xref: /openbmc/linux/arch/x86/kernel/smpboot.c (revision 2eb5f31b)
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 static inline void smpboot_setup_warm_reset_vector(unsigned long start_eip)
101 {
102 	unsigned long flags;
103 
104 	spin_lock_irqsave(&rtc_lock, flags);
105 	CMOS_WRITE(0xa, 0xf);
106 	spin_unlock_irqrestore(&rtc_lock, flags);
107 	local_flush_tlb();
108 	pr_debug("1.\n");
109 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_HIGH)) =
110 							start_eip >> 4;
111 	pr_debug("2.\n");
112 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) =
113 							start_eip & 0xf;
114 	pr_debug("3.\n");
115 }
116 
117 static inline void smpboot_restore_warm_reset_vector(void)
118 {
119 	unsigned long flags;
120 
121 	/*
122 	 * Install writable page 0 entry to set BIOS data area.
123 	 */
124 	local_flush_tlb();
125 
126 	/*
127 	 * Paranoid:  Set warm reset code and vector here back
128 	 * to default values.
129 	 */
130 	spin_lock_irqsave(&rtc_lock, flags);
131 	CMOS_WRITE(0, 0xf);
132 	spin_unlock_irqrestore(&rtc_lock, flags);
133 
134 	*((volatile u32 *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = 0;
135 }
136 
137 /*
138  * Report back to the Boot Processor during boot time or to the caller processor
139  * during CPU online.
140  */
141 static void smp_callin(void)
142 {
143 	int cpuid, phys_id;
144 
145 	/*
146 	 * If waken up by an INIT in an 82489DX configuration
147 	 * cpu_callout_mask guarantees we don't get here before
148 	 * an INIT_deassert IPI reaches our local APIC, so it is
149 	 * now safe to touch our local APIC.
150 	 */
151 	cpuid = smp_processor_id();
152 
153 	/*
154 	 * (This works even if the APIC is not enabled.)
155 	 */
156 	phys_id = read_apic_id();
157 
158 	/*
159 	 * the boot CPU has finished the init stage and is spinning
160 	 * on callin_map until we finish. We are free to set up this
161 	 * CPU, first the APIC. (this is probably redundant on most
162 	 * boards)
163 	 */
164 	apic_ap_setup();
165 
166 	/*
167 	 * Save our processor parameters. Note: this information
168 	 * is needed for clock calibration.
169 	 */
170 	smp_store_cpu_info(cpuid);
171 
172 	/*
173 	 * Get our bogomips.
174 	 * Update loops_per_jiffy in cpu_data. Previous call to
175 	 * smp_store_cpu_info() stored a value that is close but not as
176 	 * accurate as the value just calculated.
177 	 */
178 	calibrate_delay();
179 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
180 	pr_debug("Stack at about %p\n", &cpuid);
181 
182 	/*
183 	 * This must be done before setting cpu_online_mask
184 	 * or calling notify_cpu_starting.
185 	 */
186 	set_cpu_sibling_map(raw_smp_processor_id());
187 	wmb();
188 
189 	notify_cpu_starting(cpuid);
190 
191 	/*
192 	 * Allow the master to continue.
193 	 */
194 	cpumask_set_cpu(cpuid, cpu_callin_mask);
195 }
196 
197 static int cpu0_logical_apicid;
198 static int enable_start_cpu0;
199 /*
200  * Activate a secondary processor.
201  */
202 static void notrace start_secondary(void *unused)
203 {
204 	/*
205 	 * Don't put *anything* before cpu_init(), SMP booting is too
206 	 * fragile that we want to limit the things done here to the
207 	 * most necessary things.
208 	 */
209 	cpu_init();
210 	x86_cpuinit.early_percpu_clock_init();
211 	preempt_disable();
212 	smp_callin();
213 
214 	enable_start_cpu0 = 0;
215 
216 #ifdef CONFIG_X86_32
217 	/* switch away from the initial page table */
218 	load_cr3(swapper_pg_dir);
219 	__flush_tlb_all();
220 #endif
221 
222 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
223 	barrier();
224 	/*
225 	 * Check TSC synchronization with the BP:
226 	 */
227 	check_tsc_sync_target();
228 
229 	/*
230 	 * Lock vector_lock and initialize the vectors on this cpu
231 	 * before setting the cpu online. We must set it online with
232 	 * vector_lock held to prevent a concurrent setup/teardown
233 	 * from seeing a half valid vector space.
234 	 */
235 	lock_vector_lock();
236 	setup_vector_irq(smp_processor_id());
237 	set_cpu_online(smp_processor_id(), true);
238 	unlock_vector_lock();
239 	cpu_set_state_online(smp_processor_id());
240 	x86_platform.nmi_init();
241 
242 	/* enable local interrupts */
243 	local_irq_enable();
244 
245 	/* to prevent fake stack check failure in clock setup */
246 	boot_init_stack_canary();
247 
248 	x86_cpuinit.setup_percpu_clockev();
249 
250 	wmb();
251 	cpu_startup_entry(CPUHP_ONLINE);
252 }
253 
254 void __init smp_store_boot_cpu_info(void)
255 {
256 	int id = 0; /* CPU 0 */
257 	struct cpuinfo_x86 *c = &cpu_data(id);
258 
259 	*c = boot_cpu_data;
260 	c->cpu_index = id;
261 }
262 
263 /*
264  * The bootstrap kernel entry code has set these up. Save them for
265  * a given CPU
266  */
267 void smp_store_cpu_info(int id)
268 {
269 	struct cpuinfo_x86 *c = &cpu_data(id);
270 
271 	*c = boot_cpu_data;
272 	c->cpu_index = id;
273 	/*
274 	 * During boot time, CPU0 has this setup already. Save the info when
275 	 * bringing up AP or offlined CPU0.
276 	 */
277 	identify_secondary_cpu(c);
278 }
279 
280 static bool
281 topology_same_node(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
282 {
283 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
284 
285 	return (cpu_to_node(cpu1) == cpu_to_node(cpu2));
286 }
287 
288 static bool
289 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
290 {
291 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
292 
293 	return !WARN_ONCE(!topology_same_node(c, o),
294 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
295 		"[node: %d != %d]. Ignoring dependency.\n",
296 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
297 }
298 
299 #define link_mask(mfunc, c1, c2)					\
300 do {									\
301 	cpumask_set_cpu((c1), mfunc(c2));				\
302 	cpumask_set_cpu((c2), mfunc(c1));				\
303 } while (0)
304 
305 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
306 {
307 	if (cpu_has_topoext) {
308 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
309 
310 		if (c->phys_proc_id == o->phys_proc_id &&
311 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2) &&
312 		    c->compute_unit_id == o->compute_unit_id)
313 			return topology_sane(c, o, "smt");
314 
315 	} else if (c->phys_proc_id == o->phys_proc_id &&
316 		   c->cpu_core_id == o->cpu_core_id) {
317 		return topology_sane(c, o, "smt");
318 	}
319 
320 	return false;
321 }
322 
323 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
324 {
325 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
326 
327 	if (per_cpu(cpu_llc_id, cpu1) != BAD_APICID &&
328 	    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2))
329 		return topology_sane(c, o, "llc");
330 
331 	return false;
332 }
333 
334 /*
335  * Unlike the other levels, we do not enforce keeping a
336  * multicore group inside a NUMA node.  If this happens, we will
337  * discard the MC level of the topology later.
338  */
339 static bool match_die(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
340 {
341 	if (c->phys_proc_id == o->phys_proc_id)
342 		return true;
343 	return false;
344 }
345 
346 static struct sched_domain_topology_level numa_inside_package_topology[] = {
347 #ifdef CONFIG_SCHED_SMT
348 	{ cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
349 #endif
350 #ifdef CONFIG_SCHED_MC
351 	{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
352 #endif
353 	{ NULL, },
354 };
355 /*
356  * set_sched_topology() sets the topology internal to a CPU.  The
357  * NUMA topologies are layered on top of it to build the full
358  * system topology.
359  *
360  * If NUMA nodes are observed to occur within a CPU package, this
361  * function should be called.  It forces the sched domain code to
362  * only use the SMT level for the CPU portion of the topology.
363  * This essentially falls back to relying on NUMA information
364  * from the SRAT table to describe the entire system topology
365  * (except for hyperthreads).
366  */
367 static void primarily_use_numa_for_topology(void)
368 {
369 	set_sched_topology(numa_inside_package_topology);
370 }
371 
372 void set_cpu_sibling_map(int cpu)
373 {
374 	bool has_smt = smp_num_siblings > 1;
375 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
376 	struct cpuinfo_x86 *c = &cpu_data(cpu);
377 	struct cpuinfo_x86 *o;
378 	int i;
379 
380 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
381 
382 	if (!has_mp) {
383 		cpumask_set_cpu(cpu, topology_sibling_cpumask(cpu));
384 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
385 		cpumask_set_cpu(cpu, topology_core_cpumask(cpu));
386 		c->booted_cores = 1;
387 		return;
388 	}
389 
390 	for_each_cpu(i, cpu_sibling_setup_mask) {
391 		o = &cpu_data(i);
392 
393 		if ((i == cpu) || (has_smt && match_smt(c, o)))
394 			link_mask(topology_sibling_cpumask, cpu, i);
395 
396 		if ((i == cpu) || (has_mp && match_llc(c, o)))
397 			link_mask(cpu_llc_shared_mask, cpu, i);
398 
399 	}
400 
401 	/*
402 	 * This needs a separate iteration over the cpus because we rely on all
403 	 * topology_sibling_cpumask links to be set-up.
404 	 */
405 	for_each_cpu(i, cpu_sibling_setup_mask) {
406 		o = &cpu_data(i);
407 
408 		if ((i == cpu) || (has_mp && match_die(c, o))) {
409 			link_mask(topology_core_cpumask, cpu, i);
410 
411 			/*
412 			 *  Does this new cpu bringup a new core?
413 			 */
414 			if (cpumask_weight(
415 			    topology_sibling_cpumask(cpu)) == 1) {
416 				/*
417 				 * for each core in package, increment
418 				 * the booted_cores for this new cpu
419 				 */
420 				if (cpumask_first(
421 				    topology_sibling_cpumask(i)) == i)
422 					c->booted_cores++;
423 				/*
424 				 * increment the core count for all
425 				 * the other cpus in this package
426 				 */
427 				if (i != cpu)
428 					cpu_data(i).booted_cores++;
429 			} else if (i != cpu && !c->booted_cores)
430 				c->booted_cores = cpu_data(i).booted_cores;
431 		}
432 		if (match_die(c, o) && !topology_same_node(c, o))
433 			primarily_use_numa_for_topology();
434 	}
435 }
436 
437 /* maps the cpu to the sched domain representing multi-core */
438 const struct cpumask *cpu_coregroup_mask(int cpu)
439 {
440 	return cpu_llc_shared_mask(cpu);
441 }
442 
443 static void impress_friends(void)
444 {
445 	int cpu;
446 	unsigned long bogosum = 0;
447 	/*
448 	 * Allow the user to impress friends.
449 	 */
450 	pr_debug("Before bogomips\n");
451 	for_each_possible_cpu(cpu)
452 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
453 			bogosum += cpu_data(cpu).loops_per_jiffy;
454 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
455 		num_online_cpus(),
456 		bogosum/(500000/HZ),
457 		(bogosum/(5000/HZ))%100);
458 
459 	pr_debug("Before bogocount - setting activated=1\n");
460 }
461 
462 void __inquire_remote_apic(int apicid)
463 {
464 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
465 	const char * const names[] = { "ID", "VERSION", "SPIV" };
466 	int timeout;
467 	u32 status;
468 
469 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
470 
471 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
472 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
473 
474 		/*
475 		 * Wait for idle.
476 		 */
477 		status = safe_apic_wait_icr_idle();
478 		if (status)
479 			pr_cont("a previous APIC delivery may have failed\n");
480 
481 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
482 
483 		timeout = 0;
484 		do {
485 			udelay(100);
486 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
487 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
488 
489 		switch (status) {
490 		case APIC_ICR_RR_VALID:
491 			status = apic_read(APIC_RRR);
492 			pr_cont("%08x\n", status);
493 			break;
494 		default:
495 			pr_cont("failed\n");
496 		}
497 	}
498 }
499 
500 /*
501  * The Multiprocessor Specification 1.4 (1997) example code suggests
502  * that there should be a 10ms delay between the BSP asserting INIT
503  * and de-asserting INIT, when starting a remote processor.
504  * But that slows boot and resume on modern processors, which include
505  * many cores and don't require that delay.
506  *
507  * Cmdline "init_cpu_udelay=" is available to over-ride this delay.
508  * Modern processor families are quirked to remove the delay entirely.
509  */
510 #define UDELAY_10MS_DEFAULT 10000
511 
512 static unsigned int init_udelay = INT_MAX;
513 
514 static int __init cpu_init_udelay(char *str)
515 {
516 	get_option(&str, &init_udelay);
517 
518 	return 0;
519 }
520 early_param("cpu_init_udelay", cpu_init_udelay);
521 
522 static void __init smp_quirk_init_udelay(void)
523 {
524 	/* if cmdline changed it from default, leave it alone */
525 	if (init_udelay != INT_MAX)
526 		return;
527 
528 	/* if modern processor, use no delay */
529 	if (((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && (boot_cpu_data.x86 == 6)) ||
530 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && (boot_cpu_data.x86 >= 0xF)))
531 		init_udelay = 0;
532 
533 	/* else, use legacy delay */
534 	init_udelay = UDELAY_10MS_DEFAULT;
535 }
536 
537 /*
538  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
539  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
540  * won't ... remember to clear down the APIC, etc later.
541  */
542 int
543 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
544 {
545 	unsigned long send_status, accept_status = 0;
546 	int maxlvt;
547 
548 	/* Target chip */
549 	/* Boot on the stack */
550 	/* Kick the second */
551 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, apicid);
552 
553 	pr_debug("Waiting for send to finish...\n");
554 	send_status = safe_apic_wait_icr_idle();
555 
556 	/*
557 	 * Give the other CPU some time to accept the IPI.
558 	 */
559 	udelay(200);
560 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
561 		maxlvt = lapic_get_maxlvt();
562 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
563 			apic_write(APIC_ESR, 0);
564 		accept_status = (apic_read(APIC_ESR) & 0xEF);
565 	}
566 	pr_debug("NMI sent\n");
567 
568 	if (send_status)
569 		pr_err("APIC never delivered???\n");
570 	if (accept_status)
571 		pr_err("APIC delivery error (%lx)\n", accept_status);
572 
573 	return (send_status | accept_status);
574 }
575 
576 static int
577 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
578 {
579 	unsigned long send_status = 0, accept_status = 0;
580 	int maxlvt, num_starts, j;
581 
582 	maxlvt = lapic_get_maxlvt();
583 
584 	/*
585 	 * Be paranoid about clearing APIC errors.
586 	 */
587 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
588 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
589 			apic_write(APIC_ESR, 0);
590 		apic_read(APIC_ESR);
591 	}
592 
593 	pr_debug("Asserting INIT\n");
594 
595 	/*
596 	 * Turn INIT on target chip
597 	 */
598 	/*
599 	 * Send IPI
600 	 */
601 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
602 		       phys_apicid);
603 
604 	pr_debug("Waiting for send to finish...\n");
605 	send_status = safe_apic_wait_icr_idle();
606 
607 	udelay(init_udelay);
608 
609 	pr_debug("Deasserting INIT\n");
610 
611 	/* Target chip */
612 	/* Send IPI */
613 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
614 
615 	pr_debug("Waiting for send to finish...\n");
616 	send_status = safe_apic_wait_icr_idle();
617 
618 	mb();
619 
620 	/*
621 	 * Should we send STARTUP IPIs ?
622 	 *
623 	 * Determine this based on the APIC version.
624 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
625 	 */
626 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
627 		num_starts = 2;
628 	else
629 		num_starts = 0;
630 
631 	/*
632 	 * Paravirt / VMI wants a startup IPI hook here to set up the
633 	 * target processor state.
634 	 */
635 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
636 			 stack_start);
637 
638 	/*
639 	 * Run STARTUP IPI loop.
640 	 */
641 	pr_debug("#startup loops: %d\n", num_starts);
642 
643 	for (j = 1; j <= num_starts; j++) {
644 		pr_debug("Sending STARTUP #%d\n", j);
645 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
646 			apic_write(APIC_ESR, 0);
647 		apic_read(APIC_ESR);
648 		pr_debug("After apic_write\n");
649 
650 		/*
651 		 * STARTUP IPI
652 		 */
653 
654 		/* Target chip */
655 		/* Boot on the stack */
656 		/* Kick the second */
657 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
658 			       phys_apicid);
659 
660 		/*
661 		 * Give the other CPU some time to accept the IPI.
662 		 */
663 		if (init_udelay == 0)
664 			udelay(10);
665 		else
666 			udelay(300);
667 
668 		pr_debug("Startup point 1\n");
669 
670 		pr_debug("Waiting for send to finish...\n");
671 		send_status = safe_apic_wait_icr_idle();
672 
673 		/*
674 		 * Give the other CPU some time to accept the IPI.
675 		 */
676 		if (init_udelay == 0)
677 			udelay(10);
678 		else
679 			udelay(200);
680 
681 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
682 			apic_write(APIC_ESR, 0);
683 		accept_status = (apic_read(APIC_ESR) & 0xEF);
684 		if (send_status || accept_status)
685 			break;
686 	}
687 	pr_debug("After Startup\n");
688 
689 	if (send_status)
690 		pr_err("APIC never delivered???\n");
691 	if (accept_status)
692 		pr_err("APIC delivery error (%lx)\n", accept_status);
693 
694 	return (send_status | accept_status);
695 }
696 
697 void smp_announce(void)
698 {
699 	int num_nodes = num_online_nodes();
700 
701 	printk(KERN_INFO "x86: Booted up %d node%s, %d CPUs\n",
702 	       num_nodes, (num_nodes > 1 ? "s" : ""), num_online_cpus());
703 }
704 
705 /* reduce the number of lines printed when booting a large cpu count system */
706 static void announce_cpu(int cpu, int apicid)
707 {
708 	static int current_node = -1;
709 	int node = early_cpu_to_node(cpu);
710 	static int width, node_width;
711 
712 	if (!width)
713 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
714 
715 	if (!node_width)
716 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
717 
718 	if (cpu == 1)
719 		printk(KERN_INFO "x86: Booting SMP configuration:\n");
720 
721 	if (system_state == SYSTEM_BOOTING) {
722 		if (node != current_node) {
723 			if (current_node > (-1))
724 				pr_cont("\n");
725 			current_node = node;
726 
727 			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
728 			       node_width - num_digits(node), " ", node);
729 		}
730 
731 		/* Add padding for the BSP */
732 		if (cpu == 1)
733 			pr_cont("%*s", width + 1, " ");
734 
735 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
736 
737 	} else
738 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
739 			node, cpu, apicid);
740 }
741 
742 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
743 {
744 	int cpu;
745 
746 	cpu = smp_processor_id();
747 	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
748 		return NMI_HANDLED;
749 
750 	return NMI_DONE;
751 }
752 
753 /*
754  * Wake up AP by INIT, INIT, STARTUP sequence.
755  *
756  * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
757  * boot-strap code which is not a desired behavior for waking up BSP. To
758  * void the boot-strap code, wake up CPU0 by NMI instead.
759  *
760  * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
761  * (i.e. physically hot removed and then hot added), NMI won't wake it up.
762  * We'll change this code in the future to wake up hard offlined CPU0 if
763  * real platform and request are available.
764  */
765 static int
766 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
767 	       int *cpu0_nmi_registered)
768 {
769 	int id;
770 	int boot_error;
771 
772 	preempt_disable();
773 
774 	/*
775 	 * Wake up AP by INIT, INIT, STARTUP sequence.
776 	 */
777 	if (cpu) {
778 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
779 		goto out;
780 	}
781 
782 	/*
783 	 * Wake up BSP by nmi.
784 	 *
785 	 * Register a NMI handler to help wake up CPU0.
786 	 */
787 	boot_error = register_nmi_handler(NMI_LOCAL,
788 					  wakeup_cpu0_nmi, 0, "wake_cpu0");
789 
790 	if (!boot_error) {
791 		enable_start_cpu0 = 1;
792 		*cpu0_nmi_registered = 1;
793 		if (apic->dest_logical == APIC_DEST_LOGICAL)
794 			id = cpu0_logical_apicid;
795 		else
796 			id = apicid;
797 		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
798 	}
799 
800 out:
801 	preempt_enable();
802 
803 	return boot_error;
804 }
805 
806 void common_cpu_up(unsigned int cpu, struct task_struct *idle)
807 {
808 	/* Just in case we booted with a single CPU. */
809 	alternatives_enable_smp();
810 
811 	per_cpu(current_task, cpu) = idle;
812 
813 #ifdef CONFIG_X86_32
814 	/* Stack for startup_32 can be just as for start_secondary onwards */
815 	irq_ctx_init(cpu);
816 	per_cpu(cpu_current_top_of_stack, cpu) =
817 		(unsigned long)task_stack_page(idle) + THREAD_SIZE;
818 #else
819 	clear_tsk_thread_flag(idle, TIF_FORK);
820 	initial_gs = per_cpu_offset(cpu);
821 #endif
822 }
823 
824 /*
825  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
826  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
827  * Returns zero if CPU booted OK, else error code from
828  * ->wakeup_secondary_cpu.
829  */
830 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
831 {
832 	volatile u32 *trampoline_status =
833 		(volatile u32 *) __va(real_mode_header->trampoline_status);
834 	/* start_ip had better be page-aligned! */
835 	unsigned long start_ip = real_mode_header->trampoline_start;
836 
837 	unsigned long boot_error = 0;
838 	int cpu0_nmi_registered = 0;
839 	unsigned long timeout;
840 
841 	idle->thread.sp = (unsigned long) (((struct pt_regs *)
842 			  (THREAD_SIZE +  task_stack_page(idle))) - 1);
843 
844 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
845 	initial_code = (unsigned long)start_secondary;
846 	stack_start  = idle->thread.sp;
847 
848 	/*
849 	 * Enable the espfix hack for this CPU
850 	*/
851 #ifdef CONFIG_X86_ESPFIX64
852 	init_espfix_ap(cpu);
853 #endif
854 
855 	/* So we see what's up */
856 	announce_cpu(cpu, apicid);
857 
858 	/*
859 	 * This grunge runs the startup process for
860 	 * the targeted processor.
861 	 */
862 
863 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
864 
865 		pr_debug("Setting warm reset code and vector.\n");
866 
867 		smpboot_setup_warm_reset_vector(start_ip);
868 		/*
869 		 * Be paranoid about clearing APIC errors.
870 		*/
871 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
872 			apic_write(APIC_ESR, 0);
873 			apic_read(APIC_ESR);
874 		}
875 	}
876 
877 	/*
878 	 * AP might wait on cpu_callout_mask in cpu_init() with
879 	 * cpu_initialized_mask set if previous attempt to online
880 	 * it timed-out. Clear cpu_initialized_mask so that after
881 	 * INIT/SIPI it could start with a clean state.
882 	 */
883 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
884 	smp_mb();
885 
886 	/*
887 	 * Wake up a CPU in difference cases:
888 	 * - Use the method in the APIC driver if it's defined
889 	 * Otherwise,
890 	 * - Use an INIT boot APIC message for APs or NMI for BSP.
891 	 */
892 	if (apic->wakeup_secondary_cpu)
893 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
894 	else
895 		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
896 						     &cpu0_nmi_registered);
897 
898 	if (!boot_error) {
899 		/*
900 		 * Wait 10s total for first sign of life from AP
901 		 */
902 		boot_error = -1;
903 		timeout = jiffies + 10*HZ;
904 		while (time_before(jiffies, timeout)) {
905 			if (cpumask_test_cpu(cpu, cpu_initialized_mask)) {
906 				/*
907 				 * Tell AP to proceed with initialization
908 				 */
909 				cpumask_set_cpu(cpu, cpu_callout_mask);
910 				boot_error = 0;
911 				break;
912 			}
913 			schedule();
914 		}
915 	}
916 
917 	if (!boot_error) {
918 		/*
919 		 * Wait till AP completes initial initialization
920 		 */
921 		while (!cpumask_test_cpu(cpu, cpu_callin_mask)) {
922 			/*
923 			 * Allow other tasks to run while we wait for the
924 			 * AP to come online. This also gives a chance
925 			 * for the MTRR work(triggered by the AP coming online)
926 			 * to be completed in the stop machine context.
927 			 */
928 			schedule();
929 		}
930 	}
931 
932 	/* mark "stuck" area as not stuck */
933 	*trampoline_status = 0;
934 
935 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
936 		/*
937 		 * Cleanup possible dangling ends...
938 		 */
939 		smpboot_restore_warm_reset_vector();
940 	}
941 	/*
942 	 * Clean up the nmi handler. Do this after the callin and callout sync
943 	 * to avoid impact of possible long unregister time.
944 	 */
945 	if (cpu0_nmi_registered)
946 		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
947 
948 	return boot_error;
949 }
950 
951 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
952 {
953 	int apicid = apic->cpu_present_to_apicid(cpu);
954 	unsigned long flags;
955 	int err;
956 
957 	WARN_ON(irqs_disabled());
958 
959 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
960 
961 	if (apicid == BAD_APICID ||
962 	    !physid_isset(apicid, phys_cpu_present_map) ||
963 	    !apic->apic_id_valid(apicid)) {
964 		pr_err("%s: bad cpu %d\n", __func__, cpu);
965 		return -EINVAL;
966 	}
967 
968 	/*
969 	 * Already booted CPU?
970 	 */
971 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
972 		pr_debug("do_boot_cpu %d Already started\n", cpu);
973 		return -ENOSYS;
974 	}
975 
976 	/*
977 	 * Save current MTRR state in case it was changed since early boot
978 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
979 	 */
980 	mtrr_save_state();
981 
982 	/* x86 CPUs take themselves offline, so delayed offline is OK. */
983 	err = cpu_check_up_prepare(cpu);
984 	if (err && err != -EBUSY)
985 		return err;
986 
987 	/* the FPU context is blank, nobody can own it */
988 	__cpu_disable_lazy_restore(cpu);
989 
990 	common_cpu_up(cpu, tidle);
991 
992 	/*
993 	 * We have to walk the irq descriptors to setup the vector
994 	 * space for the cpu which comes online.  Prevent irq
995 	 * alloc/free across the bringup.
996 	 */
997 	irq_lock_sparse();
998 
999 	err = do_boot_cpu(apicid, cpu, tidle);
1000 
1001 	if (err) {
1002 		irq_unlock_sparse();
1003 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
1004 		return -EIO;
1005 	}
1006 
1007 	/*
1008 	 * Check TSC synchronization with the AP (keep irqs disabled
1009 	 * while doing so):
1010 	 */
1011 	local_irq_save(flags);
1012 	check_tsc_sync_source(cpu);
1013 	local_irq_restore(flags);
1014 
1015 	while (!cpu_online(cpu)) {
1016 		cpu_relax();
1017 		touch_nmi_watchdog();
1018 	}
1019 
1020 	irq_unlock_sparse();
1021 
1022 	return 0;
1023 }
1024 
1025 /**
1026  * arch_disable_smp_support() - disables SMP support for x86 at runtime
1027  */
1028 void arch_disable_smp_support(void)
1029 {
1030 	disable_ioapic_support();
1031 }
1032 
1033 /*
1034  * Fall back to non SMP mode after errors.
1035  *
1036  * RED-PEN audit/test this more. I bet there is more state messed up here.
1037  */
1038 static __init void disable_smp(void)
1039 {
1040 	pr_info("SMP disabled\n");
1041 
1042 	disable_ioapic_support();
1043 
1044 	init_cpu_present(cpumask_of(0));
1045 	init_cpu_possible(cpumask_of(0));
1046 
1047 	if (smp_found_config)
1048 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1049 	else
1050 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1051 	cpumask_set_cpu(0, topology_sibling_cpumask(0));
1052 	cpumask_set_cpu(0, topology_core_cpumask(0));
1053 }
1054 
1055 enum {
1056 	SMP_OK,
1057 	SMP_NO_CONFIG,
1058 	SMP_NO_APIC,
1059 	SMP_FORCE_UP,
1060 };
1061 
1062 /*
1063  * Various sanity checks.
1064  */
1065 static int __init smp_sanity_check(unsigned max_cpus)
1066 {
1067 	preempt_disable();
1068 
1069 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
1070 	if (def_to_bigsmp && nr_cpu_ids > 8) {
1071 		unsigned int cpu;
1072 		unsigned nr;
1073 
1074 		pr_warn("More than 8 CPUs detected - skipping them\n"
1075 			"Use CONFIG_X86_BIGSMP\n");
1076 
1077 		nr = 0;
1078 		for_each_present_cpu(cpu) {
1079 			if (nr >= 8)
1080 				set_cpu_present(cpu, false);
1081 			nr++;
1082 		}
1083 
1084 		nr = 0;
1085 		for_each_possible_cpu(cpu) {
1086 			if (nr >= 8)
1087 				set_cpu_possible(cpu, false);
1088 			nr++;
1089 		}
1090 
1091 		nr_cpu_ids = 8;
1092 	}
1093 #endif
1094 
1095 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1096 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1097 			hard_smp_processor_id());
1098 
1099 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1100 	}
1101 
1102 	/*
1103 	 * If we couldn't find an SMP configuration at boot time,
1104 	 * get out of here now!
1105 	 */
1106 	if (!smp_found_config && !acpi_lapic) {
1107 		preempt_enable();
1108 		pr_notice("SMP motherboard not detected\n");
1109 		return SMP_NO_CONFIG;
1110 	}
1111 
1112 	/*
1113 	 * Should not be necessary because the MP table should list the boot
1114 	 * CPU too, but we do it for the sake of robustness anyway.
1115 	 */
1116 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1117 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1118 			  boot_cpu_physical_apicid);
1119 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1120 	}
1121 	preempt_enable();
1122 
1123 	/*
1124 	 * If we couldn't find a local APIC, then get out of here now!
1125 	 */
1126 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1127 	    !cpu_has_apic) {
1128 		if (!disable_apic) {
1129 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
1130 				boot_cpu_physical_apicid);
1131 			pr_err("... forcing use of dummy APIC emulation (tell your hw vendor)\n");
1132 		}
1133 		return SMP_NO_APIC;
1134 	}
1135 
1136 	/*
1137 	 * If SMP should be disabled, then really disable it!
1138 	 */
1139 	if (!max_cpus) {
1140 		pr_info("SMP mode deactivated\n");
1141 		return SMP_FORCE_UP;
1142 	}
1143 
1144 	return SMP_OK;
1145 }
1146 
1147 static void __init smp_cpu_index_default(void)
1148 {
1149 	int i;
1150 	struct cpuinfo_x86 *c;
1151 
1152 	for_each_possible_cpu(i) {
1153 		c = &cpu_data(i);
1154 		/* mark all to hotplug */
1155 		c->cpu_index = nr_cpu_ids;
1156 	}
1157 }
1158 
1159 /*
1160  * Prepare for SMP bootup.  The MP table or ACPI has been read
1161  * earlier.  Just do some sanity checking here and enable APIC mode.
1162  */
1163 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1164 {
1165 	unsigned int i;
1166 
1167 	smp_cpu_index_default();
1168 
1169 	/*
1170 	 * Setup boot CPU information
1171 	 */
1172 	smp_store_boot_cpu_info(); /* Final full version of the data */
1173 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1174 	mb();
1175 
1176 	current_thread_info()->cpu = 0;  /* needed? */
1177 	for_each_possible_cpu(i) {
1178 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1179 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1180 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1181 	}
1182 	set_cpu_sibling_map(0);
1183 
1184 	switch (smp_sanity_check(max_cpus)) {
1185 	case SMP_NO_CONFIG:
1186 		disable_smp();
1187 		if (APIC_init_uniprocessor())
1188 			pr_notice("Local APIC not detected. Using dummy APIC emulation.\n");
1189 		return;
1190 	case SMP_NO_APIC:
1191 		disable_smp();
1192 		return;
1193 	case SMP_FORCE_UP:
1194 		disable_smp();
1195 		apic_bsp_setup(false);
1196 		return;
1197 	case SMP_OK:
1198 		break;
1199 	}
1200 
1201 	default_setup_apic_routing();
1202 
1203 	if (read_apic_id() != boot_cpu_physical_apicid) {
1204 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1205 		     read_apic_id(), boot_cpu_physical_apicid);
1206 		/* Or can we switch back to PIC here? */
1207 	}
1208 
1209 	cpu0_logical_apicid = apic_bsp_setup(false);
1210 
1211 	pr_info("CPU%d: ", 0);
1212 	print_cpu_info(&cpu_data(0));
1213 
1214 	if (is_uv_system())
1215 		uv_system_init();
1216 
1217 	set_mtrr_aps_delayed_init();
1218 
1219 	smp_quirk_init_udelay();
1220 }
1221 
1222 void arch_enable_nonboot_cpus_begin(void)
1223 {
1224 	set_mtrr_aps_delayed_init();
1225 }
1226 
1227 void arch_enable_nonboot_cpus_end(void)
1228 {
1229 	mtrr_aps_init();
1230 }
1231 
1232 /*
1233  * Early setup to make printk work.
1234  */
1235 void __init native_smp_prepare_boot_cpu(void)
1236 {
1237 	int me = smp_processor_id();
1238 	switch_to_new_gdt(me);
1239 	/* already set me in cpu_online_mask in boot_cpu_init() */
1240 	cpumask_set_cpu(me, cpu_callout_mask);
1241 	cpu_set_state_online(me);
1242 }
1243 
1244 void __init native_smp_cpus_done(unsigned int max_cpus)
1245 {
1246 	pr_debug("Boot done\n");
1247 
1248 	nmi_selftest();
1249 	impress_friends();
1250 	setup_ioapic_dest();
1251 	mtrr_aps_init();
1252 }
1253 
1254 static int __initdata setup_possible_cpus = -1;
1255 static int __init _setup_possible_cpus(char *str)
1256 {
1257 	get_option(&str, &setup_possible_cpus);
1258 	return 0;
1259 }
1260 early_param("possible_cpus", _setup_possible_cpus);
1261 
1262 
1263 /*
1264  * cpu_possible_mask should be static, it cannot change as cpu's
1265  * are onlined, or offlined. The reason is per-cpu data-structures
1266  * are allocated by some modules at init time, and dont expect to
1267  * do this dynamically on cpu arrival/departure.
1268  * cpu_present_mask on the other hand can change dynamically.
1269  * In case when cpu_hotplug is not compiled, then we resort to current
1270  * behaviour, which is cpu_possible == cpu_present.
1271  * - Ashok Raj
1272  *
1273  * Three ways to find out the number of additional hotplug CPUs:
1274  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1275  * - The user can overwrite it with possible_cpus=NUM
1276  * - Otherwise don't reserve additional CPUs.
1277  * We do this because additional CPUs waste a lot of memory.
1278  * -AK
1279  */
1280 __init void prefill_possible_map(void)
1281 {
1282 	int i, possible;
1283 
1284 	/* no processor from mptable or madt */
1285 	if (!num_processors)
1286 		num_processors = 1;
1287 
1288 	i = setup_max_cpus ?: 1;
1289 	if (setup_possible_cpus == -1) {
1290 		possible = num_processors;
1291 #ifdef CONFIG_HOTPLUG_CPU
1292 		if (setup_max_cpus)
1293 			possible += disabled_cpus;
1294 #else
1295 		if (possible > i)
1296 			possible = i;
1297 #endif
1298 	} else
1299 		possible = setup_possible_cpus;
1300 
1301 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1302 
1303 	/* nr_cpu_ids could be reduced via nr_cpus= */
1304 	if (possible > nr_cpu_ids) {
1305 		pr_warn("%d Processors exceeds NR_CPUS limit of %d\n",
1306 			possible, nr_cpu_ids);
1307 		possible = nr_cpu_ids;
1308 	}
1309 
1310 #ifdef CONFIG_HOTPLUG_CPU
1311 	if (!setup_max_cpus)
1312 #endif
1313 	if (possible > i) {
1314 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1315 			possible, setup_max_cpus);
1316 		possible = i;
1317 	}
1318 
1319 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1320 		possible, max_t(int, possible - num_processors, 0));
1321 
1322 	for (i = 0; i < possible; i++)
1323 		set_cpu_possible(i, true);
1324 	for (; i < NR_CPUS; i++)
1325 		set_cpu_possible(i, false);
1326 
1327 	nr_cpu_ids = possible;
1328 }
1329 
1330 #ifdef CONFIG_HOTPLUG_CPU
1331 
1332 static void remove_siblinginfo(int cpu)
1333 {
1334 	int sibling;
1335 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1336 
1337 	for_each_cpu(sibling, topology_core_cpumask(cpu)) {
1338 		cpumask_clear_cpu(cpu, topology_core_cpumask(sibling));
1339 		/*/
1340 		 * last thread sibling in this cpu core going down
1341 		 */
1342 		if (cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
1343 			cpu_data(sibling).booted_cores--;
1344 	}
1345 
1346 	for_each_cpu(sibling, topology_sibling_cpumask(cpu))
1347 		cpumask_clear_cpu(cpu, topology_sibling_cpumask(sibling));
1348 	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1349 		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1350 	cpumask_clear(cpu_llc_shared_mask(cpu));
1351 	cpumask_clear(topology_sibling_cpumask(cpu));
1352 	cpumask_clear(topology_core_cpumask(cpu));
1353 	c->phys_proc_id = 0;
1354 	c->cpu_core_id = 0;
1355 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1356 }
1357 
1358 static void remove_cpu_from_maps(int cpu)
1359 {
1360 	set_cpu_online(cpu, false);
1361 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1362 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1363 	/* was set by cpu_init() */
1364 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1365 	numa_remove_cpu(cpu);
1366 }
1367 
1368 void cpu_disable_common(void)
1369 {
1370 	int cpu = smp_processor_id();
1371 
1372 	remove_siblinginfo(cpu);
1373 
1374 	/* It's now safe to remove this processor from the online map */
1375 	lock_vector_lock();
1376 	remove_cpu_from_maps(cpu);
1377 	unlock_vector_lock();
1378 	fixup_irqs();
1379 }
1380 
1381 int native_cpu_disable(void)
1382 {
1383 	int ret;
1384 
1385 	ret = check_irq_vectors_for_cpu_disable();
1386 	if (ret)
1387 		return ret;
1388 
1389 	clear_local_APIC();
1390 	cpu_disable_common();
1391 
1392 	return 0;
1393 }
1394 
1395 int common_cpu_die(unsigned int cpu)
1396 {
1397 	int ret = 0;
1398 
1399 	/* We don't do anything here: idle task is faking death itself. */
1400 
1401 	/* They ack this in play_dead() by setting CPU_DEAD */
1402 	if (cpu_wait_death(cpu, 5)) {
1403 		if (system_state == SYSTEM_RUNNING)
1404 			pr_info("CPU %u is now offline\n", cpu);
1405 	} else {
1406 		pr_err("CPU %u didn't die...\n", cpu);
1407 		ret = -1;
1408 	}
1409 
1410 	return ret;
1411 }
1412 
1413 void native_cpu_die(unsigned int cpu)
1414 {
1415 	common_cpu_die(cpu);
1416 }
1417 
1418 void play_dead_common(void)
1419 {
1420 	idle_task_exit();
1421 	reset_lazy_tlbstate();
1422 	amd_e400_remove_cpu(raw_smp_processor_id());
1423 
1424 	/* Ack it */
1425 	(void)cpu_report_death();
1426 
1427 	/*
1428 	 * With physical CPU hotplug, we should halt the cpu
1429 	 */
1430 	local_irq_disable();
1431 }
1432 
1433 static bool wakeup_cpu0(void)
1434 {
1435 	if (smp_processor_id() == 0 && enable_start_cpu0)
1436 		return true;
1437 
1438 	return false;
1439 }
1440 
1441 /*
1442  * We need to flush the caches before going to sleep, lest we have
1443  * dirty data in our caches when we come back up.
1444  */
1445 static inline void mwait_play_dead(void)
1446 {
1447 	unsigned int eax, ebx, ecx, edx;
1448 	unsigned int highest_cstate = 0;
1449 	unsigned int highest_subcstate = 0;
1450 	void *mwait_ptr;
1451 	int i;
1452 
1453 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1454 		return;
1455 	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1456 		return;
1457 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1458 		return;
1459 
1460 	eax = CPUID_MWAIT_LEAF;
1461 	ecx = 0;
1462 	native_cpuid(&eax, &ebx, &ecx, &edx);
1463 
1464 	/*
1465 	 * eax will be 0 if EDX enumeration is not valid.
1466 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1467 	 */
1468 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1469 		eax = 0;
1470 	} else {
1471 		edx >>= MWAIT_SUBSTATE_SIZE;
1472 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1473 			if (edx & MWAIT_SUBSTATE_MASK) {
1474 				highest_cstate = i;
1475 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1476 			}
1477 		}
1478 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1479 			(highest_subcstate - 1);
1480 	}
1481 
1482 	/*
1483 	 * This should be a memory location in a cache line which is
1484 	 * unlikely to be touched by other processors.  The actual
1485 	 * content is immaterial as it is not actually modified in any way.
1486 	 */
1487 	mwait_ptr = &current_thread_info()->flags;
1488 
1489 	wbinvd();
1490 
1491 	while (1) {
1492 		/*
1493 		 * The CLFLUSH is a workaround for erratum AAI65 for
1494 		 * the Xeon 7400 series.  It's not clear it is actually
1495 		 * needed, but it should be harmless in either case.
1496 		 * The WBINVD is insufficient due to the spurious-wakeup
1497 		 * case where we return around the loop.
1498 		 */
1499 		mb();
1500 		clflush(mwait_ptr);
1501 		mb();
1502 		__monitor(mwait_ptr, 0, 0);
1503 		mb();
1504 		__mwait(eax, 0);
1505 		/*
1506 		 * If NMI wants to wake up CPU0, start CPU0.
1507 		 */
1508 		if (wakeup_cpu0())
1509 			start_cpu0();
1510 	}
1511 }
1512 
1513 static inline void hlt_play_dead(void)
1514 {
1515 	if (__this_cpu_read(cpu_info.x86) >= 4)
1516 		wbinvd();
1517 
1518 	while (1) {
1519 		native_halt();
1520 		/*
1521 		 * If NMI wants to wake up CPU0, start CPU0.
1522 		 */
1523 		if (wakeup_cpu0())
1524 			start_cpu0();
1525 	}
1526 }
1527 
1528 void native_play_dead(void)
1529 {
1530 	play_dead_common();
1531 	tboot_shutdown(TB_SHUTDOWN_WFS);
1532 
1533 	mwait_play_dead();	/* Only returns on failure */
1534 	if (cpuidle_play_dead())
1535 		hlt_play_dead();
1536 }
1537 
1538 #else /* ... !CONFIG_HOTPLUG_CPU */
1539 int native_cpu_disable(void)
1540 {
1541 	return -ENOSYS;
1542 }
1543 
1544 void native_cpu_die(unsigned int cpu)
1545 {
1546 	/* We said "no" in __cpu_disable */
1547 	BUG();
1548 }
1549 
1550 void native_play_dead(void)
1551 {
1552 	BUG();
1553 }
1554 
1555 #endif
1556