xref: /openbmc/linux/arch/x86/kernel/smpboot.c (revision aac5987a)
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 
81 /* Number of siblings per CPU package */
82 int smp_num_siblings = 1;
83 EXPORT_SYMBOL(smp_num_siblings);
84 
85 /* Last level cache ID of each logical CPU */
86 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
87 
88 /* representing HT siblings of each logical CPU */
89 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
90 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
91 
92 /* representing HT and core siblings of each logical CPU */
93 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_map);
94 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
95 
96 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_llc_shared_map);
97 
98 /* Per CPU bogomips and other parameters */
99 DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
100 EXPORT_PER_CPU_SYMBOL(cpu_info);
101 
102 /* Logical package management. We might want to allocate that dynamically */
103 static int *physical_to_logical_pkg __read_mostly;
104 static unsigned long *physical_package_map __read_mostly;;
105 static unsigned int max_physical_pkg_id __read_mostly;
106 unsigned int __max_logical_packages __read_mostly;
107 EXPORT_SYMBOL(__max_logical_packages);
108 static unsigned int logical_packages __read_mostly;
109 
110 /* Maximum number of SMT threads on any online core */
111 int __max_smt_threads __read_mostly;
112 
113 /* Flag to indicate if a complete sched domain rebuild is required */
114 bool x86_topology_update;
115 
116 int arch_update_cpu_topology(void)
117 {
118 	int retval = x86_topology_update;
119 
120 	x86_topology_update = false;
121 	return retval;
122 }
123 
124 static inline void smpboot_setup_warm_reset_vector(unsigned long start_eip)
125 {
126 	unsigned long flags;
127 
128 	spin_lock_irqsave(&rtc_lock, flags);
129 	CMOS_WRITE(0xa, 0xf);
130 	spin_unlock_irqrestore(&rtc_lock, flags);
131 	local_flush_tlb();
132 	pr_debug("1.\n");
133 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_HIGH)) =
134 							start_eip >> 4;
135 	pr_debug("2.\n");
136 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) =
137 							start_eip & 0xf;
138 	pr_debug("3.\n");
139 }
140 
141 static inline void smpboot_restore_warm_reset_vector(void)
142 {
143 	unsigned long flags;
144 
145 	/*
146 	 * Install writable page 0 entry to set BIOS data area.
147 	 */
148 	local_flush_tlb();
149 
150 	/*
151 	 * Paranoid:  Set warm reset code and vector here back
152 	 * to default values.
153 	 */
154 	spin_lock_irqsave(&rtc_lock, flags);
155 	CMOS_WRITE(0, 0xf);
156 	spin_unlock_irqrestore(&rtc_lock, flags);
157 
158 	*((volatile u32 *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = 0;
159 }
160 
161 /*
162  * Report back to the Boot Processor during boot time or to the caller processor
163  * during CPU online.
164  */
165 static void smp_callin(void)
166 {
167 	int cpuid, phys_id;
168 
169 	/*
170 	 * If waken up by an INIT in an 82489DX configuration
171 	 * cpu_callout_mask guarantees we don't get here before
172 	 * an INIT_deassert IPI reaches our local APIC, so it is
173 	 * now safe to touch our local APIC.
174 	 */
175 	cpuid = smp_processor_id();
176 
177 	/*
178 	 * (This works even if the APIC is not enabled.)
179 	 */
180 	phys_id = read_apic_id();
181 
182 	/*
183 	 * the boot CPU has finished the init stage and is spinning
184 	 * on callin_map until we finish. We are free to set up this
185 	 * CPU, first the APIC. (this is probably redundant on most
186 	 * boards)
187 	 */
188 	apic_ap_setup();
189 
190 	/*
191 	 * Save our processor parameters. Note: this information
192 	 * is needed for clock calibration.
193 	 */
194 	smp_store_cpu_info(cpuid);
195 
196 	/*
197 	 * Get our bogomips.
198 	 * Update loops_per_jiffy in cpu_data. Previous call to
199 	 * smp_store_cpu_info() stored a value that is close but not as
200 	 * accurate as the value just calculated.
201 	 */
202 	calibrate_delay();
203 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
204 	pr_debug("Stack at about %p\n", &cpuid);
205 
206 	/*
207 	 * This must be done before setting cpu_online_mask
208 	 * or calling notify_cpu_starting.
209 	 */
210 	set_cpu_sibling_map(raw_smp_processor_id());
211 	wmb();
212 
213 	notify_cpu_starting(cpuid);
214 
215 	/*
216 	 * Allow the master to continue.
217 	 */
218 	cpumask_set_cpu(cpuid, cpu_callin_mask);
219 }
220 
221 static int cpu0_logical_apicid;
222 static int enable_start_cpu0;
223 /*
224  * Activate a secondary processor.
225  */
226 static void notrace start_secondary(void *unused)
227 {
228 	/*
229 	 * Don't put *anything* before cpu_init(), SMP booting is too
230 	 * fragile that we want to limit the things done here to the
231 	 * most necessary things.
232 	 */
233 	cpu_init();
234 	x86_cpuinit.early_percpu_clock_init();
235 	preempt_disable();
236 	smp_callin();
237 
238 	enable_start_cpu0 = 0;
239 
240 #ifdef CONFIG_X86_32
241 	/* switch away from the initial page table */
242 	load_cr3(swapper_pg_dir);
243 	__flush_tlb_all();
244 #endif
245 
246 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
247 	barrier();
248 	/*
249 	 * Check TSC synchronization with the BP:
250 	 */
251 	check_tsc_sync_target();
252 
253 	/*
254 	 * Lock vector_lock and initialize the vectors on this cpu
255 	 * before setting the cpu online. We must set it online with
256 	 * vector_lock held to prevent a concurrent setup/teardown
257 	 * from seeing a half valid vector space.
258 	 */
259 	lock_vector_lock();
260 	setup_vector_irq(smp_processor_id());
261 	set_cpu_online(smp_processor_id(), true);
262 	unlock_vector_lock();
263 	cpu_set_state_online(smp_processor_id());
264 	x86_platform.nmi_init();
265 
266 	/* enable local interrupts */
267 	local_irq_enable();
268 
269 	/* to prevent fake stack check failure in clock setup */
270 	boot_init_stack_canary();
271 
272 	x86_cpuinit.setup_percpu_clockev();
273 
274 	wmb();
275 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
276 }
277 
278 /**
279  * topology_update_package_map - Update the physical to logical package map
280  * @pkg:	The physical package id as retrieved via CPUID
281  * @cpu:	The cpu for which this is updated
282  */
283 int topology_update_package_map(unsigned int pkg, unsigned int cpu)
284 {
285 	unsigned int new;
286 
287 	/* Called from early boot ? */
288 	if (!physical_package_map)
289 		return 0;
290 
291 	if (pkg >= max_physical_pkg_id)
292 		return -EINVAL;
293 
294 	/* Set the logical package id */
295 	if (test_and_set_bit(pkg, physical_package_map))
296 		goto found;
297 
298 	if (logical_packages >= __max_logical_packages) {
299 		pr_warn("Package %u of CPU %u exceeds BIOS package data %u.\n",
300 			logical_packages, cpu, __max_logical_packages);
301 		return -ENOSPC;
302 	}
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 	physical_to_logical_pkg[pkg] = new;
310 
311 found:
312 	cpu_data(cpu).logical_proc_id = physical_to_logical_pkg[pkg];
313 	return 0;
314 }
315 
316 /**
317  * topology_phys_to_logical_pkg - Map a physical package id to a logical
318  *
319  * Returns logical package id or -1 if not found
320  */
321 int topology_phys_to_logical_pkg(unsigned int phys_pkg)
322 {
323 	if (phys_pkg >= max_physical_pkg_id)
324 		return -1;
325 	return physical_to_logical_pkg[phys_pkg];
326 }
327 EXPORT_SYMBOL(topology_phys_to_logical_pkg);
328 
329 static void __init smp_init_package_map(struct cpuinfo_x86 *c, unsigned int cpu)
330 {
331 	unsigned int ncpus;
332 	size_t size;
333 
334 	/*
335 	 * Today neither Intel nor AMD support heterogenous systems. That
336 	 * might change in the future....
337 	 *
338 	 * While ideally we'd want '* smp_num_siblings' in the below @ncpus
339 	 * computation, this won't actually work since some Intel BIOSes
340 	 * report inconsistent HT data when they disable HT.
341 	 *
342 	 * In particular, they reduce the APIC-IDs to only include the cores,
343 	 * but leave the CPUID topology to say there are (2) siblings.
344 	 * This means we don't know how many threads there will be until
345 	 * after the APIC enumeration.
346 	 *
347 	 * By not including this we'll sometimes over-estimate the number of
348 	 * logical packages by the amount of !present siblings, but this is
349 	 * still better than MAX_LOCAL_APIC.
350 	 *
351 	 * We use total_cpus not nr_cpu_ids because nr_cpu_ids can be limited
352 	 * on the command line leading to a similar issue as the HT disable
353 	 * problem because the hyperthreads are usually enumerated after the
354 	 * primary cores.
355 	 */
356 	ncpus = boot_cpu_data.x86_max_cores;
357 	if (!ncpus) {
358 		pr_warn("x86_max_cores == zero !?!?");
359 		ncpus = 1;
360 	}
361 
362 	__max_logical_packages = DIV_ROUND_UP(total_cpus, ncpus);
363 	logical_packages = 0;
364 
365 	/*
366 	 * Possibly larger than what we need as the number of apic ids per
367 	 * package can be smaller than the actual used apic ids.
368 	 */
369 	max_physical_pkg_id = DIV_ROUND_UP(MAX_LOCAL_APIC, ncpus);
370 	size = max_physical_pkg_id * sizeof(unsigned int);
371 	physical_to_logical_pkg = kmalloc(size, GFP_KERNEL);
372 	memset(physical_to_logical_pkg, 0xff, size);
373 	size = BITS_TO_LONGS(max_physical_pkg_id) * sizeof(unsigned long);
374 	physical_package_map = kzalloc(size, GFP_KERNEL);
375 
376 	pr_info("Max logical packages: %u\n", __max_logical_packages);
377 
378 	topology_update_package_map(c->phys_proc_id, cpu);
379 }
380 
381 void __init smp_store_boot_cpu_info(void)
382 {
383 	int id = 0; /* CPU 0 */
384 	struct cpuinfo_x86 *c = &cpu_data(id);
385 
386 	*c = boot_cpu_data;
387 	c->cpu_index = id;
388 	smp_init_package_map(c, id);
389 }
390 
391 /*
392  * The bootstrap kernel entry code has set these up. Save them for
393  * a given CPU
394  */
395 void smp_store_cpu_info(int id)
396 {
397 	struct cpuinfo_x86 *c = &cpu_data(id);
398 
399 	*c = boot_cpu_data;
400 	c->cpu_index = id;
401 	/*
402 	 * During boot time, CPU0 has this setup already. Save the info when
403 	 * bringing up AP or offlined CPU0.
404 	 */
405 	identify_secondary_cpu(c);
406 }
407 
408 static bool
409 topology_same_node(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
410 {
411 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
412 
413 	return (cpu_to_node(cpu1) == cpu_to_node(cpu2));
414 }
415 
416 static bool
417 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
418 {
419 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
420 
421 	return !WARN_ONCE(!topology_same_node(c, o),
422 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
423 		"[node: %d != %d]. Ignoring dependency.\n",
424 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
425 }
426 
427 #define link_mask(mfunc, c1, c2)					\
428 do {									\
429 	cpumask_set_cpu((c1), mfunc(c2));				\
430 	cpumask_set_cpu((c2), mfunc(c1));				\
431 } while (0)
432 
433 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
434 {
435 	if (boot_cpu_has(X86_FEATURE_TOPOEXT)) {
436 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
437 
438 		if (c->phys_proc_id == o->phys_proc_id &&
439 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2)) {
440 			if (c->cpu_core_id == o->cpu_core_id)
441 				return topology_sane(c, o, "smt");
442 
443 			if ((c->cu_id != 0xff) &&
444 			    (o->cu_id != 0xff) &&
445 			    (c->cu_id == o->cu_id))
446 				return topology_sane(c, o, "smt");
447 		}
448 
449 	} else if (c->phys_proc_id == o->phys_proc_id &&
450 		   c->cpu_core_id == o->cpu_core_id) {
451 		return topology_sane(c, o, "smt");
452 	}
453 
454 	return false;
455 }
456 
457 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
458 {
459 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
460 
461 	if (per_cpu(cpu_llc_id, cpu1) != BAD_APICID &&
462 	    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2))
463 		return topology_sane(c, o, "llc");
464 
465 	return false;
466 }
467 
468 /*
469  * Unlike the other levels, we do not enforce keeping a
470  * multicore group inside a NUMA node.  If this happens, we will
471  * discard the MC level of the topology later.
472  */
473 static bool match_die(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
474 {
475 	if (c->phys_proc_id == o->phys_proc_id)
476 		return true;
477 	return false;
478 }
479 
480 #if defined(CONFIG_SCHED_SMT) || defined(CONFIG_SCHED_MC)
481 static inline int x86_sched_itmt_flags(void)
482 {
483 	return sysctl_sched_itmt_enabled ? SD_ASYM_PACKING : 0;
484 }
485 
486 #ifdef CONFIG_SCHED_MC
487 static int x86_core_flags(void)
488 {
489 	return cpu_core_flags() | x86_sched_itmt_flags();
490 }
491 #endif
492 #ifdef CONFIG_SCHED_SMT
493 static int x86_smt_flags(void)
494 {
495 	return cpu_smt_flags() | x86_sched_itmt_flags();
496 }
497 #endif
498 #endif
499 
500 static struct sched_domain_topology_level x86_numa_in_package_topology[] = {
501 #ifdef CONFIG_SCHED_SMT
502 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
503 #endif
504 #ifdef CONFIG_SCHED_MC
505 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
506 #endif
507 	{ NULL, },
508 };
509 
510 static struct sched_domain_topology_level x86_topology[] = {
511 #ifdef CONFIG_SCHED_SMT
512 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
513 #endif
514 #ifdef CONFIG_SCHED_MC
515 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
516 #endif
517 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
518 	{ NULL, },
519 };
520 
521 /*
522  * Set if a package/die has multiple NUMA nodes inside.
523  * AMD Magny-Cours and Intel Cluster-on-Die have this.
524  */
525 static bool x86_has_numa_in_package;
526 
527 void set_cpu_sibling_map(int cpu)
528 {
529 	bool has_smt = smp_num_siblings > 1;
530 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
531 	struct cpuinfo_x86 *c = &cpu_data(cpu);
532 	struct cpuinfo_x86 *o;
533 	int i, threads;
534 
535 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
536 
537 	if (!has_mp) {
538 		cpumask_set_cpu(cpu, topology_sibling_cpumask(cpu));
539 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
540 		cpumask_set_cpu(cpu, topology_core_cpumask(cpu));
541 		c->booted_cores = 1;
542 		return;
543 	}
544 
545 	for_each_cpu(i, cpu_sibling_setup_mask) {
546 		o = &cpu_data(i);
547 
548 		if ((i == cpu) || (has_smt && match_smt(c, o)))
549 			link_mask(topology_sibling_cpumask, cpu, i);
550 
551 		if ((i == cpu) || (has_mp && match_llc(c, o)))
552 			link_mask(cpu_llc_shared_mask, cpu, i);
553 
554 	}
555 
556 	/*
557 	 * This needs a separate iteration over the cpus because we rely on all
558 	 * topology_sibling_cpumask links to be set-up.
559 	 */
560 	for_each_cpu(i, cpu_sibling_setup_mask) {
561 		o = &cpu_data(i);
562 
563 		if ((i == cpu) || (has_mp && match_die(c, o))) {
564 			link_mask(topology_core_cpumask, cpu, i);
565 
566 			/*
567 			 *  Does this new cpu bringup a new core?
568 			 */
569 			if (cpumask_weight(
570 			    topology_sibling_cpumask(cpu)) == 1) {
571 				/*
572 				 * for each core in package, increment
573 				 * the booted_cores for this new cpu
574 				 */
575 				if (cpumask_first(
576 				    topology_sibling_cpumask(i)) == i)
577 					c->booted_cores++;
578 				/*
579 				 * increment the core count for all
580 				 * the other cpus in this package
581 				 */
582 				if (i != cpu)
583 					cpu_data(i).booted_cores++;
584 			} else if (i != cpu && !c->booted_cores)
585 				c->booted_cores = cpu_data(i).booted_cores;
586 		}
587 		if (match_die(c, o) && !topology_same_node(c, o))
588 			x86_has_numa_in_package = true;
589 	}
590 
591 	threads = cpumask_weight(topology_sibling_cpumask(cpu));
592 	if (threads > __max_smt_threads)
593 		__max_smt_threads = threads;
594 }
595 
596 /* maps the cpu to the sched domain representing multi-core */
597 const struct cpumask *cpu_coregroup_mask(int cpu)
598 {
599 	return cpu_llc_shared_mask(cpu);
600 }
601 
602 static void impress_friends(void)
603 {
604 	int cpu;
605 	unsigned long bogosum = 0;
606 	/*
607 	 * Allow the user to impress friends.
608 	 */
609 	pr_debug("Before bogomips\n");
610 	for_each_possible_cpu(cpu)
611 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
612 			bogosum += cpu_data(cpu).loops_per_jiffy;
613 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
614 		num_online_cpus(),
615 		bogosum/(500000/HZ),
616 		(bogosum/(5000/HZ))%100);
617 
618 	pr_debug("Before bogocount - setting activated=1\n");
619 }
620 
621 void __inquire_remote_apic(int apicid)
622 {
623 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
624 	const char * const names[] = { "ID", "VERSION", "SPIV" };
625 	int timeout;
626 	u32 status;
627 
628 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
629 
630 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
631 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
632 
633 		/*
634 		 * Wait for idle.
635 		 */
636 		status = safe_apic_wait_icr_idle();
637 		if (status)
638 			pr_cont("a previous APIC delivery may have failed\n");
639 
640 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
641 
642 		timeout = 0;
643 		do {
644 			udelay(100);
645 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
646 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
647 
648 		switch (status) {
649 		case APIC_ICR_RR_VALID:
650 			status = apic_read(APIC_RRR);
651 			pr_cont("%08x\n", status);
652 			break;
653 		default:
654 			pr_cont("failed\n");
655 		}
656 	}
657 }
658 
659 /*
660  * The Multiprocessor Specification 1.4 (1997) example code suggests
661  * that there should be a 10ms delay between the BSP asserting INIT
662  * and de-asserting INIT, when starting a remote processor.
663  * But that slows boot and resume on modern processors, which include
664  * many cores and don't require that delay.
665  *
666  * Cmdline "init_cpu_udelay=" is available to over-ride this delay.
667  * Modern processor families are quirked to remove the delay entirely.
668  */
669 #define UDELAY_10MS_DEFAULT 10000
670 
671 static unsigned int init_udelay = UINT_MAX;
672 
673 static int __init cpu_init_udelay(char *str)
674 {
675 	get_option(&str, &init_udelay);
676 
677 	return 0;
678 }
679 early_param("cpu_init_udelay", cpu_init_udelay);
680 
681 static void __init smp_quirk_init_udelay(void)
682 {
683 	/* if cmdline changed it from default, leave it alone */
684 	if (init_udelay != UINT_MAX)
685 		return;
686 
687 	/* if modern processor, use no delay */
688 	if (((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && (boot_cpu_data.x86 == 6)) ||
689 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && (boot_cpu_data.x86 >= 0xF))) {
690 		init_udelay = 0;
691 		return;
692 	}
693 	/* else, use legacy delay */
694 	init_udelay = UDELAY_10MS_DEFAULT;
695 }
696 
697 /*
698  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
699  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
700  * won't ... remember to clear down the APIC, etc later.
701  */
702 int
703 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
704 {
705 	unsigned long send_status, accept_status = 0;
706 	int maxlvt;
707 
708 	/* Target chip */
709 	/* Boot on the stack */
710 	/* Kick the second */
711 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, apicid);
712 
713 	pr_debug("Waiting for send to finish...\n");
714 	send_status = safe_apic_wait_icr_idle();
715 
716 	/*
717 	 * Give the other CPU some time to accept the IPI.
718 	 */
719 	udelay(200);
720 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
721 		maxlvt = lapic_get_maxlvt();
722 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
723 			apic_write(APIC_ESR, 0);
724 		accept_status = (apic_read(APIC_ESR) & 0xEF);
725 	}
726 	pr_debug("NMI sent\n");
727 
728 	if (send_status)
729 		pr_err("APIC never delivered???\n");
730 	if (accept_status)
731 		pr_err("APIC delivery error (%lx)\n", accept_status);
732 
733 	return (send_status | accept_status);
734 }
735 
736 static int
737 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
738 {
739 	unsigned long send_status = 0, accept_status = 0;
740 	int maxlvt, num_starts, j;
741 
742 	maxlvt = lapic_get_maxlvt();
743 
744 	/*
745 	 * Be paranoid about clearing APIC errors.
746 	 */
747 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
748 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
749 			apic_write(APIC_ESR, 0);
750 		apic_read(APIC_ESR);
751 	}
752 
753 	pr_debug("Asserting INIT\n");
754 
755 	/*
756 	 * Turn INIT on target chip
757 	 */
758 	/*
759 	 * Send IPI
760 	 */
761 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
762 		       phys_apicid);
763 
764 	pr_debug("Waiting for send to finish...\n");
765 	send_status = safe_apic_wait_icr_idle();
766 
767 	udelay(init_udelay);
768 
769 	pr_debug("Deasserting INIT\n");
770 
771 	/* Target chip */
772 	/* Send IPI */
773 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
774 
775 	pr_debug("Waiting for send to finish...\n");
776 	send_status = safe_apic_wait_icr_idle();
777 
778 	mb();
779 
780 	/*
781 	 * Should we send STARTUP IPIs ?
782 	 *
783 	 * Determine this based on the APIC version.
784 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
785 	 */
786 	if (APIC_INTEGRATED(boot_cpu_apic_version))
787 		num_starts = 2;
788 	else
789 		num_starts = 0;
790 
791 	/*
792 	 * Run STARTUP IPI loop.
793 	 */
794 	pr_debug("#startup loops: %d\n", num_starts);
795 
796 	for (j = 1; j <= num_starts; j++) {
797 		pr_debug("Sending STARTUP #%d\n", j);
798 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
799 			apic_write(APIC_ESR, 0);
800 		apic_read(APIC_ESR);
801 		pr_debug("After apic_write\n");
802 
803 		/*
804 		 * STARTUP IPI
805 		 */
806 
807 		/* Target chip */
808 		/* Boot on the stack */
809 		/* Kick the second */
810 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
811 			       phys_apicid);
812 
813 		/*
814 		 * Give the other CPU some time to accept the IPI.
815 		 */
816 		if (init_udelay == 0)
817 			udelay(10);
818 		else
819 			udelay(300);
820 
821 		pr_debug("Startup point 1\n");
822 
823 		pr_debug("Waiting for send to finish...\n");
824 		send_status = safe_apic_wait_icr_idle();
825 
826 		/*
827 		 * Give the other CPU some time to accept the IPI.
828 		 */
829 		if (init_udelay == 0)
830 			udelay(10);
831 		else
832 			udelay(200);
833 
834 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
835 			apic_write(APIC_ESR, 0);
836 		accept_status = (apic_read(APIC_ESR) & 0xEF);
837 		if (send_status || accept_status)
838 			break;
839 	}
840 	pr_debug("After Startup\n");
841 
842 	if (send_status)
843 		pr_err("APIC never delivered???\n");
844 	if (accept_status)
845 		pr_err("APIC delivery error (%lx)\n", accept_status);
846 
847 	return (send_status | accept_status);
848 }
849 
850 /* reduce the number of lines printed when booting a large cpu count system */
851 static void announce_cpu(int cpu, int apicid)
852 {
853 	static int current_node = -1;
854 	int node = early_cpu_to_node(cpu);
855 	static int width, node_width;
856 
857 	if (!width)
858 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
859 
860 	if (!node_width)
861 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
862 
863 	if (cpu == 1)
864 		printk(KERN_INFO "x86: Booting SMP configuration:\n");
865 
866 	if (system_state == SYSTEM_BOOTING) {
867 		if (node != current_node) {
868 			if (current_node > (-1))
869 				pr_cont("\n");
870 			current_node = node;
871 
872 			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
873 			       node_width - num_digits(node), " ", node);
874 		}
875 
876 		/* Add padding for the BSP */
877 		if (cpu == 1)
878 			pr_cont("%*s", width + 1, " ");
879 
880 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
881 
882 	} else
883 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
884 			node, cpu, apicid);
885 }
886 
887 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
888 {
889 	int cpu;
890 
891 	cpu = smp_processor_id();
892 	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
893 		return NMI_HANDLED;
894 
895 	return NMI_DONE;
896 }
897 
898 /*
899  * Wake up AP by INIT, INIT, STARTUP sequence.
900  *
901  * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
902  * boot-strap code which is not a desired behavior for waking up BSP. To
903  * void the boot-strap code, wake up CPU0 by NMI instead.
904  *
905  * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
906  * (i.e. physically hot removed and then hot added), NMI won't wake it up.
907  * We'll change this code in the future to wake up hard offlined CPU0 if
908  * real platform and request are available.
909  */
910 static int
911 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
912 	       int *cpu0_nmi_registered)
913 {
914 	int id;
915 	int boot_error;
916 
917 	preempt_disable();
918 
919 	/*
920 	 * Wake up AP by INIT, INIT, STARTUP sequence.
921 	 */
922 	if (cpu) {
923 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
924 		goto out;
925 	}
926 
927 	/*
928 	 * Wake up BSP by nmi.
929 	 *
930 	 * Register a NMI handler to help wake up CPU0.
931 	 */
932 	boot_error = register_nmi_handler(NMI_LOCAL,
933 					  wakeup_cpu0_nmi, 0, "wake_cpu0");
934 
935 	if (!boot_error) {
936 		enable_start_cpu0 = 1;
937 		*cpu0_nmi_registered = 1;
938 		if (apic->dest_logical == APIC_DEST_LOGICAL)
939 			id = cpu0_logical_apicid;
940 		else
941 			id = apicid;
942 		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
943 	}
944 
945 out:
946 	preempt_enable();
947 
948 	return boot_error;
949 }
950 
951 void common_cpu_up(unsigned int cpu, struct task_struct *idle)
952 {
953 	/* Just in case we booted with a single CPU. */
954 	alternatives_enable_smp();
955 
956 	per_cpu(current_task, cpu) = idle;
957 
958 #ifdef CONFIG_X86_32
959 	/* Stack for startup_32 can be just as for start_secondary onwards */
960 	irq_ctx_init(cpu);
961 	per_cpu(cpu_current_top_of_stack, cpu) =
962 		(unsigned long)task_stack_page(idle) + THREAD_SIZE;
963 #else
964 	initial_gs = per_cpu_offset(cpu);
965 #endif
966 }
967 
968 /*
969  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
970  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
971  * Returns zero if CPU booted OK, else error code from
972  * ->wakeup_secondary_cpu.
973  */
974 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
975 {
976 	volatile u32 *trampoline_status =
977 		(volatile u32 *) __va(real_mode_header->trampoline_status);
978 	/* start_ip had better be page-aligned! */
979 	unsigned long start_ip = real_mode_header->trampoline_start;
980 
981 	unsigned long boot_error = 0;
982 	int cpu0_nmi_registered = 0;
983 	unsigned long timeout;
984 
985 	idle->thread.sp = (unsigned long)task_pt_regs(idle);
986 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
987 	initial_code = (unsigned long)start_secondary;
988 	initial_stack  = idle->thread.sp;
989 
990 	/*
991 	 * Enable the espfix hack for this CPU
992 	*/
993 #ifdef CONFIG_X86_ESPFIX64
994 	init_espfix_ap(cpu);
995 #endif
996 
997 	/* So we see what's up */
998 	announce_cpu(cpu, apicid);
999 
1000 	/*
1001 	 * This grunge runs the startup process for
1002 	 * the targeted processor.
1003 	 */
1004 
1005 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
1006 
1007 		pr_debug("Setting warm reset code and vector.\n");
1008 
1009 		smpboot_setup_warm_reset_vector(start_ip);
1010 		/*
1011 		 * Be paranoid about clearing APIC errors.
1012 		*/
1013 		if (APIC_INTEGRATED(boot_cpu_apic_version)) {
1014 			apic_write(APIC_ESR, 0);
1015 			apic_read(APIC_ESR);
1016 		}
1017 	}
1018 
1019 	/*
1020 	 * AP might wait on cpu_callout_mask in cpu_init() with
1021 	 * cpu_initialized_mask set if previous attempt to online
1022 	 * it timed-out. Clear cpu_initialized_mask so that after
1023 	 * INIT/SIPI it could start with a clean state.
1024 	 */
1025 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1026 	smp_mb();
1027 
1028 	/*
1029 	 * Wake up a CPU in difference cases:
1030 	 * - Use the method in the APIC driver if it's defined
1031 	 * Otherwise,
1032 	 * - Use an INIT boot APIC message for APs or NMI for BSP.
1033 	 */
1034 	if (apic->wakeup_secondary_cpu)
1035 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
1036 	else
1037 		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
1038 						     &cpu0_nmi_registered);
1039 
1040 	if (!boot_error) {
1041 		/*
1042 		 * Wait 10s total for first sign of life from AP
1043 		 */
1044 		boot_error = -1;
1045 		timeout = jiffies + 10*HZ;
1046 		while (time_before(jiffies, timeout)) {
1047 			if (cpumask_test_cpu(cpu, cpu_initialized_mask)) {
1048 				/*
1049 				 * Tell AP to proceed with initialization
1050 				 */
1051 				cpumask_set_cpu(cpu, cpu_callout_mask);
1052 				boot_error = 0;
1053 				break;
1054 			}
1055 			schedule();
1056 		}
1057 	}
1058 
1059 	if (!boot_error) {
1060 		/*
1061 		 * Wait till AP completes initial initialization
1062 		 */
1063 		while (!cpumask_test_cpu(cpu, cpu_callin_mask)) {
1064 			/*
1065 			 * Allow other tasks to run while we wait for the
1066 			 * AP to come online. This also gives a chance
1067 			 * for the MTRR work(triggered by the AP coming online)
1068 			 * to be completed in the stop machine context.
1069 			 */
1070 			schedule();
1071 		}
1072 	}
1073 
1074 	/* mark "stuck" area as not stuck */
1075 	*trampoline_status = 0;
1076 
1077 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
1078 		/*
1079 		 * Cleanup possible dangling ends...
1080 		 */
1081 		smpboot_restore_warm_reset_vector();
1082 	}
1083 	/*
1084 	 * Clean up the nmi handler. Do this after the callin and callout sync
1085 	 * to avoid impact of possible long unregister time.
1086 	 */
1087 	if (cpu0_nmi_registered)
1088 		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
1089 
1090 	return boot_error;
1091 }
1092 
1093 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
1094 {
1095 	int apicid = apic->cpu_present_to_apicid(cpu);
1096 	unsigned long flags;
1097 	int err;
1098 
1099 	WARN_ON(irqs_disabled());
1100 
1101 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
1102 
1103 	if (apicid == BAD_APICID ||
1104 	    !physid_isset(apicid, phys_cpu_present_map) ||
1105 	    !apic->apic_id_valid(apicid)) {
1106 		pr_err("%s: bad cpu %d\n", __func__, cpu);
1107 		return -EINVAL;
1108 	}
1109 
1110 	/*
1111 	 * Already booted CPU?
1112 	 */
1113 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
1114 		pr_debug("do_boot_cpu %d Already started\n", cpu);
1115 		return -ENOSYS;
1116 	}
1117 
1118 	/*
1119 	 * Save current MTRR state in case it was changed since early boot
1120 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
1121 	 */
1122 	mtrr_save_state();
1123 
1124 	/* x86 CPUs take themselves offline, so delayed offline is OK. */
1125 	err = cpu_check_up_prepare(cpu);
1126 	if (err && err != -EBUSY)
1127 		return err;
1128 
1129 	/* the FPU context is blank, nobody can own it */
1130 	per_cpu(fpu_fpregs_owner_ctx, cpu) = NULL;
1131 
1132 	common_cpu_up(cpu, tidle);
1133 
1134 	err = do_boot_cpu(apicid, cpu, tidle);
1135 	if (err) {
1136 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
1137 		return -EIO;
1138 	}
1139 
1140 	/*
1141 	 * Check TSC synchronization with the AP (keep irqs disabled
1142 	 * while doing so):
1143 	 */
1144 	local_irq_save(flags);
1145 	check_tsc_sync_source(cpu);
1146 	local_irq_restore(flags);
1147 
1148 	while (!cpu_online(cpu)) {
1149 		cpu_relax();
1150 		touch_nmi_watchdog();
1151 	}
1152 
1153 	return 0;
1154 }
1155 
1156 /**
1157  * arch_disable_smp_support() - disables SMP support for x86 at runtime
1158  */
1159 void arch_disable_smp_support(void)
1160 {
1161 	disable_ioapic_support();
1162 }
1163 
1164 /*
1165  * Fall back to non SMP mode after errors.
1166  *
1167  * RED-PEN audit/test this more. I bet there is more state messed up here.
1168  */
1169 static __init void disable_smp(void)
1170 {
1171 	pr_info("SMP disabled\n");
1172 
1173 	disable_ioapic_support();
1174 
1175 	init_cpu_present(cpumask_of(0));
1176 	init_cpu_possible(cpumask_of(0));
1177 
1178 	if (smp_found_config)
1179 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1180 	else
1181 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1182 	cpumask_set_cpu(0, topology_sibling_cpumask(0));
1183 	cpumask_set_cpu(0, topology_core_cpumask(0));
1184 }
1185 
1186 enum {
1187 	SMP_OK,
1188 	SMP_NO_CONFIG,
1189 	SMP_NO_APIC,
1190 	SMP_FORCE_UP,
1191 };
1192 
1193 /*
1194  * Various sanity checks.
1195  */
1196 static int __init smp_sanity_check(unsigned max_cpus)
1197 {
1198 	preempt_disable();
1199 
1200 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
1201 	if (def_to_bigsmp && nr_cpu_ids > 8) {
1202 		unsigned int cpu;
1203 		unsigned nr;
1204 
1205 		pr_warn("More than 8 CPUs detected - skipping them\n"
1206 			"Use CONFIG_X86_BIGSMP\n");
1207 
1208 		nr = 0;
1209 		for_each_present_cpu(cpu) {
1210 			if (nr >= 8)
1211 				set_cpu_present(cpu, false);
1212 			nr++;
1213 		}
1214 
1215 		nr = 0;
1216 		for_each_possible_cpu(cpu) {
1217 			if (nr >= 8)
1218 				set_cpu_possible(cpu, false);
1219 			nr++;
1220 		}
1221 
1222 		nr_cpu_ids = 8;
1223 	}
1224 #endif
1225 
1226 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1227 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1228 			hard_smp_processor_id());
1229 
1230 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1231 	}
1232 
1233 	/*
1234 	 * If we couldn't find an SMP configuration at boot time,
1235 	 * get out of here now!
1236 	 */
1237 	if (!smp_found_config && !acpi_lapic) {
1238 		preempt_enable();
1239 		pr_notice("SMP motherboard not detected\n");
1240 		return SMP_NO_CONFIG;
1241 	}
1242 
1243 	/*
1244 	 * Should not be necessary because the MP table should list the boot
1245 	 * CPU too, but we do it for the sake of robustness anyway.
1246 	 */
1247 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1248 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1249 			  boot_cpu_physical_apicid);
1250 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1251 	}
1252 	preempt_enable();
1253 
1254 	/*
1255 	 * If we couldn't find a local APIC, then get out of here now!
1256 	 */
1257 	if (APIC_INTEGRATED(boot_cpu_apic_version) &&
1258 	    !boot_cpu_has(X86_FEATURE_APIC)) {
1259 		if (!disable_apic) {
1260 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
1261 				boot_cpu_physical_apicid);
1262 			pr_err("... forcing use of dummy APIC emulation (tell your hw vendor)\n");
1263 		}
1264 		return SMP_NO_APIC;
1265 	}
1266 
1267 	/*
1268 	 * If SMP should be disabled, then really disable it!
1269 	 */
1270 	if (!max_cpus) {
1271 		pr_info("SMP mode deactivated\n");
1272 		return SMP_FORCE_UP;
1273 	}
1274 
1275 	return SMP_OK;
1276 }
1277 
1278 static void __init smp_cpu_index_default(void)
1279 {
1280 	int i;
1281 	struct cpuinfo_x86 *c;
1282 
1283 	for_each_possible_cpu(i) {
1284 		c = &cpu_data(i);
1285 		/* mark all to hotplug */
1286 		c->cpu_index = nr_cpu_ids;
1287 	}
1288 }
1289 
1290 /*
1291  * Prepare for SMP bootup.  The MP table or ACPI has been read
1292  * earlier.  Just do some sanity checking here and enable APIC mode.
1293  */
1294 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1295 {
1296 	unsigned int i;
1297 
1298 	smp_cpu_index_default();
1299 
1300 	/*
1301 	 * Setup boot CPU information
1302 	 */
1303 	smp_store_boot_cpu_info(); /* Final full version of the data */
1304 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1305 	mb();
1306 
1307 	for_each_possible_cpu(i) {
1308 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1309 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1310 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1311 	}
1312 
1313 	/*
1314 	 * Set 'default' x86 topology, this matches default_topology() in that
1315 	 * it has NUMA nodes as a topology level. See also
1316 	 * native_smp_cpus_done().
1317 	 *
1318 	 * Must be done before set_cpus_sibling_map() is ran.
1319 	 */
1320 	set_sched_topology(x86_topology);
1321 
1322 	set_cpu_sibling_map(0);
1323 
1324 	switch (smp_sanity_check(max_cpus)) {
1325 	case SMP_NO_CONFIG:
1326 		disable_smp();
1327 		if (APIC_init_uniprocessor())
1328 			pr_notice("Local APIC not detected. Using dummy APIC emulation.\n");
1329 		return;
1330 	case SMP_NO_APIC:
1331 		disable_smp();
1332 		return;
1333 	case SMP_FORCE_UP:
1334 		disable_smp();
1335 		apic_bsp_setup(false);
1336 		return;
1337 	case SMP_OK:
1338 		break;
1339 	}
1340 
1341 	if (read_apic_id() != boot_cpu_physical_apicid) {
1342 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1343 		     read_apic_id(), boot_cpu_physical_apicid);
1344 		/* Or can we switch back to PIC here? */
1345 	}
1346 
1347 	default_setup_apic_routing();
1348 	cpu0_logical_apicid = apic_bsp_setup(false);
1349 
1350 	pr_info("CPU0: ");
1351 	print_cpu_info(&cpu_data(0));
1352 
1353 	uv_system_init();
1354 
1355 	set_mtrr_aps_delayed_init();
1356 
1357 	smp_quirk_init_udelay();
1358 }
1359 
1360 void arch_enable_nonboot_cpus_begin(void)
1361 {
1362 	set_mtrr_aps_delayed_init();
1363 }
1364 
1365 void arch_enable_nonboot_cpus_end(void)
1366 {
1367 	mtrr_aps_init();
1368 }
1369 
1370 /*
1371  * Early setup to make printk work.
1372  */
1373 void __init native_smp_prepare_boot_cpu(void)
1374 {
1375 	int me = smp_processor_id();
1376 	switch_to_new_gdt(me);
1377 	/* already set me in cpu_online_mask in boot_cpu_init() */
1378 	cpumask_set_cpu(me, cpu_callout_mask);
1379 	cpu_set_state_online(me);
1380 }
1381 
1382 void __init native_smp_cpus_done(unsigned int max_cpus)
1383 {
1384 	pr_debug("Boot done\n");
1385 
1386 	if (x86_has_numa_in_package)
1387 		set_sched_topology(x86_numa_in_package_topology);
1388 
1389 	nmi_selftest();
1390 	impress_friends();
1391 	setup_ioapic_dest();
1392 	mtrr_aps_init();
1393 }
1394 
1395 static int __initdata setup_possible_cpus = -1;
1396 static int __init _setup_possible_cpus(char *str)
1397 {
1398 	get_option(&str, &setup_possible_cpus);
1399 	return 0;
1400 }
1401 early_param("possible_cpus", _setup_possible_cpus);
1402 
1403 
1404 /*
1405  * cpu_possible_mask should be static, it cannot change as cpu's
1406  * are onlined, or offlined. The reason is per-cpu data-structures
1407  * are allocated by some modules at init time, and dont expect to
1408  * do this dynamically on cpu arrival/departure.
1409  * cpu_present_mask on the other hand can change dynamically.
1410  * In case when cpu_hotplug is not compiled, then we resort to current
1411  * behaviour, which is cpu_possible == cpu_present.
1412  * - Ashok Raj
1413  *
1414  * Three ways to find out the number of additional hotplug CPUs:
1415  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1416  * - The user can overwrite it with possible_cpus=NUM
1417  * - Otherwise don't reserve additional CPUs.
1418  * We do this because additional CPUs waste a lot of memory.
1419  * -AK
1420  */
1421 __init void prefill_possible_map(void)
1422 {
1423 	int i, possible;
1424 
1425 	/* No boot processor was found in mptable or ACPI MADT */
1426 	if (!num_processors) {
1427 		if (boot_cpu_has(X86_FEATURE_APIC)) {
1428 			int apicid = boot_cpu_physical_apicid;
1429 			int cpu = hard_smp_processor_id();
1430 
1431 			pr_warn("Boot CPU (id %d) not listed by BIOS\n", cpu);
1432 
1433 			/* Make sure boot cpu is enumerated */
1434 			if (apic->cpu_present_to_apicid(0) == BAD_APICID &&
1435 			    apic->apic_id_valid(apicid))
1436 				generic_processor_info(apicid, boot_cpu_apic_version);
1437 		}
1438 
1439 		if (!num_processors)
1440 			num_processors = 1;
1441 	}
1442 
1443 	i = setup_max_cpus ?: 1;
1444 	if (setup_possible_cpus == -1) {
1445 		possible = num_processors;
1446 #ifdef CONFIG_HOTPLUG_CPU
1447 		if (setup_max_cpus)
1448 			possible += disabled_cpus;
1449 #else
1450 		if (possible > i)
1451 			possible = i;
1452 #endif
1453 	} else
1454 		possible = setup_possible_cpus;
1455 
1456 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1457 
1458 	/* nr_cpu_ids could be reduced via nr_cpus= */
1459 	if (possible > nr_cpu_ids) {
1460 		pr_warn("%d Processors exceeds NR_CPUS limit of %d\n",
1461 			possible, nr_cpu_ids);
1462 		possible = nr_cpu_ids;
1463 	}
1464 
1465 #ifdef CONFIG_HOTPLUG_CPU
1466 	if (!setup_max_cpus)
1467 #endif
1468 	if (possible > i) {
1469 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1470 			possible, setup_max_cpus);
1471 		possible = i;
1472 	}
1473 
1474 	nr_cpu_ids = possible;
1475 
1476 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1477 		possible, max_t(int, possible - num_processors, 0));
1478 
1479 	reset_cpu_possible_mask();
1480 
1481 	for (i = 0; i < possible; i++)
1482 		set_cpu_possible(i, true);
1483 }
1484 
1485 #ifdef CONFIG_HOTPLUG_CPU
1486 
1487 /* Recompute SMT state for all CPUs on offline */
1488 static void recompute_smt_state(void)
1489 {
1490 	int max_threads, cpu;
1491 
1492 	max_threads = 0;
1493 	for_each_online_cpu (cpu) {
1494 		int threads = cpumask_weight(topology_sibling_cpumask(cpu));
1495 
1496 		if (threads > max_threads)
1497 			max_threads = threads;
1498 	}
1499 	__max_smt_threads = max_threads;
1500 }
1501 
1502 static void remove_siblinginfo(int cpu)
1503 {
1504 	int sibling;
1505 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1506 
1507 	for_each_cpu(sibling, topology_core_cpumask(cpu)) {
1508 		cpumask_clear_cpu(cpu, topology_core_cpumask(sibling));
1509 		/*/
1510 		 * last thread sibling in this cpu core going down
1511 		 */
1512 		if (cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
1513 			cpu_data(sibling).booted_cores--;
1514 	}
1515 
1516 	for_each_cpu(sibling, topology_sibling_cpumask(cpu))
1517 		cpumask_clear_cpu(cpu, topology_sibling_cpumask(sibling));
1518 	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1519 		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1520 	cpumask_clear(cpu_llc_shared_mask(cpu));
1521 	cpumask_clear(topology_sibling_cpumask(cpu));
1522 	cpumask_clear(topology_core_cpumask(cpu));
1523 	c->phys_proc_id = 0;
1524 	c->cpu_core_id = 0;
1525 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1526 	recompute_smt_state();
1527 }
1528 
1529 static void remove_cpu_from_maps(int cpu)
1530 {
1531 	set_cpu_online(cpu, false);
1532 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1533 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1534 	/* was set by cpu_init() */
1535 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1536 	numa_remove_cpu(cpu);
1537 }
1538 
1539 void cpu_disable_common(void)
1540 {
1541 	int cpu = smp_processor_id();
1542 
1543 	remove_siblinginfo(cpu);
1544 
1545 	/* It's now safe to remove this processor from the online map */
1546 	lock_vector_lock();
1547 	remove_cpu_from_maps(cpu);
1548 	unlock_vector_lock();
1549 	fixup_irqs();
1550 }
1551 
1552 int native_cpu_disable(void)
1553 {
1554 	int ret;
1555 
1556 	ret = check_irq_vectors_for_cpu_disable();
1557 	if (ret)
1558 		return ret;
1559 
1560 	clear_local_APIC();
1561 	cpu_disable_common();
1562 
1563 	return 0;
1564 }
1565 
1566 int common_cpu_die(unsigned int cpu)
1567 {
1568 	int ret = 0;
1569 
1570 	/* We don't do anything here: idle task is faking death itself. */
1571 
1572 	/* They ack this in play_dead() by setting CPU_DEAD */
1573 	if (cpu_wait_death(cpu, 5)) {
1574 		if (system_state == SYSTEM_RUNNING)
1575 			pr_info("CPU %u is now offline\n", cpu);
1576 	} else {
1577 		pr_err("CPU %u didn't die...\n", cpu);
1578 		ret = -1;
1579 	}
1580 
1581 	return ret;
1582 }
1583 
1584 void native_cpu_die(unsigned int cpu)
1585 {
1586 	common_cpu_die(cpu);
1587 }
1588 
1589 void play_dead_common(void)
1590 {
1591 	idle_task_exit();
1592 	reset_lazy_tlbstate();
1593 
1594 	/* Ack it */
1595 	(void)cpu_report_death();
1596 
1597 	/*
1598 	 * With physical CPU hotplug, we should halt the cpu
1599 	 */
1600 	local_irq_disable();
1601 }
1602 
1603 static bool wakeup_cpu0(void)
1604 {
1605 	if (smp_processor_id() == 0 && enable_start_cpu0)
1606 		return true;
1607 
1608 	return false;
1609 }
1610 
1611 /*
1612  * We need to flush the caches before going to sleep, lest we have
1613  * dirty data in our caches when we come back up.
1614  */
1615 static inline void mwait_play_dead(void)
1616 {
1617 	unsigned int eax, ebx, ecx, edx;
1618 	unsigned int highest_cstate = 0;
1619 	unsigned int highest_subcstate = 0;
1620 	void *mwait_ptr;
1621 	int i;
1622 
1623 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1624 		return;
1625 	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1626 		return;
1627 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1628 		return;
1629 
1630 	eax = CPUID_MWAIT_LEAF;
1631 	ecx = 0;
1632 	native_cpuid(&eax, &ebx, &ecx, &edx);
1633 
1634 	/*
1635 	 * eax will be 0 if EDX enumeration is not valid.
1636 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1637 	 */
1638 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1639 		eax = 0;
1640 	} else {
1641 		edx >>= MWAIT_SUBSTATE_SIZE;
1642 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1643 			if (edx & MWAIT_SUBSTATE_MASK) {
1644 				highest_cstate = i;
1645 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1646 			}
1647 		}
1648 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1649 			(highest_subcstate - 1);
1650 	}
1651 
1652 	/*
1653 	 * This should be a memory location in a cache line which is
1654 	 * unlikely to be touched by other processors.  The actual
1655 	 * content is immaterial as it is not actually modified in any way.
1656 	 */
1657 	mwait_ptr = &current_thread_info()->flags;
1658 
1659 	wbinvd();
1660 
1661 	while (1) {
1662 		/*
1663 		 * The CLFLUSH is a workaround for erratum AAI65 for
1664 		 * the Xeon 7400 series.  It's not clear it is actually
1665 		 * needed, but it should be harmless in either case.
1666 		 * The WBINVD is insufficient due to the spurious-wakeup
1667 		 * case where we return around the loop.
1668 		 */
1669 		mb();
1670 		clflush(mwait_ptr);
1671 		mb();
1672 		__monitor(mwait_ptr, 0, 0);
1673 		mb();
1674 		__mwait(eax, 0);
1675 		/*
1676 		 * If NMI wants to wake up CPU0, start CPU0.
1677 		 */
1678 		if (wakeup_cpu0())
1679 			start_cpu0();
1680 	}
1681 }
1682 
1683 void hlt_play_dead(void)
1684 {
1685 	if (__this_cpu_read(cpu_info.x86) >= 4)
1686 		wbinvd();
1687 
1688 	while (1) {
1689 		native_halt();
1690 		/*
1691 		 * If NMI wants to wake up CPU0, start CPU0.
1692 		 */
1693 		if (wakeup_cpu0())
1694 			start_cpu0();
1695 	}
1696 }
1697 
1698 void native_play_dead(void)
1699 {
1700 	play_dead_common();
1701 	tboot_shutdown(TB_SHUTDOWN_WFS);
1702 
1703 	mwait_play_dead();	/* Only returns on failure */
1704 	if (cpuidle_play_dead())
1705 		hlt_play_dead();
1706 }
1707 
1708 #else /* ... !CONFIG_HOTPLUG_CPU */
1709 int native_cpu_disable(void)
1710 {
1711 	return -ENOSYS;
1712 }
1713 
1714 void native_cpu_die(unsigned int cpu)
1715 {
1716 	/* We said "no" in __cpu_disable */
1717 	BUG();
1718 }
1719 
1720 void native_play_dead(void)
1721 {
1722 	BUG();
1723 }
1724 
1725 #endif
1726