xref: /linux/arch/x86/kernel/smpboot.c (revision c1aac62f36c1e37ee81c9e09ee9ee733eef05dcb)
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/percpu.h>
49 #include <linux/bootmem.h>
50 #include <linux/err.h>
51 #include <linux/nmi.h>
52 #include <linux/tboot.h>
53 #include <linux/stackprotector.h>
54 #include <linux/gfp.h>
55 #include <linux/cpuidle.h>
56 
57 #include <asm/acpi.h>
58 #include <asm/desc.h>
59 #include <asm/nmi.h>
60 #include <asm/irq.h>
61 #include <asm/realmode.h>
62 #include <asm/cpu.h>
63 #include <asm/numa.h>
64 #include <asm/pgtable.h>
65 #include <asm/tlbflush.h>
66 #include <asm/mtrr.h>
67 #include <asm/mwait.h>
68 #include <asm/apic.h>
69 #include <asm/io_apic.h>
70 #include <asm/fpu/internal.h>
71 #include <asm/setup.h>
72 #include <asm/uv/uv.h>
73 #include <linux/mc146818rtc.h>
74 #include <asm/i8259.h>
75 #include <asm/realmode.h>
76 #include <asm/misc.h>
77 
78 /* Number of siblings per CPU package */
79 int smp_num_siblings = 1;
80 EXPORT_SYMBOL(smp_num_siblings);
81 
82 /* Last level cache ID of each logical CPU */
83 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
84 
85 /* representing HT siblings of each logical CPU */
86 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
87 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
88 
89 /* representing HT and core siblings of each logical CPU */
90 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_map);
91 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
92 
93 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_llc_shared_map);
94 
95 /* Per CPU bogomips and other parameters */
96 DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
97 EXPORT_PER_CPU_SYMBOL(cpu_info);
98 
99 /* Logical package management. We might want to allocate that dynamically */
100 static int *physical_to_logical_pkg __read_mostly;
101 static unsigned long *physical_package_map __read_mostly;;
102 static unsigned int max_physical_pkg_id __read_mostly;
103 unsigned int __max_logical_packages __read_mostly;
104 EXPORT_SYMBOL(__max_logical_packages);
105 static unsigned int logical_packages __read_mostly;
106 
107 /* Maximum number of SMT threads on any online core */
108 int __max_smt_threads __read_mostly;
109 
110 /* Flag to indicate if a complete sched domain rebuild is required */
111 bool x86_topology_update;
112 
113 int arch_update_cpu_topology(void)
114 {
115 	int retval = x86_topology_update;
116 
117 	x86_topology_update = false;
118 	return retval;
119 }
120 
121 static inline void smpboot_setup_warm_reset_vector(unsigned long start_eip)
122 {
123 	unsigned long flags;
124 
125 	spin_lock_irqsave(&rtc_lock, flags);
126 	CMOS_WRITE(0xa, 0xf);
127 	spin_unlock_irqrestore(&rtc_lock, flags);
128 	local_flush_tlb();
129 	pr_debug("1.\n");
130 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_HIGH)) =
131 							start_eip >> 4;
132 	pr_debug("2.\n");
133 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) =
134 							start_eip & 0xf;
135 	pr_debug("3.\n");
136 }
137 
138 static inline void smpboot_restore_warm_reset_vector(void)
139 {
140 	unsigned long flags;
141 
142 	/*
143 	 * Install writable page 0 entry to set BIOS data area.
144 	 */
145 	local_flush_tlb();
146 
147 	/*
148 	 * Paranoid:  Set warm reset code and vector here back
149 	 * to default values.
150 	 */
151 	spin_lock_irqsave(&rtc_lock, flags);
152 	CMOS_WRITE(0, 0xf);
153 	spin_unlock_irqrestore(&rtc_lock, flags);
154 
155 	*((volatile u32 *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = 0;
156 }
157 
158 /*
159  * Report back to the Boot Processor during boot time or to the caller processor
160  * during CPU online.
161  */
162 static void smp_callin(void)
163 {
164 	int cpuid, phys_id;
165 
166 	/*
167 	 * If waken up by an INIT in an 82489DX configuration
168 	 * cpu_callout_mask guarantees we don't get here before
169 	 * an INIT_deassert IPI reaches our local APIC, so it is
170 	 * now safe to touch our local APIC.
171 	 */
172 	cpuid = smp_processor_id();
173 
174 	/*
175 	 * (This works even if the APIC is not enabled.)
176 	 */
177 	phys_id = read_apic_id();
178 
179 	/*
180 	 * the boot CPU has finished the init stage and is spinning
181 	 * on callin_map until we finish. We are free to set up this
182 	 * CPU, first the APIC. (this is probably redundant on most
183 	 * boards)
184 	 */
185 	apic_ap_setup();
186 
187 	/*
188 	 * Save our processor parameters. Note: this information
189 	 * is needed for clock calibration.
190 	 */
191 	smp_store_cpu_info(cpuid);
192 
193 	/*
194 	 * Get our bogomips.
195 	 * Update loops_per_jiffy in cpu_data. Previous call to
196 	 * smp_store_cpu_info() stored a value that is close but not as
197 	 * accurate as the value just calculated.
198 	 */
199 	calibrate_delay();
200 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
201 	pr_debug("Stack at about %p\n", &cpuid);
202 
203 	/*
204 	 * This must be done before setting cpu_online_mask
205 	 * or calling notify_cpu_starting.
206 	 */
207 	set_cpu_sibling_map(raw_smp_processor_id());
208 	wmb();
209 
210 	notify_cpu_starting(cpuid);
211 
212 	/*
213 	 * Allow the master to continue.
214 	 */
215 	cpumask_set_cpu(cpuid, cpu_callin_mask);
216 }
217 
218 static int cpu0_logical_apicid;
219 static int enable_start_cpu0;
220 /*
221  * Activate a secondary processor.
222  */
223 static void notrace start_secondary(void *unused)
224 {
225 	/*
226 	 * Don't put *anything* before cpu_init(), SMP booting is too
227 	 * fragile that we want to limit the things done here to the
228 	 * most necessary things.
229 	 */
230 	cpu_init();
231 	x86_cpuinit.early_percpu_clock_init();
232 	preempt_disable();
233 	smp_callin();
234 
235 	enable_start_cpu0 = 0;
236 
237 #ifdef CONFIG_X86_32
238 	/* switch away from the initial page table */
239 	load_cr3(swapper_pg_dir);
240 	__flush_tlb_all();
241 #endif
242 
243 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
244 	barrier();
245 	/*
246 	 * Check TSC synchronization with the BP:
247 	 */
248 	check_tsc_sync_target();
249 
250 	/*
251 	 * Lock vector_lock and initialize the vectors on this cpu
252 	 * before setting the cpu online. We must set it online with
253 	 * vector_lock held to prevent a concurrent setup/teardown
254 	 * from seeing a half valid vector space.
255 	 */
256 	lock_vector_lock();
257 	setup_vector_irq(smp_processor_id());
258 	set_cpu_online(smp_processor_id(), true);
259 	unlock_vector_lock();
260 	cpu_set_state_online(smp_processor_id());
261 	x86_platform.nmi_init();
262 
263 	/* enable local interrupts */
264 	local_irq_enable();
265 
266 	/* to prevent fake stack check failure in clock setup */
267 	boot_init_stack_canary();
268 
269 	x86_cpuinit.setup_percpu_clockev();
270 
271 	wmb();
272 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
273 }
274 
275 /**
276  * topology_update_package_map - Update the physical to logical package map
277  * @pkg:	The physical package id as retrieved via CPUID
278  * @cpu:	The cpu for which this is updated
279  */
280 int topology_update_package_map(unsigned int pkg, unsigned int cpu)
281 {
282 	unsigned int new;
283 
284 	/* Called from early boot ? */
285 	if (!physical_package_map)
286 		return 0;
287 
288 	if (pkg >= max_physical_pkg_id)
289 		return -EINVAL;
290 
291 	/* Set the logical package id */
292 	if (test_and_set_bit(pkg, physical_package_map))
293 		goto found;
294 
295 	if (logical_packages >= __max_logical_packages) {
296 		pr_warn("Package %u of CPU %u exceeds BIOS package data %u.\n",
297 			logical_packages, cpu, __max_logical_packages);
298 		return -ENOSPC;
299 	}
300 
301 	new = logical_packages++;
302 	if (new != pkg) {
303 		pr_info("CPU %u Converting physical %u to logical package %u\n",
304 			cpu, pkg, new);
305 	}
306 	physical_to_logical_pkg[pkg] = new;
307 
308 found:
309 	cpu_data(cpu).logical_proc_id = physical_to_logical_pkg[pkg];
310 	return 0;
311 }
312 
313 /**
314  * topology_phys_to_logical_pkg - Map a physical package id to a logical
315  *
316  * Returns logical package id or -1 if not found
317  */
318 int topology_phys_to_logical_pkg(unsigned int phys_pkg)
319 {
320 	if (phys_pkg >= max_physical_pkg_id)
321 		return -1;
322 	return physical_to_logical_pkg[phys_pkg];
323 }
324 EXPORT_SYMBOL(topology_phys_to_logical_pkg);
325 
326 static void __init smp_init_package_map(struct cpuinfo_x86 *c, unsigned int cpu)
327 {
328 	unsigned int ncpus;
329 	size_t size;
330 
331 	/*
332 	 * Today neither Intel nor AMD support heterogenous systems. That
333 	 * might change in the future....
334 	 *
335 	 * While ideally we'd want '* smp_num_siblings' in the below @ncpus
336 	 * computation, this won't actually work since some Intel BIOSes
337 	 * report inconsistent HT data when they disable HT.
338 	 *
339 	 * In particular, they reduce the APIC-IDs to only include the cores,
340 	 * but leave the CPUID topology to say there are (2) siblings.
341 	 * This means we don't know how many threads there will be until
342 	 * after the APIC enumeration.
343 	 *
344 	 * By not including this we'll sometimes over-estimate the number of
345 	 * logical packages by the amount of !present siblings, but this is
346 	 * still better than MAX_LOCAL_APIC.
347 	 *
348 	 * We use total_cpus not nr_cpu_ids because nr_cpu_ids can be limited
349 	 * on the command line leading to a similar issue as the HT disable
350 	 * problem because the hyperthreads are usually enumerated after the
351 	 * primary cores.
352 	 */
353 	ncpus = boot_cpu_data.x86_max_cores;
354 	if (!ncpus) {
355 		pr_warn("x86_max_cores == zero !?!?");
356 		ncpus = 1;
357 	}
358 
359 	__max_logical_packages = DIV_ROUND_UP(total_cpus, ncpus);
360 	logical_packages = 0;
361 
362 	/*
363 	 * Possibly larger than what we need as the number of apic ids per
364 	 * package can be smaller than the actual used apic ids.
365 	 */
366 	max_physical_pkg_id = DIV_ROUND_UP(MAX_LOCAL_APIC, ncpus);
367 	size = max_physical_pkg_id * sizeof(unsigned int);
368 	physical_to_logical_pkg = kmalloc(size, GFP_KERNEL);
369 	memset(physical_to_logical_pkg, 0xff, size);
370 	size = BITS_TO_LONGS(max_physical_pkg_id) * sizeof(unsigned long);
371 	physical_package_map = kzalloc(size, GFP_KERNEL);
372 
373 	pr_info("Max logical packages: %u\n", __max_logical_packages);
374 
375 	topology_update_package_map(c->phys_proc_id, cpu);
376 }
377 
378 void __init smp_store_boot_cpu_info(void)
379 {
380 	int id = 0; /* CPU 0 */
381 	struct cpuinfo_x86 *c = &cpu_data(id);
382 
383 	*c = boot_cpu_data;
384 	c->cpu_index = id;
385 	smp_init_package_map(c, id);
386 }
387 
388 /*
389  * The bootstrap kernel entry code has set these up. Save them for
390  * a given CPU
391  */
392 void smp_store_cpu_info(int id)
393 {
394 	struct cpuinfo_x86 *c = &cpu_data(id);
395 
396 	*c = boot_cpu_data;
397 	c->cpu_index = id;
398 	/*
399 	 * During boot time, CPU0 has this setup already. Save the info when
400 	 * bringing up AP or offlined CPU0.
401 	 */
402 	identify_secondary_cpu(c);
403 }
404 
405 static bool
406 topology_same_node(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
407 {
408 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
409 
410 	return (cpu_to_node(cpu1) == cpu_to_node(cpu2));
411 }
412 
413 static bool
414 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
415 {
416 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
417 
418 	return !WARN_ONCE(!topology_same_node(c, o),
419 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
420 		"[node: %d != %d]. Ignoring dependency.\n",
421 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
422 }
423 
424 #define link_mask(mfunc, c1, c2)					\
425 do {									\
426 	cpumask_set_cpu((c1), mfunc(c2));				\
427 	cpumask_set_cpu((c2), mfunc(c1));				\
428 } while (0)
429 
430 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
431 {
432 	if (boot_cpu_has(X86_FEATURE_TOPOEXT)) {
433 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
434 
435 		if (c->phys_proc_id == o->phys_proc_id &&
436 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2)) {
437 			if (c->cpu_core_id == o->cpu_core_id)
438 				return topology_sane(c, o, "smt");
439 
440 			if ((c->cu_id != 0xff) &&
441 			    (o->cu_id != 0xff) &&
442 			    (c->cu_id == o->cu_id))
443 				return topology_sane(c, o, "smt");
444 		}
445 
446 	} else if (c->phys_proc_id == o->phys_proc_id &&
447 		   c->cpu_core_id == o->cpu_core_id) {
448 		return topology_sane(c, o, "smt");
449 	}
450 
451 	return false;
452 }
453 
454 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
455 {
456 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
457 
458 	if (per_cpu(cpu_llc_id, cpu1) != BAD_APICID &&
459 	    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2))
460 		return topology_sane(c, o, "llc");
461 
462 	return false;
463 }
464 
465 /*
466  * Unlike the other levels, we do not enforce keeping a
467  * multicore group inside a NUMA node.  If this happens, we will
468  * discard the MC level of the topology later.
469  */
470 static bool match_die(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
471 {
472 	if (c->phys_proc_id == o->phys_proc_id)
473 		return true;
474 	return false;
475 }
476 
477 #if defined(CONFIG_SCHED_SMT) || defined(CONFIG_SCHED_MC)
478 static inline int x86_sched_itmt_flags(void)
479 {
480 	return sysctl_sched_itmt_enabled ? SD_ASYM_PACKING : 0;
481 }
482 
483 #ifdef CONFIG_SCHED_MC
484 static int x86_core_flags(void)
485 {
486 	return cpu_core_flags() | x86_sched_itmt_flags();
487 }
488 #endif
489 #ifdef CONFIG_SCHED_SMT
490 static int x86_smt_flags(void)
491 {
492 	return cpu_smt_flags() | x86_sched_itmt_flags();
493 }
494 #endif
495 #endif
496 
497 static struct sched_domain_topology_level x86_numa_in_package_topology[] = {
498 #ifdef CONFIG_SCHED_SMT
499 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
500 #endif
501 #ifdef CONFIG_SCHED_MC
502 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
503 #endif
504 	{ NULL, },
505 };
506 
507 static struct sched_domain_topology_level x86_topology[] = {
508 #ifdef CONFIG_SCHED_SMT
509 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
510 #endif
511 #ifdef CONFIG_SCHED_MC
512 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
513 #endif
514 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
515 	{ NULL, },
516 };
517 
518 /*
519  * Set if a package/die has multiple NUMA nodes inside.
520  * AMD Magny-Cours and Intel Cluster-on-Die have this.
521  */
522 static bool x86_has_numa_in_package;
523 
524 void set_cpu_sibling_map(int cpu)
525 {
526 	bool has_smt = smp_num_siblings > 1;
527 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
528 	struct cpuinfo_x86 *c = &cpu_data(cpu);
529 	struct cpuinfo_x86 *o;
530 	int i, threads;
531 
532 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
533 
534 	if (!has_mp) {
535 		cpumask_set_cpu(cpu, topology_sibling_cpumask(cpu));
536 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
537 		cpumask_set_cpu(cpu, topology_core_cpumask(cpu));
538 		c->booted_cores = 1;
539 		return;
540 	}
541 
542 	for_each_cpu(i, cpu_sibling_setup_mask) {
543 		o = &cpu_data(i);
544 
545 		if ((i == cpu) || (has_smt && match_smt(c, o)))
546 			link_mask(topology_sibling_cpumask, cpu, i);
547 
548 		if ((i == cpu) || (has_mp && match_llc(c, o)))
549 			link_mask(cpu_llc_shared_mask, cpu, i);
550 
551 	}
552 
553 	/*
554 	 * This needs a separate iteration over the cpus because we rely on all
555 	 * topology_sibling_cpumask links to be set-up.
556 	 */
557 	for_each_cpu(i, cpu_sibling_setup_mask) {
558 		o = &cpu_data(i);
559 
560 		if ((i == cpu) || (has_mp && match_die(c, o))) {
561 			link_mask(topology_core_cpumask, cpu, i);
562 
563 			/*
564 			 *  Does this new cpu bringup a new core?
565 			 */
566 			if (cpumask_weight(
567 			    topology_sibling_cpumask(cpu)) == 1) {
568 				/*
569 				 * for each core in package, increment
570 				 * the booted_cores for this new cpu
571 				 */
572 				if (cpumask_first(
573 				    topology_sibling_cpumask(i)) == i)
574 					c->booted_cores++;
575 				/*
576 				 * increment the core count for all
577 				 * the other cpus in this package
578 				 */
579 				if (i != cpu)
580 					cpu_data(i).booted_cores++;
581 			} else if (i != cpu && !c->booted_cores)
582 				c->booted_cores = cpu_data(i).booted_cores;
583 		}
584 		if (match_die(c, o) && !topology_same_node(c, o))
585 			x86_has_numa_in_package = true;
586 	}
587 
588 	threads = cpumask_weight(topology_sibling_cpumask(cpu));
589 	if (threads > __max_smt_threads)
590 		__max_smt_threads = threads;
591 }
592 
593 /* maps the cpu to the sched domain representing multi-core */
594 const struct cpumask *cpu_coregroup_mask(int cpu)
595 {
596 	return cpu_llc_shared_mask(cpu);
597 }
598 
599 static void impress_friends(void)
600 {
601 	int cpu;
602 	unsigned long bogosum = 0;
603 	/*
604 	 * Allow the user to impress friends.
605 	 */
606 	pr_debug("Before bogomips\n");
607 	for_each_possible_cpu(cpu)
608 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
609 			bogosum += cpu_data(cpu).loops_per_jiffy;
610 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
611 		num_online_cpus(),
612 		bogosum/(500000/HZ),
613 		(bogosum/(5000/HZ))%100);
614 
615 	pr_debug("Before bogocount - setting activated=1\n");
616 }
617 
618 void __inquire_remote_apic(int apicid)
619 {
620 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
621 	const char * const names[] = { "ID", "VERSION", "SPIV" };
622 	int timeout;
623 	u32 status;
624 
625 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
626 
627 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
628 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
629 
630 		/*
631 		 * Wait for idle.
632 		 */
633 		status = safe_apic_wait_icr_idle();
634 		if (status)
635 			pr_cont("a previous APIC delivery may have failed\n");
636 
637 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
638 
639 		timeout = 0;
640 		do {
641 			udelay(100);
642 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
643 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
644 
645 		switch (status) {
646 		case APIC_ICR_RR_VALID:
647 			status = apic_read(APIC_RRR);
648 			pr_cont("%08x\n", status);
649 			break;
650 		default:
651 			pr_cont("failed\n");
652 		}
653 	}
654 }
655 
656 /*
657  * The Multiprocessor Specification 1.4 (1997) example code suggests
658  * that there should be a 10ms delay between the BSP asserting INIT
659  * and de-asserting INIT, when starting a remote processor.
660  * But that slows boot and resume on modern processors, which include
661  * many cores and don't require that delay.
662  *
663  * Cmdline "init_cpu_udelay=" is available to over-ride this delay.
664  * Modern processor families are quirked to remove the delay entirely.
665  */
666 #define UDELAY_10MS_DEFAULT 10000
667 
668 static unsigned int init_udelay = UINT_MAX;
669 
670 static int __init cpu_init_udelay(char *str)
671 {
672 	get_option(&str, &init_udelay);
673 
674 	return 0;
675 }
676 early_param("cpu_init_udelay", cpu_init_udelay);
677 
678 static void __init smp_quirk_init_udelay(void)
679 {
680 	/* if cmdline changed it from default, leave it alone */
681 	if (init_udelay != UINT_MAX)
682 		return;
683 
684 	/* if modern processor, use no delay */
685 	if (((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && (boot_cpu_data.x86 == 6)) ||
686 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && (boot_cpu_data.x86 >= 0xF))) {
687 		init_udelay = 0;
688 		return;
689 	}
690 	/* else, use legacy delay */
691 	init_udelay = UDELAY_10MS_DEFAULT;
692 }
693 
694 /*
695  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
696  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
697  * won't ... remember to clear down the APIC, etc later.
698  */
699 int
700 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
701 {
702 	unsigned long send_status, accept_status = 0;
703 	int maxlvt;
704 
705 	/* Target chip */
706 	/* Boot on the stack */
707 	/* Kick the second */
708 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, apicid);
709 
710 	pr_debug("Waiting for send to finish...\n");
711 	send_status = safe_apic_wait_icr_idle();
712 
713 	/*
714 	 * Give the other CPU some time to accept the IPI.
715 	 */
716 	udelay(200);
717 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
718 		maxlvt = lapic_get_maxlvt();
719 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
720 			apic_write(APIC_ESR, 0);
721 		accept_status = (apic_read(APIC_ESR) & 0xEF);
722 	}
723 	pr_debug("NMI sent\n");
724 
725 	if (send_status)
726 		pr_err("APIC never delivered???\n");
727 	if (accept_status)
728 		pr_err("APIC delivery error (%lx)\n", accept_status);
729 
730 	return (send_status | accept_status);
731 }
732 
733 static int
734 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
735 {
736 	unsigned long send_status = 0, accept_status = 0;
737 	int maxlvt, num_starts, j;
738 
739 	maxlvt = lapic_get_maxlvt();
740 
741 	/*
742 	 * Be paranoid about clearing APIC errors.
743 	 */
744 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
745 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
746 			apic_write(APIC_ESR, 0);
747 		apic_read(APIC_ESR);
748 	}
749 
750 	pr_debug("Asserting INIT\n");
751 
752 	/*
753 	 * Turn INIT on target chip
754 	 */
755 	/*
756 	 * Send IPI
757 	 */
758 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
759 		       phys_apicid);
760 
761 	pr_debug("Waiting for send to finish...\n");
762 	send_status = safe_apic_wait_icr_idle();
763 
764 	udelay(init_udelay);
765 
766 	pr_debug("Deasserting INIT\n");
767 
768 	/* Target chip */
769 	/* Send IPI */
770 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
771 
772 	pr_debug("Waiting for send to finish...\n");
773 	send_status = safe_apic_wait_icr_idle();
774 
775 	mb();
776 
777 	/*
778 	 * Should we send STARTUP IPIs ?
779 	 *
780 	 * Determine this based on the APIC version.
781 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
782 	 */
783 	if (APIC_INTEGRATED(boot_cpu_apic_version))
784 		num_starts = 2;
785 	else
786 		num_starts = 0;
787 
788 	/*
789 	 * Run STARTUP IPI loop.
790 	 */
791 	pr_debug("#startup loops: %d\n", num_starts);
792 
793 	for (j = 1; j <= num_starts; j++) {
794 		pr_debug("Sending STARTUP #%d\n", j);
795 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
796 			apic_write(APIC_ESR, 0);
797 		apic_read(APIC_ESR);
798 		pr_debug("After apic_write\n");
799 
800 		/*
801 		 * STARTUP IPI
802 		 */
803 
804 		/* Target chip */
805 		/* Boot on the stack */
806 		/* Kick the second */
807 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
808 			       phys_apicid);
809 
810 		/*
811 		 * Give the other CPU some time to accept the IPI.
812 		 */
813 		if (init_udelay == 0)
814 			udelay(10);
815 		else
816 			udelay(300);
817 
818 		pr_debug("Startup point 1\n");
819 
820 		pr_debug("Waiting for send to finish...\n");
821 		send_status = safe_apic_wait_icr_idle();
822 
823 		/*
824 		 * Give the other CPU some time to accept the IPI.
825 		 */
826 		if (init_udelay == 0)
827 			udelay(10);
828 		else
829 			udelay(200);
830 
831 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
832 			apic_write(APIC_ESR, 0);
833 		accept_status = (apic_read(APIC_ESR) & 0xEF);
834 		if (send_status || accept_status)
835 			break;
836 	}
837 	pr_debug("After Startup\n");
838 
839 	if (send_status)
840 		pr_err("APIC never delivered???\n");
841 	if (accept_status)
842 		pr_err("APIC delivery error (%lx)\n", accept_status);
843 
844 	return (send_status | accept_status);
845 }
846 
847 /* reduce the number of lines printed when booting a large cpu count system */
848 static void announce_cpu(int cpu, int apicid)
849 {
850 	static int current_node = -1;
851 	int node = early_cpu_to_node(cpu);
852 	static int width, node_width;
853 
854 	if (!width)
855 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
856 
857 	if (!node_width)
858 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
859 
860 	if (cpu == 1)
861 		printk(KERN_INFO "x86: Booting SMP configuration:\n");
862 
863 	if (system_state == SYSTEM_BOOTING) {
864 		if (node != current_node) {
865 			if (current_node > (-1))
866 				pr_cont("\n");
867 			current_node = node;
868 
869 			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
870 			       node_width - num_digits(node), " ", node);
871 		}
872 
873 		/* Add padding for the BSP */
874 		if (cpu == 1)
875 			pr_cont("%*s", width + 1, " ");
876 
877 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
878 
879 	} else
880 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
881 			node, cpu, apicid);
882 }
883 
884 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
885 {
886 	int cpu;
887 
888 	cpu = smp_processor_id();
889 	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
890 		return NMI_HANDLED;
891 
892 	return NMI_DONE;
893 }
894 
895 /*
896  * Wake up AP by INIT, INIT, STARTUP sequence.
897  *
898  * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
899  * boot-strap code which is not a desired behavior for waking up BSP. To
900  * void the boot-strap code, wake up CPU0 by NMI instead.
901  *
902  * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
903  * (i.e. physically hot removed and then hot added), NMI won't wake it up.
904  * We'll change this code in the future to wake up hard offlined CPU0 if
905  * real platform and request are available.
906  */
907 static int
908 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
909 	       int *cpu0_nmi_registered)
910 {
911 	int id;
912 	int boot_error;
913 
914 	preempt_disable();
915 
916 	/*
917 	 * Wake up AP by INIT, INIT, STARTUP sequence.
918 	 */
919 	if (cpu) {
920 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
921 		goto out;
922 	}
923 
924 	/*
925 	 * Wake up BSP by nmi.
926 	 *
927 	 * Register a NMI handler to help wake up CPU0.
928 	 */
929 	boot_error = register_nmi_handler(NMI_LOCAL,
930 					  wakeup_cpu0_nmi, 0, "wake_cpu0");
931 
932 	if (!boot_error) {
933 		enable_start_cpu0 = 1;
934 		*cpu0_nmi_registered = 1;
935 		if (apic->dest_logical == APIC_DEST_LOGICAL)
936 			id = cpu0_logical_apicid;
937 		else
938 			id = apicid;
939 		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
940 	}
941 
942 out:
943 	preempt_enable();
944 
945 	return boot_error;
946 }
947 
948 void common_cpu_up(unsigned int cpu, struct task_struct *idle)
949 {
950 	/* Just in case we booted with a single CPU. */
951 	alternatives_enable_smp();
952 
953 	per_cpu(current_task, cpu) = idle;
954 
955 #ifdef CONFIG_X86_32
956 	/* Stack for startup_32 can be just as for start_secondary onwards */
957 	irq_ctx_init(cpu);
958 	per_cpu(cpu_current_top_of_stack, cpu) =
959 		(unsigned long)task_stack_page(idle) + THREAD_SIZE;
960 #else
961 	initial_gs = per_cpu_offset(cpu);
962 #endif
963 }
964 
965 /*
966  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
967  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
968  * Returns zero if CPU booted OK, else error code from
969  * ->wakeup_secondary_cpu.
970  */
971 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
972 {
973 	volatile u32 *trampoline_status =
974 		(volatile u32 *) __va(real_mode_header->trampoline_status);
975 	/* start_ip had better be page-aligned! */
976 	unsigned long start_ip = real_mode_header->trampoline_start;
977 
978 	unsigned long boot_error = 0;
979 	int cpu0_nmi_registered = 0;
980 	unsigned long timeout;
981 
982 	idle->thread.sp = (unsigned long)task_pt_regs(idle);
983 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
984 	initial_code = (unsigned long)start_secondary;
985 	initial_stack  = idle->thread.sp;
986 
987 	/*
988 	 * Enable the espfix hack for this CPU
989 	*/
990 #ifdef CONFIG_X86_ESPFIX64
991 	init_espfix_ap(cpu);
992 #endif
993 
994 	/* So we see what's up */
995 	announce_cpu(cpu, apicid);
996 
997 	/*
998 	 * This grunge runs the startup process for
999 	 * the targeted processor.
1000 	 */
1001 
1002 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
1003 
1004 		pr_debug("Setting warm reset code and vector.\n");
1005 
1006 		smpboot_setup_warm_reset_vector(start_ip);
1007 		/*
1008 		 * Be paranoid about clearing APIC errors.
1009 		*/
1010 		if (APIC_INTEGRATED(boot_cpu_apic_version)) {
1011 			apic_write(APIC_ESR, 0);
1012 			apic_read(APIC_ESR);
1013 		}
1014 	}
1015 
1016 	/*
1017 	 * AP might wait on cpu_callout_mask in cpu_init() with
1018 	 * cpu_initialized_mask set if previous attempt to online
1019 	 * it timed-out. Clear cpu_initialized_mask so that after
1020 	 * INIT/SIPI it could start with a clean state.
1021 	 */
1022 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1023 	smp_mb();
1024 
1025 	/*
1026 	 * Wake up a CPU in difference cases:
1027 	 * - Use the method in the APIC driver if it's defined
1028 	 * Otherwise,
1029 	 * - Use an INIT boot APIC message for APs or NMI for BSP.
1030 	 */
1031 	if (apic->wakeup_secondary_cpu)
1032 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
1033 	else
1034 		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
1035 						     &cpu0_nmi_registered);
1036 
1037 	if (!boot_error) {
1038 		/*
1039 		 * Wait 10s total for first sign of life from AP
1040 		 */
1041 		boot_error = -1;
1042 		timeout = jiffies + 10*HZ;
1043 		while (time_before(jiffies, timeout)) {
1044 			if (cpumask_test_cpu(cpu, cpu_initialized_mask)) {
1045 				/*
1046 				 * Tell AP to proceed with initialization
1047 				 */
1048 				cpumask_set_cpu(cpu, cpu_callout_mask);
1049 				boot_error = 0;
1050 				break;
1051 			}
1052 			schedule();
1053 		}
1054 	}
1055 
1056 	if (!boot_error) {
1057 		/*
1058 		 * Wait till AP completes initial initialization
1059 		 */
1060 		while (!cpumask_test_cpu(cpu, cpu_callin_mask)) {
1061 			/*
1062 			 * Allow other tasks to run while we wait for the
1063 			 * AP to come online. This also gives a chance
1064 			 * for the MTRR work(triggered by the AP coming online)
1065 			 * to be completed in the stop machine context.
1066 			 */
1067 			schedule();
1068 		}
1069 	}
1070 
1071 	/* mark "stuck" area as not stuck */
1072 	*trampoline_status = 0;
1073 
1074 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
1075 		/*
1076 		 * Cleanup possible dangling ends...
1077 		 */
1078 		smpboot_restore_warm_reset_vector();
1079 	}
1080 	/*
1081 	 * Clean up the nmi handler. Do this after the callin and callout sync
1082 	 * to avoid impact of possible long unregister time.
1083 	 */
1084 	if (cpu0_nmi_registered)
1085 		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
1086 
1087 	return boot_error;
1088 }
1089 
1090 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
1091 {
1092 	int apicid = apic->cpu_present_to_apicid(cpu);
1093 	unsigned long flags;
1094 	int err;
1095 
1096 	WARN_ON(irqs_disabled());
1097 
1098 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
1099 
1100 	if (apicid == BAD_APICID ||
1101 	    !physid_isset(apicid, phys_cpu_present_map) ||
1102 	    !apic->apic_id_valid(apicid)) {
1103 		pr_err("%s: bad cpu %d\n", __func__, cpu);
1104 		return -EINVAL;
1105 	}
1106 
1107 	/*
1108 	 * Already booted CPU?
1109 	 */
1110 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
1111 		pr_debug("do_boot_cpu %d Already started\n", cpu);
1112 		return -ENOSYS;
1113 	}
1114 
1115 	/*
1116 	 * Save current MTRR state in case it was changed since early boot
1117 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
1118 	 */
1119 	mtrr_save_state();
1120 
1121 	/* x86 CPUs take themselves offline, so delayed offline is OK. */
1122 	err = cpu_check_up_prepare(cpu);
1123 	if (err && err != -EBUSY)
1124 		return err;
1125 
1126 	/* the FPU context is blank, nobody can own it */
1127 	per_cpu(fpu_fpregs_owner_ctx, cpu) = NULL;
1128 
1129 	common_cpu_up(cpu, tidle);
1130 
1131 	err = do_boot_cpu(apicid, cpu, tidle);
1132 	if (err) {
1133 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
1134 		return -EIO;
1135 	}
1136 
1137 	/*
1138 	 * Check TSC synchronization with the AP (keep irqs disabled
1139 	 * while doing so):
1140 	 */
1141 	local_irq_save(flags);
1142 	check_tsc_sync_source(cpu);
1143 	local_irq_restore(flags);
1144 
1145 	while (!cpu_online(cpu)) {
1146 		cpu_relax();
1147 		touch_nmi_watchdog();
1148 	}
1149 
1150 	return 0;
1151 }
1152 
1153 /**
1154  * arch_disable_smp_support() - disables SMP support for x86 at runtime
1155  */
1156 void arch_disable_smp_support(void)
1157 {
1158 	disable_ioapic_support();
1159 }
1160 
1161 /*
1162  * Fall back to non SMP mode after errors.
1163  *
1164  * RED-PEN audit/test this more. I bet there is more state messed up here.
1165  */
1166 static __init void disable_smp(void)
1167 {
1168 	pr_info("SMP disabled\n");
1169 
1170 	disable_ioapic_support();
1171 
1172 	init_cpu_present(cpumask_of(0));
1173 	init_cpu_possible(cpumask_of(0));
1174 
1175 	if (smp_found_config)
1176 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1177 	else
1178 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1179 	cpumask_set_cpu(0, topology_sibling_cpumask(0));
1180 	cpumask_set_cpu(0, topology_core_cpumask(0));
1181 }
1182 
1183 enum {
1184 	SMP_OK,
1185 	SMP_NO_CONFIG,
1186 	SMP_NO_APIC,
1187 	SMP_FORCE_UP,
1188 };
1189 
1190 /*
1191  * Various sanity checks.
1192  */
1193 static int __init smp_sanity_check(unsigned max_cpus)
1194 {
1195 	preempt_disable();
1196 
1197 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
1198 	if (def_to_bigsmp && nr_cpu_ids > 8) {
1199 		unsigned int cpu;
1200 		unsigned nr;
1201 
1202 		pr_warn("More than 8 CPUs detected - skipping them\n"
1203 			"Use CONFIG_X86_BIGSMP\n");
1204 
1205 		nr = 0;
1206 		for_each_present_cpu(cpu) {
1207 			if (nr >= 8)
1208 				set_cpu_present(cpu, false);
1209 			nr++;
1210 		}
1211 
1212 		nr = 0;
1213 		for_each_possible_cpu(cpu) {
1214 			if (nr >= 8)
1215 				set_cpu_possible(cpu, false);
1216 			nr++;
1217 		}
1218 
1219 		nr_cpu_ids = 8;
1220 	}
1221 #endif
1222 
1223 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1224 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1225 			hard_smp_processor_id());
1226 
1227 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1228 	}
1229 
1230 	/*
1231 	 * If we couldn't find an SMP configuration at boot time,
1232 	 * get out of here now!
1233 	 */
1234 	if (!smp_found_config && !acpi_lapic) {
1235 		preempt_enable();
1236 		pr_notice("SMP motherboard not detected\n");
1237 		return SMP_NO_CONFIG;
1238 	}
1239 
1240 	/*
1241 	 * Should not be necessary because the MP table should list the boot
1242 	 * CPU too, but we do it for the sake of robustness anyway.
1243 	 */
1244 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1245 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1246 			  boot_cpu_physical_apicid);
1247 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1248 	}
1249 	preempt_enable();
1250 
1251 	/*
1252 	 * If we couldn't find a local APIC, then get out of here now!
1253 	 */
1254 	if (APIC_INTEGRATED(boot_cpu_apic_version) &&
1255 	    !boot_cpu_has(X86_FEATURE_APIC)) {
1256 		if (!disable_apic) {
1257 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
1258 				boot_cpu_physical_apicid);
1259 			pr_err("... forcing use of dummy APIC emulation (tell your hw vendor)\n");
1260 		}
1261 		return SMP_NO_APIC;
1262 	}
1263 
1264 	/*
1265 	 * If SMP should be disabled, then really disable it!
1266 	 */
1267 	if (!max_cpus) {
1268 		pr_info("SMP mode deactivated\n");
1269 		return SMP_FORCE_UP;
1270 	}
1271 
1272 	return SMP_OK;
1273 }
1274 
1275 static void __init smp_cpu_index_default(void)
1276 {
1277 	int i;
1278 	struct cpuinfo_x86 *c;
1279 
1280 	for_each_possible_cpu(i) {
1281 		c = &cpu_data(i);
1282 		/* mark all to hotplug */
1283 		c->cpu_index = nr_cpu_ids;
1284 	}
1285 }
1286 
1287 /*
1288  * Prepare for SMP bootup.  The MP table or ACPI has been read
1289  * earlier.  Just do some sanity checking here and enable APIC mode.
1290  */
1291 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1292 {
1293 	unsigned int i;
1294 
1295 	smp_cpu_index_default();
1296 
1297 	/*
1298 	 * Setup boot CPU information
1299 	 */
1300 	smp_store_boot_cpu_info(); /* Final full version of the data */
1301 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1302 	mb();
1303 
1304 	for_each_possible_cpu(i) {
1305 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1306 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1307 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1308 	}
1309 
1310 	/*
1311 	 * Set 'default' x86 topology, this matches default_topology() in that
1312 	 * it has NUMA nodes as a topology level. See also
1313 	 * native_smp_cpus_done().
1314 	 *
1315 	 * Must be done before set_cpus_sibling_map() is ran.
1316 	 */
1317 	set_sched_topology(x86_topology);
1318 
1319 	set_cpu_sibling_map(0);
1320 
1321 	switch (smp_sanity_check(max_cpus)) {
1322 	case SMP_NO_CONFIG:
1323 		disable_smp();
1324 		if (APIC_init_uniprocessor())
1325 			pr_notice("Local APIC not detected. Using dummy APIC emulation.\n");
1326 		return;
1327 	case SMP_NO_APIC:
1328 		disable_smp();
1329 		return;
1330 	case SMP_FORCE_UP:
1331 		disable_smp();
1332 		apic_bsp_setup(false);
1333 		return;
1334 	case SMP_OK:
1335 		break;
1336 	}
1337 
1338 	if (read_apic_id() != boot_cpu_physical_apicid) {
1339 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1340 		     read_apic_id(), boot_cpu_physical_apicid);
1341 		/* Or can we switch back to PIC here? */
1342 	}
1343 
1344 	default_setup_apic_routing();
1345 	cpu0_logical_apicid = apic_bsp_setup(false);
1346 
1347 	pr_info("CPU0: ");
1348 	print_cpu_info(&cpu_data(0));
1349 
1350 	uv_system_init();
1351 
1352 	set_mtrr_aps_delayed_init();
1353 
1354 	smp_quirk_init_udelay();
1355 }
1356 
1357 void arch_enable_nonboot_cpus_begin(void)
1358 {
1359 	set_mtrr_aps_delayed_init();
1360 }
1361 
1362 void arch_enable_nonboot_cpus_end(void)
1363 {
1364 	mtrr_aps_init();
1365 }
1366 
1367 /*
1368  * Early setup to make printk work.
1369  */
1370 void __init native_smp_prepare_boot_cpu(void)
1371 {
1372 	int me = smp_processor_id();
1373 	switch_to_new_gdt(me);
1374 	/* already set me in cpu_online_mask in boot_cpu_init() */
1375 	cpumask_set_cpu(me, cpu_callout_mask);
1376 	cpu_set_state_online(me);
1377 }
1378 
1379 void __init native_smp_cpus_done(unsigned int max_cpus)
1380 {
1381 	pr_debug("Boot done\n");
1382 
1383 	if (x86_has_numa_in_package)
1384 		set_sched_topology(x86_numa_in_package_topology);
1385 
1386 	nmi_selftest();
1387 	impress_friends();
1388 	setup_ioapic_dest();
1389 	mtrr_aps_init();
1390 }
1391 
1392 static int __initdata setup_possible_cpus = -1;
1393 static int __init _setup_possible_cpus(char *str)
1394 {
1395 	get_option(&str, &setup_possible_cpus);
1396 	return 0;
1397 }
1398 early_param("possible_cpus", _setup_possible_cpus);
1399 
1400 
1401 /*
1402  * cpu_possible_mask should be static, it cannot change as cpu's
1403  * are onlined, or offlined. The reason is per-cpu data-structures
1404  * are allocated by some modules at init time, and dont expect to
1405  * do this dynamically on cpu arrival/departure.
1406  * cpu_present_mask on the other hand can change dynamically.
1407  * In case when cpu_hotplug is not compiled, then we resort to current
1408  * behaviour, which is cpu_possible == cpu_present.
1409  * - Ashok Raj
1410  *
1411  * Three ways to find out the number of additional hotplug CPUs:
1412  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1413  * - The user can overwrite it with possible_cpus=NUM
1414  * - Otherwise don't reserve additional CPUs.
1415  * We do this because additional CPUs waste a lot of memory.
1416  * -AK
1417  */
1418 __init void prefill_possible_map(void)
1419 {
1420 	int i, possible;
1421 
1422 	/* No boot processor was found in mptable or ACPI MADT */
1423 	if (!num_processors) {
1424 		if (boot_cpu_has(X86_FEATURE_APIC)) {
1425 			int apicid = boot_cpu_physical_apicid;
1426 			int cpu = hard_smp_processor_id();
1427 
1428 			pr_warn("Boot CPU (id %d) not listed by BIOS\n", cpu);
1429 
1430 			/* Make sure boot cpu is enumerated */
1431 			if (apic->cpu_present_to_apicid(0) == BAD_APICID &&
1432 			    apic->apic_id_valid(apicid))
1433 				generic_processor_info(apicid, boot_cpu_apic_version);
1434 		}
1435 
1436 		if (!num_processors)
1437 			num_processors = 1;
1438 	}
1439 
1440 	i = setup_max_cpus ?: 1;
1441 	if (setup_possible_cpus == -1) {
1442 		possible = num_processors;
1443 #ifdef CONFIG_HOTPLUG_CPU
1444 		if (setup_max_cpus)
1445 			possible += disabled_cpus;
1446 #else
1447 		if (possible > i)
1448 			possible = i;
1449 #endif
1450 	} else
1451 		possible = setup_possible_cpus;
1452 
1453 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1454 
1455 	/* nr_cpu_ids could be reduced via nr_cpus= */
1456 	if (possible > nr_cpu_ids) {
1457 		pr_warn("%d Processors exceeds NR_CPUS limit of %d\n",
1458 			possible, nr_cpu_ids);
1459 		possible = nr_cpu_ids;
1460 	}
1461 
1462 #ifdef CONFIG_HOTPLUG_CPU
1463 	if (!setup_max_cpus)
1464 #endif
1465 	if (possible > i) {
1466 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1467 			possible, setup_max_cpus);
1468 		possible = i;
1469 	}
1470 
1471 	nr_cpu_ids = possible;
1472 
1473 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1474 		possible, max_t(int, possible - num_processors, 0));
1475 
1476 	reset_cpu_possible_mask();
1477 
1478 	for (i = 0; i < possible; i++)
1479 		set_cpu_possible(i, true);
1480 }
1481 
1482 #ifdef CONFIG_HOTPLUG_CPU
1483 
1484 /* Recompute SMT state for all CPUs on offline */
1485 static void recompute_smt_state(void)
1486 {
1487 	int max_threads, cpu;
1488 
1489 	max_threads = 0;
1490 	for_each_online_cpu (cpu) {
1491 		int threads = cpumask_weight(topology_sibling_cpumask(cpu));
1492 
1493 		if (threads > max_threads)
1494 			max_threads = threads;
1495 	}
1496 	__max_smt_threads = max_threads;
1497 }
1498 
1499 static void remove_siblinginfo(int cpu)
1500 {
1501 	int sibling;
1502 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1503 
1504 	for_each_cpu(sibling, topology_core_cpumask(cpu)) {
1505 		cpumask_clear_cpu(cpu, topology_core_cpumask(sibling));
1506 		/*/
1507 		 * last thread sibling in this cpu core going down
1508 		 */
1509 		if (cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
1510 			cpu_data(sibling).booted_cores--;
1511 	}
1512 
1513 	for_each_cpu(sibling, topology_sibling_cpumask(cpu))
1514 		cpumask_clear_cpu(cpu, topology_sibling_cpumask(sibling));
1515 	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1516 		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1517 	cpumask_clear(cpu_llc_shared_mask(cpu));
1518 	cpumask_clear(topology_sibling_cpumask(cpu));
1519 	cpumask_clear(topology_core_cpumask(cpu));
1520 	c->phys_proc_id = 0;
1521 	c->cpu_core_id = 0;
1522 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1523 	recompute_smt_state();
1524 }
1525 
1526 static void remove_cpu_from_maps(int cpu)
1527 {
1528 	set_cpu_online(cpu, false);
1529 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1530 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1531 	/* was set by cpu_init() */
1532 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1533 	numa_remove_cpu(cpu);
1534 }
1535 
1536 void cpu_disable_common(void)
1537 {
1538 	int cpu = smp_processor_id();
1539 
1540 	remove_siblinginfo(cpu);
1541 
1542 	/* It's now safe to remove this processor from the online map */
1543 	lock_vector_lock();
1544 	remove_cpu_from_maps(cpu);
1545 	unlock_vector_lock();
1546 	fixup_irqs();
1547 }
1548 
1549 int native_cpu_disable(void)
1550 {
1551 	int ret;
1552 
1553 	ret = check_irq_vectors_for_cpu_disable();
1554 	if (ret)
1555 		return ret;
1556 
1557 	clear_local_APIC();
1558 	cpu_disable_common();
1559 
1560 	return 0;
1561 }
1562 
1563 int common_cpu_die(unsigned int cpu)
1564 {
1565 	int ret = 0;
1566 
1567 	/* We don't do anything here: idle task is faking death itself. */
1568 
1569 	/* They ack this in play_dead() by setting CPU_DEAD */
1570 	if (cpu_wait_death(cpu, 5)) {
1571 		if (system_state == SYSTEM_RUNNING)
1572 			pr_info("CPU %u is now offline\n", cpu);
1573 	} else {
1574 		pr_err("CPU %u didn't die...\n", cpu);
1575 		ret = -1;
1576 	}
1577 
1578 	return ret;
1579 }
1580 
1581 void native_cpu_die(unsigned int cpu)
1582 {
1583 	common_cpu_die(cpu);
1584 }
1585 
1586 void play_dead_common(void)
1587 {
1588 	idle_task_exit();
1589 	reset_lazy_tlbstate();
1590 
1591 	/* Ack it */
1592 	(void)cpu_report_death();
1593 
1594 	/*
1595 	 * With physical CPU hotplug, we should halt the cpu
1596 	 */
1597 	local_irq_disable();
1598 }
1599 
1600 static bool wakeup_cpu0(void)
1601 {
1602 	if (smp_processor_id() == 0 && enable_start_cpu0)
1603 		return true;
1604 
1605 	return false;
1606 }
1607 
1608 /*
1609  * We need to flush the caches before going to sleep, lest we have
1610  * dirty data in our caches when we come back up.
1611  */
1612 static inline void mwait_play_dead(void)
1613 {
1614 	unsigned int eax, ebx, ecx, edx;
1615 	unsigned int highest_cstate = 0;
1616 	unsigned int highest_subcstate = 0;
1617 	void *mwait_ptr;
1618 	int i;
1619 
1620 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1621 		return;
1622 	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1623 		return;
1624 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1625 		return;
1626 
1627 	eax = CPUID_MWAIT_LEAF;
1628 	ecx = 0;
1629 	native_cpuid(&eax, &ebx, &ecx, &edx);
1630 
1631 	/*
1632 	 * eax will be 0 if EDX enumeration is not valid.
1633 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1634 	 */
1635 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1636 		eax = 0;
1637 	} else {
1638 		edx >>= MWAIT_SUBSTATE_SIZE;
1639 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1640 			if (edx & MWAIT_SUBSTATE_MASK) {
1641 				highest_cstate = i;
1642 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1643 			}
1644 		}
1645 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1646 			(highest_subcstate - 1);
1647 	}
1648 
1649 	/*
1650 	 * This should be a memory location in a cache line which is
1651 	 * unlikely to be touched by other processors.  The actual
1652 	 * content is immaterial as it is not actually modified in any way.
1653 	 */
1654 	mwait_ptr = &current_thread_info()->flags;
1655 
1656 	wbinvd();
1657 
1658 	while (1) {
1659 		/*
1660 		 * The CLFLUSH is a workaround for erratum AAI65 for
1661 		 * the Xeon 7400 series.  It's not clear it is actually
1662 		 * needed, but it should be harmless in either case.
1663 		 * The WBINVD is insufficient due to the spurious-wakeup
1664 		 * case where we return around the loop.
1665 		 */
1666 		mb();
1667 		clflush(mwait_ptr);
1668 		mb();
1669 		__monitor(mwait_ptr, 0, 0);
1670 		mb();
1671 		__mwait(eax, 0);
1672 		/*
1673 		 * If NMI wants to wake up CPU0, start CPU0.
1674 		 */
1675 		if (wakeup_cpu0())
1676 			start_cpu0();
1677 	}
1678 }
1679 
1680 void hlt_play_dead(void)
1681 {
1682 	if (__this_cpu_read(cpu_info.x86) >= 4)
1683 		wbinvd();
1684 
1685 	while (1) {
1686 		native_halt();
1687 		/*
1688 		 * If NMI wants to wake up CPU0, start CPU0.
1689 		 */
1690 		if (wakeup_cpu0())
1691 			start_cpu0();
1692 	}
1693 }
1694 
1695 void native_play_dead(void)
1696 {
1697 	play_dead_common();
1698 	tboot_shutdown(TB_SHUTDOWN_WFS);
1699 
1700 	mwait_play_dead();	/* Only returns on failure */
1701 	if (cpuidle_play_dead())
1702 		hlt_play_dead();
1703 }
1704 
1705 #else /* ... !CONFIG_HOTPLUG_CPU */
1706 int native_cpu_disable(void)
1707 {
1708 	return -ENOSYS;
1709 }
1710 
1711 void native_cpu_die(unsigned int cpu)
1712 {
1713 	/* We said "no" in __cpu_disable */
1714 	BUG();
1715 }
1716 
1717 void native_play_dead(void)
1718 {
1719 	BUG();
1720 }
1721 
1722 #endif
1723