
932 }
933 
934 /**
935  * arch_disable_smp_support() - disables SMP support for x86 at runtime
936  */
937 void arch_disable_smp_support(void)
938 {
939 	disable_ioapic_support();
940 }
941 
942 /*
943  * Fall back to non SMP mode after errors.
944  *
945  * RED-PEN audit/test this more. I bet there is more state messed up here.
946  */
947 static __init void disable_smp(void)
948 {
949 	init_cpu_present(cpumask_of(0));
950 	init_cpu_possible(cpumask_of(0));
951 	smpboot_clear_io_apic_irqs();
952 
953 	if (smp_found_config)
954 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
955 	else
956 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
957 	cpumask_set_cpu(0, cpu_sibling_mask(0));
958 	cpumask_set_cpu(0, cpu_core_mask(0));
959 }
960 
961 /*
962  * Various sanity checks.
963  */
964 static int __init smp_sanity_check(unsigned max_cpus)
965 {
966 	preempt_disable();
967 
968 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
969 	if (def_to_bigsmp && nr_cpu_ids > 8) {
970 		unsigned int cpu;
971 		unsigned nr;
972 
973 		pr_warn("More than 8 CPUs detected - skipping them\n"
974 			"Use CONFIG_X86_BIGSMP\n");
975 
976 		nr = 0;
977 		for_each_present_cpu(cpu) {
978 			if (nr >= 8)
979 				set_cpu_present(cpu, false);
980 			nr++;
981 		}
982 
983 		nr = 0;
984 		for_each_possible_cpu(cpu) {
985 			if (nr >= 8)
986 				set_cpu_possible(cpu, false);
987 			nr++;
988 		}
989 
990 		nr_cpu_ids = 8;
991 	}
992 #endif
993 
994 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
995 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
996 			hard_smp_processor_id());
997 
998 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
999 	}
1000 
1001 	/*
1002 	 * If we couldn't find an SMP configuration at boot time,
1003 	 * get out of here now!
1004 	 */
1005 	if (!smp_found_config && !acpi_lapic) {
1006 		preempt_enable();
1007 		pr_notice("SMP motherboard not detected\n");
1008 		disable_smp();
1009 		if (APIC_init_uniprocessor())
1010 			pr_notice("Local APIC not detected. Using dummy APIC emulation.\n");
1011 		return -1;
1012 	}
1013 
1014 	/*
1015 	 * Should not be necessary because the MP table should list the boot
1016 	 * CPU too, but we do it for the sake of robustness anyway.
1017 	 */
1018 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1019 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1020 			  boot_cpu_physical_apicid);
1021 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1022 	}
1023 	preempt_enable();
1024 
1025 	/*
1026 	 * If we couldn't find a local APIC, then get out of here now!
1027 	 */
1028 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1029 	    !cpu_has_apic) {
1030 		if (!disable_apic) {
1031 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
1032 				boot_cpu_physical_apicid);
1033 			pr_err("... forcing use of dummy APIC emulation (tell your hw vendor)\n");
1034 		}
1035 		smpboot_clear_io_apic();
1036 		disable_ioapic_support();
1037 		return -1;
1038 	}
1039 
1040 	verify_local_APIC();
1041 
1042 	/*
1043 	 * If SMP should be disabled, then really disable it!
1044 	 */
1045 	if (!max_cpus) {
1046 		pr_info("SMP mode deactivated\n");
1047 		smpboot_clear_io_apic();
1048 
1049 		connect_bsp_APIC();
1050 		setup_local_APIC();
1051 		bsp_end_local_APIC_setup();
1052 		return -1;
1053 	}
1054 
1055 	return 0;
1056 }
1057 
1058 static void __init smp_cpu_index_default(void)
1059 {
1060 	int i;
1061 	struct cpuinfo_x86 *c;
1062 
1063 	for_each_possible_cpu(i) {
1064 		c = &cpu_data(i);
1065 		/* mark all to hotplug */
1066 		c->cpu_index = nr_cpu_ids;
1067 	}
1068 }
1069 
1070 /*
1071  * Prepare for SMP bootup.  The MP table or ACPI has been read
1072  * earlier.  Just do some sanity checking here and enable APIC mode.
1073  */
1074 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1075 {
1076 	unsigned int i;
1077 
1078 	preempt_disable();
1079 	smp_cpu_index_default();
1080 
1081 	/*
1082 	 * Setup boot CPU information
1083 	 */
1084 	smp_store_boot_cpu_info(); /* Final full version of the data */
1085 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1086 	mb();
1087 
1088 	current_thread_info()->cpu = 0;  /* needed? */
1089 	for_each_possible_cpu(i) {
1090 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1091 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1092 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1093 	}
1094 	set_cpu_sibling_map(0);
1095 
1096 
1097 	if (smp_sanity_check(max_cpus) < 0) {
1098 		pr_info("SMP disabled\n");
1099 		disable_smp();
1100 		goto out;
1101 	}
1102 
1103 	default_setup_apic_routing();
1104 
1105 	preempt_disable();
1106 	if (read_apic_id() != boot_cpu_physical_apicid) {
1107 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1108 		     read_apic_id(), boot_cpu_physical_apicid);
1109 		/* Or can we switch back to PIC here? */
1110 	}
1111 	preempt_enable();
1112 
1113 	connect_bsp_APIC();
1114 
1115 	/*
1116 	 * Switch from PIC to APIC mode.
1117 	 */
1118 	setup_local_APIC();
1119 
1120 	if (x2apic_mode)
1121 		cpu0_logical_apicid = apic_read(APIC_LDR);
1122 	else
1123 		cpu0_logical_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1124 
1125 	/*
1126 	 * Enable IO APIC before setting up error vector
1127 	 */
1128 	if (!skip_ioapic_setup && nr_ioapics)
1129 		enable_IO_APIC();
1130 
1131 	bsp_end_local_APIC_setup();
1132 
1133 	if (apic->setup_portio_remap)
1134 		apic->setup_portio_remap();
1135 
1136 	smpboot_setup_io_apic();
1137 	/*
1138 	 * Set up local APIC timer on boot CPU.
1139 	 */
1140 
1141 	pr_info("CPU%d: ", 0);
1142 	print_cpu_info(&cpu_data(0));
1143 	x86_init.timers.setup_percpu_clockev();
1144 
1145 	if (is_uv_system())
1146 		uv_system_init();
1147 
1148 	set_mtrr_aps_delayed_init();
1149 out:
1150 	preempt_enable();
1151 }
1152 
1153 void arch_enable_nonboot_cpus_begin(void)
1154 {
1155 	set_mtrr_aps_delayed_init();
1156 }
1157 
1158 void arch_enable_nonboot_cpus_end(void)
1159 {
1160 	mtrr_aps_init();
1161 }
1162 
1163 /*
1164  * Early setup to make printk work.
1165  */
1166 void __init native_smp_prepare_boot_cpu(void)
1167 {
1168 	int me = smp_processor_id();
1169 	switch_to_new_gdt(me);
1170 	/* already set me in cpu_online_mask in boot_cpu_init() */
1171 	cpumask_set_cpu(me, cpu_callout_mask);
1172 	per_cpu(cpu_state, me) = CPU_ONLINE;
1173 }
1174 
1175 void __init native_smp_cpus_done(unsigned int max_cpus)
1176 {
1177 	pr_debug("Boot done\n");
1178 
1179 	nmi_selftest();
1180 	impress_friends();
1181 #ifdef CONFIG_X86_IO_APIC
1182 	setup_ioapic_dest();
1183 #endif
1184 	mtrr_aps_init();
1185 }
1186 
1187 static int __initdata setup_possible_cpus = -1;
1188 static int __init _setup_possible_cpus(char *str)
1189 {
1190 	get_option(&str, &setup_possible_cpus);
1191 	return 0;
1192 }
1193 early_param("possible_cpus", _setup_possible_cpus);
1194 
1195 
1196 /*
1197  * cpu_possible_mask should be static, it cannot change as cpu's
1198  * are onlined, or offlined. The reason is per-cpu data-structures
1199  * are allocated by some modules at init time, and dont expect to
1200  * do this dynamically on cpu arrival/departure.
1201  * cpu_present_mask on the other hand can change dynamically.
1202  * In case when cpu_hotplug is not compiled, then we resort to current
1203  * behaviour, which is cpu_possible == cpu_present.
1204  * - Ashok Raj
1205  *
1206  * Three ways to find out the number of additional hotplug CPUs:
1207  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1208  * - The user can overwrite it with possible_cpus=NUM
1209  * - Otherwise don't reserve additional CPUs.
1210  * We do this because additional CPUs waste a lot of memory.
1211  * -AK
1212  */
1213 __init void prefill_possible_map(void)
1214 {
1215 	int i, possible;
1216 
1217 	/* no processor from mptable or madt */
1218 	if (!num_processors)
1219 		num_processors = 1;
1220 
1221 	i = setup_max_cpus ?: 1;
1222 	if (setup_possible_cpus == -1) {
1223 		possible = num_processors;
1224 #ifdef CONFIG_HOTPLUG_CPU
1225 		if (setup_max_cpus)
1226 			possible += disabled_cpus;
1227 #else
1228 		if (possible > i)
1229 			possible = i;
1230 #endif
1231 	} else
1232 		possible = setup_possible_cpus;
1233 
1234 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1235 
1236 	/* nr_cpu_ids could be reduced via nr_cpus= */
1237 	if (possible > nr_cpu_ids) {
1238 		pr_warn("%d Processors exceeds NR_CPUS limit of %d\n",
1239 			possible, nr_cpu_ids);
1240 		possible = nr_cpu_ids;
1241 	}
1242 
1243 #ifdef CONFIG_HOTPLUG_CPU
1244 	if (!setup_max_cpus)
1245 #endif
1246 	if (possible > i) {
1247 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1248 			possible, setup_max_cpus);
1249 		possible = i;
1250 	}
1251 
1252 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1253 		possible, max_t(int, possible - num_processors, 0));
1254 
1255 	for (i = 0; i < possible; i++)
1256 		set_cpu_possible(i, true);
1257 	for (; i < NR_CPUS; i++)
1258 		set_cpu_possible(i, false);
1259 
1260 	nr_cpu_ids = possible;
1261 }
1262 
1263 #ifdef CONFIG_HOTPLUG_CPU
1264 
1265 static void remove_siblinginfo(int cpu)
1266 {
1267 	int sibling;
1268 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1269 
1270 	for_each_cpu(sibling, cpu_core_mask(cpu)) {
1271 		cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1272 		/*/
1273 		 * last thread sibling in this cpu core going down
1274 		 */
1275 		if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1276 			cpu_data(sibling).booted_cores--;
1277 	}
1278 
1279 	for_each_cpu(sibling, cpu_sibling_mask(cpu))
1280 		cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1281 	cpumask_clear(cpu_sibling_mask(cpu));
1282 	cpumask_clear(cpu_core_mask(cpu));
1283 	c->phys_proc_id = 0;
1284 	c->cpu_core_id = 0;
1285 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1286 }
1287 
1288 static void __ref remove_cpu_from_maps(int cpu)
1289 {
1290 	set_cpu_online(cpu, false);
1291 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1292 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1293 	/* was set by cpu_init() */
1294 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1295 	numa_remove_cpu(cpu);
1296 }
1297 
1298 void cpu_disable_common(void)
1299 {
1300 	int cpu = smp_processor_id();
1301 
1302 	remove_siblinginfo(cpu);
1303 
1304 	/* It's now safe to remove this processor from the online map */
1305 	lock_vector_lock();
1306 	remove_cpu_from_maps(cpu);
1307 	unlock_vector_lock();
1308 	fixup_irqs();
1309 }
1310 
1311 int native_cpu_disable(void)
1312 {
1313 	clear_local_APIC();
1314 
1315 	cpu_disable_common();
1316 	return 0;
1317 }
1318 
1319 void native_cpu_die(unsigned int cpu)
1320 {
1321 	/* We don't do anything here: idle task is faking death itself. */
1322 	unsigned int i;
1323 
1324 	for (i = 0; i < 10; i++) {
1325 		/* They ack this in play_dead by setting CPU_DEAD */
1326 		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1327 			if (system_state == SYSTEM_RUNNING)
1328 				pr_info("CPU %u is now offline\n", cpu);
1329 			return;
1330 		}
1331 		msleep(100);
1332 	}
1333 	pr_err("CPU %u didn't die...\n", cpu);
1334 }
1335 
1336 void play_dead_common(void)
1337 {
1338 	idle_task_exit();
1339 	reset_lazy_tlbstate();
1340 	amd_e400_remove_cpu(raw_smp_processor_id());
1341 
1342 	mb();
1343 	/* Ack it */
1344 	__this_cpu_write(cpu_state, CPU_DEAD);
1345 
1346 	/*
1347 	 * With physical CPU hotplug, we should halt the cpu
1348 	 */
1349 	local_irq_disable();
1350 }
1351 
1352 static bool wakeup_cpu0(void)
1353 {
1354 	if (smp_processor_id() == 0 && enable_start_cpu0)
1355 		return true;
1356 
1357 	return false;
1358 }
1359 
1360 /*
1361  * We need to flush the caches before going to sleep, lest we have
1362  * dirty data in our caches when we come back up.
1363  */
1364 static inline void mwait_play_dead(void)
1365 {
1366 	unsigned int eax, ebx, ecx, edx;
1367 	unsigned int highest_cstate = 0;
1368 	unsigned int highest_subcstate = 0;
1369 	void *mwait_ptr;
1370 	int i;
1371 
1372 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1373 		return;
1374 	if (!this_cpu_has(X86_FEATURE_CLFLSH))
1375 		return;
1376 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1377 		return;
1378 
1379 	eax = CPUID_MWAIT_LEAF;
1380 	ecx = 0;
1381 	native_cpuid(&eax, &ebx, &ecx, &edx);
1382 
1383 	/*
1384 	 * eax will be 0 if EDX enumeration is not valid.
1385 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1386 	 */
1387 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1388 		eax = 0;
1389 	} else {
1390 		edx >>= MWAIT_SUBSTATE_SIZE;
1391 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1392 			if (edx & MWAIT_SUBSTATE_MASK) {
1393 				highest_cstate = i;
1394 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1395 			}
1396 		}
1397 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1398 			(highest_subcstate - 1);
1399 	}
1400 
1401 	/*
1402 	 * This should be a memory location in a cache line which is
1403 	 * unlikely to be touched by other processors.  The actual
1404 	 * content is immaterial as it is not actually modified in any way.
1405 	 */
1406 	mwait_ptr = &current_thread_info()->flags;
1407 
1408 	wbinvd();
1409 
1410 	while (1) {
1411 		/*
1412 		 * The CLFLUSH is a workaround for erratum AAI65 for
1413 		 * the Xeon 7400 series.  It's not clear it is actually
1414 		 * needed, but it should be harmless in either case.
1415 		 * The WBINVD is insufficient due to the spurious-wakeup
1416 		 * case where we return around the loop.
1417 		 */
1418 		clflush(mwait_ptr);
1419 		__monitor(mwait_ptr, 0, 0);
1420 		mb();
1421 		__mwait(eax, 0);
1422 		/*
1423 		 * If NMI wants to wake up CPU0, start CPU0.
1424 		 */
1425 		if (wakeup_cpu0())
1426 			start_cpu0();
1427 	}
1428 }
1429 
1430 static inline void hlt_play_dead(void)
1431 {
1432 	if (__this_cpu_read(cpu_info.x86) >= 4)
1433 		wbinvd();
1434 
1435 	while (1) {
1436 		native_halt();
1437 		/*
1438 		 * If NMI wants to wake up CPU0, start CPU0.
1439 		 */
1440 		if (wakeup_cpu0())
1441 			start_cpu0();
1442 	}
1443 }
1444 
1445 void native_play_dead(void)
1446 {
1447 	play_dead_common();
1448 	tboot_shutdown(TB_SHUTDOWN_WFS);
1449 
1450 	mwait_play_dead();	/* Only returns on failure */
1451 	if (cpuidle_play_dead())
1452 		hlt_play_dead();
1453 }
1454 
1455 #else /* ... !CONFIG_HOTPLUG_CPU */
1456 int native_cpu_disable(void)
1457 {
1458 	return -ENOSYS;
1459 }
1460 
1461 void native_cpu_die(unsigned int cpu)
1462 {
1463 	/* We said "no" in __cpu_disable */
1464 	BUG();
1465 }
1466 
1467 void native_play_dead(void)
1468 {
1469 	BUG();
1470 }
1471 
1472 #endif
1473