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/i8259.h>
77#include <asm/realmode.h>
78#include <asm/misc.h>
79
80/* Number of siblings per CPU package */
81int smp_num_siblings = 1;
82EXPORT_SYMBOL(smp_num_siblings);
83
84/* Last level cache ID of each logical CPU */
85DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
86
87/* representing HT siblings of each logical CPU */
88DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
89EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
90
91/* representing HT and core siblings of each logical CPU */
92DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_map);
93EXPORT_PER_CPU_SYMBOL(cpu_core_map);
94
95DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_llc_shared_map);
96
97/* Per CPU bogomips and other parameters */
98DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
99EXPORT_PER_CPU_SYMBOL(cpu_info);
100
101atomic_t init_deasserted;
102
103static inline void smpboot_setup_warm_reset_vector(unsigned long start_eip)
104{
105	unsigned long flags;
106
107	spin_lock_irqsave(&rtc_lock, flags);
108	CMOS_WRITE(0xa, 0xf);
109	spin_unlock_irqrestore(&rtc_lock, flags);
110	local_flush_tlb();
111	pr_debug("1.\n");
112	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_HIGH)) =
113							start_eip >> 4;
114	pr_debug("2.\n");
115	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) =
116							start_eip & 0xf;
117	pr_debug("3.\n");
118}
119
120static inline void smpboot_restore_warm_reset_vector(void)
121{
122	unsigned long flags;
123
124	/*
125	 * Install writable page 0 entry to set BIOS data area.
126	 */
127	local_flush_tlb();
128
129	/*
130	 * Paranoid:  Set warm reset code and vector here back
131	 * to default values.
132	 */
133	spin_lock_irqsave(&rtc_lock, flags);
134	CMOS_WRITE(0, 0xf);
135	spin_unlock_irqrestore(&rtc_lock, flags);
136
137	*((volatile u32 *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = 0;
138}
139
140/*
141 * Report back to the Boot Processor during boot time or to the caller processor
142 * during CPU online.
143 */
144static void smp_callin(void)
145{
146	int cpuid, phys_id;
147
148	/*
149	 * If waken up by an INIT in an 82489DX configuration
150	 * we may get here before an INIT-deassert IPI reaches
151	 * our local APIC.  We have to wait for the IPI or we'll
152	 * lock up on an APIC access.
153	 *
154	 * Since CPU0 is not wakened up by INIT, it doesn't wait for the IPI.
155	 */
156	cpuid = smp_processor_id();
157	if (apic->wait_for_init_deassert && cpuid)
158		while (!atomic_read(&init_deasserted))
159			cpu_relax();
160
161	/*
162	 * (This works even if the APIC is not enabled.)
163	 */
164	phys_id = read_apic_id();
165
166	/*
167	 * the boot CPU has finished the init stage and is spinning
168	 * on callin_map until we finish. We are free to set up this
169	 * CPU, first the APIC. (this is probably redundant on most
170	 * boards)
171	 */
172	apic_ap_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
210static int cpu0_logical_apicid;
211static int enable_start_cpu0;
212/*
213 * Activate a secondary processor.
214 */
215static 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	cpu_set_state_online(smp_processor_id());
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
272void __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 */
285void 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
298static bool
299topology_same_node(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
300{
301	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
302
303	return (cpu_to_node(cpu1) == cpu_to_node(cpu2));
304}
305
306static bool
307topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
308{
309	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
310
311	return !WARN_ONCE(!topology_same_node(c, o),
312		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
313		"[node: %d != %d]. Ignoring dependency.\n",
314		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
315}
316
317#define link_mask(_m, c1, c2)						\
318do {									\
319	cpumask_set_cpu((c1), cpu_##_m##_mask(c2));			\
320	cpumask_set_cpu((c2), cpu_##_m##_mask(c1));			\
321} while (0)
322
323static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
324{
325	if (cpu_has_topoext) {
326		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
327
328		if (c->phys_proc_id == o->phys_proc_id &&
329		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2) &&
330		    c->compute_unit_id == o->compute_unit_id)
331			return topology_sane(c, o, "smt");
332
333	} else if (c->phys_proc_id == o->phys_proc_id &&
334		   c->cpu_core_id == o->cpu_core_id) {
335		return topology_sane(c, o, "smt");
336	}
337
338	return false;
339}
340
341static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
342{
343	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
344
345	if (per_cpu(cpu_llc_id, cpu1) != BAD_APICID &&
346	    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2))
347		return topology_sane(c, o, "llc");
348
349	return false;
350}
351
352/*
353 * Unlike the other levels, we do not enforce keeping a
354 * multicore group inside a NUMA node.  If this happens, we will
355 * discard the MC level of the topology later.
356 */
357static bool match_die(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
358{
359	if (c->phys_proc_id == o->phys_proc_id)
360		return true;
361	return false;
362}
363
364static struct sched_domain_topology_level numa_inside_package_topology[] = {
365#ifdef CONFIG_SCHED_SMT
366	{ cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
367#endif
368#ifdef CONFIG_SCHED_MC
369	{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
370#endif
371	{ NULL, },
372};
373/*
374 * set_sched_topology() sets the topology internal to a CPU.  The
375 * NUMA topologies are layered on top of it to build the full
376 * system topology.
377 *
378 * If NUMA nodes are observed to occur within a CPU package, this
379 * function should be called.  It forces the sched domain code to
380 * only use the SMT level for the CPU portion of the topology.
381 * This essentially falls back to relying on NUMA information
382 * from the SRAT table to describe the entire system topology
383 * (except for hyperthreads).
384 */
385static void primarily_use_numa_for_topology(void)
386{
387	set_sched_topology(numa_inside_package_topology);
388}
389
390void set_cpu_sibling_map(int cpu)
391{
392	bool has_smt = smp_num_siblings > 1;
393	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
394	struct cpuinfo_x86 *c = &cpu_data(cpu);
395	struct cpuinfo_x86 *o;
396	int i;
397
398	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
399
400	if (!has_mp) {
401		cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
402		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
403		cpumask_set_cpu(cpu, cpu_core_mask(cpu));
404		c->booted_cores = 1;
405		return;
406	}
407
408	for_each_cpu(i, cpu_sibling_setup_mask) {
409		o = &cpu_data(i);
410
411		if ((i == cpu) || (has_smt && match_smt(c, o)))
412			link_mask(sibling, cpu, i);
413
414		if ((i == cpu) || (has_mp && match_llc(c, o)))
415			link_mask(llc_shared, cpu, i);
416
417	}
418
419	/*
420	 * This needs a separate iteration over the cpus because we rely on all
421	 * cpu_sibling_mask links to be set-up.
422	 */
423	for_each_cpu(i, cpu_sibling_setup_mask) {
424		o = &cpu_data(i);
425
426		if ((i == cpu) || (has_mp && match_die(c, o))) {
427			link_mask(core, cpu, i);
428
429			/*
430			 *  Does this new cpu bringup a new core?
431			 */
432			if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
433				/*
434				 * for each core in package, increment
435				 * the booted_cores for this new cpu
436				 */
437				if (cpumask_first(cpu_sibling_mask(i)) == i)
438					c->booted_cores++;
439				/*
440				 * increment the core count for all
441				 * the other cpus in this package
442				 */
443				if (i != cpu)
444					cpu_data(i).booted_cores++;
445			} else if (i != cpu && !c->booted_cores)
446				c->booted_cores = cpu_data(i).booted_cores;
447		}
448		if (match_die(c, o) && !topology_same_node(c, o))
449			primarily_use_numa_for_topology();
450	}
451}
452
453/* maps the cpu to the sched domain representing multi-core */
454const struct cpumask *cpu_coregroup_mask(int cpu)
455{
456	return cpu_llc_shared_mask(cpu);
457}
458
459static void impress_friends(void)
460{
461	int cpu;
462	unsigned long bogosum = 0;
463	/*
464	 * Allow the user to impress friends.
465	 */
466	pr_debug("Before bogomips\n");
467	for_each_possible_cpu(cpu)
468		if (cpumask_test_cpu(cpu, cpu_callout_mask))
469			bogosum += cpu_data(cpu).loops_per_jiffy;
470	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
471		num_online_cpus(),
472		bogosum/(500000/HZ),
473		(bogosum/(5000/HZ))%100);
474
475	pr_debug("Before bogocount - setting activated=1\n");
476}
477
478void __inquire_remote_apic(int apicid)
479{
480	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
481	const char * const names[] = { "ID", "VERSION", "SPIV" };
482	int timeout;
483	u32 status;
484
485	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
486
487	for (i = 0; i < ARRAY_SIZE(regs); i++) {
488		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
489
490		/*
491		 * Wait for idle.
492		 */
493		status = safe_apic_wait_icr_idle();
494		if (status)
495			pr_cont("a previous APIC delivery may have failed\n");
496
497		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
498
499		timeout = 0;
500		do {
501			udelay(100);
502			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
503		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
504
505		switch (status) {
506		case APIC_ICR_RR_VALID:
507			status = apic_read(APIC_RRR);
508			pr_cont("%08x\n", status);
509			break;
510		default:
511			pr_cont("failed\n");
512		}
513	}
514}
515
516/*
517 * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
518 * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
519 * won't ... remember to clear down the APIC, etc later.
520 */
521int
522wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
523{
524	unsigned long send_status, accept_status = 0;
525	int maxlvt;
526
527	/* Target chip */
528	/* Boot on the stack */
529	/* Kick the second */
530	apic_icr_write(APIC_DM_NMI | apic->dest_logical, apicid);
531
532	pr_debug("Waiting for send to finish...\n");
533	send_status = safe_apic_wait_icr_idle();
534
535	/*
536	 * Give the other CPU some time to accept the IPI.
537	 */
538	udelay(200);
539	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
540		maxlvt = lapic_get_maxlvt();
541		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
542			apic_write(APIC_ESR, 0);
543		accept_status = (apic_read(APIC_ESR) & 0xEF);
544	}
545	pr_debug("NMI sent\n");
546
547	if (send_status)
548		pr_err("APIC never delivered???\n");
549	if (accept_status)
550		pr_err("APIC delivery error (%lx)\n", accept_status);
551
552	return (send_status | accept_status);
553}
554
555static int
556wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
557{
558	unsigned long send_status, accept_status = 0;
559	int maxlvt, num_starts, j;
560
561	maxlvt = lapic_get_maxlvt();
562
563	/*
564	 * Be paranoid about clearing APIC errors.
565	 */
566	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
567		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
568			apic_write(APIC_ESR, 0);
569		apic_read(APIC_ESR);
570	}
571
572	pr_debug("Asserting INIT\n");
573
574	/*
575	 * Turn INIT on target chip
576	 */
577	/*
578	 * Send IPI
579	 */
580	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
581		       phys_apicid);
582
583	pr_debug("Waiting for send to finish...\n");
584	send_status = safe_apic_wait_icr_idle();
585
586	mdelay(10);
587
588	pr_debug("Deasserting INIT\n");
589
590	/* Target chip */
591	/* Send IPI */
592	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
593
594	pr_debug("Waiting for send to finish...\n");
595	send_status = safe_apic_wait_icr_idle();
596
597	mb();
598	atomic_set(&init_deasserted, 1);
599
600	/*
601	 * Should we send STARTUP IPIs ?
602	 *
603	 * Determine this based on the APIC version.
604	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
605	 */
606	if (APIC_INTEGRATED(apic_version[phys_apicid]))
607		num_starts = 2;
608	else
609		num_starts = 0;
610
611	/*
612	 * Paravirt / VMI wants a startup IPI hook here to set up the
613	 * target processor state.
614	 */
615	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
616			 stack_start);
617
618	/*
619	 * Run STARTUP IPI loop.
620	 */
621	pr_debug("#startup loops: %d\n", num_starts);
622
623	for (j = 1; j <= num_starts; j++) {
624		pr_debug("Sending STARTUP #%d\n", j);
625		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
626			apic_write(APIC_ESR, 0);
627		apic_read(APIC_ESR);
628		pr_debug("After apic_write\n");
629
630		/*
631		 * STARTUP IPI
632		 */
633
634		/* Target chip */
635		/* Boot on the stack */
636		/* Kick the second */
637		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
638			       phys_apicid);
639
640		/*
641		 * Give the other CPU some time to accept the IPI.
642		 */
643		udelay(300);
644
645		pr_debug("Startup point 1\n");
646
647		pr_debug("Waiting for send to finish...\n");
648		send_status = safe_apic_wait_icr_idle();
649
650		/*
651		 * Give the other CPU some time to accept the IPI.
652		 */
653		udelay(200);
654		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
655			apic_write(APIC_ESR, 0);
656		accept_status = (apic_read(APIC_ESR) & 0xEF);
657		if (send_status || accept_status)
658			break;
659	}
660	pr_debug("After Startup\n");
661
662	if (send_status)
663		pr_err("APIC never delivered???\n");
664	if (accept_status)
665		pr_err("APIC delivery error (%lx)\n", accept_status);
666
667	return (send_status | accept_status);
668}
669
670void smp_announce(void)
671{
672	int num_nodes = num_online_nodes();
673
674	printk(KERN_INFO "x86: Booted up %d node%s, %d CPUs\n",
675	       num_nodes, (num_nodes > 1 ? "s" : ""), num_online_cpus());
676}
677
678/* reduce the number of lines printed when booting a large cpu count system */
679static void announce_cpu(int cpu, int apicid)
680{
681	static int current_node = -1;
682	int node = early_cpu_to_node(cpu);
683	static int width, node_width;
684
685	if (!width)
686		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
687
688	if (!node_width)
689		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
690
691	if (cpu == 1)
692		printk(KERN_INFO "x86: Booting SMP configuration:\n");
693
694	if (system_state == SYSTEM_BOOTING) {
695		if (node != current_node) {
696			if (current_node > (-1))
697				pr_cont("\n");
698			current_node = node;
699
700			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
701			       node_width - num_digits(node), " ", node);
702		}
703
704		/* Add padding for the BSP */
705		if (cpu == 1)
706			pr_cont("%*s", width + 1, " ");
707
708		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
709
710	} else
711		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
712			node, cpu, apicid);
713}
714
715static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
716{
717	int cpu;
718
719	cpu = smp_processor_id();
720	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
721		return NMI_HANDLED;
722
723	return NMI_DONE;
724}
725
726/*
727 * Wake up AP by INIT, INIT, STARTUP sequence.
728 *
729 * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
730 * boot-strap code which is not a desired behavior for waking up BSP. To
731 * void the boot-strap code, wake up CPU0 by NMI instead.
732 *
733 * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
734 * (i.e. physically hot removed and then hot added), NMI won't wake it up.
735 * We'll change this code in the future to wake up hard offlined CPU0 if
736 * real platform and request are available.
737 */
738static int
739wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
740	       int *cpu0_nmi_registered)
741{
742	int id;
743	int boot_error;
744
745	preempt_disable();
746
747	/*
748	 * Wake up AP by INIT, INIT, STARTUP sequence.
749	 */
750	if (cpu) {
751		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
752		goto out;
753	}
754
755	/*
756	 * Wake up BSP by nmi.
757	 *
758	 * Register a NMI handler to help wake up CPU0.
759	 */
760	boot_error = register_nmi_handler(NMI_LOCAL,
761					  wakeup_cpu0_nmi, 0, "wake_cpu0");
762
763	if (!boot_error) {
764		enable_start_cpu0 = 1;
765		*cpu0_nmi_registered = 1;
766		if (apic->dest_logical == APIC_DEST_LOGICAL)
767			id = cpu0_logical_apicid;
768		else
769			id = apicid;
770		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
771	}
772
773out:
774	preempt_enable();
775
776	return boot_error;
777}
778
779void common_cpu_up(unsigned int cpu, struct task_struct *idle)
780{
781	/* Just in case we booted with a single CPU. */
782	alternatives_enable_smp();
783
784	per_cpu(current_task, cpu) = idle;
785
786#ifdef CONFIG_X86_32
787	/* Stack for startup_32 can be just as for start_secondary onwards */
788	irq_ctx_init(cpu);
789	per_cpu(cpu_current_top_of_stack, cpu) =
790		(unsigned long)task_stack_page(idle) + THREAD_SIZE;
791#else
792	clear_tsk_thread_flag(idle, TIF_FORK);
793	initial_gs = per_cpu_offset(cpu);
794#endif
795	per_cpu(kernel_stack, cpu) =
796		(unsigned long)task_stack_page(idle) + THREAD_SIZE;
797}
798
799/*
800 * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
801 * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
802 * Returns zero if CPU booted OK, else error code from
803 * ->wakeup_secondary_cpu.
804 */
805static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
806{
807	volatile u32 *trampoline_status =
808		(volatile u32 *) __va(real_mode_header->trampoline_status);
809	/* start_ip had better be page-aligned! */
810	unsigned long start_ip = real_mode_header->trampoline_start;
811
812	unsigned long boot_error = 0;
813	int cpu0_nmi_registered = 0;
814	unsigned long timeout;
815
816	idle->thread.sp = (unsigned long) (((struct pt_regs *)
817			  (THREAD_SIZE +  task_stack_page(idle))) - 1);
818
819	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
820	initial_code = (unsigned long)start_secondary;
821	stack_start  = idle->thread.sp;
822
823	/* So we see what's up */
824	announce_cpu(cpu, apicid);
825
826	/*
827	 * This grunge runs the startup process for
828	 * the targeted processor.
829	 */
830
831	atomic_set(&init_deasserted, 0);
832
833	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
834
835		pr_debug("Setting warm reset code and vector.\n");
836
837		smpboot_setup_warm_reset_vector(start_ip);
838		/*
839		 * Be paranoid about clearing APIC errors.
840		*/
841		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
842			apic_write(APIC_ESR, 0);
843			apic_read(APIC_ESR);
844		}
845	}
846
847	/*
848	 * AP might wait on cpu_callout_mask in cpu_init() with
849	 * cpu_initialized_mask set if previous attempt to online
850	 * it timed-out. Clear cpu_initialized_mask so that after
851	 * INIT/SIPI it could start with a clean state.
852	 */
853	cpumask_clear_cpu(cpu, cpu_initialized_mask);
854	smp_mb();
855
856	/*
857	 * Wake up a CPU in difference cases:
858	 * - Use the method in the APIC driver if it's defined
859	 * Otherwise,
860	 * - Use an INIT boot APIC message for APs or NMI for BSP.
861	 */
862	if (apic->wakeup_secondary_cpu)
863		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
864	else
865		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
866						     &cpu0_nmi_registered);
867
868	if (!boot_error) {
869		/*
870		 * Wait 10s total for a response from AP
871		 */
872		boot_error = -1;
873		timeout = jiffies + 10*HZ;
874		while (time_before(jiffies, timeout)) {
875			if (cpumask_test_cpu(cpu, cpu_initialized_mask)) {
876				/*
877				 * Tell AP to proceed with initialization
878				 */
879				cpumask_set_cpu(cpu, cpu_callout_mask);
880				boot_error = 0;
881				break;
882			}
883			udelay(100);
884			schedule();
885		}
886	}
887
888	if (!boot_error) {
889		/*
890		 * Wait till AP completes initial initialization
891		 */
892		while (!cpumask_test_cpu(cpu, cpu_callin_mask)) {
893			/*
894			 * Allow other tasks to run while we wait for the
895			 * AP to come online. This also gives a chance
896			 * for the MTRR work(triggered by the AP coming online)
897			 * to be completed in the stop machine context.
898			 */
899			udelay(100);
900			schedule();
901		}
902	}
903
904	/* mark "stuck" area as not stuck */
905	*trampoline_status = 0;
906
907	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
908		/*
909		 * Cleanup possible dangling ends...
910		 */
911		smpboot_restore_warm_reset_vector();
912	}
913	/*
914	 * Clean up the nmi handler. Do this after the callin and callout sync
915	 * to avoid impact of possible long unregister time.
916	 */
917	if (cpu0_nmi_registered)
918		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
919
920	return boot_error;
921}
922
923int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
924{
925	int apicid = apic->cpu_present_to_apicid(cpu);
926	unsigned long flags;
927	int err;
928
929	WARN_ON(irqs_disabled());
930
931	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
932
933	if (apicid == BAD_APICID ||
934	    !physid_isset(apicid, phys_cpu_present_map) ||
935	    !apic->apic_id_valid(apicid)) {
936		pr_err("%s: bad cpu %d\n", __func__, cpu);
937		return -EINVAL;
938	}
939
940	/*
941	 * Already booted CPU?
942	 */
943	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
944		pr_debug("do_boot_cpu %d Already started\n", cpu);
945		return -ENOSYS;
946	}
947
948	/*
949	 * Save current MTRR state in case it was changed since early boot
950	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
951	 */
952	mtrr_save_state();
953
954	/* x86 CPUs take themselves offline, so delayed offline is OK. */
955	err = cpu_check_up_prepare(cpu);
956	if (err && err != -EBUSY)
957		return err;
958
959	/* the FPU context is blank, nobody can own it */
960	__cpu_disable_lazy_restore(cpu);
961
962	common_cpu_up(cpu, tidle);
963
964	err = do_boot_cpu(apicid, cpu, tidle);
965	if (err) {
966		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
967		return -EIO;
968	}
969
970	/*
971	 * Check TSC synchronization with the AP (keep irqs disabled
972	 * while doing so):
973	 */
974	local_irq_save(flags);
975	check_tsc_sync_source(cpu);
976	local_irq_restore(flags);
977
978	while (!cpu_online(cpu)) {
979		cpu_relax();
980		touch_nmi_watchdog();
981	}
982
983	return 0;
984}
985
986/**
987 * arch_disable_smp_support() - disables SMP support for x86 at runtime
988 */
989void arch_disable_smp_support(void)
990{
991	disable_ioapic_support();
992}
993
994/*
995 * Fall back to non SMP mode after errors.
996 *
997 * RED-PEN audit/test this more. I bet there is more state messed up here.
998 */
999static __init void disable_smp(void)
1000{
1001	pr_info("SMP disabled\n");
1002
1003	disable_ioapic_support();
1004
1005	init_cpu_present(cpumask_of(0));
1006	init_cpu_possible(cpumask_of(0));
1007
1008	if (smp_found_config)
1009		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1010	else
1011		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1012	cpumask_set_cpu(0, cpu_sibling_mask(0));
1013	cpumask_set_cpu(0, cpu_core_mask(0));
1014}
1015
1016enum {
1017	SMP_OK,
1018	SMP_NO_CONFIG,
1019	SMP_NO_APIC,
1020	SMP_FORCE_UP,
1021};
1022
1023/*
1024 * Various sanity checks.
1025 */
1026static int __init smp_sanity_check(unsigned max_cpus)
1027{
1028	preempt_disable();
1029
1030#if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
1031	if (def_to_bigsmp && nr_cpu_ids > 8) {
1032		unsigned int cpu;
1033		unsigned nr;
1034
1035		pr_warn("More than 8 CPUs detected - skipping them\n"
1036			"Use CONFIG_X86_BIGSMP\n");
1037
1038		nr = 0;
1039		for_each_present_cpu(cpu) {
1040			if (nr >= 8)
1041				set_cpu_present(cpu, false);
1042			nr++;
1043		}
1044
1045		nr = 0;
1046		for_each_possible_cpu(cpu) {
1047			if (nr >= 8)
1048				set_cpu_possible(cpu, false);
1049			nr++;
1050		}
1051
1052		nr_cpu_ids = 8;
1053	}
1054#endif
1055
1056	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1057		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1058			hard_smp_processor_id());
1059
1060		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1061	}
1062
1063	/*
1064	 * If we couldn't find an SMP configuration at boot time,
1065	 * get out of here now!
1066	 */
1067	if (!smp_found_config && !acpi_lapic) {
1068		preempt_enable();
1069		pr_notice("SMP motherboard not detected\n");
1070		return SMP_NO_CONFIG;
1071	}
1072
1073	/*
1074	 * Should not be necessary because the MP table should list the boot
1075	 * CPU too, but we do it for the sake of robustness anyway.
1076	 */
1077	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1078		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1079			  boot_cpu_physical_apicid);
1080		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1081	}
1082	preempt_enable();
1083
1084	/*
1085	 * If we couldn't find a local APIC, then get out of here now!
1086	 */
1087	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1088	    !cpu_has_apic) {
1089		if (!disable_apic) {
1090			pr_err("BIOS bug, local APIC #%d not detected!...\n",
1091				boot_cpu_physical_apicid);
1092			pr_err("... forcing use of dummy APIC emulation (tell your hw vendor)\n");
1093		}
1094		return SMP_NO_APIC;
1095	}
1096
1097	/*
1098	 * If SMP should be disabled, then really disable it!
1099	 */
1100	if (!max_cpus) {
1101		pr_info("SMP mode deactivated\n");
1102		return SMP_FORCE_UP;
1103	}
1104
1105	return SMP_OK;
1106}
1107
1108static void __init smp_cpu_index_default(void)
1109{
1110	int i;
1111	struct cpuinfo_x86 *c;
1112
1113	for_each_possible_cpu(i) {
1114		c = &cpu_data(i);
1115		/* mark all to hotplug */
1116		c->cpu_index = nr_cpu_ids;
1117	}
1118}
1119
1120/*
1121 * Prepare for SMP bootup.  The MP table or ACPI has been read
1122 * earlier.  Just do some sanity checking here and enable APIC mode.
1123 */
1124void __init native_smp_prepare_cpus(unsigned int max_cpus)
1125{
1126	unsigned int i;
1127
1128	smp_cpu_index_default();
1129
1130	/*
1131	 * Setup boot CPU information
1132	 */
1133	smp_store_boot_cpu_info(); /* Final full version of the data */
1134	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1135	mb();
1136
1137	current_thread_info()->cpu = 0;  /* needed? */
1138	for_each_possible_cpu(i) {
1139		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1140		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1141		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1142	}
1143	set_cpu_sibling_map(0);
1144
1145	switch (smp_sanity_check(max_cpus)) {
1146	case SMP_NO_CONFIG:
1147		disable_smp();
1148		if (APIC_init_uniprocessor())
1149			pr_notice("Local APIC not detected. Using dummy APIC emulation.\n");
1150		return;
1151	case SMP_NO_APIC:
1152		disable_smp();
1153		return;
1154	case SMP_FORCE_UP:
1155		disable_smp();
1156		apic_bsp_setup(false);
1157		return;
1158	case SMP_OK:
1159		break;
1160	}
1161
1162	default_setup_apic_routing();
1163
1164	if (read_apic_id() != boot_cpu_physical_apicid) {
1165		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1166		     read_apic_id(), boot_cpu_physical_apicid);
1167		/* Or can we switch back to PIC here? */
1168	}
1169
1170	cpu0_logical_apicid = apic_bsp_setup(false);
1171
1172	pr_info("CPU%d: ", 0);
1173	print_cpu_info(&cpu_data(0));
1174
1175	if (is_uv_system())
1176		uv_system_init();
1177
1178	set_mtrr_aps_delayed_init();
1179}
1180
1181void arch_enable_nonboot_cpus_begin(void)
1182{
1183	set_mtrr_aps_delayed_init();
1184}
1185
1186void arch_enable_nonboot_cpus_end(void)
1187{
1188	mtrr_aps_init();
1189}
1190
1191/*
1192 * Early setup to make printk work.
1193 */
1194void __init native_smp_prepare_boot_cpu(void)
1195{
1196	int me = smp_processor_id();
1197	switch_to_new_gdt(me);
1198	/* already set me in cpu_online_mask in boot_cpu_init() */
1199	cpumask_set_cpu(me, cpu_callout_mask);
1200	cpu_set_state_online(me);
1201}
1202
1203void __init native_smp_cpus_done(unsigned int max_cpus)
1204{
1205	pr_debug("Boot done\n");
1206
1207	nmi_selftest();
1208	impress_friends();
1209	setup_ioapic_dest();
1210	mtrr_aps_init();
1211}
1212
1213static int __initdata setup_possible_cpus = -1;
1214static int __init _setup_possible_cpus(char *str)
1215{
1216	get_option(&str, &setup_possible_cpus);
1217	return 0;
1218}
1219early_param("possible_cpus", _setup_possible_cpus);
1220
1221
1222/*
1223 * cpu_possible_mask should be static, it cannot change as cpu's
1224 * are onlined, or offlined. The reason is per-cpu data-structures
1225 * are allocated by some modules at init time, and dont expect to
1226 * do this dynamically on cpu arrival/departure.
1227 * cpu_present_mask on the other hand can change dynamically.
1228 * In case when cpu_hotplug is not compiled, then we resort to current
1229 * behaviour, which is cpu_possible == cpu_present.
1230 * - Ashok Raj
1231 *
1232 * Three ways to find out the number of additional hotplug CPUs:
1233 * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1234 * - The user can overwrite it with possible_cpus=NUM
1235 * - Otherwise don't reserve additional CPUs.
1236 * We do this because additional CPUs waste a lot of memory.
1237 * -AK
1238 */
1239__init void prefill_possible_map(void)
1240{
1241	int i, possible;
1242
1243	/* no processor from mptable or madt */
1244	if (!num_processors)
1245		num_processors = 1;
1246
1247	i = setup_max_cpus ?: 1;
1248	if (setup_possible_cpus == -1) {
1249		possible = num_processors;
1250#ifdef CONFIG_HOTPLUG_CPU
1251		if (setup_max_cpus)
1252			possible += disabled_cpus;
1253#else
1254		if (possible > i)
1255			possible = i;
1256#endif
1257	} else
1258		possible = setup_possible_cpus;
1259
1260	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1261
1262	/* nr_cpu_ids could be reduced via nr_cpus= */
1263	if (possible > nr_cpu_ids) {
1264		pr_warn("%d Processors exceeds NR_CPUS limit of %d\n",
1265			possible, nr_cpu_ids);
1266		possible = nr_cpu_ids;
1267	}
1268
1269#ifdef CONFIG_HOTPLUG_CPU
1270	if (!setup_max_cpus)
1271#endif
1272	if (possible > i) {
1273		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1274			possible, setup_max_cpus);
1275		possible = i;
1276	}
1277
1278	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1279		possible, max_t(int, possible - num_processors, 0));
1280
1281	for (i = 0; i < possible; i++)
1282		set_cpu_possible(i, true);
1283	for (; i < NR_CPUS; i++)
1284		set_cpu_possible(i, false);
1285
1286	nr_cpu_ids = possible;
1287}
1288
1289#ifdef CONFIG_HOTPLUG_CPU
1290
1291static void remove_siblinginfo(int cpu)
1292{
1293	int sibling;
1294	struct cpuinfo_x86 *c = &cpu_data(cpu);
1295
1296	for_each_cpu(sibling, cpu_core_mask(cpu)) {
1297		cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1298		/*/
1299		 * last thread sibling in this cpu core going down
1300		 */
1301		if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1302			cpu_data(sibling).booted_cores--;
1303	}
1304
1305	for_each_cpu(sibling, cpu_sibling_mask(cpu))
1306		cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1307	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1308		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1309	cpumask_clear(cpu_llc_shared_mask(cpu));
1310	cpumask_clear(cpu_sibling_mask(cpu));
1311	cpumask_clear(cpu_core_mask(cpu));
1312	c->phys_proc_id = 0;
1313	c->cpu_core_id = 0;
1314	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1315}
1316
1317static void __ref remove_cpu_from_maps(int cpu)
1318{
1319	set_cpu_online(cpu, false);
1320	cpumask_clear_cpu(cpu, cpu_callout_mask);
1321	cpumask_clear_cpu(cpu, cpu_callin_mask);
1322	/* was set by cpu_init() */
1323	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1324	numa_remove_cpu(cpu);
1325}
1326
1327void cpu_disable_common(void)
1328{
1329	int cpu = smp_processor_id();
1330
1331	remove_siblinginfo(cpu);
1332
1333	/* It's now safe to remove this processor from the online map */
1334	lock_vector_lock();
1335	remove_cpu_from_maps(cpu);
1336	unlock_vector_lock();
1337	fixup_irqs();
1338}
1339
1340int native_cpu_disable(void)
1341{
1342	int ret;
1343
1344	ret = check_irq_vectors_for_cpu_disable();
1345	if (ret)
1346		return ret;
1347
1348	clear_local_APIC();
1349	cpu_disable_common();
1350
1351	return 0;
1352}
1353
1354int common_cpu_die(unsigned int cpu)
1355{
1356	int ret = 0;
1357
1358	/* We don't do anything here: idle task is faking death itself. */
1359
1360	/* They ack this in play_dead() by setting CPU_DEAD */
1361	if (cpu_wait_death(cpu, 5)) {
1362		if (system_state == SYSTEM_RUNNING)
1363			pr_info("CPU %u is now offline\n", cpu);
1364	} else {
1365		pr_err("CPU %u didn't die...\n", cpu);
1366		ret = -1;
1367	}
1368
1369	return ret;
1370}
1371
1372void native_cpu_die(unsigned int cpu)
1373{
1374	common_cpu_die(cpu);
1375}
1376
1377void play_dead_common(void)
1378{
1379	idle_task_exit();
1380	reset_lazy_tlbstate();
1381	amd_e400_remove_cpu(raw_smp_processor_id());
1382
1383	/* Ack it */
1384	(void)cpu_report_death();
1385
1386	/*
1387	 * With physical CPU hotplug, we should halt the cpu
1388	 */
1389	local_irq_disable();
1390}
1391
1392static bool wakeup_cpu0(void)
1393{
1394	if (smp_processor_id() == 0 && enable_start_cpu0)
1395		return true;
1396
1397	return false;
1398}
1399
1400/*
1401 * We need to flush the caches before going to sleep, lest we have
1402 * dirty data in our caches when we come back up.
1403 */
1404static inline void mwait_play_dead(void)
1405{
1406	unsigned int eax, ebx, ecx, edx;
1407	unsigned int highest_cstate = 0;
1408	unsigned int highest_subcstate = 0;
1409	void *mwait_ptr;
1410	int i;
1411
1412	if (!this_cpu_has(X86_FEATURE_MWAIT))
1413		return;
1414	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1415		return;
1416	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1417		return;
1418
1419	eax = CPUID_MWAIT_LEAF;
1420	ecx = 0;
1421	native_cpuid(&eax, &ebx, &ecx, &edx);
1422
1423	/*
1424	 * eax will be 0 if EDX enumeration is not valid.
1425	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1426	 */
1427	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1428		eax = 0;
1429	} else {
1430		edx >>= MWAIT_SUBSTATE_SIZE;
1431		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1432			if (edx & MWAIT_SUBSTATE_MASK) {
1433				highest_cstate = i;
1434				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1435			}
1436		}
1437		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1438			(highest_subcstate - 1);
1439	}
1440
1441	/*
1442	 * This should be a memory location in a cache line which is
1443	 * unlikely to be touched by other processors.  The actual
1444	 * content is immaterial as it is not actually modified in any way.
1445	 */
1446	mwait_ptr = &current_thread_info()->flags;
1447
1448	wbinvd();
1449
1450	while (1) {
1451		/*
1452		 * The CLFLUSH is a workaround for erratum AAI65 for
1453		 * the Xeon 7400 series.  It's not clear it is actually
1454		 * needed, but it should be harmless in either case.
1455		 * The WBINVD is insufficient due to the spurious-wakeup
1456		 * case where we return around the loop.
1457		 */
1458		mb();
1459		clflush(mwait_ptr);
1460		mb();
1461		__monitor(mwait_ptr, 0, 0);
1462		mb();
1463		__mwait(eax, 0);
1464		/*
1465		 * If NMI wants to wake up CPU0, start CPU0.
1466		 */
1467		if (wakeup_cpu0())
1468			start_cpu0();
1469	}
1470}
1471
1472static inline void hlt_play_dead(void)
1473{
1474	if (__this_cpu_read(cpu_info.x86) >= 4)
1475		wbinvd();
1476
1477	while (1) {
1478		native_halt();
1479		/*
1480		 * If NMI wants to wake up CPU0, start CPU0.
1481		 */
1482		if (wakeup_cpu0())
1483			start_cpu0();
1484	}
1485}
1486
1487void native_play_dead(void)
1488{
1489	play_dead_common();
1490	tboot_shutdown(TB_SHUTDOWN_WFS);
1491
1492	mwait_play_dead();	/* Only returns on failure */
1493	if (cpuidle_play_dead())
1494		hlt_play_dead();
1495}
1496
1497#else /* ... !CONFIG_HOTPLUG_CPU */
1498int native_cpu_disable(void)
1499{
1500	return -ENOSYS;
1501}
1502
1503void native_cpu_die(unsigned int cpu)
1504{
1505	/* We said "no" in __cpu_disable */
1506	BUG();
1507}
1508
1509void native_play_dead(void)
1510{
1511	BUG();
1512}
1513
1514#endif
1515