xref: /openbmc/linux/arch/x86/kernel/smpboot.c (revision 4a44a19b)
1  /*
2  *	x86 SMP booting functions
3  *
4  *	(c) 1995 Alan Cox, Building #3 <alan@lxorguk.ukuu.org.uk>
5  *	(c) 1998, 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
6  *	Copyright 2001 Andi Kleen, SuSE Labs.
7  *
8  *	Much of the core SMP work is based on previous work by Thomas Radke, to
9  *	whom a great many thanks are extended.
10  *
11  *	Thanks to Intel for making available several different Pentium,
12  *	Pentium Pro and Pentium-II/Xeon MP machines.
13  *	Original development of Linux SMP code supported by Caldera.
14  *
15  *	This code is released under the GNU General Public License version 2 or
16  *	later.
17  *
18  *	Fixes
19  *		Felix Koop	:	NR_CPUS used properly
20  *		Jose Renau	:	Handle single CPU case.
21  *		Alan Cox	:	By repeated request 8) - Total BogoMIPS report.
22  *		Greg Wright	:	Fix for kernel stacks panic.
23  *		Erich Boleyn	:	MP v1.4 and additional changes.
24  *	Matthias Sattler	:	Changes for 2.1 kernel map.
25  *	Michel Lespinasse	:	Changes for 2.1 kernel map.
26  *	Michael Chastain	:	Change trampoline.S to gnu as.
27  *		Alan Cox	:	Dumb bug: 'B' step PPro's are fine
28  *		Ingo Molnar	:	Added APIC timers, based on code
29  *					from Jose Renau
30  *		Ingo Molnar	:	various cleanups and rewrites
31  *		Tigran Aivazian	:	fixed "0.00 in /proc/uptime on SMP" bug.
32  *	Maciej W. Rozycki	:	Bits for genuine 82489DX APICs
33  *	Andi Kleen		:	Changed for SMP boot into long mode.
34  *		Martin J. Bligh	: 	Added support for multi-quad systems
35  *		Dave Jones	:	Report invalid combinations of Athlon CPUs.
36  *		Rusty Russell	:	Hacked into shape for new "hotplug" boot process.
37  *      Andi Kleen              :       Converted to new state machine.
38  *	Ashok Raj		: 	CPU hotplug support
39  *	Glauber Costa		:	i386 and x86_64 integration
40  */
41 
42 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
43 
44 #include <linux/init.h>
45 #include <linux/smp.h>
46 #include <linux/module.h>
47 #include <linux/sched.h>
48 #include <linux/percpu.h>
49 #include <linux/bootmem.h>
50 #include <linux/err.h>
51 #include <linux/nmi.h>
52 #include <linux/tboot.h>
53 #include <linux/stackprotector.h>
54 #include <linux/gfp.h>
55 #include <linux/cpuidle.h>
56 
57 #include <asm/acpi.h>
58 #include <asm/desc.h>
59 #include <asm/nmi.h>
60 #include <asm/irq.h>
61 #include <asm/idle.h>
62 #include <asm/realmode.h>
63 #include <asm/cpu.h>
64 #include <asm/numa.h>
65 #include <asm/pgtable.h>
66 #include <asm/tlbflush.h>
67 #include <asm/mtrr.h>
68 #include <asm/mwait.h>
69 #include <asm/apic.h>
70 #include <asm/io_apic.h>
71 #include <asm/i387.h>
72 #include <asm/fpu-internal.h>
73 #include <asm/setup.h>
74 #include <asm/uv/uv.h>
75 #include <linux/mc146818rtc.h>
76 #include <asm/smpboot_hooks.h>
77 #include <asm/i8259.h>
78 #include <asm/realmode.h>
79 #include <asm/misc.h>
80 
81 /* State of each CPU */
82 DEFINE_PER_CPU(int, cpu_state) = { 0 };
83 
84 /* Number of siblings per CPU package */
85 int smp_num_siblings = 1;
86 EXPORT_SYMBOL(smp_num_siblings);
87 
88 /* Last level cache ID of each logical CPU */
89 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
90 
91 /* representing HT siblings of each logical CPU */
92 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
93 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
94 
95 /* representing HT and core siblings of each logical CPU */
96 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_map);
97 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
98 
99 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_llc_shared_map);
100 
101 /* Per CPU bogomips and other parameters */
102 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
103 EXPORT_PER_CPU_SYMBOL(cpu_info);
104 
105 atomic_t init_deasserted;
106 
107 /*
108  * Report back to the Boot Processor during boot time or to the caller processor
109  * during CPU online.
110  */
111 static void smp_callin(void)
112 {
113 	int cpuid, phys_id;
114 
115 	/*
116 	 * If waken up by an INIT in an 82489DX configuration
117 	 * we may get here before an INIT-deassert IPI reaches
118 	 * our local APIC.  We have to wait for the IPI or we'll
119 	 * lock up on an APIC access.
120 	 *
121 	 * Since CPU0 is not wakened up by INIT, it doesn't wait for the IPI.
122 	 */
123 	cpuid = smp_processor_id();
124 	if (apic->wait_for_init_deassert && cpuid)
125 		while (!atomic_read(&init_deasserted))
126 			cpu_relax();
127 
128 	/*
129 	 * (This works even if the APIC is not enabled.)
130 	 */
131 	phys_id = read_apic_id();
132 
133 	/*
134 	 * the boot CPU has finished the init stage and is spinning
135 	 * on callin_map until we finish. We are free to set up this
136 	 * CPU, first the APIC. (this is probably redundant on most
137 	 * boards)
138 	 */
139 	setup_local_APIC();
140 	end_local_APIC_setup();
141 
142 	/*
143 	 * Need to setup vector mappings before we enable interrupts.
144 	 */
145 	setup_vector_irq(smp_processor_id());
146 
147 	/*
148 	 * Save our processor parameters. Note: this information
149 	 * is needed for clock calibration.
150 	 */
151 	smp_store_cpu_info(cpuid);
152 
153 	/*
154 	 * Get our bogomips.
155 	 * Update loops_per_jiffy in cpu_data. Previous call to
156 	 * smp_store_cpu_info() stored a value that is close but not as
157 	 * accurate as the value just calculated.
158 	 */
159 	calibrate_delay();
160 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
161 	pr_debug("Stack at about %p\n", &cpuid);
162 
163 	/*
164 	 * This must be done before setting cpu_online_mask
165 	 * or calling notify_cpu_starting.
166 	 */
167 	set_cpu_sibling_map(raw_smp_processor_id());
168 	wmb();
169 
170 	notify_cpu_starting(cpuid);
171 
172 	/*
173 	 * Allow the master to continue.
174 	 */
175 	cpumask_set_cpu(cpuid, cpu_callin_mask);
176 }
177 
178 static int cpu0_logical_apicid;
179 static int enable_start_cpu0;
180 /*
181  * Activate a secondary processor.
182  */
183 static void notrace start_secondary(void *unused)
184 {
185 	/*
186 	 * Don't put *anything* before cpu_init(), SMP booting is too
187 	 * fragile that we want to limit the things done here to the
188 	 * most necessary things.
189 	 */
190 	cpu_init();
191 	x86_cpuinit.early_percpu_clock_init();
192 	preempt_disable();
193 	smp_callin();
194 
195 	enable_start_cpu0 = 0;
196 
197 #ifdef CONFIG_X86_32
198 	/* switch away from the initial page table */
199 	load_cr3(swapper_pg_dir);
200 	__flush_tlb_all();
201 #endif
202 
203 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
204 	barrier();
205 	/*
206 	 * Check TSC synchronization with the BP:
207 	 */
208 	check_tsc_sync_target();
209 
210 	/*
211 	 * Enable the espfix hack for this CPU
212 	 */
213 #ifdef CONFIG_X86_ESPFIX64
214 	init_espfix_ap();
215 #endif
216 
217 	/*
218 	 * We need to hold vector_lock so there the set of online cpus
219 	 * does not change while we are assigning vectors to cpus.  Holding
220 	 * this lock ensures we don't half assign or remove an irq from a cpu.
221 	 */
222 	lock_vector_lock();
223 	set_cpu_online(smp_processor_id(), true);
224 	unlock_vector_lock();
225 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
226 	x86_platform.nmi_init();
227 
228 	/* enable local interrupts */
229 	local_irq_enable();
230 
231 	/* to prevent fake stack check failure in clock setup */
232 	boot_init_stack_canary();
233 
234 	x86_cpuinit.setup_percpu_clockev();
235 
236 	wmb();
237 	cpu_startup_entry(CPUHP_ONLINE);
238 }
239 
240 void __init smp_store_boot_cpu_info(void)
241 {
242 	int id = 0; /* CPU 0 */
243 	struct cpuinfo_x86 *c = &cpu_data(id);
244 
245 	*c = boot_cpu_data;
246 	c->cpu_index = id;
247 }
248 
249 /*
250  * The bootstrap kernel entry code has set these up. Save them for
251  * a given CPU
252  */
253 void smp_store_cpu_info(int id)
254 {
255 	struct cpuinfo_x86 *c = &cpu_data(id);
256 
257 	*c = boot_cpu_data;
258 	c->cpu_index = id;
259 	/*
260 	 * During boot time, CPU0 has this setup already. Save the info when
261 	 * bringing up AP or offlined CPU0.
262 	 */
263 	identify_secondary_cpu(c);
264 }
265 
266 static bool
267 topology_same_node(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
268 {
269 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
270 
271 	return (cpu_to_node(cpu1) == cpu_to_node(cpu2));
272 }
273 
274 static bool
275 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
276 {
277 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
278 
279 	return !WARN_ONCE(!topology_same_node(c, o),
280 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
281 		"[node: %d != %d]. Ignoring dependency.\n",
282 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
283 }
284 
285 #define link_mask(_m, c1, c2)						\
286 do {									\
287 	cpumask_set_cpu((c1), cpu_##_m##_mask(c2));			\
288 	cpumask_set_cpu((c2), cpu_##_m##_mask(c1));			\
289 } while (0)
290 
291 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
292 {
293 	if (cpu_has_topoext) {
294 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
295 
296 		if (c->phys_proc_id == o->phys_proc_id &&
297 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2) &&
298 		    c->compute_unit_id == o->compute_unit_id)
299 			return topology_sane(c, o, "smt");
300 
301 	} else if (c->phys_proc_id == o->phys_proc_id &&
302 		   c->cpu_core_id == o->cpu_core_id) {
303 		return topology_sane(c, o, "smt");
304 	}
305 
306 	return false;
307 }
308 
309 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
310 {
311 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
312 
313 	if (per_cpu(cpu_llc_id, cpu1) != BAD_APICID &&
314 	    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2))
315 		return topology_sane(c, o, "llc");
316 
317 	return false;
318 }
319 
320 /*
321  * Unlike the other levels, we do not enforce keeping a
322  * multicore group inside a NUMA node.  If this happens, we will
323  * discard the MC level of the topology later.
324  */
325 static bool match_die(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
326 {
327 	if (c->phys_proc_id == o->phys_proc_id)
328 		return true;
329 	return false;
330 }
331 
332 static struct sched_domain_topology_level numa_inside_package_topology[] = {
333 #ifdef CONFIG_SCHED_SMT
334 	{ cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
335 #endif
336 #ifdef CONFIG_SCHED_MC
337 	{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
338 #endif
339 	{ NULL, },
340 };
341 /*
342  * set_sched_topology() sets the topology internal to a CPU.  The
343  * NUMA topologies are layered on top of it to build the full
344  * system topology.
345  *
346  * If NUMA nodes are observed to occur within a CPU package, this
347  * function should be called.  It forces the sched domain code to
348  * only use the SMT level for the CPU portion of the topology.
349  * This essentially falls back to relying on NUMA information
350  * from the SRAT table to describe the entire system topology
351  * (except for hyperthreads).
352  */
353 static void primarily_use_numa_for_topology(void)
354 {
355 	set_sched_topology(numa_inside_package_topology);
356 }
357 
358 void set_cpu_sibling_map(int cpu)
359 {
360 	bool has_smt = smp_num_siblings > 1;
361 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
362 	struct cpuinfo_x86 *c = &cpu_data(cpu);
363 	struct cpuinfo_x86 *o;
364 	int i;
365 
366 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
367 
368 	if (!has_mp) {
369 		cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
370 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
371 		cpumask_set_cpu(cpu, cpu_core_mask(cpu));
372 		c->booted_cores = 1;
373 		return;
374 	}
375 
376 	for_each_cpu(i, cpu_sibling_setup_mask) {
377 		o = &cpu_data(i);
378 
379 		if ((i == cpu) || (has_smt && match_smt(c, o)))
380 			link_mask(sibling, cpu, i);
381 
382 		if ((i == cpu) || (has_mp && match_llc(c, o)))
383 			link_mask(llc_shared, cpu, i);
384 
385 	}
386 
387 	/*
388 	 * This needs a separate iteration over the cpus because we rely on all
389 	 * cpu_sibling_mask links to be set-up.
390 	 */
391 	for_each_cpu(i, cpu_sibling_setup_mask) {
392 		o = &cpu_data(i);
393 
394 		if ((i == cpu) || (has_mp && match_die(c, o))) {
395 			link_mask(core, cpu, i);
396 
397 			/*
398 			 *  Does this new cpu bringup a new core?
399 			 */
400 			if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
401 				/*
402 				 * for each core in package, increment
403 				 * the booted_cores for this new cpu
404 				 */
405 				if (cpumask_first(cpu_sibling_mask(i)) == i)
406 					c->booted_cores++;
407 				/*
408 				 * increment the core count for all
409 				 * the other cpus in this package
410 				 */
411 				if (i != cpu)
412 					cpu_data(i).booted_cores++;
413 			} else if (i != cpu && !c->booted_cores)
414 				c->booted_cores = cpu_data(i).booted_cores;
415 		}
416 		if (match_die(c, o) && !topology_same_node(c, o))
417 			primarily_use_numa_for_topology();
418 	}
419 }
420 
421 /* maps the cpu to the sched domain representing multi-core */
422 const struct cpumask *cpu_coregroup_mask(int cpu)
423 {
424 	return cpu_llc_shared_mask(cpu);
425 }
426 
427 static void impress_friends(void)
428 {
429 	int cpu;
430 	unsigned long bogosum = 0;
431 	/*
432 	 * Allow the user to impress friends.
433 	 */
434 	pr_debug("Before bogomips\n");
435 	for_each_possible_cpu(cpu)
436 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
437 			bogosum += cpu_data(cpu).loops_per_jiffy;
438 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
439 		num_online_cpus(),
440 		bogosum/(500000/HZ),
441 		(bogosum/(5000/HZ))%100);
442 
443 	pr_debug("Before bogocount - setting activated=1\n");
444 }
445 
446 void __inquire_remote_apic(int apicid)
447 {
448 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
449 	const char * const names[] = { "ID", "VERSION", "SPIV" };
450 	int timeout;
451 	u32 status;
452 
453 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
454 
455 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
456 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
457 
458 		/*
459 		 * Wait for idle.
460 		 */
461 		status = safe_apic_wait_icr_idle();
462 		if (status)
463 			pr_cont("a previous APIC delivery may have failed\n");
464 
465 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
466 
467 		timeout = 0;
468 		do {
469 			udelay(100);
470 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
471 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
472 
473 		switch (status) {
474 		case APIC_ICR_RR_VALID:
475 			status = apic_read(APIC_RRR);
476 			pr_cont("%08x\n", status);
477 			break;
478 		default:
479 			pr_cont("failed\n");
480 		}
481 	}
482 }
483 
484 /*
485  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
486  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
487  * won't ... remember to clear down the APIC, etc later.
488  */
489 int
490 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
491 {
492 	unsigned long send_status, accept_status = 0;
493 	int maxlvt;
494 
495 	/* Target chip */
496 	/* Boot on the stack */
497 	/* Kick the second */
498 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, apicid);
499 
500 	pr_debug("Waiting for send to finish...\n");
501 	send_status = safe_apic_wait_icr_idle();
502 
503 	/*
504 	 * Give the other CPU some time to accept the IPI.
505 	 */
506 	udelay(200);
507 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
508 		maxlvt = lapic_get_maxlvt();
509 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
510 			apic_write(APIC_ESR, 0);
511 		accept_status = (apic_read(APIC_ESR) & 0xEF);
512 	}
513 	pr_debug("NMI sent\n");
514 
515 	if (send_status)
516 		pr_err("APIC never delivered???\n");
517 	if (accept_status)
518 		pr_err("APIC delivery error (%lx)\n", accept_status);
519 
520 	return (send_status | accept_status);
521 }
522 
523 static int
524 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
525 {
526 	unsigned long send_status, accept_status = 0;
527 	int maxlvt, num_starts, j;
528 
529 	maxlvt = lapic_get_maxlvt();
530 
531 	/*
532 	 * Be paranoid about clearing APIC errors.
533 	 */
534 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
535 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
536 			apic_write(APIC_ESR, 0);
537 		apic_read(APIC_ESR);
538 	}
539 
540 	pr_debug("Asserting INIT\n");
541 
542 	/*
543 	 * Turn INIT on target chip
544 	 */
545 	/*
546 	 * Send IPI
547 	 */
548 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
549 		       phys_apicid);
550 
551 	pr_debug("Waiting for send to finish...\n");
552 	send_status = safe_apic_wait_icr_idle();
553 
554 	mdelay(10);
555 
556 	pr_debug("Deasserting INIT\n");
557 
558 	/* Target chip */
559 	/* Send IPI */
560 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
561 
562 	pr_debug("Waiting for send to finish...\n");
563 	send_status = safe_apic_wait_icr_idle();
564 
565 	mb();
566 	atomic_set(&init_deasserted, 1);
567 
568 	/*
569 	 * Should we send STARTUP IPIs ?
570 	 *
571 	 * Determine this based on the APIC version.
572 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
573 	 */
574 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
575 		num_starts = 2;
576 	else
577 		num_starts = 0;
578 
579 	/*
580 	 * Paravirt / VMI wants a startup IPI hook here to set up the
581 	 * target processor state.
582 	 */
583 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
584 			 stack_start);
585 
586 	/*
587 	 * Run STARTUP IPI loop.
588 	 */
589 	pr_debug("#startup loops: %d\n", num_starts);
590 
591 	for (j = 1; j <= num_starts; j++) {
592 		pr_debug("Sending STARTUP #%d\n", j);
593 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
594 			apic_write(APIC_ESR, 0);
595 		apic_read(APIC_ESR);
596 		pr_debug("After apic_write\n");
597 
598 		/*
599 		 * STARTUP IPI
600 		 */
601 
602 		/* Target chip */
603 		/* Boot on the stack */
604 		/* Kick the second */
605 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
606 			       phys_apicid);
607 
608 		/*
609 		 * Give the other CPU some time to accept the IPI.
610 		 */
611 		udelay(300);
612 
613 		pr_debug("Startup point 1\n");
614 
615 		pr_debug("Waiting for send to finish...\n");
616 		send_status = safe_apic_wait_icr_idle();
617 
618 		/*
619 		 * Give the other CPU some time to accept the IPI.
620 		 */
621 		udelay(200);
622 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
623 			apic_write(APIC_ESR, 0);
624 		accept_status = (apic_read(APIC_ESR) & 0xEF);
625 		if (send_status || accept_status)
626 			break;
627 	}
628 	pr_debug("After Startup\n");
629 
630 	if (send_status)
631 		pr_err("APIC never delivered???\n");
632 	if (accept_status)
633 		pr_err("APIC delivery error (%lx)\n", accept_status);
634 
635 	return (send_status | accept_status);
636 }
637 
638 void smp_announce(void)
639 {
640 	int num_nodes = num_online_nodes();
641 
642 	printk(KERN_INFO "x86: Booted up %d node%s, %d CPUs\n",
643 	       num_nodes, (num_nodes > 1 ? "s" : ""), num_online_cpus());
644 }
645 
646 /* reduce the number of lines printed when booting a large cpu count system */
647 static void announce_cpu(int cpu, int apicid)
648 {
649 	static int current_node = -1;
650 	int node = early_cpu_to_node(cpu);
651 	static int width, node_width;
652 
653 	if (!width)
654 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
655 
656 	if (!node_width)
657 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
658 
659 	if (cpu == 1)
660 		printk(KERN_INFO "x86: Booting SMP configuration:\n");
661 
662 	if (system_state == SYSTEM_BOOTING) {
663 		if (node != current_node) {
664 			if (current_node > (-1))
665 				pr_cont("\n");
666 			current_node = node;
667 
668 			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
669 			       node_width - num_digits(node), " ", node);
670 		}
671 
672 		/* Add padding for the BSP */
673 		if (cpu == 1)
674 			pr_cont("%*s", width + 1, " ");
675 
676 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
677 
678 	} else
679 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
680 			node, cpu, apicid);
681 }
682 
683 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
684 {
685 	int cpu;
686 
687 	cpu = smp_processor_id();
688 	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
689 		return NMI_HANDLED;
690 
691 	return NMI_DONE;
692 }
693 
694 /*
695  * Wake up AP by INIT, INIT, STARTUP sequence.
696  *
697  * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
698  * boot-strap code which is not a desired behavior for waking up BSP. To
699  * void the boot-strap code, wake up CPU0 by NMI instead.
700  *
701  * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
702  * (i.e. physically hot removed and then hot added), NMI won't wake it up.
703  * We'll change this code in the future to wake up hard offlined CPU0 if
704  * real platform and request are available.
705  */
706 static int
707 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
708 	       int *cpu0_nmi_registered)
709 {
710 	int id;
711 	int boot_error;
712 
713 	preempt_disable();
714 
715 	/*
716 	 * Wake up AP by INIT, INIT, STARTUP sequence.
717 	 */
718 	if (cpu) {
719 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
720 		goto out;
721 	}
722 
723 	/*
724 	 * Wake up BSP by nmi.
725 	 *
726 	 * Register a NMI handler to help wake up CPU0.
727 	 */
728 	boot_error = register_nmi_handler(NMI_LOCAL,
729 					  wakeup_cpu0_nmi, 0, "wake_cpu0");
730 
731 	if (!boot_error) {
732 		enable_start_cpu0 = 1;
733 		*cpu0_nmi_registered = 1;
734 		if (apic->dest_logical == APIC_DEST_LOGICAL)
735 			id = cpu0_logical_apicid;
736 		else
737 			id = apicid;
738 		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
739 	}
740 
741 out:
742 	preempt_enable();
743 
744 	return boot_error;
745 }
746 
747 /*
748  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
749  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
750  * Returns zero if CPU booted OK, else error code from
751  * ->wakeup_secondary_cpu.
752  */
753 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
754 {
755 	volatile u32 *trampoline_status =
756 		(volatile u32 *) __va(real_mode_header->trampoline_status);
757 	/* start_ip had better be page-aligned! */
758 	unsigned long start_ip = real_mode_header->trampoline_start;
759 
760 	unsigned long boot_error = 0;
761 	int cpu0_nmi_registered = 0;
762 	unsigned long timeout;
763 
764 	/* Just in case we booted with a single CPU. */
765 	alternatives_enable_smp();
766 
767 	idle->thread.sp = (unsigned long) (((struct pt_regs *)
768 			  (THREAD_SIZE +  task_stack_page(idle))) - 1);
769 	per_cpu(current_task, cpu) = idle;
770 
771 #ifdef CONFIG_X86_32
772 	/* Stack for startup_32 can be just as for start_secondary onwards */
773 	irq_ctx_init(cpu);
774 #else
775 	clear_tsk_thread_flag(idle, TIF_FORK);
776 	initial_gs = per_cpu_offset(cpu);
777 #endif
778 	per_cpu(kernel_stack, cpu) =
779 		(unsigned long)task_stack_page(idle) -
780 		KERNEL_STACK_OFFSET + THREAD_SIZE;
781 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
782 	initial_code = (unsigned long)start_secondary;
783 	stack_start  = idle->thread.sp;
784 
785 	/* So we see what's up */
786 	announce_cpu(cpu, apicid);
787 
788 	/*
789 	 * This grunge runs the startup process for
790 	 * the targeted processor.
791 	 */
792 
793 	atomic_set(&init_deasserted, 0);
794 
795 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
796 
797 		pr_debug("Setting warm reset code and vector.\n");
798 
799 		smpboot_setup_warm_reset_vector(start_ip);
800 		/*
801 		 * Be paranoid about clearing APIC errors.
802 		*/
803 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
804 			apic_write(APIC_ESR, 0);
805 			apic_read(APIC_ESR);
806 		}
807 	}
808 
809 	/*
810 	 * AP might wait on cpu_callout_mask in cpu_init() with
811 	 * cpu_initialized_mask set if previous attempt to online
812 	 * it timed-out. Clear cpu_initialized_mask so that after
813 	 * INIT/SIPI it could start with a clean state.
814 	 */
815 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
816 	smp_mb();
817 
818 	/*
819 	 * Wake up a CPU in difference cases:
820 	 * - Use the method in the APIC driver if it's defined
821 	 * Otherwise,
822 	 * - Use an INIT boot APIC message for APs or NMI for BSP.
823 	 */
824 	if (apic->wakeup_secondary_cpu)
825 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
826 	else
827 		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
828 						     &cpu0_nmi_registered);
829 
830 	if (!boot_error) {
831 		/*
832 		 * Wait 10s total for a response from AP
833 		 */
834 		boot_error = -1;
835 		timeout = jiffies + 10*HZ;
836 		while (time_before(jiffies, timeout)) {
837 			if (cpumask_test_cpu(cpu, cpu_initialized_mask)) {
838 				/*
839 				 * Tell AP to proceed with initialization
840 				 */
841 				cpumask_set_cpu(cpu, cpu_callout_mask);
842 				boot_error = 0;
843 				break;
844 			}
845 			udelay(100);
846 			schedule();
847 		}
848 	}
849 
850 	if (!boot_error) {
851 		/*
852 		 * Wait till AP completes initial initialization
853 		 */
854 		while (!cpumask_test_cpu(cpu, cpu_callin_mask)) {
855 			/*
856 			 * Allow other tasks to run while we wait for the
857 			 * AP to come online. This also gives a chance
858 			 * for the MTRR work(triggered by the AP coming online)
859 			 * to be completed in the stop machine context.
860 			 */
861 			udelay(100);
862 			schedule();
863 		}
864 	}
865 
866 	/* mark "stuck" area as not stuck */
867 	*trampoline_status = 0;
868 
869 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
870 		/*
871 		 * Cleanup possible dangling ends...
872 		 */
873 		smpboot_restore_warm_reset_vector();
874 	}
875 	/*
876 	 * Clean up the nmi handler. Do this after the callin and callout sync
877 	 * to avoid impact of possible long unregister time.
878 	 */
879 	if (cpu0_nmi_registered)
880 		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
881 
882 	return boot_error;
883 }
884 
885 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
886 {
887 	int apicid = apic->cpu_present_to_apicid(cpu);
888 	unsigned long flags;
889 	int err;
890 
891 	WARN_ON(irqs_disabled());
892 
893 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
894 
895 	if (apicid == BAD_APICID ||
896 	    !physid_isset(apicid, phys_cpu_present_map) ||
897 	    !apic->apic_id_valid(apicid)) {
898 		pr_err("%s: bad cpu %d\n", __func__, cpu);
899 		return -EINVAL;
900 	}
901 
902 	/*
903 	 * Already booted CPU?
904 	 */
905 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
906 		pr_debug("do_boot_cpu %d Already started\n", cpu);
907 		return -ENOSYS;
908 	}
909 
910 	/*
911 	 * Save current MTRR state in case it was changed since early boot
912 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
913 	 */
914 	mtrr_save_state();
915 
916 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
917 
918 	/* the FPU context is blank, nobody can own it */
919 	__cpu_disable_lazy_restore(cpu);
920 
921 	err = do_boot_cpu(apicid, cpu, tidle);
922 	if (err) {
923 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
924 		return -EIO;
925 	}
926 
927 	/*
928 	 * Check TSC synchronization with the AP (keep irqs disabled
929 	 * while doing so):
930 	 */
931 	local_irq_save(flags);
932 	check_tsc_sync_source(cpu);
933 	local_irq_restore(flags);
934 
935 	while (!cpu_online(cpu)) {
936 		cpu_relax();
937 		touch_nmi_watchdog();
938 	}
939 
940 	return 0;
941 }
942 
943 /**
944  * arch_disable_smp_support() - disables SMP support for x86 at runtime
945  */
946 void arch_disable_smp_support(void)
947 {
948 	disable_ioapic_support();
949 }
950 
951 /*
952  * Fall back to non SMP mode after errors.
953  *
954  * RED-PEN audit/test this more. I bet there is more state messed up here.
955  */
956 static __init void disable_smp(void)
957 {
958 	init_cpu_present(cpumask_of(0));
959 	init_cpu_possible(cpumask_of(0));
960 	smpboot_clear_io_apic_irqs();
961 
962 	if (smp_found_config)
963 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
964 	else
965 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
966 	cpumask_set_cpu(0, cpu_sibling_mask(0));
967 	cpumask_set_cpu(0, cpu_core_mask(0));
968 }
969 
970 /*
971  * Various sanity checks.
972  */
973 static int __init smp_sanity_check(unsigned max_cpus)
974 {
975 	preempt_disable();
976 
977 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
978 	if (def_to_bigsmp && nr_cpu_ids > 8) {
979 		unsigned int cpu;
980 		unsigned nr;
981 
982 		pr_warn("More than 8 CPUs detected - skipping them\n"
983 			"Use CONFIG_X86_BIGSMP\n");
984 
985 		nr = 0;
986 		for_each_present_cpu(cpu) {
987 			if (nr >= 8)
988 				set_cpu_present(cpu, false);
989 			nr++;
990 		}
991 
992 		nr = 0;
993 		for_each_possible_cpu(cpu) {
994 			if (nr >= 8)
995 				set_cpu_possible(cpu, false);
996 			nr++;
997 		}
998 
999 		nr_cpu_ids = 8;
1000 	}
1001 #endif
1002 
1003 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1004 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1005 			hard_smp_processor_id());
1006 
1007 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1008 	}
1009 
1010 	/*
1011 	 * If we couldn't find an SMP configuration at boot time,
1012 	 * get out of here now!
1013 	 */
1014 	if (!smp_found_config && !acpi_lapic) {
1015 		preempt_enable();
1016 		pr_notice("SMP motherboard not detected\n");
1017 		disable_smp();
1018 		if (APIC_init_uniprocessor())
1019 			pr_notice("Local APIC not detected. Using dummy APIC emulation.\n");
1020 		return -1;
1021 	}
1022 
1023 	/*
1024 	 * Should not be necessary because the MP table should list the boot
1025 	 * CPU too, but we do it for the sake of robustness anyway.
1026 	 */
1027 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1028 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1029 			  boot_cpu_physical_apicid);
1030 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1031 	}
1032 	preempt_enable();
1033 
1034 	/*
1035 	 * If we couldn't find a local APIC, then get out of here now!
1036 	 */
1037 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1038 	    !cpu_has_apic) {
1039 		if (!disable_apic) {
1040 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
1041 				boot_cpu_physical_apicid);
1042 			pr_err("... forcing use of dummy APIC emulation (tell your hw vendor)\n");
1043 		}
1044 		smpboot_clear_io_apic();
1045 		disable_ioapic_support();
1046 		return -1;
1047 	}
1048 
1049 	verify_local_APIC();
1050 
1051 	/*
1052 	 * If SMP should be disabled, then really disable it!
1053 	 */
1054 	if (!max_cpus) {
1055 		pr_info("SMP mode deactivated\n");
1056 		smpboot_clear_io_apic();
1057 
1058 		connect_bsp_APIC();
1059 		setup_local_APIC();
1060 		bsp_end_local_APIC_setup();
1061 		return -1;
1062 	}
1063 
1064 	return 0;
1065 }
1066 
1067 static void __init smp_cpu_index_default(void)
1068 {
1069 	int i;
1070 	struct cpuinfo_x86 *c;
1071 
1072 	for_each_possible_cpu(i) {
1073 		c = &cpu_data(i);
1074 		/* mark all to hotplug */
1075 		c->cpu_index = nr_cpu_ids;
1076 	}
1077 }
1078 
1079 /*
1080  * Prepare for SMP bootup.  The MP table or ACPI has been read
1081  * earlier.  Just do some sanity checking here and enable APIC mode.
1082  */
1083 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1084 {
1085 	unsigned int i;
1086 
1087 	preempt_disable();
1088 	smp_cpu_index_default();
1089 
1090 	/*
1091 	 * Setup boot CPU information
1092 	 */
1093 	smp_store_boot_cpu_info(); /* Final full version of the data */
1094 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1095 	mb();
1096 
1097 	current_thread_info()->cpu = 0;  /* needed? */
1098 	for_each_possible_cpu(i) {
1099 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1100 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1101 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1102 	}
1103 	set_cpu_sibling_map(0);
1104 
1105 
1106 	if (smp_sanity_check(max_cpus) < 0) {
1107 		pr_info("SMP disabled\n");
1108 		disable_smp();
1109 		goto out;
1110 	}
1111 
1112 	default_setup_apic_routing();
1113 
1114 	preempt_disable();
1115 	if (read_apic_id() != boot_cpu_physical_apicid) {
1116 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1117 		     read_apic_id(), boot_cpu_physical_apicid);
1118 		/* Or can we switch back to PIC here? */
1119 	}
1120 	preempt_enable();
1121 
1122 	connect_bsp_APIC();
1123 
1124 	/*
1125 	 * Switch from PIC to APIC mode.
1126 	 */
1127 	setup_local_APIC();
1128 
1129 	if (x2apic_mode)
1130 		cpu0_logical_apicid = apic_read(APIC_LDR);
1131 	else
1132 		cpu0_logical_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1133 
1134 	/*
1135 	 * Enable IO APIC before setting up error vector
1136 	 */
1137 	if (!skip_ioapic_setup && nr_ioapics)
1138 		enable_IO_APIC();
1139 
1140 	bsp_end_local_APIC_setup();
1141 	smpboot_setup_io_apic();
1142 	/*
1143 	 * Set up local APIC timer on boot CPU.
1144 	 */
1145 
1146 	pr_info("CPU%d: ", 0);
1147 	print_cpu_info(&cpu_data(0));
1148 	x86_init.timers.setup_percpu_clockev();
1149 
1150 	if (is_uv_system())
1151 		uv_system_init();
1152 
1153 	set_mtrr_aps_delayed_init();
1154 out:
1155 	preempt_enable();
1156 }
1157 
1158 void arch_enable_nonboot_cpus_begin(void)
1159 {
1160 	set_mtrr_aps_delayed_init();
1161 }
1162 
1163 void arch_enable_nonboot_cpus_end(void)
1164 {
1165 	mtrr_aps_init();
1166 }
1167 
1168 /*
1169  * Early setup to make printk work.
1170  */
1171 void __init native_smp_prepare_boot_cpu(void)
1172 {
1173 	int me = smp_processor_id();
1174 	switch_to_new_gdt(me);
1175 	/* already set me in cpu_online_mask in boot_cpu_init() */
1176 	cpumask_set_cpu(me, cpu_callout_mask);
1177 	per_cpu(cpu_state, me) = CPU_ONLINE;
1178 }
1179 
1180 void __init native_smp_cpus_done(unsigned int max_cpus)
1181 {
1182 	pr_debug("Boot done\n");
1183 
1184 	nmi_selftest();
1185 	impress_friends();
1186 #ifdef CONFIG_X86_IO_APIC
1187 	setup_ioapic_dest();
1188 #endif
1189 	mtrr_aps_init();
1190 }
1191 
1192 static int __initdata setup_possible_cpus = -1;
1193 static int __init _setup_possible_cpus(char *str)
1194 {
1195 	get_option(&str, &setup_possible_cpus);
1196 	return 0;
1197 }
1198 early_param("possible_cpus", _setup_possible_cpus);
1199 
1200 
1201 /*
1202  * cpu_possible_mask should be static, it cannot change as cpu's
1203  * are onlined, or offlined. The reason is per-cpu data-structures
1204  * are allocated by some modules at init time, and dont expect to
1205  * do this dynamically on cpu arrival/departure.
1206  * cpu_present_mask on the other hand can change dynamically.
1207  * In case when cpu_hotplug is not compiled, then we resort to current
1208  * behaviour, which is cpu_possible == cpu_present.
1209  * - Ashok Raj
1210  *
1211  * Three ways to find out the number of additional hotplug CPUs:
1212  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1213  * - The user can overwrite it with possible_cpus=NUM
1214  * - Otherwise don't reserve additional CPUs.
1215  * We do this because additional CPUs waste a lot of memory.
1216  * -AK
1217  */
1218 __init void prefill_possible_map(void)
1219 {
1220 	int i, possible;
1221 
1222 	/* no processor from mptable or madt */
1223 	if (!num_processors)
1224 		num_processors = 1;
1225 
1226 	i = setup_max_cpus ?: 1;
1227 	if (setup_possible_cpus == -1) {
1228 		possible = num_processors;
1229 #ifdef CONFIG_HOTPLUG_CPU
1230 		if (setup_max_cpus)
1231 			possible += disabled_cpus;
1232 #else
1233 		if (possible > i)
1234 			possible = i;
1235 #endif
1236 	} else
1237 		possible = setup_possible_cpus;
1238 
1239 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1240 
1241 	/* nr_cpu_ids could be reduced via nr_cpus= */
1242 	if (possible > nr_cpu_ids) {
1243 		pr_warn("%d Processors exceeds NR_CPUS limit of %d\n",
1244 			possible, nr_cpu_ids);
1245 		possible = nr_cpu_ids;
1246 	}
1247 
1248 #ifdef CONFIG_HOTPLUG_CPU
1249 	if (!setup_max_cpus)
1250 #endif
1251 	if (possible > i) {
1252 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1253 			possible, setup_max_cpus);
1254 		possible = i;
1255 	}
1256 
1257 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1258 		possible, max_t(int, possible - num_processors, 0));
1259 
1260 	for (i = 0; i < possible; i++)
1261 		set_cpu_possible(i, true);
1262 	for (; i < NR_CPUS; i++)
1263 		set_cpu_possible(i, false);
1264 
1265 	nr_cpu_ids = possible;
1266 }
1267 
1268 #ifdef CONFIG_HOTPLUG_CPU
1269 
1270 static void remove_siblinginfo(int cpu)
1271 {
1272 	int sibling;
1273 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1274 
1275 	for_each_cpu(sibling, cpu_core_mask(cpu)) {
1276 		cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1277 		/*/
1278 		 * last thread sibling in this cpu core going down
1279 		 */
1280 		if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1281 			cpu_data(sibling).booted_cores--;
1282 	}
1283 
1284 	for_each_cpu(sibling, cpu_sibling_mask(cpu))
1285 		cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1286 	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1287 		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1288 	cpumask_clear(cpu_llc_shared_mask(cpu));
1289 	cpumask_clear(cpu_sibling_mask(cpu));
1290 	cpumask_clear(cpu_core_mask(cpu));
1291 	c->phys_proc_id = 0;
1292 	c->cpu_core_id = 0;
1293 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1294 }
1295 
1296 static void __ref remove_cpu_from_maps(int cpu)
1297 {
1298 	set_cpu_online(cpu, false);
1299 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1300 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1301 	/* was set by cpu_init() */
1302 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1303 	numa_remove_cpu(cpu);
1304 }
1305 
1306 static DEFINE_PER_CPU(struct completion, die_complete);
1307 
1308 void cpu_disable_common(void)
1309 {
1310 	int cpu = smp_processor_id();
1311 
1312 	init_completion(&per_cpu(die_complete, smp_processor_id()));
1313 
1314 	remove_siblinginfo(cpu);
1315 
1316 	/* It's now safe to remove this processor from the online map */
1317 	lock_vector_lock();
1318 	remove_cpu_from_maps(cpu);
1319 	unlock_vector_lock();
1320 	fixup_irqs();
1321 }
1322 
1323 int native_cpu_disable(void)
1324 {
1325 	int ret;
1326 
1327 	ret = check_irq_vectors_for_cpu_disable();
1328 	if (ret)
1329 		return ret;
1330 
1331 	clear_local_APIC();
1332 	cpu_disable_common();
1333 
1334 	return 0;
1335 }
1336 
1337 void cpu_die_common(unsigned int cpu)
1338 {
1339 	wait_for_completion_timeout(&per_cpu(die_complete, cpu), HZ);
1340 }
1341 
1342 void native_cpu_die(unsigned int cpu)
1343 {
1344 	/* We don't do anything here: idle task is faking death itself. */
1345 
1346 	cpu_die_common(cpu);
1347 
1348 	/* They ack this in play_dead() by setting CPU_DEAD */
1349 	if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1350 		if (system_state == SYSTEM_RUNNING)
1351 			pr_info("CPU %u is now offline\n", cpu);
1352 	} else {
1353 		pr_err("CPU %u didn't die...\n", cpu);
1354 	}
1355 }
1356 
1357 void play_dead_common(void)
1358 {
1359 	idle_task_exit();
1360 	reset_lazy_tlbstate();
1361 	amd_e400_remove_cpu(raw_smp_processor_id());
1362 
1363 	mb();
1364 	/* Ack it */
1365 	__this_cpu_write(cpu_state, CPU_DEAD);
1366 	complete(&per_cpu(die_complete, smp_processor_id()));
1367 
1368 	/*
1369 	 * With physical CPU hotplug, we should halt the cpu
1370 	 */
1371 	local_irq_disable();
1372 }
1373 
1374 static bool wakeup_cpu0(void)
1375 {
1376 	if (smp_processor_id() == 0 && enable_start_cpu0)
1377 		return true;
1378 
1379 	return false;
1380 }
1381 
1382 /*
1383  * We need to flush the caches before going to sleep, lest we have
1384  * dirty data in our caches when we come back up.
1385  */
1386 static inline void mwait_play_dead(void)
1387 {
1388 	unsigned int eax, ebx, ecx, edx;
1389 	unsigned int highest_cstate = 0;
1390 	unsigned int highest_subcstate = 0;
1391 	void *mwait_ptr;
1392 	int i;
1393 
1394 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1395 		return;
1396 	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1397 		return;
1398 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1399 		return;
1400 
1401 	eax = CPUID_MWAIT_LEAF;
1402 	ecx = 0;
1403 	native_cpuid(&eax, &ebx, &ecx, &edx);
1404 
1405 	/*
1406 	 * eax will be 0 if EDX enumeration is not valid.
1407 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1408 	 */
1409 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1410 		eax = 0;
1411 	} else {
1412 		edx >>= MWAIT_SUBSTATE_SIZE;
1413 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1414 			if (edx & MWAIT_SUBSTATE_MASK) {
1415 				highest_cstate = i;
1416 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1417 			}
1418 		}
1419 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1420 			(highest_subcstate - 1);
1421 	}
1422 
1423 	/*
1424 	 * This should be a memory location in a cache line which is
1425 	 * unlikely to be touched by other processors.  The actual
1426 	 * content is immaterial as it is not actually modified in any way.
1427 	 */
1428 	mwait_ptr = &current_thread_info()->flags;
1429 
1430 	wbinvd();
1431 
1432 	while (1) {
1433 		/*
1434 		 * The CLFLUSH is a workaround for erratum AAI65 for
1435 		 * the Xeon 7400 series.  It's not clear it is actually
1436 		 * needed, but it should be harmless in either case.
1437 		 * The WBINVD is insufficient due to the spurious-wakeup
1438 		 * case where we return around the loop.
1439 		 */
1440 		mb();
1441 		clflush(mwait_ptr);
1442 		mb();
1443 		__monitor(mwait_ptr, 0, 0);
1444 		mb();
1445 		__mwait(eax, 0);
1446 		/*
1447 		 * If NMI wants to wake up CPU0, start CPU0.
1448 		 */
1449 		if (wakeup_cpu0())
1450 			start_cpu0();
1451 	}
1452 }
1453 
1454 static inline void hlt_play_dead(void)
1455 {
1456 	if (__this_cpu_read(cpu_info.x86) >= 4)
1457 		wbinvd();
1458 
1459 	while (1) {
1460 		native_halt();
1461 		/*
1462 		 * If NMI wants to wake up CPU0, start CPU0.
1463 		 */
1464 		if (wakeup_cpu0())
1465 			start_cpu0();
1466 	}
1467 }
1468 
1469 void native_play_dead(void)
1470 {
1471 	play_dead_common();
1472 	tboot_shutdown(TB_SHUTDOWN_WFS);
1473 
1474 	mwait_play_dead();	/* Only returns on failure */
1475 	if (cpuidle_play_dead())
1476 		hlt_play_dead();
1477 }
1478 
1479 #else /* ... !CONFIG_HOTPLUG_CPU */
1480 int native_cpu_disable(void)
1481 {
1482 	return -ENOSYS;
1483 }
1484 
1485 void native_cpu_die(unsigned int cpu)
1486 {
1487 	/* We said "no" in __cpu_disable */
1488 	BUG();
1489 }
1490 
1491 void native_play_dead(void)
1492 {
1493 	BUG();
1494 }
1495 
1496 #endif
1497