xref: /linux/arch/x86/kernel/smpboot.c (revision 005438a8eef063495ac059d128eea71b58de50e5)
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/fpu/internal.h>
72 #include <asm/setup.h>
73 #include <asm/uv/uv.h>
74 #include <linux/mc146818rtc.h>
75 #include <asm/i8259.h>
76 #include <asm/realmode.h>
77 #include <asm/misc.h>
78 
79 /* Number of siblings per CPU package */
80 int smp_num_siblings = 1;
81 EXPORT_SYMBOL(smp_num_siblings);
82 
83 /* Last level cache ID of each logical CPU */
84 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
85 
86 /* representing HT siblings of each logical CPU */
87 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
88 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
89 
90 /* representing HT and core siblings of each logical CPU */
91 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_map);
92 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
93 
94 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_llc_shared_map);
95 
96 /* Per CPU bogomips and other parameters */
97 DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
98 EXPORT_PER_CPU_SYMBOL(cpu_info);
99 
100 atomic_t init_deasserted;
101 
102 static inline void smpboot_setup_warm_reset_vector(unsigned long start_eip)
103 {
104 	unsigned long flags;
105 
106 	spin_lock_irqsave(&rtc_lock, flags);
107 	CMOS_WRITE(0xa, 0xf);
108 	spin_unlock_irqrestore(&rtc_lock, flags);
109 	local_flush_tlb();
110 	pr_debug("1.\n");
111 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_HIGH)) =
112 							start_eip >> 4;
113 	pr_debug("2.\n");
114 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) =
115 							start_eip & 0xf;
116 	pr_debug("3.\n");
117 }
118 
119 static inline void smpboot_restore_warm_reset_vector(void)
120 {
121 	unsigned long flags;
122 
123 	/*
124 	 * Install writable page 0 entry to set BIOS data area.
125 	 */
126 	local_flush_tlb();
127 
128 	/*
129 	 * Paranoid:  Set warm reset code and vector here back
130 	 * to default values.
131 	 */
132 	spin_lock_irqsave(&rtc_lock, flags);
133 	CMOS_WRITE(0, 0xf);
134 	spin_unlock_irqrestore(&rtc_lock, flags);
135 
136 	*((volatile u32 *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = 0;
137 }
138 
139 /*
140  * Report back to the Boot Processor during boot time or to the caller processor
141  * during CPU online.
142  */
143 static void smp_callin(void)
144 {
145 	int cpuid, phys_id;
146 
147 	/*
148 	 * If waken up by an INIT in an 82489DX configuration
149 	 * we may get here before an INIT-deassert IPI reaches
150 	 * our local APIC.  We have to wait for the IPI or we'll
151 	 * lock up on an APIC access.
152 	 *
153 	 * Since CPU0 is not wakened up by INIT, it doesn't wait for the IPI.
154 	 */
155 	cpuid = smp_processor_id();
156 	if (apic->wait_for_init_deassert && cpuid)
157 		while (!atomic_read(&init_deasserted))
158 			cpu_relax();
159 
160 	/*
161 	 * (This works even if the APIC is not enabled.)
162 	 */
163 	phys_id = read_apic_id();
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 	apic_ap_setup();
172 
173 	/*
174 	 * Need to setup vector mappings before we enable interrupts.
175 	 */
176 	setup_vector_irq(smp_processor_id());
177 
178 	/*
179 	 * Save our processor parameters. Note: this information
180 	 * is needed for clock calibration.
181 	 */
182 	smp_store_cpu_info(cpuid);
183 
184 	/*
185 	 * Get our bogomips.
186 	 * Update loops_per_jiffy in cpu_data. Previous call to
187 	 * smp_store_cpu_info() stored a value that is close but not as
188 	 * accurate as the value just calculated.
189 	 */
190 	calibrate_delay();
191 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
192 	pr_debug("Stack at about %p\n", &cpuid);
193 
194 	/*
195 	 * This must be done before setting cpu_online_mask
196 	 * or calling notify_cpu_starting.
197 	 */
198 	set_cpu_sibling_map(raw_smp_processor_id());
199 	wmb();
200 
201 	notify_cpu_starting(cpuid);
202 
203 	/*
204 	 * Allow the master to continue.
205 	 */
206 	cpumask_set_cpu(cpuid, cpu_callin_mask);
207 }
208 
209 static int cpu0_logical_apicid;
210 static int enable_start_cpu0;
211 /*
212  * Activate a secondary processor.
213  */
214 static void notrace start_secondary(void *unused)
215 {
216 	/*
217 	 * Don't put *anything* before cpu_init(), SMP booting is too
218 	 * fragile that we want to limit the things done here to the
219 	 * most necessary things.
220 	 */
221 	cpu_init();
222 	x86_cpuinit.early_percpu_clock_init();
223 	preempt_disable();
224 	smp_callin();
225 
226 	enable_start_cpu0 = 0;
227 
228 #ifdef CONFIG_X86_32
229 	/* switch away from the initial page table */
230 	load_cr3(swapper_pg_dir);
231 	__flush_tlb_all();
232 #endif
233 
234 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
235 	barrier();
236 	/*
237 	 * Check TSC synchronization with the BP:
238 	 */
239 	check_tsc_sync_target();
240 
241 	/*
242 	 * Enable the espfix hack for this CPU
243 	 */
244 #ifdef CONFIG_X86_ESPFIX64
245 	init_espfix_ap();
246 #endif
247 
248 	/*
249 	 * We need to hold vector_lock so there the set of online cpus
250 	 * does not change while we are assigning vectors to cpus.  Holding
251 	 * this lock ensures we don't half assign or remove an irq from a cpu.
252 	 */
253 	lock_vector_lock();
254 	set_cpu_online(smp_processor_id(), true);
255 	unlock_vector_lock();
256 	cpu_set_state_online(smp_processor_id());
257 	x86_platform.nmi_init();
258 
259 	/* enable local interrupts */
260 	local_irq_enable();
261 
262 	/* to prevent fake stack check failure in clock setup */
263 	boot_init_stack_canary();
264 
265 	x86_cpuinit.setup_percpu_clockev();
266 
267 	wmb();
268 	cpu_startup_entry(CPUHP_ONLINE);
269 }
270 
271 void __init smp_store_boot_cpu_info(void)
272 {
273 	int id = 0; /* CPU 0 */
274 	struct cpuinfo_x86 *c = &cpu_data(id);
275 
276 	*c = boot_cpu_data;
277 	c->cpu_index = id;
278 }
279 
280 /*
281  * The bootstrap kernel entry code has set these up. Save them for
282  * a given CPU
283  */
284 void smp_store_cpu_info(int id)
285 {
286 	struct cpuinfo_x86 *c = &cpu_data(id);
287 
288 	*c = boot_cpu_data;
289 	c->cpu_index = id;
290 	/*
291 	 * During boot time, CPU0 has this setup already. Save the info when
292 	 * bringing up AP or offlined CPU0.
293 	 */
294 	identify_secondary_cpu(c);
295 }
296 
297 static bool
298 topology_same_node(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
299 {
300 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
301 
302 	return (cpu_to_node(cpu1) == cpu_to_node(cpu2));
303 }
304 
305 static bool
306 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
307 {
308 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
309 
310 	return !WARN_ONCE(!topology_same_node(c, o),
311 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
312 		"[node: %d != %d]. Ignoring dependency.\n",
313 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
314 }
315 
316 #define link_mask(mfunc, c1, c2)					\
317 do {									\
318 	cpumask_set_cpu((c1), mfunc(c2));				\
319 	cpumask_set_cpu((c2), mfunc(c1));				\
320 } while (0)
321 
322 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
323 {
324 	if (cpu_has_topoext) {
325 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
326 
327 		if (c->phys_proc_id == o->phys_proc_id &&
328 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2) &&
329 		    c->compute_unit_id == o->compute_unit_id)
330 			return topology_sane(c, o, "smt");
331 
332 	} else if (c->phys_proc_id == o->phys_proc_id &&
333 		   c->cpu_core_id == o->cpu_core_id) {
334 		return topology_sane(c, o, "smt");
335 	}
336 
337 	return false;
338 }
339 
340 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
341 {
342 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
343 
344 	if (per_cpu(cpu_llc_id, cpu1) != BAD_APICID &&
345 	    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2))
346 		return topology_sane(c, o, "llc");
347 
348 	return false;
349 }
350 
351 /*
352  * Unlike the other levels, we do not enforce keeping a
353  * multicore group inside a NUMA node.  If this happens, we will
354  * discard the MC level of the topology later.
355  */
356 static bool match_die(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
357 {
358 	if (c->phys_proc_id == o->phys_proc_id)
359 		return true;
360 	return false;
361 }
362 
363 static struct sched_domain_topology_level numa_inside_package_topology[] = {
364 #ifdef CONFIG_SCHED_SMT
365 	{ cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
366 #endif
367 #ifdef CONFIG_SCHED_MC
368 	{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
369 #endif
370 	{ NULL, },
371 };
372 /*
373  * set_sched_topology() sets the topology internal to a CPU.  The
374  * NUMA topologies are layered on top of it to build the full
375  * system topology.
376  *
377  * If NUMA nodes are observed to occur within a CPU package, this
378  * function should be called.  It forces the sched domain code to
379  * only use the SMT level for the CPU portion of the topology.
380  * This essentially falls back to relying on NUMA information
381  * from the SRAT table to describe the entire system topology
382  * (except for hyperthreads).
383  */
384 static void primarily_use_numa_for_topology(void)
385 {
386 	set_sched_topology(numa_inside_package_topology);
387 }
388 
389 void set_cpu_sibling_map(int cpu)
390 {
391 	bool has_smt = smp_num_siblings > 1;
392 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
393 	struct cpuinfo_x86 *c = &cpu_data(cpu);
394 	struct cpuinfo_x86 *o;
395 	int i;
396 
397 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
398 
399 	if (!has_mp) {
400 		cpumask_set_cpu(cpu, topology_sibling_cpumask(cpu));
401 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
402 		cpumask_set_cpu(cpu, topology_core_cpumask(cpu));
403 		c->booted_cores = 1;
404 		return;
405 	}
406 
407 	for_each_cpu(i, cpu_sibling_setup_mask) {
408 		o = &cpu_data(i);
409 
410 		if ((i == cpu) || (has_smt && match_smt(c, o)))
411 			link_mask(topology_sibling_cpumask, cpu, i);
412 
413 		if ((i == cpu) || (has_mp && match_llc(c, o)))
414 			link_mask(cpu_llc_shared_mask, cpu, i);
415 
416 	}
417 
418 	/*
419 	 * This needs a separate iteration over the cpus because we rely on all
420 	 * topology_sibling_cpumask links to be set-up.
421 	 */
422 	for_each_cpu(i, cpu_sibling_setup_mask) {
423 		o = &cpu_data(i);
424 
425 		if ((i == cpu) || (has_mp && match_die(c, o))) {
426 			link_mask(topology_core_cpumask, cpu, i);
427 
428 			/*
429 			 *  Does this new cpu bringup a new core?
430 			 */
431 			if (cpumask_weight(
432 			    topology_sibling_cpumask(cpu)) == 1) {
433 				/*
434 				 * for each core in package, increment
435 				 * the booted_cores for this new cpu
436 				 */
437 				if (cpumask_first(
438 				    topology_sibling_cpumask(i)) == i)
439 					c->booted_cores++;
440 				/*
441 				 * increment the core count for all
442 				 * the other cpus in this package
443 				 */
444 				if (i != cpu)
445 					cpu_data(i).booted_cores++;
446 			} else if (i != cpu && !c->booted_cores)
447 				c->booted_cores = cpu_data(i).booted_cores;
448 		}
449 		if (match_die(c, o) && !topology_same_node(c, o))
450 			primarily_use_numa_for_topology();
451 	}
452 }
453 
454 /* maps the cpu to the sched domain representing multi-core */
455 const struct cpumask *cpu_coregroup_mask(int cpu)
456 {
457 	return cpu_llc_shared_mask(cpu);
458 }
459 
460 static void impress_friends(void)
461 {
462 	int cpu;
463 	unsigned long bogosum = 0;
464 	/*
465 	 * Allow the user to impress friends.
466 	 */
467 	pr_debug("Before bogomips\n");
468 	for_each_possible_cpu(cpu)
469 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
470 			bogosum += cpu_data(cpu).loops_per_jiffy;
471 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
472 		num_online_cpus(),
473 		bogosum/(500000/HZ),
474 		(bogosum/(5000/HZ))%100);
475 
476 	pr_debug("Before bogocount - setting activated=1\n");
477 }
478 
479 void __inquire_remote_apic(int apicid)
480 {
481 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
482 	const char * const names[] = { "ID", "VERSION", "SPIV" };
483 	int timeout;
484 	u32 status;
485 
486 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
487 
488 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
489 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
490 
491 		/*
492 		 * Wait for idle.
493 		 */
494 		status = safe_apic_wait_icr_idle();
495 		if (status)
496 			pr_cont("a previous APIC delivery may have failed\n");
497 
498 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
499 
500 		timeout = 0;
501 		do {
502 			udelay(100);
503 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
504 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
505 
506 		switch (status) {
507 		case APIC_ICR_RR_VALID:
508 			status = apic_read(APIC_RRR);
509 			pr_cont("%08x\n", status);
510 			break;
511 		default:
512 			pr_cont("failed\n");
513 		}
514 	}
515 }
516 
517 /*
518  * The Multiprocessor Specification 1.4 (1997) example code suggests
519  * that there should be a 10ms delay between the BSP asserting INIT
520  * and de-asserting INIT, when starting a remote processor.
521  * But that slows boot and resume on modern processors, which include
522  * many cores and don't require that delay.
523  *
524  * Cmdline "init_cpu_udelay=" is available to over-ride this delay.
525  * Modern processor families are quirked to remove the delay entirely.
526  */
527 #define UDELAY_10MS_DEFAULT 10000
528 
529 static unsigned int init_udelay = UDELAY_10MS_DEFAULT;
530 
531 static int __init cpu_init_udelay(char *str)
532 {
533 	get_option(&str, &init_udelay);
534 
535 	return 0;
536 }
537 early_param("cpu_init_udelay", cpu_init_udelay);
538 
539 static void __init smp_quirk_init_udelay(void)
540 {
541 	/* if cmdline changed it from default, leave it alone */
542 	if (init_udelay != UDELAY_10MS_DEFAULT)
543 		return;
544 
545 	/* if modern processor, use no delay */
546 	if (((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && (boot_cpu_data.x86 == 6)) ||
547 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && (boot_cpu_data.x86 >= 0xF)))
548 		init_udelay = 0;
549 }
550 
551 /*
552  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
553  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
554  * won't ... remember to clear down the APIC, etc later.
555  */
556 int
557 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
558 {
559 	unsigned long send_status, accept_status = 0;
560 	int maxlvt;
561 
562 	/* Target chip */
563 	/* Boot on the stack */
564 	/* Kick the second */
565 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, apicid);
566 
567 	pr_debug("Waiting for send to finish...\n");
568 	send_status = safe_apic_wait_icr_idle();
569 
570 	/*
571 	 * Give the other CPU some time to accept the IPI.
572 	 */
573 	udelay(200);
574 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
575 		maxlvt = lapic_get_maxlvt();
576 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
577 			apic_write(APIC_ESR, 0);
578 		accept_status = (apic_read(APIC_ESR) & 0xEF);
579 	}
580 	pr_debug("NMI sent\n");
581 
582 	if (send_status)
583 		pr_err("APIC never delivered???\n");
584 	if (accept_status)
585 		pr_err("APIC delivery error (%lx)\n", accept_status);
586 
587 	return (send_status | accept_status);
588 }
589 
590 static int
591 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
592 {
593 	unsigned long send_status = 0, accept_status = 0;
594 	int maxlvt, num_starts, j;
595 
596 	maxlvt = lapic_get_maxlvt();
597 
598 	/*
599 	 * Be paranoid about clearing APIC errors.
600 	 */
601 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
602 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
603 			apic_write(APIC_ESR, 0);
604 		apic_read(APIC_ESR);
605 	}
606 
607 	pr_debug("Asserting INIT\n");
608 
609 	/*
610 	 * Turn INIT on target chip
611 	 */
612 	/*
613 	 * Send IPI
614 	 */
615 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
616 		       phys_apicid);
617 
618 	pr_debug("Waiting for send to finish...\n");
619 	send_status = safe_apic_wait_icr_idle();
620 
621 	udelay(init_udelay);
622 
623 	pr_debug("Deasserting INIT\n");
624 
625 	/* Target chip */
626 	/* Send IPI */
627 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
628 
629 	pr_debug("Waiting for send to finish...\n");
630 	send_status = safe_apic_wait_icr_idle();
631 
632 	mb();
633 	atomic_set(&init_deasserted, 1);
634 
635 	/*
636 	 * Should we send STARTUP IPIs ?
637 	 *
638 	 * Determine this based on the APIC version.
639 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
640 	 */
641 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
642 		num_starts = 2;
643 	else
644 		num_starts = 0;
645 
646 	/*
647 	 * Paravirt / VMI wants a startup IPI hook here to set up the
648 	 * target processor state.
649 	 */
650 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
651 			 stack_start);
652 
653 	/*
654 	 * Run STARTUP IPI loop.
655 	 */
656 	pr_debug("#startup loops: %d\n", num_starts);
657 
658 	for (j = 1; j <= num_starts; j++) {
659 		pr_debug("Sending STARTUP #%d\n", j);
660 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
661 			apic_write(APIC_ESR, 0);
662 		apic_read(APIC_ESR);
663 		pr_debug("After apic_write\n");
664 
665 		/*
666 		 * STARTUP IPI
667 		 */
668 
669 		/* Target chip */
670 		/* Boot on the stack */
671 		/* Kick the second */
672 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
673 			       phys_apicid);
674 
675 		/*
676 		 * Give the other CPU some time to accept the IPI.
677 		 */
678 		udelay(300);
679 
680 		pr_debug("Startup point 1\n");
681 
682 		pr_debug("Waiting for send to finish...\n");
683 		send_status = safe_apic_wait_icr_idle();
684 
685 		/*
686 		 * Give the other CPU some time to accept the IPI.
687 		 */
688 		udelay(200);
689 
690 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
691 			apic_write(APIC_ESR, 0);
692 		accept_status = (apic_read(APIC_ESR) & 0xEF);
693 		if (send_status || accept_status)
694 			break;
695 	}
696 	pr_debug("After Startup\n");
697 
698 	if (send_status)
699 		pr_err("APIC never delivered???\n");
700 	if (accept_status)
701 		pr_err("APIC delivery error (%lx)\n", accept_status);
702 
703 	return (send_status | accept_status);
704 }
705 
706 void smp_announce(void)
707 {
708 	int num_nodes = num_online_nodes();
709 
710 	printk(KERN_INFO "x86: Booted up %d node%s, %d CPUs\n",
711 	       num_nodes, (num_nodes > 1 ? "s" : ""), num_online_cpus());
712 }
713 
714 /* reduce the number of lines printed when booting a large cpu count system */
715 static void announce_cpu(int cpu, int apicid)
716 {
717 	static int current_node = -1;
718 	int node = early_cpu_to_node(cpu);
719 	static int width, node_width;
720 
721 	if (!width)
722 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
723 
724 	if (!node_width)
725 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
726 
727 	if (cpu == 1)
728 		printk(KERN_INFO "x86: Booting SMP configuration:\n");
729 
730 	if (system_state == SYSTEM_BOOTING) {
731 		if (node != current_node) {
732 			if (current_node > (-1))
733 				pr_cont("\n");
734 			current_node = node;
735 
736 			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
737 			       node_width - num_digits(node), " ", node);
738 		}
739 
740 		/* Add padding for the BSP */
741 		if (cpu == 1)
742 			pr_cont("%*s", width + 1, " ");
743 
744 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
745 
746 	} else
747 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
748 			node, cpu, apicid);
749 }
750 
751 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
752 {
753 	int cpu;
754 
755 	cpu = smp_processor_id();
756 	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
757 		return NMI_HANDLED;
758 
759 	return NMI_DONE;
760 }
761 
762 /*
763  * Wake up AP by INIT, INIT, STARTUP sequence.
764  *
765  * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
766  * boot-strap code which is not a desired behavior for waking up BSP. To
767  * void the boot-strap code, wake up CPU0 by NMI instead.
768  *
769  * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
770  * (i.e. physically hot removed and then hot added), NMI won't wake it up.
771  * We'll change this code in the future to wake up hard offlined CPU0 if
772  * real platform and request are available.
773  */
774 static int
775 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
776 	       int *cpu0_nmi_registered)
777 {
778 	int id;
779 	int boot_error;
780 
781 	preempt_disable();
782 
783 	/*
784 	 * Wake up AP by INIT, INIT, STARTUP sequence.
785 	 */
786 	if (cpu) {
787 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
788 		goto out;
789 	}
790 
791 	/*
792 	 * Wake up BSP by nmi.
793 	 *
794 	 * Register a NMI handler to help wake up CPU0.
795 	 */
796 	boot_error = register_nmi_handler(NMI_LOCAL,
797 					  wakeup_cpu0_nmi, 0, "wake_cpu0");
798 
799 	if (!boot_error) {
800 		enable_start_cpu0 = 1;
801 		*cpu0_nmi_registered = 1;
802 		if (apic->dest_logical == APIC_DEST_LOGICAL)
803 			id = cpu0_logical_apicid;
804 		else
805 			id = apicid;
806 		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
807 	}
808 
809 out:
810 	preempt_enable();
811 
812 	return boot_error;
813 }
814 
815 void common_cpu_up(unsigned int cpu, struct task_struct *idle)
816 {
817 	/* Just in case we booted with a single CPU. */
818 	alternatives_enable_smp();
819 
820 	per_cpu(current_task, cpu) = idle;
821 
822 #ifdef CONFIG_X86_32
823 	/* Stack for startup_32 can be just as for start_secondary onwards */
824 	irq_ctx_init(cpu);
825 	per_cpu(cpu_current_top_of_stack, cpu) =
826 		(unsigned long)task_stack_page(idle) + THREAD_SIZE;
827 #else
828 	clear_tsk_thread_flag(idle, TIF_FORK);
829 	initial_gs = per_cpu_offset(cpu);
830 #endif
831 }
832 
833 /*
834  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
835  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
836  * Returns zero if CPU booted OK, else error code from
837  * ->wakeup_secondary_cpu.
838  */
839 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
840 {
841 	volatile u32 *trampoline_status =
842 		(volatile u32 *) __va(real_mode_header->trampoline_status);
843 	/* start_ip had better be page-aligned! */
844 	unsigned long start_ip = real_mode_header->trampoline_start;
845 
846 	unsigned long boot_error = 0;
847 	int cpu0_nmi_registered = 0;
848 	unsigned long timeout;
849 
850 	idle->thread.sp = (unsigned long) (((struct pt_regs *)
851 			  (THREAD_SIZE +  task_stack_page(idle))) - 1);
852 
853 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
854 	initial_code = (unsigned long)start_secondary;
855 	stack_start  = idle->thread.sp;
856 
857 	/* So we see what's up */
858 	announce_cpu(cpu, apicid);
859 
860 	/*
861 	 * This grunge runs the startup process for
862 	 * the targeted processor.
863 	 */
864 
865 	atomic_set(&init_deasserted, 0);
866 
867 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
868 
869 		pr_debug("Setting warm reset code and vector.\n");
870 
871 		smpboot_setup_warm_reset_vector(start_ip);
872 		/*
873 		 * Be paranoid about clearing APIC errors.
874 		*/
875 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
876 			apic_write(APIC_ESR, 0);
877 			apic_read(APIC_ESR);
878 		}
879 	}
880 
881 	/*
882 	 * AP might wait on cpu_callout_mask in cpu_init() with
883 	 * cpu_initialized_mask set if previous attempt to online
884 	 * it timed-out. Clear cpu_initialized_mask so that after
885 	 * INIT/SIPI it could start with a clean state.
886 	 */
887 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
888 	smp_mb();
889 
890 	/*
891 	 * Wake up a CPU in difference cases:
892 	 * - Use the method in the APIC driver if it's defined
893 	 * Otherwise,
894 	 * - Use an INIT boot APIC message for APs or NMI for BSP.
895 	 */
896 	if (apic->wakeup_secondary_cpu)
897 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
898 	else
899 		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
900 						     &cpu0_nmi_registered);
901 
902 	if (!boot_error) {
903 		/*
904 		 * Wait 10s total for a response from AP
905 		 */
906 		boot_error = -1;
907 		timeout = jiffies + 10*HZ;
908 		while (time_before(jiffies, timeout)) {
909 			if (cpumask_test_cpu(cpu, cpu_initialized_mask)) {
910 				/*
911 				 * Tell AP to proceed with initialization
912 				 */
913 				cpumask_set_cpu(cpu, cpu_callout_mask);
914 				boot_error = 0;
915 				break;
916 			}
917 			udelay(100);
918 			schedule();
919 		}
920 	}
921 
922 	if (!boot_error) {
923 		/*
924 		 * Wait till AP completes initial initialization
925 		 */
926 		while (!cpumask_test_cpu(cpu, cpu_callin_mask)) {
927 			/*
928 			 * Allow other tasks to run while we wait for the
929 			 * AP to come online. This also gives a chance
930 			 * for the MTRR work(triggered by the AP coming online)
931 			 * to be completed in the stop machine context.
932 			 */
933 			udelay(100);
934 			schedule();
935 		}
936 	}
937 
938 	/* mark "stuck" area as not stuck */
939 	*trampoline_status = 0;
940 
941 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
942 		/*
943 		 * Cleanup possible dangling ends...
944 		 */
945 		smpboot_restore_warm_reset_vector();
946 	}
947 	/*
948 	 * Clean up the nmi handler. Do this after the callin and callout sync
949 	 * to avoid impact of possible long unregister time.
950 	 */
951 	if (cpu0_nmi_registered)
952 		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
953 
954 	return boot_error;
955 }
956 
957 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
958 {
959 	int apicid = apic->cpu_present_to_apicid(cpu);
960 	unsigned long flags;
961 	int err;
962 
963 	WARN_ON(irqs_disabled());
964 
965 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
966 
967 	if (apicid == BAD_APICID ||
968 	    !physid_isset(apicid, phys_cpu_present_map) ||
969 	    !apic->apic_id_valid(apicid)) {
970 		pr_err("%s: bad cpu %d\n", __func__, cpu);
971 		return -EINVAL;
972 	}
973 
974 	/*
975 	 * Already booted CPU?
976 	 */
977 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
978 		pr_debug("do_boot_cpu %d Already started\n", cpu);
979 		return -ENOSYS;
980 	}
981 
982 	/*
983 	 * Save current MTRR state in case it was changed since early boot
984 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
985 	 */
986 	mtrr_save_state();
987 
988 	/* x86 CPUs take themselves offline, so delayed offline is OK. */
989 	err = cpu_check_up_prepare(cpu);
990 	if (err && err != -EBUSY)
991 		return err;
992 
993 	/* the FPU context is blank, nobody can own it */
994 	__cpu_disable_lazy_restore(cpu);
995 
996 	common_cpu_up(cpu, tidle);
997 
998 	err = do_boot_cpu(apicid, cpu, tidle);
999 	if (err) {
1000 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
1001 		return -EIO;
1002 	}
1003 
1004 	/*
1005 	 * Check TSC synchronization with the AP (keep irqs disabled
1006 	 * while doing so):
1007 	 */
1008 	local_irq_save(flags);
1009 	check_tsc_sync_source(cpu);
1010 	local_irq_restore(flags);
1011 
1012 	while (!cpu_online(cpu)) {
1013 		cpu_relax();
1014 		touch_nmi_watchdog();
1015 	}
1016 
1017 	return 0;
1018 }
1019 
1020 /**
1021  * arch_disable_smp_support() - disables SMP support for x86 at runtime
1022  */
1023 void arch_disable_smp_support(void)
1024 {
1025 	disable_ioapic_support();
1026 }
1027 
1028 /*
1029  * Fall back to non SMP mode after errors.
1030  *
1031  * RED-PEN audit/test this more. I bet there is more state messed up here.
1032  */
1033 static __init void disable_smp(void)
1034 {
1035 	pr_info("SMP disabled\n");
1036 
1037 	disable_ioapic_support();
1038 
1039 	init_cpu_present(cpumask_of(0));
1040 	init_cpu_possible(cpumask_of(0));
1041 
1042 	if (smp_found_config)
1043 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1044 	else
1045 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1046 	cpumask_set_cpu(0, topology_sibling_cpumask(0));
1047 	cpumask_set_cpu(0, topology_core_cpumask(0));
1048 }
1049 
1050 enum {
1051 	SMP_OK,
1052 	SMP_NO_CONFIG,
1053 	SMP_NO_APIC,
1054 	SMP_FORCE_UP,
1055 };
1056 
1057 /*
1058  * Various sanity checks.
1059  */
1060 static int __init smp_sanity_check(unsigned max_cpus)
1061 {
1062 	preempt_disable();
1063 
1064 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
1065 	if (def_to_bigsmp && nr_cpu_ids > 8) {
1066 		unsigned int cpu;
1067 		unsigned nr;
1068 
1069 		pr_warn("More than 8 CPUs detected - skipping them\n"
1070 			"Use CONFIG_X86_BIGSMP\n");
1071 
1072 		nr = 0;
1073 		for_each_present_cpu(cpu) {
1074 			if (nr >= 8)
1075 				set_cpu_present(cpu, false);
1076 			nr++;
1077 		}
1078 
1079 		nr = 0;
1080 		for_each_possible_cpu(cpu) {
1081 			if (nr >= 8)
1082 				set_cpu_possible(cpu, false);
1083 			nr++;
1084 		}
1085 
1086 		nr_cpu_ids = 8;
1087 	}
1088 #endif
1089 
1090 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1091 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1092 			hard_smp_processor_id());
1093 
1094 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1095 	}
1096 
1097 	/*
1098 	 * If we couldn't find an SMP configuration at boot time,
1099 	 * get out of here now!
1100 	 */
1101 	if (!smp_found_config && !acpi_lapic) {
1102 		preempt_enable();
1103 		pr_notice("SMP motherboard not detected\n");
1104 		return SMP_NO_CONFIG;
1105 	}
1106 
1107 	/*
1108 	 * Should not be necessary because the MP table should list the boot
1109 	 * CPU too, but we do it for the sake of robustness anyway.
1110 	 */
1111 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1112 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1113 			  boot_cpu_physical_apicid);
1114 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1115 	}
1116 	preempt_enable();
1117 
1118 	/*
1119 	 * If we couldn't find a local APIC, then get out of here now!
1120 	 */
1121 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1122 	    !cpu_has_apic) {
1123 		if (!disable_apic) {
1124 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
1125 				boot_cpu_physical_apicid);
1126 			pr_err("... forcing use of dummy APIC emulation (tell your hw vendor)\n");
1127 		}
1128 		return SMP_NO_APIC;
1129 	}
1130 
1131 	/*
1132 	 * If SMP should be disabled, then really disable it!
1133 	 */
1134 	if (!max_cpus) {
1135 		pr_info("SMP mode deactivated\n");
1136 		return SMP_FORCE_UP;
1137 	}
1138 
1139 	return SMP_OK;
1140 }
1141 
1142 static void __init smp_cpu_index_default(void)
1143 {
1144 	int i;
1145 	struct cpuinfo_x86 *c;
1146 
1147 	for_each_possible_cpu(i) {
1148 		c = &cpu_data(i);
1149 		/* mark all to hotplug */
1150 		c->cpu_index = nr_cpu_ids;
1151 	}
1152 }
1153 
1154 /*
1155  * Prepare for SMP bootup.  The MP table or ACPI has been read
1156  * earlier.  Just do some sanity checking here and enable APIC mode.
1157  */
1158 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1159 {
1160 	unsigned int i;
1161 
1162 	smp_cpu_index_default();
1163 
1164 	/*
1165 	 * Setup boot CPU information
1166 	 */
1167 	smp_store_boot_cpu_info(); /* Final full version of the data */
1168 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1169 	mb();
1170 
1171 	current_thread_info()->cpu = 0;  /* needed? */
1172 	for_each_possible_cpu(i) {
1173 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1174 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1175 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1176 	}
1177 	set_cpu_sibling_map(0);
1178 
1179 	switch (smp_sanity_check(max_cpus)) {
1180 	case SMP_NO_CONFIG:
1181 		disable_smp();
1182 		if (APIC_init_uniprocessor())
1183 			pr_notice("Local APIC not detected. Using dummy APIC emulation.\n");
1184 		return;
1185 	case SMP_NO_APIC:
1186 		disable_smp();
1187 		return;
1188 	case SMP_FORCE_UP:
1189 		disable_smp();
1190 		apic_bsp_setup(false);
1191 		return;
1192 	case SMP_OK:
1193 		break;
1194 	}
1195 
1196 	default_setup_apic_routing();
1197 
1198 	if (read_apic_id() != boot_cpu_physical_apicid) {
1199 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1200 		     read_apic_id(), boot_cpu_physical_apicid);
1201 		/* Or can we switch back to PIC here? */
1202 	}
1203 
1204 	cpu0_logical_apicid = apic_bsp_setup(false);
1205 
1206 	pr_info("CPU%d: ", 0);
1207 	print_cpu_info(&cpu_data(0));
1208 
1209 	if (is_uv_system())
1210 		uv_system_init();
1211 
1212 	set_mtrr_aps_delayed_init();
1213 
1214 	smp_quirk_init_udelay();
1215 }
1216 
1217 void arch_enable_nonboot_cpus_begin(void)
1218 {
1219 	set_mtrr_aps_delayed_init();
1220 }
1221 
1222 void arch_enable_nonboot_cpus_end(void)
1223 {
1224 	mtrr_aps_init();
1225 }
1226 
1227 /*
1228  * Early setup to make printk work.
1229  */
1230 void __init native_smp_prepare_boot_cpu(void)
1231 {
1232 	int me = smp_processor_id();
1233 	switch_to_new_gdt(me);
1234 	/* already set me in cpu_online_mask in boot_cpu_init() */
1235 	cpumask_set_cpu(me, cpu_callout_mask);
1236 	cpu_set_state_online(me);
1237 }
1238 
1239 void __init native_smp_cpus_done(unsigned int max_cpus)
1240 {
1241 	pr_debug("Boot done\n");
1242 
1243 	nmi_selftest();
1244 	impress_friends();
1245 	setup_ioapic_dest();
1246 	mtrr_aps_init();
1247 }
1248 
1249 static int __initdata setup_possible_cpus = -1;
1250 static int __init _setup_possible_cpus(char *str)
1251 {
1252 	get_option(&str, &setup_possible_cpus);
1253 	return 0;
1254 }
1255 early_param("possible_cpus", _setup_possible_cpus);
1256 
1257 
1258 /*
1259  * cpu_possible_mask should be static, it cannot change as cpu's
1260  * are onlined, or offlined. The reason is per-cpu data-structures
1261  * are allocated by some modules at init time, and dont expect to
1262  * do this dynamically on cpu arrival/departure.
1263  * cpu_present_mask on the other hand can change dynamically.
1264  * In case when cpu_hotplug is not compiled, then we resort to current
1265  * behaviour, which is cpu_possible == cpu_present.
1266  * - Ashok Raj
1267  *
1268  * Three ways to find out the number of additional hotplug CPUs:
1269  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1270  * - The user can overwrite it with possible_cpus=NUM
1271  * - Otherwise don't reserve additional CPUs.
1272  * We do this because additional CPUs waste a lot of memory.
1273  * -AK
1274  */
1275 __init void prefill_possible_map(void)
1276 {
1277 	int i, possible;
1278 
1279 	/* no processor from mptable or madt */
1280 	if (!num_processors)
1281 		num_processors = 1;
1282 
1283 	i = setup_max_cpus ?: 1;
1284 	if (setup_possible_cpus == -1) {
1285 		possible = num_processors;
1286 #ifdef CONFIG_HOTPLUG_CPU
1287 		if (setup_max_cpus)
1288 			possible += disabled_cpus;
1289 #else
1290 		if (possible > i)
1291 			possible = i;
1292 #endif
1293 	} else
1294 		possible = setup_possible_cpus;
1295 
1296 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1297 
1298 	/* nr_cpu_ids could be reduced via nr_cpus= */
1299 	if (possible > nr_cpu_ids) {
1300 		pr_warn("%d Processors exceeds NR_CPUS limit of %d\n",
1301 			possible, nr_cpu_ids);
1302 		possible = nr_cpu_ids;
1303 	}
1304 
1305 #ifdef CONFIG_HOTPLUG_CPU
1306 	if (!setup_max_cpus)
1307 #endif
1308 	if (possible > i) {
1309 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1310 			possible, setup_max_cpus);
1311 		possible = i;
1312 	}
1313 
1314 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1315 		possible, max_t(int, possible - num_processors, 0));
1316 
1317 	for (i = 0; i < possible; i++)
1318 		set_cpu_possible(i, true);
1319 	for (; i < NR_CPUS; i++)
1320 		set_cpu_possible(i, false);
1321 
1322 	nr_cpu_ids = possible;
1323 }
1324 
1325 #ifdef CONFIG_HOTPLUG_CPU
1326 
1327 static void remove_siblinginfo(int cpu)
1328 {
1329 	int sibling;
1330 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1331 
1332 	for_each_cpu(sibling, topology_core_cpumask(cpu)) {
1333 		cpumask_clear_cpu(cpu, topology_core_cpumask(sibling));
1334 		/*/
1335 		 * last thread sibling in this cpu core going down
1336 		 */
1337 		if (cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
1338 			cpu_data(sibling).booted_cores--;
1339 	}
1340 
1341 	for_each_cpu(sibling, topology_sibling_cpumask(cpu))
1342 		cpumask_clear_cpu(cpu, topology_sibling_cpumask(sibling));
1343 	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1344 		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1345 	cpumask_clear(cpu_llc_shared_mask(cpu));
1346 	cpumask_clear(topology_sibling_cpumask(cpu));
1347 	cpumask_clear(topology_core_cpumask(cpu));
1348 	c->phys_proc_id = 0;
1349 	c->cpu_core_id = 0;
1350 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1351 }
1352 
1353 static void __ref remove_cpu_from_maps(int cpu)
1354 {
1355 	set_cpu_online(cpu, false);
1356 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1357 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1358 	/* was set by cpu_init() */
1359 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1360 	numa_remove_cpu(cpu);
1361 }
1362 
1363 void cpu_disable_common(void)
1364 {
1365 	int cpu = smp_processor_id();
1366 
1367 	remove_siblinginfo(cpu);
1368 
1369 	/* It's now safe to remove this processor from the online map */
1370 	lock_vector_lock();
1371 	remove_cpu_from_maps(cpu);
1372 	unlock_vector_lock();
1373 	fixup_irqs();
1374 }
1375 
1376 int native_cpu_disable(void)
1377 {
1378 	int ret;
1379 
1380 	ret = check_irq_vectors_for_cpu_disable();
1381 	if (ret)
1382 		return ret;
1383 
1384 	clear_local_APIC();
1385 	cpu_disable_common();
1386 
1387 	return 0;
1388 }
1389 
1390 int common_cpu_die(unsigned int cpu)
1391 {
1392 	int ret = 0;
1393 
1394 	/* We don't do anything here: idle task is faking death itself. */
1395 
1396 	/* They ack this in play_dead() by setting CPU_DEAD */
1397 	if (cpu_wait_death(cpu, 5)) {
1398 		if (system_state == SYSTEM_RUNNING)
1399 			pr_info("CPU %u is now offline\n", cpu);
1400 	} else {
1401 		pr_err("CPU %u didn't die...\n", cpu);
1402 		ret = -1;
1403 	}
1404 
1405 	return ret;
1406 }
1407 
1408 void native_cpu_die(unsigned int cpu)
1409 {
1410 	common_cpu_die(cpu);
1411 }
1412 
1413 void play_dead_common(void)
1414 {
1415 	idle_task_exit();
1416 	reset_lazy_tlbstate();
1417 	amd_e400_remove_cpu(raw_smp_processor_id());
1418 
1419 	/* Ack it */
1420 	(void)cpu_report_death();
1421 
1422 	/*
1423 	 * With physical CPU hotplug, we should halt the cpu
1424 	 */
1425 	local_irq_disable();
1426 }
1427 
1428 static bool wakeup_cpu0(void)
1429 {
1430 	if (smp_processor_id() == 0 && enable_start_cpu0)
1431 		return true;
1432 
1433 	return false;
1434 }
1435 
1436 /*
1437  * We need to flush the caches before going to sleep, lest we have
1438  * dirty data in our caches when we come back up.
1439  */
1440 static inline void mwait_play_dead(void)
1441 {
1442 	unsigned int eax, ebx, ecx, edx;
1443 	unsigned int highest_cstate = 0;
1444 	unsigned int highest_subcstate = 0;
1445 	void *mwait_ptr;
1446 	int i;
1447 
1448 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1449 		return;
1450 	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1451 		return;
1452 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1453 		return;
1454 
1455 	eax = CPUID_MWAIT_LEAF;
1456 	ecx = 0;
1457 	native_cpuid(&eax, &ebx, &ecx, &edx);
1458 
1459 	/*
1460 	 * eax will be 0 if EDX enumeration is not valid.
1461 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1462 	 */
1463 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1464 		eax = 0;
1465 	} else {
1466 		edx >>= MWAIT_SUBSTATE_SIZE;
1467 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1468 			if (edx & MWAIT_SUBSTATE_MASK) {
1469 				highest_cstate = i;
1470 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1471 			}
1472 		}
1473 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1474 			(highest_subcstate - 1);
1475 	}
1476 
1477 	/*
1478 	 * This should be a memory location in a cache line which is
1479 	 * unlikely to be touched by other processors.  The actual
1480 	 * content is immaterial as it is not actually modified in any way.
1481 	 */
1482 	mwait_ptr = &current_thread_info()->flags;
1483 
1484 	wbinvd();
1485 
1486 	while (1) {
1487 		/*
1488 		 * The CLFLUSH is a workaround for erratum AAI65 for
1489 		 * the Xeon 7400 series.  It's not clear it is actually
1490 		 * needed, but it should be harmless in either case.
1491 		 * The WBINVD is insufficient due to the spurious-wakeup
1492 		 * case where we return around the loop.
1493 		 */
1494 		mb();
1495 		clflush(mwait_ptr);
1496 		mb();
1497 		__monitor(mwait_ptr, 0, 0);
1498 		mb();
1499 		__mwait(eax, 0);
1500 		/*
1501 		 * If NMI wants to wake up CPU0, start CPU0.
1502 		 */
1503 		if (wakeup_cpu0())
1504 			start_cpu0();
1505 	}
1506 }
1507 
1508 static inline void hlt_play_dead(void)
1509 {
1510 	if (__this_cpu_read(cpu_info.x86) >= 4)
1511 		wbinvd();
1512 
1513 	while (1) {
1514 		native_halt();
1515 		/*
1516 		 * If NMI wants to wake up CPU0, start CPU0.
1517 		 */
1518 		if (wakeup_cpu0())
1519 			start_cpu0();
1520 	}
1521 }
1522 
1523 void native_play_dead(void)
1524 {
1525 	play_dead_common();
1526 	tboot_shutdown(TB_SHUTDOWN_WFS);
1527 
1528 	mwait_play_dead();	/* Only returns on failure */
1529 	if (cpuidle_play_dead())
1530 		hlt_play_dead();
1531 }
1532 
1533 #else /* ... !CONFIG_HOTPLUG_CPU */
1534 int native_cpu_disable(void)
1535 {
1536 	return -ENOSYS;
1537 }
1538 
1539 void native_cpu_die(unsigned int cpu)
1540 {
1541 	/* We said "no" in __cpu_disable */
1542 	BUG();
1543 }
1544 
1545 void native_play_dead(void)
1546 {
1547 	BUG();
1548 }
1549 
1550 #endif
1551