xref: /linux/arch/x86/kernel/smpboot.c (revision 140eb5227767c6754742020a16d2691222b9c19b)
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/export.h>
47 #include <linux/sched.h>
48 #include <linux/sched/topology.h>
49 #include <linux/sched/hotplug.h>
50 #include <linux/sched/task_stack.h>
51 #include <linux/percpu.h>
52 #include <linux/bootmem.h>
53 #include <linux/err.h>
54 #include <linux/nmi.h>
55 #include <linux/tboot.h>
56 #include <linux/stackprotector.h>
57 #include <linux/gfp.h>
58 #include <linux/cpuidle.h>
59 
60 #include <asm/acpi.h>
61 #include <asm/desc.h>
62 #include <asm/nmi.h>
63 #include <asm/irq.h>
64 #include <asm/realmode.h>
65 #include <asm/cpu.h>
66 #include <asm/numa.h>
67 #include <asm/pgtable.h>
68 #include <asm/tlbflush.h>
69 #include <asm/mtrr.h>
70 #include <asm/mwait.h>
71 #include <asm/apic.h>
72 #include <asm/io_apic.h>
73 #include <asm/fpu/internal.h>
74 #include <asm/setup.h>
75 #include <asm/uv/uv.h>
76 #include <linux/mc146818rtc.h>
77 #include <asm/i8259.h>
78 #include <asm/realmode.h>
79 #include <asm/misc.h>
80 #include <asm/qspinlock.h>
81 
82 /* Number of siblings per CPU package */
83 int smp_num_siblings = 1;
84 EXPORT_SYMBOL(smp_num_siblings);
85 
86 /* Last level cache ID of each logical CPU */
87 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
88 
89 /* representing HT siblings of each logical CPU */
90 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
91 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
92 
93 /* representing HT and core siblings of each logical CPU */
94 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_map);
95 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
96 
97 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_llc_shared_map);
98 
99 /* Per CPU bogomips and other parameters */
100 DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
101 EXPORT_PER_CPU_SYMBOL(cpu_info);
102 
103 /* Logical package management. We might want to allocate that dynamically */
104 unsigned int __max_logical_packages __read_mostly;
105 EXPORT_SYMBOL(__max_logical_packages);
106 static unsigned int logical_packages __read_mostly;
107 
108 /* Maximum number of SMT threads on any online core */
109 int __read_mostly __max_smt_threads = 1;
110 
111 /* Flag to indicate if a complete sched domain rebuild is required */
112 bool x86_topology_update;
113 
114 int arch_update_cpu_topology(void)
115 {
116 	int retval = x86_topology_update;
117 
118 	x86_topology_update = false;
119 	return retval;
120 }
121 
122 static inline void smpboot_setup_warm_reset_vector(unsigned long start_eip)
123 {
124 	unsigned long flags;
125 
126 	spin_lock_irqsave(&rtc_lock, flags);
127 	CMOS_WRITE(0xa, 0xf);
128 	spin_unlock_irqrestore(&rtc_lock, flags);
129 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_HIGH)) =
130 							start_eip >> 4;
131 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) =
132 							start_eip & 0xf;
133 }
134 
135 static inline void smpboot_restore_warm_reset_vector(void)
136 {
137 	unsigned long flags;
138 
139 	/*
140 	 * Paranoid:  Set warm reset code and vector here back
141 	 * to default values.
142 	 */
143 	spin_lock_irqsave(&rtc_lock, flags);
144 	CMOS_WRITE(0, 0xf);
145 	spin_unlock_irqrestore(&rtc_lock, flags);
146 
147 	*((volatile u32 *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = 0;
148 }
149 
150 /*
151  * Report back to the Boot Processor during boot time or to the caller processor
152  * during CPU online.
153  */
154 static void smp_callin(void)
155 {
156 	int cpuid, phys_id;
157 
158 	/*
159 	 * If waken up by an INIT in an 82489DX configuration
160 	 * cpu_callout_mask guarantees we don't get here before
161 	 * an INIT_deassert IPI reaches our local APIC, so it is
162 	 * now safe to touch our local APIC.
163 	 */
164 	cpuid = smp_processor_id();
165 
166 	/*
167 	 * (This works even if the APIC is not enabled.)
168 	 */
169 	phys_id = read_apic_id();
170 
171 	/*
172 	 * the boot CPU has finished the init stage and is spinning
173 	 * on callin_map until we finish. We are free to set up this
174 	 * CPU, first the APIC. (this is probably redundant on most
175 	 * boards)
176 	 */
177 	apic_ap_setup();
178 
179 	/*
180 	 * Save our processor parameters. Note: this information
181 	 * is needed for clock calibration.
182 	 */
183 	smp_store_cpu_info(cpuid);
184 
185 	/*
186 	 * The topology information must be up to date before
187 	 * calibrate_delay() and notify_cpu_starting().
188 	 */
189 	set_cpu_sibling_map(raw_smp_processor_id());
190 
191 	/*
192 	 * Get our bogomips.
193 	 * Update loops_per_jiffy in cpu_data. Previous call to
194 	 * smp_store_cpu_info() stored a value that is close but not as
195 	 * accurate as the value just calculated.
196 	 */
197 	calibrate_delay();
198 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
199 	pr_debug("Stack at about %p\n", &cpuid);
200 
201 	wmb();
202 
203 	notify_cpu_starting(cpuid);
204 
205 	/*
206 	 * Allow the master to continue.
207 	 */
208 	cpumask_set_cpu(cpuid, cpu_callin_mask);
209 }
210 
211 static int cpu0_logical_apicid;
212 static int enable_start_cpu0;
213 /*
214  * Activate a secondary processor.
215  */
216 static void notrace start_secondary(void *unused)
217 {
218 	/*
219 	 * Don't put *anything* except direct CPU state initialization
220 	 * before cpu_init(), SMP booting is too fragile that we want to
221 	 * limit the things done here to the most necessary things.
222 	 */
223 	if (boot_cpu_has(X86_FEATURE_PCID))
224 		__write_cr4(__read_cr4() | X86_CR4_PCIDE);
225 
226 #ifdef CONFIG_X86_32
227 	/* switch away from the initial page table */
228 	load_cr3(swapper_pg_dir);
229 	__flush_tlb_all();
230 #endif
231 	load_current_idt();
232 	cpu_init();
233 	x86_cpuinit.early_percpu_clock_init();
234 	preempt_disable();
235 	smp_callin();
236 
237 	enable_start_cpu0 = 0;
238 
239 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
240 	barrier();
241 	/*
242 	 * Check TSC synchronization with the boot CPU:
243 	 */
244 	check_tsc_sync_target();
245 
246 	/*
247 	 * Lock vector_lock, set CPU online and bring the vector
248 	 * allocator online. Online must be set with vector_lock held
249 	 * to prevent a concurrent irq setup/teardown from seeing a
250 	 * half valid vector space.
251 	 */
252 	lock_vector_lock();
253 	set_cpu_online(smp_processor_id(), true);
254 	lapic_online();
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_AP_ONLINE_IDLE);
269 }
270 
271 /**
272  * topology_phys_to_logical_pkg - Map a physical package id to a logical
273  *
274  * Returns logical package id or -1 if not found
275  */
276 int topology_phys_to_logical_pkg(unsigned int phys_pkg)
277 {
278 	int cpu;
279 
280 	for_each_possible_cpu(cpu) {
281 		struct cpuinfo_x86 *c = &cpu_data(cpu);
282 
283 		if (c->initialized && c->phys_proc_id == phys_pkg)
284 			return c->logical_proc_id;
285 	}
286 	return -1;
287 }
288 EXPORT_SYMBOL(topology_phys_to_logical_pkg);
289 
290 /**
291  * topology_update_package_map - Update the physical to logical package map
292  * @pkg:	The physical package id as retrieved via CPUID
293  * @cpu:	The cpu for which this is updated
294  */
295 int topology_update_package_map(unsigned int pkg, unsigned int cpu)
296 {
297 	int new;
298 
299 	/* Already available somewhere? */
300 	new = topology_phys_to_logical_pkg(pkg);
301 	if (new >= 0)
302 		goto found;
303 
304 	new = logical_packages++;
305 	if (new != pkg) {
306 		pr_info("CPU %u Converting physical %u to logical package %u\n",
307 			cpu, pkg, new);
308 	}
309 found:
310 	cpu_data(cpu).logical_proc_id = new;
311 	return 0;
312 }
313 
314 void __init smp_store_boot_cpu_info(void)
315 {
316 	int id = 0; /* CPU 0 */
317 	struct cpuinfo_x86 *c = &cpu_data(id);
318 
319 	*c = boot_cpu_data;
320 	c->cpu_index = id;
321 	topology_update_package_map(c->phys_proc_id, id);
322 	c->initialized = true;
323 }
324 
325 /*
326  * The bootstrap kernel entry code has set these up. Save them for
327  * a given CPU
328  */
329 void smp_store_cpu_info(int id)
330 {
331 	struct cpuinfo_x86 *c = &cpu_data(id);
332 
333 	/* Copy boot_cpu_data only on the first bringup */
334 	if (!c->initialized)
335 		*c = boot_cpu_data;
336 	c->cpu_index = id;
337 	/*
338 	 * During boot time, CPU0 has this setup already. Save the info when
339 	 * bringing up AP or offlined CPU0.
340 	 */
341 	identify_secondary_cpu(c);
342 	c->initialized = true;
343 }
344 
345 static bool
346 topology_same_node(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
347 {
348 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
349 
350 	return (cpu_to_node(cpu1) == cpu_to_node(cpu2));
351 }
352 
353 static bool
354 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
355 {
356 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
357 
358 	return !WARN_ONCE(!topology_same_node(c, o),
359 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
360 		"[node: %d != %d]. Ignoring dependency.\n",
361 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
362 }
363 
364 #define link_mask(mfunc, c1, c2)					\
365 do {									\
366 	cpumask_set_cpu((c1), mfunc(c2));				\
367 	cpumask_set_cpu((c2), mfunc(c1));				\
368 } while (0)
369 
370 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
371 {
372 	if (boot_cpu_has(X86_FEATURE_TOPOEXT)) {
373 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
374 
375 		if (c->phys_proc_id == o->phys_proc_id &&
376 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2)) {
377 			if (c->cpu_core_id == o->cpu_core_id)
378 				return topology_sane(c, o, "smt");
379 
380 			if ((c->cu_id != 0xff) &&
381 			    (o->cu_id != 0xff) &&
382 			    (c->cu_id == o->cu_id))
383 				return topology_sane(c, o, "smt");
384 		}
385 
386 	} else if (c->phys_proc_id == o->phys_proc_id &&
387 		   c->cpu_core_id == o->cpu_core_id) {
388 		return topology_sane(c, o, "smt");
389 	}
390 
391 	return false;
392 }
393 
394 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
395 {
396 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
397 
398 	if (per_cpu(cpu_llc_id, cpu1) != BAD_APICID &&
399 	    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2))
400 		return topology_sane(c, o, "llc");
401 
402 	return false;
403 }
404 
405 /*
406  * Unlike the other levels, we do not enforce keeping a
407  * multicore group inside a NUMA node.  If this happens, we will
408  * discard the MC level of the topology later.
409  */
410 static bool match_die(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
411 {
412 	if (c->phys_proc_id == o->phys_proc_id)
413 		return true;
414 	return false;
415 }
416 
417 #if defined(CONFIG_SCHED_SMT) || defined(CONFIG_SCHED_MC)
418 static inline int x86_sched_itmt_flags(void)
419 {
420 	return sysctl_sched_itmt_enabled ? SD_ASYM_PACKING : 0;
421 }
422 
423 #ifdef CONFIG_SCHED_MC
424 static int x86_core_flags(void)
425 {
426 	return cpu_core_flags() | x86_sched_itmt_flags();
427 }
428 #endif
429 #ifdef CONFIG_SCHED_SMT
430 static int x86_smt_flags(void)
431 {
432 	return cpu_smt_flags() | x86_sched_itmt_flags();
433 }
434 #endif
435 #endif
436 
437 static struct sched_domain_topology_level x86_numa_in_package_topology[] = {
438 #ifdef CONFIG_SCHED_SMT
439 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
440 #endif
441 #ifdef CONFIG_SCHED_MC
442 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
443 #endif
444 	{ NULL, },
445 };
446 
447 static struct sched_domain_topology_level x86_topology[] = {
448 #ifdef CONFIG_SCHED_SMT
449 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
450 #endif
451 #ifdef CONFIG_SCHED_MC
452 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
453 #endif
454 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
455 	{ NULL, },
456 };
457 
458 /*
459  * Set if a package/die has multiple NUMA nodes inside.
460  * AMD Magny-Cours and Intel Cluster-on-Die have this.
461  */
462 static bool x86_has_numa_in_package;
463 
464 void set_cpu_sibling_map(int cpu)
465 {
466 	bool has_smt = smp_num_siblings > 1;
467 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
468 	struct cpuinfo_x86 *c = &cpu_data(cpu);
469 	struct cpuinfo_x86 *o;
470 	int i, threads;
471 
472 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
473 
474 	if (!has_mp) {
475 		cpumask_set_cpu(cpu, topology_sibling_cpumask(cpu));
476 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
477 		cpumask_set_cpu(cpu, topology_core_cpumask(cpu));
478 		c->booted_cores = 1;
479 		return;
480 	}
481 
482 	for_each_cpu(i, cpu_sibling_setup_mask) {
483 		o = &cpu_data(i);
484 
485 		if ((i == cpu) || (has_smt && match_smt(c, o)))
486 			link_mask(topology_sibling_cpumask, cpu, i);
487 
488 		if ((i == cpu) || (has_mp && match_llc(c, o)))
489 			link_mask(cpu_llc_shared_mask, cpu, i);
490 
491 	}
492 
493 	/*
494 	 * This needs a separate iteration over the cpus because we rely on all
495 	 * topology_sibling_cpumask links to be set-up.
496 	 */
497 	for_each_cpu(i, cpu_sibling_setup_mask) {
498 		o = &cpu_data(i);
499 
500 		if ((i == cpu) || (has_mp && match_die(c, o))) {
501 			link_mask(topology_core_cpumask, cpu, i);
502 
503 			/*
504 			 *  Does this new cpu bringup a new core?
505 			 */
506 			if (cpumask_weight(
507 			    topology_sibling_cpumask(cpu)) == 1) {
508 				/*
509 				 * for each core in package, increment
510 				 * the booted_cores for this new cpu
511 				 */
512 				if (cpumask_first(
513 				    topology_sibling_cpumask(i)) == i)
514 					c->booted_cores++;
515 				/*
516 				 * increment the core count for all
517 				 * the other cpus in this package
518 				 */
519 				if (i != cpu)
520 					cpu_data(i).booted_cores++;
521 			} else if (i != cpu && !c->booted_cores)
522 				c->booted_cores = cpu_data(i).booted_cores;
523 		}
524 		if (match_die(c, o) && !topology_same_node(c, o))
525 			x86_has_numa_in_package = true;
526 	}
527 
528 	threads = cpumask_weight(topology_sibling_cpumask(cpu));
529 	if (threads > __max_smt_threads)
530 		__max_smt_threads = threads;
531 }
532 
533 /* maps the cpu to the sched domain representing multi-core */
534 const struct cpumask *cpu_coregroup_mask(int cpu)
535 {
536 	return cpu_llc_shared_mask(cpu);
537 }
538 
539 static void impress_friends(void)
540 {
541 	int cpu;
542 	unsigned long bogosum = 0;
543 	/*
544 	 * Allow the user to impress friends.
545 	 */
546 	pr_debug("Before bogomips\n");
547 	for_each_possible_cpu(cpu)
548 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
549 			bogosum += cpu_data(cpu).loops_per_jiffy;
550 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
551 		num_online_cpus(),
552 		bogosum/(500000/HZ),
553 		(bogosum/(5000/HZ))%100);
554 
555 	pr_debug("Before bogocount - setting activated=1\n");
556 }
557 
558 void __inquire_remote_apic(int apicid)
559 {
560 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
561 	const char * const names[] = { "ID", "VERSION", "SPIV" };
562 	int timeout;
563 	u32 status;
564 
565 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
566 
567 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
568 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
569 
570 		/*
571 		 * Wait for idle.
572 		 */
573 		status = safe_apic_wait_icr_idle();
574 		if (status)
575 			pr_cont("a previous APIC delivery may have failed\n");
576 
577 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
578 
579 		timeout = 0;
580 		do {
581 			udelay(100);
582 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
583 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
584 
585 		switch (status) {
586 		case APIC_ICR_RR_VALID:
587 			status = apic_read(APIC_RRR);
588 			pr_cont("%08x\n", status);
589 			break;
590 		default:
591 			pr_cont("failed\n");
592 		}
593 	}
594 }
595 
596 /*
597  * The Multiprocessor Specification 1.4 (1997) example code suggests
598  * that there should be a 10ms delay between the BSP asserting INIT
599  * and de-asserting INIT, when starting a remote processor.
600  * But that slows boot and resume on modern processors, which include
601  * many cores and don't require that delay.
602  *
603  * Cmdline "init_cpu_udelay=" is available to over-ride this delay.
604  * Modern processor families are quirked to remove the delay entirely.
605  */
606 #define UDELAY_10MS_DEFAULT 10000
607 
608 static unsigned int init_udelay = UINT_MAX;
609 
610 static int __init cpu_init_udelay(char *str)
611 {
612 	get_option(&str, &init_udelay);
613 
614 	return 0;
615 }
616 early_param("cpu_init_udelay", cpu_init_udelay);
617 
618 static void __init smp_quirk_init_udelay(void)
619 {
620 	/* if cmdline changed it from default, leave it alone */
621 	if (init_udelay != UINT_MAX)
622 		return;
623 
624 	/* if modern processor, use no delay */
625 	if (((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && (boot_cpu_data.x86 == 6)) ||
626 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && (boot_cpu_data.x86 >= 0xF))) {
627 		init_udelay = 0;
628 		return;
629 	}
630 	/* else, use legacy delay */
631 	init_udelay = UDELAY_10MS_DEFAULT;
632 }
633 
634 /*
635  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
636  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
637  * won't ... remember to clear down the APIC, etc later.
638  */
639 int
640 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
641 {
642 	unsigned long send_status, accept_status = 0;
643 	int maxlvt;
644 
645 	/* Target chip */
646 	/* Boot on the stack */
647 	/* Kick the second */
648 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, apicid);
649 
650 	pr_debug("Waiting for send to finish...\n");
651 	send_status = safe_apic_wait_icr_idle();
652 
653 	/*
654 	 * Give the other CPU some time to accept the IPI.
655 	 */
656 	udelay(200);
657 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
658 		maxlvt = lapic_get_maxlvt();
659 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
660 			apic_write(APIC_ESR, 0);
661 		accept_status = (apic_read(APIC_ESR) & 0xEF);
662 	}
663 	pr_debug("NMI sent\n");
664 
665 	if (send_status)
666 		pr_err("APIC never delivered???\n");
667 	if (accept_status)
668 		pr_err("APIC delivery error (%lx)\n", accept_status);
669 
670 	return (send_status | accept_status);
671 }
672 
673 static int
674 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
675 {
676 	unsigned long send_status = 0, accept_status = 0;
677 	int maxlvt, num_starts, j;
678 
679 	maxlvt = lapic_get_maxlvt();
680 
681 	/*
682 	 * Be paranoid about clearing APIC errors.
683 	 */
684 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
685 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
686 			apic_write(APIC_ESR, 0);
687 		apic_read(APIC_ESR);
688 	}
689 
690 	pr_debug("Asserting INIT\n");
691 
692 	/*
693 	 * Turn INIT on target chip
694 	 */
695 	/*
696 	 * Send IPI
697 	 */
698 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
699 		       phys_apicid);
700 
701 	pr_debug("Waiting for send to finish...\n");
702 	send_status = safe_apic_wait_icr_idle();
703 
704 	udelay(init_udelay);
705 
706 	pr_debug("Deasserting INIT\n");
707 
708 	/* Target chip */
709 	/* Send IPI */
710 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
711 
712 	pr_debug("Waiting for send to finish...\n");
713 	send_status = safe_apic_wait_icr_idle();
714 
715 	mb();
716 
717 	/*
718 	 * Should we send STARTUP IPIs ?
719 	 *
720 	 * Determine this based on the APIC version.
721 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
722 	 */
723 	if (APIC_INTEGRATED(boot_cpu_apic_version))
724 		num_starts = 2;
725 	else
726 		num_starts = 0;
727 
728 	/*
729 	 * Run STARTUP IPI loop.
730 	 */
731 	pr_debug("#startup loops: %d\n", num_starts);
732 
733 	for (j = 1; j <= num_starts; j++) {
734 		pr_debug("Sending STARTUP #%d\n", j);
735 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
736 			apic_write(APIC_ESR, 0);
737 		apic_read(APIC_ESR);
738 		pr_debug("After apic_write\n");
739 
740 		/*
741 		 * STARTUP IPI
742 		 */
743 
744 		/* Target chip */
745 		/* Boot on the stack */
746 		/* Kick the second */
747 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
748 			       phys_apicid);
749 
750 		/*
751 		 * Give the other CPU some time to accept the IPI.
752 		 */
753 		if (init_udelay == 0)
754 			udelay(10);
755 		else
756 			udelay(300);
757 
758 		pr_debug("Startup point 1\n");
759 
760 		pr_debug("Waiting for send to finish...\n");
761 		send_status = safe_apic_wait_icr_idle();
762 
763 		/*
764 		 * Give the other CPU some time to accept the IPI.
765 		 */
766 		if (init_udelay == 0)
767 			udelay(10);
768 		else
769 			udelay(200);
770 
771 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
772 			apic_write(APIC_ESR, 0);
773 		accept_status = (apic_read(APIC_ESR) & 0xEF);
774 		if (send_status || accept_status)
775 			break;
776 	}
777 	pr_debug("After Startup\n");
778 
779 	if (send_status)
780 		pr_err("APIC never delivered???\n");
781 	if (accept_status)
782 		pr_err("APIC delivery error (%lx)\n", accept_status);
783 
784 	return (send_status | accept_status);
785 }
786 
787 /* reduce the number of lines printed when booting a large cpu count system */
788 static void announce_cpu(int cpu, int apicid)
789 {
790 	static int current_node = -1;
791 	int node = early_cpu_to_node(cpu);
792 	static int width, node_width;
793 
794 	if (!width)
795 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
796 
797 	if (!node_width)
798 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
799 
800 	if (cpu == 1)
801 		printk(KERN_INFO "x86: Booting SMP configuration:\n");
802 
803 	if (system_state < SYSTEM_RUNNING) {
804 		if (node != current_node) {
805 			if (current_node > (-1))
806 				pr_cont("\n");
807 			current_node = node;
808 
809 			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
810 			       node_width - num_digits(node), " ", node);
811 		}
812 
813 		/* Add padding for the BSP */
814 		if (cpu == 1)
815 			pr_cont("%*s", width + 1, " ");
816 
817 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
818 
819 	} else
820 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
821 			node, cpu, apicid);
822 }
823 
824 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
825 {
826 	int cpu;
827 
828 	cpu = smp_processor_id();
829 	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
830 		return NMI_HANDLED;
831 
832 	return NMI_DONE;
833 }
834 
835 /*
836  * Wake up AP by INIT, INIT, STARTUP sequence.
837  *
838  * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
839  * boot-strap code which is not a desired behavior for waking up BSP. To
840  * void the boot-strap code, wake up CPU0 by NMI instead.
841  *
842  * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
843  * (i.e. physically hot removed and then hot added), NMI won't wake it up.
844  * We'll change this code in the future to wake up hard offlined CPU0 if
845  * real platform and request are available.
846  */
847 static int
848 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
849 	       int *cpu0_nmi_registered)
850 {
851 	int id;
852 	int boot_error;
853 
854 	preempt_disable();
855 
856 	/*
857 	 * Wake up AP by INIT, INIT, STARTUP sequence.
858 	 */
859 	if (cpu) {
860 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
861 		goto out;
862 	}
863 
864 	/*
865 	 * Wake up BSP by nmi.
866 	 *
867 	 * Register a NMI handler to help wake up CPU0.
868 	 */
869 	boot_error = register_nmi_handler(NMI_LOCAL,
870 					  wakeup_cpu0_nmi, 0, "wake_cpu0");
871 
872 	if (!boot_error) {
873 		enable_start_cpu0 = 1;
874 		*cpu0_nmi_registered = 1;
875 		if (apic->dest_logical == APIC_DEST_LOGICAL)
876 			id = cpu0_logical_apicid;
877 		else
878 			id = apicid;
879 		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
880 	}
881 
882 out:
883 	preempt_enable();
884 
885 	return boot_error;
886 }
887 
888 void common_cpu_up(unsigned int cpu, struct task_struct *idle)
889 {
890 	/* Just in case we booted with a single CPU. */
891 	alternatives_enable_smp();
892 
893 	per_cpu(current_task, cpu) = idle;
894 
895 #ifdef CONFIG_X86_32
896 	/* Stack for startup_32 can be just as for start_secondary onwards */
897 	irq_ctx_init(cpu);
898 	per_cpu(cpu_current_top_of_stack, cpu) = task_top_of_stack(idle);
899 #else
900 	initial_gs = per_cpu_offset(cpu);
901 #endif
902 }
903 
904 /*
905  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
906  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
907  * Returns zero if CPU booted OK, else error code from
908  * ->wakeup_secondary_cpu.
909  */
910 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle,
911 		       int *cpu0_nmi_registered)
912 {
913 	volatile u32 *trampoline_status =
914 		(volatile u32 *) __va(real_mode_header->trampoline_status);
915 	/* start_ip had better be page-aligned! */
916 	unsigned long start_ip = real_mode_header->trampoline_start;
917 
918 	unsigned long boot_error = 0;
919 	unsigned long timeout;
920 
921 	idle->thread.sp = (unsigned long)task_pt_regs(idle);
922 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_rw(cpu);
923 	initial_code = (unsigned long)start_secondary;
924 	initial_stack  = idle->thread.sp;
925 
926 	/* Enable the espfix hack for this CPU */
927 	init_espfix_ap(cpu);
928 
929 	/* So we see what's up */
930 	announce_cpu(cpu, apicid);
931 
932 	/*
933 	 * This grunge runs the startup process for
934 	 * the targeted processor.
935 	 */
936 
937 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
938 
939 		pr_debug("Setting warm reset code and vector.\n");
940 
941 		smpboot_setup_warm_reset_vector(start_ip);
942 		/*
943 		 * Be paranoid about clearing APIC errors.
944 		*/
945 		if (APIC_INTEGRATED(boot_cpu_apic_version)) {
946 			apic_write(APIC_ESR, 0);
947 			apic_read(APIC_ESR);
948 		}
949 	}
950 
951 	/*
952 	 * AP might wait on cpu_callout_mask in cpu_init() with
953 	 * cpu_initialized_mask set if previous attempt to online
954 	 * it timed-out. Clear cpu_initialized_mask so that after
955 	 * INIT/SIPI it could start with a clean state.
956 	 */
957 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
958 	smp_mb();
959 
960 	/*
961 	 * Wake up a CPU in difference cases:
962 	 * - Use the method in the APIC driver if it's defined
963 	 * Otherwise,
964 	 * - Use an INIT boot APIC message for APs or NMI for BSP.
965 	 */
966 	if (apic->wakeup_secondary_cpu)
967 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
968 	else
969 		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
970 						     cpu0_nmi_registered);
971 
972 	if (!boot_error) {
973 		/*
974 		 * Wait 10s total for first sign of life from AP
975 		 */
976 		boot_error = -1;
977 		timeout = jiffies + 10*HZ;
978 		while (time_before(jiffies, timeout)) {
979 			if (cpumask_test_cpu(cpu, cpu_initialized_mask)) {
980 				/*
981 				 * Tell AP to proceed with initialization
982 				 */
983 				cpumask_set_cpu(cpu, cpu_callout_mask);
984 				boot_error = 0;
985 				break;
986 			}
987 			schedule();
988 		}
989 	}
990 
991 	if (!boot_error) {
992 		/*
993 		 * Wait till AP completes initial initialization
994 		 */
995 		while (!cpumask_test_cpu(cpu, cpu_callin_mask)) {
996 			/*
997 			 * Allow other tasks to run while we wait for the
998 			 * AP to come online. This also gives a chance
999 			 * for the MTRR work(triggered by the AP coming online)
1000 			 * to be completed in the stop machine context.
1001 			 */
1002 			schedule();
1003 		}
1004 	}
1005 
1006 	/* mark "stuck" area as not stuck */
1007 	*trampoline_status = 0;
1008 
1009 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
1010 		/*
1011 		 * Cleanup possible dangling ends...
1012 		 */
1013 		smpboot_restore_warm_reset_vector();
1014 	}
1015 
1016 	return boot_error;
1017 }
1018 
1019 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
1020 {
1021 	int apicid = apic->cpu_present_to_apicid(cpu);
1022 	int cpu0_nmi_registered = 0;
1023 	unsigned long flags;
1024 	int err, ret = 0;
1025 
1026 	lockdep_assert_irqs_enabled();
1027 
1028 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
1029 
1030 	if (apicid == BAD_APICID ||
1031 	    !physid_isset(apicid, phys_cpu_present_map) ||
1032 	    !apic->apic_id_valid(apicid)) {
1033 		pr_err("%s: bad cpu %d\n", __func__, cpu);
1034 		return -EINVAL;
1035 	}
1036 
1037 	/*
1038 	 * Already booted CPU?
1039 	 */
1040 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
1041 		pr_debug("do_boot_cpu %d Already started\n", cpu);
1042 		return -ENOSYS;
1043 	}
1044 
1045 	/*
1046 	 * Save current MTRR state in case it was changed since early boot
1047 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
1048 	 */
1049 	mtrr_save_state();
1050 
1051 	/* x86 CPUs take themselves offline, so delayed offline is OK. */
1052 	err = cpu_check_up_prepare(cpu);
1053 	if (err && err != -EBUSY)
1054 		return err;
1055 
1056 	/* the FPU context is blank, nobody can own it */
1057 	per_cpu(fpu_fpregs_owner_ctx, cpu) = NULL;
1058 
1059 	common_cpu_up(cpu, tidle);
1060 
1061 	err = do_boot_cpu(apicid, cpu, tidle, &cpu0_nmi_registered);
1062 	if (err) {
1063 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
1064 		ret = -EIO;
1065 		goto unreg_nmi;
1066 	}
1067 
1068 	/*
1069 	 * Check TSC synchronization with the AP (keep irqs disabled
1070 	 * while doing so):
1071 	 */
1072 	local_irq_save(flags);
1073 	check_tsc_sync_source(cpu);
1074 	local_irq_restore(flags);
1075 
1076 	while (!cpu_online(cpu)) {
1077 		cpu_relax();
1078 		touch_nmi_watchdog();
1079 	}
1080 
1081 unreg_nmi:
1082 	/*
1083 	 * Clean up the nmi handler. Do this after the callin and callout sync
1084 	 * to avoid impact of possible long unregister time.
1085 	 */
1086 	if (cpu0_nmi_registered)
1087 		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
1088 
1089 	return ret;
1090 }
1091 
1092 /**
1093  * arch_disable_smp_support() - disables SMP support for x86 at runtime
1094  */
1095 void arch_disable_smp_support(void)
1096 {
1097 	disable_ioapic_support();
1098 }
1099 
1100 /*
1101  * Fall back to non SMP mode after errors.
1102  *
1103  * RED-PEN audit/test this more. I bet there is more state messed up here.
1104  */
1105 static __init void disable_smp(void)
1106 {
1107 	pr_info("SMP disabled\n");
1108 
1109 	disable_ioapic_support();
1110 
1111 	init_cpu_present(cpumask_of(0));
1112 	init_cpu_possible(cpumask_of(0));
1113 
1114 	if (smp_found_config)
1115 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1116 	else
1117 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1118 	cpumask_set_cpu(0, topology_sibling_cpumask(0));
1119 	cpumask_set_cpu(0, topology_core_cpumask(0));
1120 }
1121 
1122 /*
1123  * Various sanity checks.
1124  */
1125 static void __init smp_sanity_check(void)
1126 {
1127 	preempt_disable();
1128 
1129 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
1130 	if (def_to_bigsmp && nr_cpu_ids > 8) {
1131 		unsigned int cpu;
1132 		unsigned nr;
1133 
1134 		pr_warn("More than 8 CPUs detected - skipping them\n"
1135 			"Use CONFIG_X86_BIGSMP\n");
1136 
1137 		nr = 0;
1138 		for_each_present_cpu(cpu) {
1139 			if (nr >= 8)
1140 				set_cpu_present(cpu, false);
1141 			nr++;
1142 		}
1143 
1144 		nr = 0;
1145 		for_each_possible_cpu(cpu) {
1146 			if (nr >= 8)
1147 				set_cpu_possible(cpu, false);
1148 			nr++;
1149 		}
1150 
1151 		nr_cpu_ids = 8;
1152 	}
1153 #endif
1154 
1155 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1156 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1157 			hard_smp_processor_id());
1158 
1159 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1160 	}
1161 
1162 	/*
1163 	 * Should not be necessary because the MP table should list the boot
1164 	 * CPU too, but we do it for the sake of robustness anyway.
1165 	 */
1166 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1167 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1168 			  boot_cpu_physical_apicid);
1169 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1170 	}
1171 	preempt_enable();
1172 }
1173 
1174 static void __init smp_cpu_index_default(void)
1175 {
1176 	int i;
1177 	struct cpuinfo_x86 *c;
1178 
1179 	for_each_possible_cpu(i) {
1180 		c = &cpu_data(i);
1181 		/* mark all to hotplug */
1182 		c->cpu_index = nr_cpu_ids;
1183 	}
1184 }
1185 
1186 static void __init smp_get_logical_apicid(void)
1187 {
1188 	if (x2apic_mode)
1189 		cpu0_logical_apicid = apic_read(APIC_LDR);
1190 	else
1191 		cpu0_logical_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1192 }
1193 
1194 /*
1195  * Prepare for SMP bootup.
1196  * @max_cpus: configured maximum number of CPUs, It is a legacy parameter
1197  *            for common interface support.
1198  */
1199 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1200 {
1201 	unsigned int i;
1202 
1203 	smp_cpu_index_default();
1204 
1205 	/*
1206 	 * Setup boot CPU information
1207 	 */
1208 	smp_store_boot_cpu_info(); /* Final full version of the data */
1209 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1210 	mb();
1211 
1212 	for_each_possible_cpu(i) {
1213 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1214 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1215 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1216 	}
1217 
1218 	/*
1219 	 * Set 'default' x86 topology, this matches default_topology() in that
1220 	 * it has NUMA nodes as a topology level. See also
1221 	 * native_smp_cpus_done().
1222 	 *
1223 	 * Must be done before set_cpus_sibling_map() is ran.
1224 	 */
1225 	set_sched_topology(x86_topology);
1226 
1227 	set_cpu_sibling_map(0);
1228 
1229 	smp_sanity_check();
1230 
1231 	switch (apic_intr_mode) {
1232 	case APIC_PIC:
1233 	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1234 		disable_smp();
1235 		return;
1236 	case APIC_SYMMETRIC_IO_NO_ROUTING:
1237 		disable_smp();
1238 		/* Setup local timer */
1239 		x86_init.timers.setup_percpu_clockev();
1240 		return;
1241 	case APIC_VIRTUAL_WIRE:
1242 	case APIC_SYMMETRIC_IO:
1243 		break;
1244 	}
1245 
1246 	/* Setup local timer */
1247 	x86_init.timers.setup_percpu_clockev();
1248 
1249 	smp_get_logical_apicid();
1250 
1251 	pr_info("CPU0: ");
1252 	print_cpu_info(&cpu_data(0));
1253 
1254 	native_pv_lock_init();
1255 
1256 	uv_system_init();
1257 
1258 	set_mtrr_aps_delayed_init();
1259 
1260 	smp_quirk_init_udelay();
1261 }
1262 
1263 void arch_enable_nonboot_cpus_begin(void)
1264 {
1265 	set_mtrr_aps_delayed_init();
1266 }
1267 
1268 void arch_enable_nonboot_cpus_end(void)
1269 {
1270 	mtrr_aps_init();
1271 }
1272 
1273 /*
1274  * Early setup to make printk work.
1275  */
1276 void __init native_smp_prepare_boot_cpu(void)
1277 {
1278 	int me = smp_processor_id();
1279 	switch_to_new_gdt(me);
1280 	/* already set me in cpu_online_mask in boot_cpu_init() */
1281 	cpumask_set_cpu(me, cpu_callout_mask);
1282 	cpu_set_state_online(me);
1283 }
1284 
1285 void __init native_smp_cpus_done(unsigned int max_cpus)
1286 {
1287 	int ncpus;
1288 
1289 	pr_debug("Boot done\n");
1290 	/*
1291 	 * Today neither Intel nor AMD support heterogenous systems so
1292 	 * extrapolate the boot cpu's data to all packages.
1293 	 */
1294 	ncpus = cpu_data(0).booted_cores * topology_max_smt_threads();
1295 	__max_logical_packages = DIV_ROUND_UP(nr_cpu_ids, ncpus);
1296 	pr_info("Max logical packages: %u\n", __max_logical_packages);
1297 
1298 	if (x86_has_numa_in_package)
1299 		set_sched_topology(x86_numa_in_package_topology);
1300 
1301 	nmi_selftest();
1302 	impress_friends();
1303 	mtrr_aps_init();
1304 }
1305 
1306 static int __initdata setup_possible_cpus = -1;
1307 static int __init _setup_possible_cpus(char *str)
1308 {
1309 	get_option(&str, &setup_possible_cpus);
1310 	return 0;
1311 }
1312 early_param("possible_cpus", _setup_possible_cpus);
1313 
1314 
1315 /*
1316  * cpu_possible_mask should be static, it cannot change as cpu's
1317  * are onlined, or offlined. The reason is per-cpu data-structures
1318  * are allocated by some modules at init time, and dont expect to
1319  * do this dynamically on cpu arrival/departure.
1320  * cpu_present_mask on the other hand can change dynamically.
1321  * In case when cpu_hotplug is not compiled, then we resort to current
1322  * behaviour, which is cpu_possible == cpu_present.
1323  * - Ashok Raj
1324  *
1325  * Three ways to find out the number of additional hotplug CPUs:
1326  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1327  * - The user can overwrite it with possible_cpus=NUM
1328  * - Otherwise don't reserve additional CPUs.
1329  * We do this because additional CPUs waste a lot of memory.
1330  * -AK
1331  */
1332 __init void prefill_possible_map(void)
1333 {
1334 	int i, possible;
1335 
1336 	/* No boot processor was found in mptable or ACPI MADT */
1337 	if (!num_processors) {
1338 		if (boot_cpu_has(X86_FEATURE_APIC)) {
1339 			int apicid = boot_cpu_physical_apicid;
1340 			int cpu = hard_smp_processor_id();
1341 
1342 			pr_warn("Boot CPU (id %d) not listed by BIOS\n", cpu);
1343 
1344 			/* Make sure boot cpu is enumerated */
1345 			if (apic->cpu_present_to_apicid(0) == BAD_APICID &&
1346 			    apic->apic_id_valid(apicid))
1347 				generic_processor_info(apicid, boot_cpu_apic_version);
1348 		}
1349 
1350 		if (!num_processors)
1351 			num_processors = 1;
1352 	}
1353 
1354 	i = setup_max_cpus ?: 1;
1355 	if (setup_possible_cpus == -1) {
1356 		possible = num_processors;
1357 #ifdef CONFIG_HOTPLUG_CPU
1358 		if (setup_max_cpus)
1359 			possible += disabled_cpus;
1360 #else
1361 		if (possible > i)
1362 			possible = i;
1363 #endif
1364 	} else
1365 		possible = setup_possible_cpus;
1366 
1367 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1368 
1369 	/* nr_cpu_ids could be reduced via nr_cpus= */
1370 	if (possible > nr_cpu_ids) {
1371 		pr_warn("%d Processors exceeds NR_CPUS limit of %u\n",
1372 			possible, nr_cpu_ids);
1373 		possible = nr_cpu_ids;
1374 	}
1375 
1376 #ifdef CONFIG_HOTPLUG_CPU
1377 	if (!setup_max_cpus)
1378 #endif
1379 	if (possible > i) {
1380 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1381 			possible, setup_max_cpus);
1382 		possible = i;
1383 	}
1384 
1385 	nr_cpu_ids = possible;
1386 
1387 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1388 		possible, max_t(int, possible - num_processors, 0));
1389 
1390 	reset_cpu_possible_mask();
1391 
1392 	for (i = 0; i < possible; i++)
1393 		set_cpu_possible(i, true);
1394 }
1395 
1396 #ifdef CONFIG_HOTPLUG_CPU
1397 
1398 /* Recompute SMT state for all CPUs on offline */
1399 static void recompute_smt_state(void)
1400 {
1401 	int max_threads, cpu;
1402 
1403 	max_threads = 0;
1404 	for_each_online_cpu (cpu) {
1405 		int threads = cpumask_weight(topology_sibling_cpumask(cpu));
1406 
1407 		if (threads > max_threads)
1408 			max_threads = threads;
1409 	}
1410 	__max_smt_threads = max_threads;
1411 }
1412 
1413 static void remove_siblinginfo(int cpu)
1414 {
1415 	int sibling;
1416 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1417 
1418 	for_each_cpu(sibling, topology_core_cpumask(cpu)) {
1419 		cpumask_clear_cpu(cpu, topology_core_cpumask(sibling));
1420 		/*/
1421 		 * last thread sibling in this cpu core going down
1422 		 */
1423 		if (cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
1424 			cpu_data(sibling).booted_cores--;
1425 	}
1426 
1427 	for_each_cpu(sibling, topology_sibling_cpumask(cpu))
1428 		cpumask_clear_cpu(cpu, topology_sibling_cpumask(sibling));
1429 	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1430 		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1431 	cpumask_clear(cpu_llc_shared_mask(cpu));
1432 	cpumask_clear(topology_sibling_cpumask(cpu));
1433 	cpumask_clear(topology_core_cpumask(cpu));
1434 	c->phys_proc_id = 0;
1435 	c->cpu_core_id = 0;
1436 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1437 	recompute_smt_state();
1438 }
1439 
1440 static void remove_cpu_from_maps(int cpu)
1441 {
1442 	set_cpu_online(cpu, false);
1443 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1444 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1445 	/* was set by cpu_init() */
1446 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1447 	numa_remove_cpu(cpu);
1448 }
1449 
1450 void cpu_disable_common(void)
1451 {
1452 	int cpu = smp_processor_id();
1453 
1454 	remove_siblinginfo(cpu);
1455 
1456 	/* It's now safe to remove this processor from the online map */
1457 	lock_vector_lock();
1458 	remove_cpu_from_maps(cpu);
1459 	unlock_vector_lock();
1460 	fixup_irqs();
1461 	lapic_offline();
1462 }
1463 
1464 int native_cpu_disable(void)
1465 {
1466 	int ret;
1467 
1468 	ret = lapic_can_unplug_cpu();
1469 	if (ret)
1470 		return ret;
1471 
1472 	clear_local_APIC();
1473 	cpu_disable_common();
1474 
1475 	return 0;
1476 }
1477 
1478 int common_cpu_die(unsigned int cpu)
1479 {
1480 	int ret = 0;
1481 
1482 	/* We don't do anything here: idle task is faking death itself. */
1483 
1484 	/* They ack this in play_dead() by setting CPU_DEAD */
1485 	if (cpu_wait_death(cpu, 5)) {
1486 		if (system_state == SYSTEM_RUNNING)
1487 			pr_info("CPU %u is now offline\n", cpu);
1488 	} else {
1489 		pr_err("CPU %u didn't die...\n", cpu);
1490 		ret = -1;
1491 	}
1492 
1493 	return ret;
1494 }
1495 
1496 void native_cpu_die(unsigned int cpu)
1497 {
1498 	common_cpu_die(cpu);
1499 }
1500 
1501 void play_dead_common(void)
1502 {
1503 	idle_task_exit();
1504 
1505 	/* Ack it */
1506 	(void)cpu_report_death();
1507 
1508 	/*
1509 	 * With physical CPU hotplug, we should halt the cpu
1510 	 */
1511 	local_irq_disable();
1512 }
1513 
1514 static bool wakeup_cpu0(void)
1515 {
1516 	if (smp_processor_id() == 0 && enable_start_cpu0)
1517 		return true;
1518 
1519 	return false;
1520 }
1521 
1522 /*
1523  * We need to flush the caches before going to sleep, lest we have
1524  * dirty data in our caches when we come back up.
1525  */
1526 static inline void mwait_play_dead(void)
1527 {
1528 	unsigned int eax, ebx, ecx, edx;
1529 	unsigned int highest_cstate = 0;
1530 	unsigned int highest_subcstate = 0;
1531 	void *mwait_ptr;
1532 	int i;
1533 
1534 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1535 		return;
1536 	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1537 		return;
1538 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1539 		return;
1540 
1541 	eax = CPUID_MWAIT_LEAF;
1542 	ecx = 0;
1543 	native_cpuid(&eax, &ebx, &ecx, &edx);
1544 
1545 	/*
1546 	 * eax will be 0 if EDX enumeration is not valid.
1547 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1548 	 */
1549 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1550 		eax = 0;
1551 	} else {
1552 		edx >>= MWAIT_SUBSTATE_SIZE;
1553 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1554 			if (edx & MWAIT_SUBSTATE_MASK) {
1555 				highest_cstate = i;
1556 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1557 			}
1558 		}
1559 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1560 			(highest_subcstate - 1);
1561 	}
1562 
1563 	/*
1564 	 * This should be a memory location in a cache line which is
1565 	 * unlikely to be touched by other processors.  The actual
1566 	 * content is immaterial as it is not actually modified in any way.
1567 	 */
1568 	mwait_ptr = &current_thread_info()->flags;
1569 
1570 	wbinvd();
1571 
1572 	while (1) {
1573 		/*
1574 		 * The CLFLUSH is a workaround for erratum AAI65 for
1575 		 * the Xeon 7400 series.  It's not clear it is actually
1576 		 * needed, but it should be harmless in either case.
1577 		 * The WBINVD is insufficient due to the spurious-wakeup
1578 		 * case where we return around the loop.
1579 		 */
1580 		mb();
1581 		clflush(mwait_ptr);
1582 		mb();
1583 		__monitor(mwait_ptr, 0, 0);
1584 		mb();
1585 		__mwait(eax, 0);
1586 		/*
1587 		 * If NMI wants to wake up CPU0, start CPU0.
1588 		 */
1589 		if (wakeup_cpu0())
1590 			start_cpu0();
1591 	}
1592 }
1593 
1594 void hlt_play_dead(void)
1595 {
1596 	if (__this_cpu_read(cpu_info.x86) >= 4)
1597 		wbinvd();
1598 
1599 	while (1) {
1600 		native_halt();
1601 		/*
1602 		 * If NMI wants to wake up CPU0, start CPU0.
1603 		 */
1604 		if (wakeup_cpu0())
1605 			start_cpu0();
1606 	}
1607 }
1608 
1609 void native_play_dead(void)
1610 {
1611 	play_dead_common();
1612 	tboot_shutdown(TB_SHUTDOWN_WFS);
1613 
1614 	mwait_play_dead();	/* Only returns on failure */
1615 	if (cpuidle_play_dead())
1616 		hlt_play_dead();
1617 }
1618 
1619 #else /* ... !CONFIG_HOTPLUG_CPU */
1620 int native_cpu_disable(void)
1621 {
1622 	return -ENOSYS;
1623 }
1624 
1625 void native_cpu_die(unsigned int cpu)
1626 {
1627 	/* We said "no" in __cpu_disable */
1628 	BUG();
1629 }
1630 
1631 void native_play_dead(void)
1632 {
1633 	BUG();
1634 }
1635 
1636 #endif
1637