xref: /linux/arch/x86/kernel/smpboot.c (revision a594533df0f6ca391da003f43d53b336a2d23ffa)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2  /*
3  *	x86 SMP booting functions
4  *
5  *	(c) 1995 Alan Cox, Building #3 <alan@lxorguk.ukuu.org.uk>
6  *	(c) 1998, 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
7  *	Copyright 2001 Andi Kleen, SuSE Labs.
8  *
9  *	Much of the core SMP work is based on previous work by Thomas Radke, to
10  *	whom a great many thanks are extended.
11  *
12  *	Thanks to Intel for making available several different Pentium,
13  *	Pentium Pro and Pentium-II/Xeon MP machines.
14  *	Original development of Linux SMP code supported by Caldera.
15  *
16  *	Fixes
17  *		Felix Koop	:	NR_CPUS used properly
18  *		Jose Renau	:	Handle single CPU case.
19  *		Alan Cox	:	By repeated request 8) - Total BogoMIPS report.
20  *		Greg Wright	:	Fix for kernel stacks panic.
21  *		Erich Boleyn	:	MP v1.4 and additional changes.
22  *	Matthias Sattler	:	Changes for 2.1 kernel map.
23  *	Michel Lespinasse	:	Changes for 2.1 kernel map.
24  *	Michael Chastain	:	Change trampoline.S to gnu as.
25  *		Alan Cox	:	Dumb bug: 'B' step PPro's are fine
26  *		Ingo Molnar	:	Added APIC timers, based on code
27  *					from Jose Renau
28  *		Ingo Molnar	:	various cleanups and rewrites
29  *		Tigran Aivazian	:	fixed "0.00 in /proc/uptime on SMP" bug.
30  *	Maciej W. Rozycki	:	Bits for genuine 82489DX APICs
31  *	Andi Kleen		:	Changed for SMP boot into long mode.
32  *		Martin J. Bligh	: 	Added support for multi-quad systems
33  *		Dave Jones	:	Report invalid combinations of Athlon CPUs.
34  *		Rusty Russell	:	Hacked into shape for new "hotplug" boot process.
35  *      Andi Kleen              :       Converted to new state machine.
36  *	Ashok Raj		: 	CPU hotplug support
37  *	Glauber Costa		:	i386 and x86_64 integration
38  */
39 
40 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
41 
42 #include <linux/init.h>
43 #include <linux/smp.h>
44 #include <linux/export.h>
45 #include <linux/sched.h>
46 #include <linux/sched/topology.h>
47 #include <linux/sched/hotplug.h>
48 #include <linux/sched/task_stack.h>
49 #include <linux/percpu.h>
50 #include <linux/memblock.h>
51 #include <linux/err.h>
52 #include <linux/nmi.h>
53 #include <linux/tboot.h>
54 #include <linux/gfp.h>
55 #include <linux/cpuidle.h>
56 #include <linux/numa.h>
57 #include <linux/pgtable.h>
58 #include <linux/overflow.h>
59 #include <linux/stackprotector.h>
60 
61 #include <asm/acpi.h>
62 #include <asm/desc.h>
63 #include <asm/nmi.h>
64 #include <asm/irq.h>
65 #include <asm/realmode.h>
66 #include <asm/cpu.h>
67 #include <asm/numa.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/api.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/misc.h>
79 #include <asm/qspinlock.h>
80 #include <asm/intel-family.h>
81 #include <asm/cpu_device_id.h>
82 #include <asm/spec-ctrl.h>
83 #include <asm/hw_irq.h>
84 #include <asm/stackprotector.h>
85 #include <asm/sev.h>
86 
87 /* representing HT siblings of each logical CPU */
88 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
89 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
90 
91 /* representing HT and core siblings of each logical CPU */
92 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_map);
93 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
94 
95 /* representing HT, core, and die siblings of each logical CPU */
96 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_die_map);
97 EXPORT_PER_CPU_SYMBOL(cpu_die_map);
98 
99 /* Per CPU bogomips and other parameters */
100 DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
101 EXPORT_PER_CPU_SYMBOL(cpu_info);
102 
103 /* Logical package management. We might want to allocate that dynamically */
104 unsigned int __max_logical_packages __read_mostly;
105 EXPORT_SYMBOL(__max_logical_packages);
106 static unsigned int logical_packages __read_mostly;
107 static unsigned int logical_die __read_mostly;
108 
109 /* Maximum number of SMT threads on any online core */
110 int __read_mostly __max_smt_threads = 1;
111 
112 /* Flag to indicate if a complete sched domain rebuild is required */
113 bool x86_topology_update;
114 
115 int arch_update_cpu_topology(void)
116 {
117 	int retval = x86_topology_update;
118 
119 	x86_topology_update = false;
120 	return retval;
121 }
122 
123 static inline void smpboot_setup_warm_reset_vector(unsigned long start_eip)
124 {
125 	unsigned long flags;
126 
127 	spin_lock_irqsave(&rtc_lock, flags);
128 	CMOS_WRITE(0xa, 0xf);
129 	spin_unlock_irqrestore(&rtc_lock, flags);
130 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_HIGH)) =
131 							start_eip >> 4;
132 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) =
133 							start_eip & 0xf;
134 }
135 
136 static inline void smpboot_restore_warm_reset_vector(void)
137 {
138 	unsigned long flags;
139 
140 	/*
141 	 * Paranoid:  Set warm reset code and vector here back
142 	 * to default values.
143 	 */
144 	spin_lock_irqsave(&rtc_lock, flags);
145 	CMOS_WRITE(0, 0xf);
146 	spin_unlock_irqrestore(&rtc_lock, flags);
147 
148 	*((volatile u32 *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = 0;
149 }
150 
151 /*
152  * Report back to the Boot Processor during boot time or to the caller processor
153  * during CPU online.
154  */
155 static void smp_callin(void)
156 {
157 	int cpuid;
158 
159 	/*
160 	 * If waken up by an INIT in an 82489DX configuration
161 	 * cpu_callout_mask guarantees we don't get here before
162 	 * an INIT_deassert IPI reaches our local APIC, so it is
163 	 * now safe to touch our local APIC.
164 	 */
165 	cpuid = smp_processor_id();
166 
167 	/*
168 	 * the boot CPU has finished the init stage and is spinning
169 	 * on callin_map until we finish. We are free to set up this
170 	 * CPU, first the APIC. (this is probably redundant on most
171 	 * boards)
172 	 */
173 	apic_ap_setup();
174 
175 	/*
176 	 * Save our processor parameters. Note: this information
177 	 * is needed for clock calibration.
178 	 */
179 	smp_store_cpu_info(cpuid);
180 
181 	/*
182 	 * The topology information must be up to date before
183 	 * calibrate_delay() and notify_cpu_starting().
184 	 */
185 	set_cpu_sibling_map(raw_smp_processor_id());
186 
187 	ap_init_aperfmperf();
188 
189 	/*
190 	 * Get our bogomips.
191 	 * Update loops_per_jiffy in cpu_data. Previous call to
192 	 * smp_store_cpu_info() stored a value that is close but not as
193 	 * accurate as the value just calculated.
194 	 */
195 	calibrate_delay();
196 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
197 	pr_debug("Stack at about %p\n", &cpuid);
198 
199 	wmb();
200 
201 	notify_cpu_starting(cpuid);
202 
203 	/*
204 	 * Allow the master to continue.
205 	 */
206 	cpumask_set_cpu(cpuid, cpu_callin_mask);
207 }
208 
209 static int cpu0_logical_apicid;
210 static int enable_start_cpu0;
211 /*
212  * Activate a secondary processor.
213  */
214 static void notrace start_secondary(void *unused)
215 {
216 	/*
217 	 * Don't put *anything* except direct CPU state initialization
218 	 * before cpu_init(), SMP booting is too fragile that we want to
219 	 * limit the things done here to the most necessary things.
220 	 */
221 	cr4_init();
222 
223 #ifdef CONFIG_X86_32
224 	/* switch away from the initial page table */
225 	load_cr3(swapper_pg_dir);
226 	__flush_tlb_all();
227 #endif
228 	cpu_init_secondary();
229 	rcu_cpu_starting(raw_smp_processor_id());
230 	x86_cpuinit.early_percpu_clock_init();
231 	smp_callin();
232 
233 	enable_start_cpu0 = 0;
234 
235 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
236 	barrier();
237 	/*
238 	 * Check TSC synchronization with the boot CPU:
239 	 */
240 	check_tsc_sync_target();
241 
242 	speculative_store_bypass_ht_init();
243 
244 	/*
245 	 * Lock vector_lock, set CPU online and bring the vector
246 	 * allocator online. Online must be set with vector_lock held
247 	 * to prevent a concurrent irq setup/teardown from seeing a
248 	 * half valid vector space.
249 	 */
250 	lock_vector_lock();
251 	set_cpu_online(smp_processor_id(), true);
252 	lapic_online();
253 	unlock_vector_lock();
254 	cpu_set_state_online(smp_processor_id());
255 	x86_platform.nmi_init();
256 
257 	/* enable local interrupts */
258 	local_irq_enable();
259 
260 	x86_cpuinit.setup_percpu_clockev();
261 
262 	wmb();
263 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
264 }
265 
266 /**
267  * topology_is_primary_thread - Check whether CPU is the primary SMT thread
268  * @cpu:	CPU to check
269  */
270 bool topology_is_primary_thread(unsigned int cpu)
271 {
272 	return apic_id_is_primary_thread(per_cpu(x86_cpu_to_apicid, cpu));
273 }
274 
275 /**
276  * topology_smt_supported - Check whether SMT is supported by the CPUs
277  */
278 bool topology_smt_supported(void)
279 {
280 	return smp_num_siblings > 1;
281 }
282 
283 /**
284  * topology_phys_to_logical_pkg - Map a physical package id to a logical
285  *
286  * Returns logical package id or -1 if not found
287  */
288 int topology_phys_to_logical_pkg(unsigned int phys_pkg)
289 {
290 	int cpu;
291 
292 	for_each_possible_cpu(cpu) {
293 		struct cpuinfo_x86 *c = &cpu_data(cpu);
294 
295 		if (c->initialized && c->phys_proc_id == phys_pkg)
296 			return c->logical_proc_id;
297 	}
298 	return -1;
299 }
300 EXPORT_SYMBOL(topology_phys_to_logical_pkg);
301 /**
302  * topology_phys_to_logical_die - Map a physical die id to logical
303  *
304  * Returns logical die id or -1 if not found
305  */
306 int topology_phys_to_logical_die(unsigned int die_id, unsigned int cur_cpu)
307 {
308 	int cpu;
309 	int proc_id = cpu_data(cur_cpu).phys_proc_id;
310 
311 	for_each_possible_cpu(cpu) {
312 		struct cpuinfo_x86 *c = &cpu_data(cpu);
313 
314 		if (c->initialized && c->cpu_die_id == die_id &&
315 		    c->phys_proc_id == proc_id)
316 			return c->logical_die_id;
317 	}
318 	return -1;
319 }
320 EXPORT_SYMBOL(topology_phys_to_logical_die);
321 
322 /**
323  * topology_update_package_map - Update the physical to logical package map
324  * @pkg:	The physical package id as retrieved via CPUID
325  * @cpu:	The cpu for which this is updated
326  */
327 int topology_update_package_map(unsigned int pkg, unsigned int cpu)
328 {
329 	int new;
330 
331 	/* Already available somewhere? */
332 	new = topology_phys_to_logical_pkg(pkg);
333 	if (new >= 0)
334 		goto found;
335 
336 	new = logical_packages++;
337 	if (new != pkg) {
338 		pr_info("CPU %u Converting physical %u to logical package %u\n",
339 			cpu, pkg, new);
340 	}
341 found:
342 	cpu_data(cpu).logical_proc_id = new;
343 	return 0;
344 }
345 /**
346  * topology_update_die_map - Update the physical to logical die map
347  * @die:	The die id as retrieved via CPUID
348  * @cpu:	The cpu for which this is updated
349  */
350 int topology_update_die_map(unsigned int die, unsigned int cpu)
351 {
352 	int new;
353 
354 	/* Already available somewhere? */
355 	new = topology_phys_to_logical_die(die, cpu);
356 	if (new >= 0)
357 		goto found;
358 
359 	new = logical_die++;
360 	if (new != die) {
361 		pr_info("CPU %u Converting physical %u to logical die %u\n",
362 			cpu, die, new);
363 	}
364 found:
365 	cpu_data(cpu).logical_die_id = new;
366 	return 0;
367 }
368 
369 void __init smp_store_boot_cpu_info(void)
370 {
371 	int id = 0; /* CPU 0 */
372 	struct cpuinfo_x86 *c = &cpu_data(id);
373 
374 	*c = boot_cpu_data;
375 	c->cpu_index = id;
376 	topology_update_package_map(c->phys_proc_id, id);
377 	topology_update_die_map(c->cpu_die_id, id);
378 	c->initialized = true;
379 }
380 
381 /*
382  * The bootstrap kernel entry code has set these up. Save them for
383  * a given CPU
384  */
385 void smp_store_cpu_info(int id)
386 {
387 	struct cpuinfo_x86 *c = &cpu_data(id);
388 
389 	/* Copy boot_cpu_data only on the first bringup */
390 	if (!c->initialized)
391 		*c = boot_cpu_data;
392 	c->cpu_index = id;
393 	/*
394 	 * During boot time, CPU0 has this setup already. Save the info when
395 	 * bringing up AP or offlined CPU0.
396 	 */
397 	identify_secondary_cpu(c);
398 	c->initialized = true;
399 }
400 
401 static bool
402 topology_same_node(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
403 {
404 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
405 
406 	return (cpu_to_node(cpu1) == cpu_to_node(cpu2));
407 }
408 
409 static bool
410 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
411 {
412 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
413 
414 	return !WARN_ONCE(!topology_same_node(c, o),
415 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
416 		"[node: %d != %d]. Ignoring dependency.\n",
417 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
418 }
419 
420 #define link_mask(mfunc, c1, c2)					\
421 do {									\
422 	cpumask_set_cpu((c1), mfunc(c2));				\
423 	cpumask_set_cpu((c2), mfunc(c1));				\
424 } while (0)
425 
426 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
427 {
428 	if (boot_cpu_has(X86_FEATURE_TOPOEXT)) {
429 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
430 
431 		if (c->phys_proc_id == o->phys_proc_id &&
432 		    c->cpu_die_id == o->cpu_die_id &&
433 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2)) {
434 			if (c->cpu_core_id == o->cpu_core_id)
435 				return topology_sane(c, o, "smt");
436 
437 			if ((c->cu_id != 0xff) &&
438 			    (o->cu_id != 0xff) &&
439 			    (c->cu_id == o->cu_id))
440 				return topology_sane(c, o, "smt");
441 		}
442 
443 	} else if (c->phys_proc_id == o->phys_proc_id &&
444 		   c->cpu_die_id == o->cpu_die_id &&
445 		   c->cpu_core_id == o->cpu_core_id) {
446 		return topology_sane(c, o, "smt");
447 	}
448 
449 	return false;
450 }
451 
452 static bool match_die(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
453 {
454 	if (c->phys_proc_id == o->phys_proc_id &&
455 	    c->cpu_die_id == o->cpu_die_id)
456 		return true;
457 	return false;
458 }
459 
460 static bool match_l2c(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
461 {
462 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
463 
464 	/* If the arch didn't set up l2c_id, fall back to SMT */
465 	if (per_cpu(cpu_l2c_id, cpu1) == BAD_APICID)
466 		return match_smt(c, o);
467 
468 	/* Do not match if L2 cache id does not match: */
469 	if (per_cpu(cpu_l2c_id, cpu1) != per_cpu(cpu_l2c_id, cpu2))
470 		return false;
471 
472 	return topology_sane(c, o, "l2c");
473 }
474 
475 /*
476  * Unlike the other levels, we do not enforce keeping a
477  * multicore group inside a NUMA node.  If this happens, we will
478  * discard the MC level of the topology later.
479  */
480 static bool match_pkg(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
481 {
482 	if (c->phys_proc_id == o->phys_proc_id)
483 		return true;
484 	return false;
485 }
486 
487 /*
488  * Define intel_cod_cpu[] for Intel COD (Cluster-on-Die) CPUs.
489  *
490  * Any Intel CPU that has multiple nodes per package and does not
491  * match intel_cod_cpu[] has the SNC (Sub-NUMA Cluster) topology.
492  *
493  * When in SNC mode, these CPUs enumerate an LLC that is shared
494  * by multiple NUMA nodes. The LLC is shared for off-package data
495  * access but private to the NUMA node (half of the package) for
496  * on-package access. CPUID (the source of the information about
497  * the LLC) can only enumerate the cache as shared or unshared,
498  * but not this particular configuration.
499  */
500 
501 static const struct x86_cpu_id intel_cod_cpu[] = {
502 	X86_MATCH_INTEL_FAM6_MODEL(HASWELL_X, 0),	/* COD */
503 	X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_X, 0),	/* COD */
504 	X86_MATCH_INTEL_FAM6_MODEL(ANY, 1),		/* SNC */
505 	{}
506 };
507 
508 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
509 {
510 	const struct x86_cpu_id *id = x86_match_cpu(intel_cod_cpu);
511 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
512 	bool intel_snc = id && id->driver_data;
513 
514 	/* Do not match if we do not have a valid APICID for cpu: */
515 	if (per_cpu(cpu_llc_id, cpu1) == BAD_APICID)
516 		return false;
517 
518 	/* Do not match if LLC id does not match: */
519 	if (per_cpu(cpu_llc_id, cpu1) != per_cpu(cpu_llc_id, cpu2))
520 		return false;
521 
522 	/*
523 	 * Allow the SNC topology without warning. Return of false
524 	 * means 'c' does not share the LLC of 'o'. This will be
525 	 * reflected to userspace.
526 	 */
527 	if (match_pkg(c, o) && !topology_same_node(c, o) && intel_snc)
528 		return false;
529 
530 	return topology_sane(c, o, "llc");
531 }
532 
533 
534 #if defined(CONFIG_SCHED_SMT) || defined(CONFIG_SCHED_CLUSTER) || defined(CONFIG_SCHED_MC)
535 static inline int x86_sched_itmt_flags(void)
536 {
537 	return sysctl_sched_itmt_enabled ? SD_ASYM_PACKING : 0;
538 }
539 
540 #ifdef CONFIG_SCHED_MC
541 static int x86_core_flags(void)
542 {
543 	return cpu_core_flags() | x86_sched_itmt_flags();
544 }
545 #endif
546 #ifdef CONFIG_SCHED_SMT
547 static int x86_smt_flags(void)
548 {
549 	return cpu_smt_flags() | x86_sched_itmt_flags();
550 }
551 #endif
552 #ifdef CONFIG_SCHED_CLUSTER
553 static int x86_cluster_flags(void)
554 {
555 	return cpu_cluster_flags() | x86_sched_itmt_flags();
556 }
557 #endif
558 #endif
559 
560 static struct sched_domain_topology_level x86_numa_in_package_topology[] = {
561 #ifdef CONFIG_SCHED_SMT
562 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
563 #endif
564 #ifdef CONFIG_SCHED_CLUSTER
565 	{ cpu_clustergroup_mask, x86_cluster_flags, SD_INIT_NAME(CLS) },
566 #endif
567 #ifdef CONFIG_SCHED_MC
568 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
569 #endif
570 	{ NULL, },
571 };
572 
573 static struct sched_domain_topology_level x86_hybrid_topology[] = {
574 #ifdef CONFIG_SCHED_SMT
575 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
576 #endif
577 #ifdef CONFIG_SCHED_MC
578 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
579 #endif
580 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
581 	{ NULL, },
582 };
583 
584 static struct sched_domain_topology_level x86_topology[] = {
585 #ifdef CONFIG_SCHED_SMT
586 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
587 #endif
588 #ifdef CONFIG_SCHED_CLUSTER
589 	{ cpu_clustergroup_mask, x86_cluster_flags, SD_INIT_NAME(CLS) },
590 #endif
591 #ifdef CONFIG_SCHED_MC
592 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
593 #endif
594 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
595 	{ NULL, },
596 };
597 
598 /*
599  * Set if a package/die has multiple NUMA nodes inside.
600  * AMD Magny-Cours, Intel Cluster-on-Die, and Intel
601  * Sub-NUMA Clustering have this.
602  */
603 static bool x86_has_numa_in_package;
604 
605 void set_cpu_sibling_map(int cpu)
606 {
607 	bool has_smt = smp_num_siblings > 1;
608 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
609 	struct cpuinfo_x86 *c = &cpu_data(cpu);
610 	struct cpuinfo_x86 *o;
611 	int i, threads;
612 
613 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
614 
615 	if (!has_mp) {
616 		cpumask_set_cpu(cpu, topology_sibling_cpumask(cpu));
617 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
618 		cpumask_set_cpu(cpu, cpu_l2c_shared_mask(cpu));
619 		cpumask_set_cpu(cpu, topology_core_cpumask(cpu));
620 		cpumask_set_cpu(cpu, topology_die_cpumask(cpu));
621 		c->booted_cores = 1;
622 		return;
623 	}
624 
625 	for_each_cpu(i, cpu_sibling_setup_mask) {
626 		o = &cpu_data(i);
627 
628 		if (match_pkg(c, o) && !topology_same_node(c, o))
629 			x86_has_numa_in_package = true;
630 
631 		if ((i == cpu) || (has_smt && match_smt(c, o)))
632 			link_mask(topology_sibling_cpumask, cpu, i);
633 
634 		if ((i == cpu) || (has_mp && match_llc(c, o)))
635 			link_mask(cpu_llc_shared_mask, cpu, i);
636 
637 		if ((i == cpu) || (has_mp && match_l2c(c, o)))
638 			link_mask(cpu_l2c_shared_mask, cpu, i);
639 
640 		if ((i == cpu) || (has_mp && match_die(c, o)))
641 			link_mask(topology_die_cpumask, cpu, i);
642 	}
643 
644 	threads = cpumask_weight(topology_sibling_cpumask(cpu));
645 	if (threads > __max_smt_threads)
646 		__max_smt_threads = threads;
647 
648 	for_each_cpu(i, topology_sibling_cpumask(cpu))
649 		cpu_data(i).smt_active = threads > 1;
650 
651 	/*
652 	 * This needs a separate iteration over the cpus because we rely on all
653 	 * topology_sibling_cpumask links to be set-up.
654 	 */
655 	for_each_cpu(i, cpu_sibling_setup_mask) {
656 		o = &cpu_data(i);
657 
658 		if ((i == cpu) || (has_mp && match_pkg(c, o))) {
659 			link_mask(topology_core_cpumask, cpu, i);
660 
661 			/*
662 			 *  Does this new cpu bringup a new core?
663 			 */
664 			if (threads == 1) {
665 				/*
666 				 * for each core in package, increment
667 				 * the booted_cores for this new cpu
668 				 */
669 				if (cpumask_first(
670 				    topology_sibling_cpumask(i)) == i)
671 					c->booted_cores++;
672 				/*
673 				 * increment the core count for all
674 				 * the other cpus in this package
675 				 */
676 				if (i != cpu)
677 					cpu_data(i).booted_cores++;
678 			} else if (i != cpu && !c->booted_cores)
679 				c->booted_cores = cpu_data(i).booted_cores;
680 		}
681 	}
682 }
683 
684 /* maps the cpu to the sched domain representing multi-core */
685 const struct cpumask *cpu_coregroup_mask(int cpu)
686 {
687 	return cpu_llc_shared_mask(cpu);
688 }
689 
690 const struct cpumask *cpu_clustergroup_mask(int cpu)
691 {
692 	return cpu_l2c_shared_mask(cpu);
693 }
694 
695 static void impress_friends(void)
696 {
697 	int cpu;
698 	unsigned long bogosum = 0;
699 	/*
700 	 * Allow the user to impress friends.
701 	 */
702 	pr_debug("Before bogomips\n");
703 	for_each_possible_cpu(cpu)
704 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
705 			bogosum += cpu_data(cpu).loops_per_jiffy;
706 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
707 		num_online_cpus(),
708 		bogosum/(500000/HZ),
709 		(bogosum/(5000/HZ))%100);
710 
711 	pr_debug("Before bogocount - setting activated=1\n");
712 }
713 
714 void __inquire_remote_apic(int apicid)
715 {
716 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
717 	const char * const names[] = { "ID", "VERSION", "SPIV" };
718 	int timeout;
719 	u32 status;
720 
721 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
722 
723 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
724 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
725 
726 		/*
727 		 * Wait for idle.
728 		 */
729 		status = safe_apic_wait_icr_idle();
730 		if (status)
731 			pr_cont("a previous APIC delivery may have failed\n");
732 
733 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
734 
735 		timeout = 0;
736 		do {
737 			udelay(100);
738 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
739 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
740 
741 		switch (status) {
742 		case APIC_ICR_RR_VALID:
743 			status = apic_read(APIC_RRR);
744 			pr_cont("%08x\n", status);
745 			break;
746 		default:
747 			pr_cont("failed\n");
748 		}
749 	}
750 }
751 
752 /*
753  * The Multiprocessor Specification 1.4 (1997) example code suggests
754  * that there should be a 10ms delay between the BSP asserting INIT
755  * and de-asserting INIT, when starting a remote processor.
756  * But that slows boot and resume on modern processors, which include
757  * many cores and don't require that delay.
758  *
759  * Cmdline "init_cpu_udelay=" is available to over-ride this delay.
760  * Modern processor families are quirked to remove the delay entirely.
761  */
762 #define UDELAY_10MS_DEFAULT 10000
763 
764 static unsigned int init_udelay = UINT_MAX;
765 
766 static int __init cpu_init_udelay(char *str)
767 {
768 	get_option(&str, &init_udelay);
769 
770 	return 0;
771 }
772 early_param("cpu_init_udelay", cpu_init_udelay);
773 
774 static void __init smp_quirk_init_udelay(void)
775 {
776 	/* if cmdline changed it from default, leave it alone */
777 	if (init_udelay != UINT_MAX)
778 		return;
779 
780 	/* if modern processor, use no delay */
781 	if (((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && (boot_cpu_data.x86 == 6)) ||
782 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) && (boot_cpu_data.x86 >= 0x18)) ||
783 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && (boot_cpu_data.x86 >= 0xF))) {
784 		init_udelay = 0;
785 		return;
786 	}
787 	/* else, use legacy delay */
788 	init_udelay = UDELAY_10MS_DEFAULT;
789 }
790 
791 /*
792  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
793  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
794  * won't ... remember to clear down the APIC, etc later.
795  */
796 int
797 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
798 {
799 	u32 dm = apic->dest_mode_logical ? APIC_DEST_LOGICAL : APIC_DEST_PHYSICAL;
800 	unsigned long send_status, accept_status = 0;
801 	int maxlvt;
802 
803 	/* Target chip */
804 	/* Boot on the stack */
805 	/* Kick the second */
806 	apic_icr_write(APIC_DM_NMI | dm, apicid);
807 
808 	pr_debug("Waiting for send to finish...\n");
809 	send_status = safe_apic_wait_icr_idle();
810 
811 	/*
812 	 * Give the other CPU some time to accept the IPI.
813 	 */
814 	udelay(200);
815 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
816 		maxlvt = lapic_get_maxlvt();
817 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
818 			apic_write(APIC_ESR, 0);
819 		accept_status = (apic_read(APIC_ESR) & 0xEF);
820 	}
821 	pr_debug("NMI sent\n");
822 
823 	if (send_status)
824 		pr_err("APIC never delivered???\n");
825 	if (accept_status)
826 		pr_err("APIC delivery error (%lx)\n", accept_status);
827 
828 	return (send_status | accept_status);
829 }
830 
831 static int
832 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
833 {
834 	unsigned long send_status = 0, accept_status = 0;
835 	int maxlvt, num_starts, j;
836 
837 	maxlvt = lapic_get_maxlvt();
838 
839 	/*
840 	 * Be paranoid about clearing APIC errors.
841 	 */
842 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
843 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
844 			apic_write(APIC_ESR, 0);
845 		apic_read(APIC_ESR);
846 	}
847 
848 	pr_debug("Asserting INIT\n");
849 
850 	/*
851 	 * Turn INIT on target chip
852 	 */
853 	/*
854 	 * Send IPI
855 	 */
856 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
857 		       phys_apicid);
858 
859 	pr_debug("Waiting for send to finish...\n");
860 	send_status = safe_apic_wait_icr_idle();
861 
862 	udelay(init_udelay);
863 
864 	pr_debug("Deasserting INIT\n");
865 
866 	/* Target chip */
867 	/* Send IPI */
868 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
869 
870 	pr_debug("Waiting for send to finish...\n");
871 	send_status = safe_apic_wait_icr_idle();
872 
873 	mb();
874 
875 	/*
876 	 * Should we send STARTUP IPIs ?
877 	 *
878 	 * Determine this based on the APIC version.
879 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
880 	 */
881 	if (APIC_INTEGRATED(boot_cpu_apic_version))
882 		num_starts = 2;
883 	else
884 		num_starts = 0;
885 
886 	/*
887 	 * Run STARTUP IPI loop.
888 	 */
889 	pr_debug("#startup loops: %d\n", num_starts);
890 
891 	for (j = 1; j <= num_starts; j++) {
892 		pr_debug("Sending STARTUP #%d\n", j);
893 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
894 			apic_write(APIC_ESR, 0);
895 		apic_read(APIC_ESR);
896 		pr_debug("After apic_write\n");
897 
898 		/*
899 		 * STARTUP IPI
900 		 */
901 
902 		/* Target chip */
903 		/* Boot on the stack */
904 		/* Kick the second */
905 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
906 			       phys_apicid);
907 
908 		/*
909 		 * Give the other CPU some time to accept the IPI.
910 		 */
911 		if (init_udelay == 0)
912 			udelay(10);
913 		else
914 			udelay(300);
915 
916 		pr_debug("Startup point 1\n");
917 
918 		pr_debug("Waiting for send to finish...\n");
919 		send_status = safe_apic_wait_icr_idle();
920 
921 		/*
922 		 * Give the other CPU some time to accept the IPI.
923 		 */
924 		if (init_udelay == 0)
925 			udelay(10);
926 		else
927 			udelay(200);
928 
929 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
930 			apic_write(APIC_ESR, 0);
931 		accept_status = (apic_read(APIC_ESR) & 0xEF);
932 		if (send_status || accept_status)
933 			break;
934 	}
935 	pr_debug("After Startup\n");
936 
937 	if (send_status)
938 		pr_err("APIC never delivered???\n");
939 	if (accept_status)
940 		pr_err("APIC delivery error (%lx)\n", accept_status);
941 
942 	return (send_status | accept_status);
943 }
944 
945 /* reduce the number of lines printed when booting a large cpu count system */
946 static void announce_cpu(int cpu, int apicid)
947 {
948 	static int current_node = NUMA_NO_NODE;
949 	int node = early_cpu_to_node(cpu);
950 	static int width, node_width;
951 
952 	if (!width)
953 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
954 
955 	if (!node_width)
956 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
957 
958 	if (cpu == 1)
959 		printk(KERN_INFO "x86: Booting SMP configuration:\n");
960 
961 	if (system_state < SYSTEM_RUNNING) {
962 		if (node != current_node) {
963 			if (current_node > (-1))
964 				pr_cont("\n");
965 			current_node = node;
966 
967 			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
968 			       node_width - num_digits(node), " ", node);
969 		}
970 
971 		/* Add padding for the BSP */
972 		if (cpu == 1)
973 			pr_cont("%*s", width + 1, " ");
974 
975 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
976 
977 	} else
978 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
979 			node, cpu, apicid);
980 }
981 
982 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
983 {
984 	int cpu;
985 
986 	cpu = smp_processor_id();
987 	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
988 		return NMI_HANDLED;
989 
990 	return NMI_DONE;
991 }
992 
993 /*
994  * Wake up AP by INIT, INIT, STARTUP sequence.
995  *
996  * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
997  * boot-strap code which is not a desired behavior for waking up BSP. To
998  * void the boot-strap code, wake up CPU0 by NMI instead.
999  *
1000  * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
1001  * (i.e. physically hot removed and then hot added), NMI won't wake it up.
1002  * We'll change this code in the future to wake up hard offlined CPU0 if
1003  * real platform and request are available.
1004  */
1005 static int
1006 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
1007 	       int *cpu0_nmi_registered)
1008 {
1009 	int id;
1010 	int boot_error;
1011 
1012 	preempt_disable();
1013 
1014 	/*
1015 	 * Wake up AP by INIT, INIT, STARTUP sequence.
1016 	 */
1017 	if (cpu) {
1018 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
1019 		goto out;
1020 	}
1021 
1022 	/*
1023 	 * Wake up BSP by nmi.
1024 	 *
1025 	 * Register a NMI handler to help wake up CPU0.
1026 	 */
1027 	boot_error = register_nmi_handler(NMI_LOCAL,
1028 					  wakeup_cpu0_nmi, 0, "wake_cpu0");
1029 
1030 	if (!boot_error) {
1031 		enable_start_cpu0 = 1;
1032 		*cpu0_nmi_registered = 1;
1033 		id = apic->dest_mode_logical ? cpu0_logical_apicid : apicid;
1034 		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
1035 	}
1036 
1037 out:
1038 	preempt_enable();
1039 
1040 	return boot_error;
1041 }
1042 
1043 int common_cpu_up(unsigned int cpu, struct task_struct *idle)
1044 {
1045 	int ret;
1046 
1047 	/* Just in case we booted with a single CPU. */
1048 	alternatives_enable_smp();
1049 
1050 	per_cpu(current_task, cpu) = idle;
1051 	cpu_init_stack_canary(cpu, idle);
1052 
1053 	/* Initialize the interrupt stack(s) */
1054 	ret = irq_init_percpu_irqstack(cpu);
1055 	if (ret)
1056 		return ret;
1057 
1058 #ifdef CONFIG_X86_32
1059 	/* Stack for startup_32 can be just as for start_secondary onwards */
1060 	per_cpu(cpu_current_top_of_stack, cpu) = task_top_of_stack(idle);
1061 #else
1062 	initial_gs = per_cpu_offset(cpu);
1063 #endif
1064 	return 0;
1065 }
1066 
1067 /*
1068  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
1069  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
1070  * Returns zero if CPU booted OK, else error code from
1071  * ->wakeup_secondary_cpu.
1072  */
1073 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle,
1074 		       int *cpu0_nmi_registered)
1075 {
1076 	/* start_ip had better be page-aligned! */
1077 	unsigned long start_ip = real_mode_header->trampoline_start;
1078 
1079 	unsigned long boot_error = 0;
1080 	unsigned long timeout;
1081 
1082 #ifdef CONFIG_X86_64
1083 	/* If 64-bit wakeup method exists, use the 64-bit mode trampoline IP */
1084 	if (apic->wakeup_secondary_cpu_64)
1085 		start_ip = real_mode_header->trampoline_start64;
1086 #endif
1087 	idle->thread.sp = (unsigned long)task_pt_regs(idle);
1088 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_rw(cpu);
1089 	initial_code = (unsigned long)start_secondary;
1090 	initial_stack  = idle->thread.sp;
1091 
1092 	/* Enable the espfix hack for this CPU */
1093 	init_espfix_ap(cpu);
1094 
1095 	/* So we see what's up */
1096 	announce_cpu(cpu, apicid);
1097 
1098 	/*
1099 	 * This grunge runs the startup process for
1100 	 * the targeted processor.
1101 	 */
1102 
1103 	if (x86_platform.legacy.warm_reset) {
1104 
1105 		pr_debug("Setting warm reset code and vector.\n");
1106 
1107 		smpboot_setup_warm_reset_vector(start_ip);
1108 		/*
1109 		 * Be paranoid about clearing APIC errors.
1110 		*/
1111 		if (APIC_INTEGRATED(boot_cpu_apic_version)) {
1112 			apic_write(APIC_ESR, 0);
1113 			apic_read(APIC_ESR);
1114 		}
1115 	}
1116 
1117 	/*
1118 	 * AP might wait on cpu_callout_mask in cpu_init() with
1119 	 * cpu_initialized_mask set if previous attempt to online
1120 	 * it timed-out. Clear cpu_initialized_mask so that after
1121 	 * INIT/SIPI it could start with a clean state.
1122 	 */
1123 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1124 	smp_mb();
1125 
1126 	/*
1127 	 * Wake up a CPU in difference cases:
1128 	 * - Use a method from the APIC driver if one defined, with wakeup
1129 	 *   straight to 64-bit mode preferred over wakeup to RM.
1130 	 * Otherwise,
1131 	 * - Use an INIT boot APIC message for APs or NMI for BSP.
1132 	 */
1133 	if (apic->wakeup_secondary_cpu_64)
1134 		boot_error = apic->wakeup_secondary_cpu_64(apicid, start_ip);
1135 	else if (apic->wakeup_secondary_cpu)
1136 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
1137 	else
1138 		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
1139 						     cpu0_nmi_registered);
1140 
1141 	if (!boot_error) {
1142 		/*
1143 		 * Wait 10s total for first sign of life from AP
1144 		 */
1145 		boot_error = -1;
1146 		timeout = jiffies + 10*HZ;
1147 		while (time_before(jiffies, timeout)) {
1148 			if (cpumask_test_cpu(cpu, cpu_initialized_mask)) {
1149 				/*
1150 				 * Tell AP to proceed with initialization
1151 				 */
1152 				cpumask_set_cpu(cpu, cpu_callout_mask);
1153 				boot_error = 0;
1154 				break;
1155 			}
1156 			schedule();
1157 		}
1158 	}
1159 
1160 	if (!boot_error) {
1161 		/*
1162 		 * Wait till AP completes initial initialization
1163 		 */
1164 		while (!cpumask_test_cpu(cpu, cpu_callin_mask)) {
1165 			/*
1166 			 * Allow other tasks to run while we wait for the
1167 			 * AP to come online. This also gives a chance
1168 			 * for the MTRR work(triggered by the AP coming online)
1169 			 * to be completed in the stop machine context.
1170 			 */
1171 			schedule();
1172 		}
1173 	}
1174 
1175 	if (x86_platform.legacy.warm_reset) {
1176 		/*
1177 		 * Cleanup possible dangling ends...
1178 		 */
1179 		smpboot_restore_warm_reset_vector();
1180 	}
1181 
1182 	return boot_error;
1183 }
1184 
1185 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
1186 {
1187 	int apicid = apic->cpu_present_to_apicid(cpu);
1188 	int cpu0_nmi_registered = 0;
1189 	unsigned long flags;
1190 	int err, ret = 0;
1191 
1192 	lockdep_assert_irqs_enabled();
1193 
1194 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
1195 
1196 	if (apicid == BAD_APICID ||
1197 	    !physid_isset(apicid, phys_cpu_present_map) ||
1198 	    !apic->apic_id_valid(apicid)) {
1199 		pr_err("%s: bad cpu %d\n", __func__, cpu);
1200 		return -EINVAL;
1201 	}
1202 
1203 	/*
1204 	 * Already booted CPU?
1205 	 */
1206 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
1207 		pr_debug("do_boot_cpu %d Already started\n", cpu);
1208 		return -ENOSYS;
1209 	}
1210 
1211 	/*
1212 	 * Save current MTRR state in case it was changed since early boot
1213 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
1214 	 */
1215 	mtrr_save_state();
1216 
1217 	/* x86 CPUs take themselves offline, so delayed offline is OK. */
1218 	err = cpu_check_up_prepare(cpu);
1219 	if (err && err != -EBUSY)
1220 		return err;
1221 
1222 	/* the FPU context is blank, nobody can own it */
1223 	per_cpu(fpu_fpregs_owner_ctx, cpu) = NULL;
1224 
1225 	err = common_cpu_up(cpu, tidle);
1226 	if (err)
1227 		return err;
1228 
1229 	err = do_boot_cpu(apicid, cpu, tidle, &cpu0_nmi_registered);
1230 	if (err) {
1231 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
1232 		ret = -EIO;
1233 		goto unreg_nmi;
1234 	}
1235 
1236 	/*
1237 	 * Check TSC synchronization with the AP (keep irqs disabled
1238 	 * while doing so):
1239 	 */
1240 	local_irq_save(flags);
1241 	check_tsc_sync_source(cpu);
1242 	local_irq_restore(flags);
1243 
1244 	while (!cpu_online(cpu)) {
1245 		cpu_relax();
1246 		touch_nmi_watchdog();
1247 	}
1248 
1249 unreg_nmi:
1250 	/*
1251 	 * Clean up the nmi handler. Do this after the callin and callout sync
1252 	 * to avoid impact of possible long unregister time.
1253 	 */
1254 	if (cpu0_nmi_registered)
1255 		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
1256 
1257 	return ret;
1258 }
1259 
1260 /**
1261  * arch_disable_smp_support() - disables SMP support for x86 at runtime
1262  */
1263 void arch_disable_smp_support(void)
1264 {
1265 	disable_ioapic_support();
1266 }
1267 
1268 /*
1269  * Fall back to non SMP mode after errors.
1270  *
1271  * RED-PEN audit/test this more. I bet there is more state messed up here.
1272  */
1273 static __init void disable_smp(void)
1274 {
1275 	pr_info("SMP disabled\n");
1276 
1277 	disable_ioapic_support();
1278 
1279 	init_cpu_present(cpumask_of(0));
1280 	init_cpu_possible(cpumask_of(0));
1281 
1282 	if (smp_found_config)
1283 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1284 	else
1285 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1286 	cpumask_set_cpu(0, topology_sibling_cpumask(0));
1287 	cpumask_set_cpu(0, topology_core_cpumask(0));
1288 	cpumask_set_cpu(0, topology_die_cpumask(0));
1289 }
1290 
1291 /*
1292  * Various sanity checks.
1293  */
1294 static void __init smp_sanity_check(void)
1295 {
1296 	preempt_disable();
1297 
1298 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
1299 	if (def_to_bigsmp && nr_cpu_ids > 8) {
1300 		unsigned int cpu;
1301 		unsigned nr;
1302 
1303 		pr_warn("More than 8 CPUs detected - skipping them\n"
1304 			"Use CONFIG_X86_BIGSMP\n");
1305 
1306 		nr = 0;
1307 		for_each_present_cpu(cpu) {
1308 			if (nr >= 8)
1309 				set_cpu_present(cpu, false);
1310 			nr++;
1311 		}
1312 
1313 		nr = 0;
1314 		for_each_possible_cpu(cpu) {
1315 			if (nr >= 8)
1316 				set_cpu_possible(cpu, false);
1317 			nr++;
1318 		}
1319 
1320 		set_nr_cpu_ids(8);
1321 	}
1322 #endif
1323 
1324 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1325 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1326 			hard_smp_processor_id());
1327 
1328 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1329 	}
1330 
1331 	/*
1332 	 * Should not be necessary because the MP table should list the boot
1333 	 * CPU too, but we do it for the sake of robustness anyway.
1334 	 */
1335 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1336 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1337 			  boot_cpu_physical_apicid);
1338 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1339 	}
1340 	preempt_enable();
1341 }
1342 
1343 static void __init smp_cpu_index_default(void)
1344 {
1345 	int i;
1346 	struct cpuinfo_x86 *c;
1347 
1348 	for_each_possible_cpu(i) {
1349 		c = &cpu_data(i);
1350 		/* mark all to hotplug */
1351 		c->cpu_index = nr_cpu_ids;
1352 	}
1353 }
1354 
1355 static void __init smp_get_logical_apicid(void)
1356 {
1357 	if (x2apic_mode)
1358 		cpu0_logical_apicid = apic_read(APIC_LDR);
1359 	else
1360 		cpu0_logical_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1361 }
1362 
1363 void __init smp_prepare_cpus_common(void)
1364 {
1365 	unsigned int i;
1366 
1367 	smp_cpu_index_default();
1368 
1369 	/*
1370 	 * Setup boot CPU information
1371 	 */
1372 	smp_store_boot_cpu_info(); /* Final full version of the data */
1373 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1374 	mb();
1375 
1376 	for_each_possible_cpu(i) {
1377 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1378 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1379 		zalloc_cpumask_var(&per_cpu(cpu_die_map, i), GFP_KERNEL);
1380 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1381 		zalloc_cpumask_var(&per_cpu(cpu_l2c_shared_map, i), GFP_KERNEL);
1382 	}
1383 
1384 	/*
1385 	 * Set 'default' x86 topology, this matches default_topology() in that
1386 	 * it has NUMA nodes as a topology level. See also
1387 	 * native_smp_cpus_done().
1388 	 *
1389 	 * Must be done before set_cpus_sibling_map() is ran.
1390 	 */
1391 	set_sched_topology(x86_topology);
1392 
1393 	set_cpu_sibling_map(0);
1394 }
1395 
1396 /*
1397  * Prepare for SMP bootup.
1398  * @max_cpus: configured maximum number of CPUs, It is a legacy parameter
1399  *            for common interface support.
1400  */
1401 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1402 {
1403 	smp_prepare_cpus_common();
1404 
1405 	smp_sanity_check();
1406 
1407 	switch (apic_intr_mode) {
1408 	case APIC_PIC:
1409 	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1410 		disable_smp();
1411 		return;
1412 	case APIC_SYMMETRIC_IO_NO_ROUTING:
1413 		disable_smp();
1414 		/* Setup local timer */
1415 		x86_init.timers.setup_percpu_clockev();
1416 		return;
1417 	case APIC_VIRTUAL_WIRE:
1418 	case APIC_SYMMETRIC_IO:
1419 		break;
1420 	}
1421 
1422 	/* Setup local timer */
1423 	x86_init.timers.setup_percpu_clockev();
1424 
1425 	smp_get_logical_apicid();
1426 
1427 	pr_info("CPU0: ");
1428 	print_cpu_info(&cpu_data(0));
1429 
1430 	uv_system_init();
1431 
1432 	set_mtrr_aps_delayed_init();
1433 
1434 	smp_quirk_init_udelay();
1435 
1436 	speculative_store_bypass_ht_init();
1437 
1438 	snp_set_wakeup_secondary_cpu();
1439 }
1440 
1441 void arch_thaw_secondary_cpus_begin(void)
1442 {
1443 	set_mtrr_aps_delayed_init();
1444 }
1445 
1446 void arch_thaw_secondary_cpus_end(void)
1447 {
1448 	mtrr_aps_init();
1449 }
1450 
1451 /*
1452  * Early setup to make printk work.
1453  */
1454 void __init native_smp_prepare_boot_cpu(void)
1455 {
1456 	int me = smp_processor_id();
1457 	switch_to_new_gdt(me);
1458 	/* already set me in cpu_online_mask in boot_cpu_init() */
1459 	cpumask_set_cpu(me, cpu_callout_mask);
1460 	cpu_set_state_online(me);
1461 	native_pv_lock_init();
1462 }
1463 
1464 void __init calculate_max_logical_packages(void)
1465 {
1466 	int ncpus;
1467 
1468 	/*
1469 	 * Today neither Intel nor AMD support heterogeneous systems so
1470 	 * extrapolate the boot cpu's data to all packages.
1471 	 */
1472 	ncpus = cpu_data(0).booted_cores * topology_max_smt_threads();
1473 	__max_logical_packages = DIV_ROUND_UP(total_cpus, ncpus);
1474 	pr_info("Max logical packages: %u\n", __max_logical_packages);
1475 }
1476 
1477 void __init native_smp_cpus_done(unsigned int max_cpus)
1478 {
1479 	pr_debug("Boot done\n");
1480 
1481 	calculate_max_logical_packages();
1482 
1483 	/* XXX for now assume numa-in-package and hybrid don't overlap */
1484 	if (x86_has_numa_in_package)
1485 		set_sched_topology(x86_numa_in_package_topology);
1486 	if (cpu_feature_enabled(X86_FEATURE_HYBRID_CPU))
1487 		set_sched_topology(x86_hybrid_topology);
1488 
1489 	nmi_selftest();
1490 	impress_friends();
1491 	mtrr_aps_init();
1492 }
1493 
1494 static int __initdata setup_possible_cpus = -1;
1495 static int __init _setup_possible_cpus(char *str)
1496 {
1497 	get_option(&str, &setup_possible_cpus);
1498 	return 0;
1499 }
1500 early_param("possible_cpus", _setup_possible_cpus);
1501 
1502 
1503 /*
1504  * cpu_possible_mask should be static, it cannot change as cpu's
1505  * are onlined, or offlined. The reason is per-cpu data-structures
1506  * are allocated by some modules at init time, and don't expect to
1507  * do this dynamically on cpu arrival/departure.
1508  * cpu_present_mask on the other hand can change dynamically.
1509  * In case when cpu_hotplug is not compiled, then we resort to current
1510  * behaviour, which is cpu_possible == cpu_present.
1511  * - Ashok Raj
1512  *
1513  * Three ways to find out the number of additional hotplug CPUs:
1514  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1515  * - The user can overwrite it with possible_cpus=NUM
1516  * - Otherwise don't reserve additional CPUs.
1517  * We do this because additional CPUs waste a lot of memory.
1518  * -AK
1519  */
1520 __init void prefill_possible_map(void)
1521 {
1522 	int i, possible;
1523 
1524 	/* No boot processor was found in mptable or ACPI MADT */
1525 	if (!num_processors) {
1526 		if (boot_cpu_has(X86_FEATURE_APIC)) {
1527 			int apicid = boot_cpu_physical_apicid;
1528 			int cpu = hard_smp_processor_id();
1529 
1530 			pr_warn("Boot CPU (id %d) not listed by BIOS\n", cpu);
1531 
1532 			/* Make sure boot cpu is enumerated */
1533 			if (apic->cpu_present_to_apicid(0) == BAD_APICID &&
1534 			    apic->apic_id_valid(apicid))
1535 				generic_processor_info(apicid, boot_cpu_apic_version);
1536 		}
1537 
1538 		if (!num_processors)
1539 			num_processors = 1;
1540 	}
1541 
1542 	i = setup_max_cpus ?: 1;
1543 	if (setup_possible_cpus == -1) {
1544 		possible = num_processors;
1545 #ifdef CONFIG_HOTPLUG_CPU
1546 		if (setup_max_cpus)
1547 			possible += disabled_cpus;
1548 #else
1549 		if (possible > i)
1550 			possible = i;
1551 #endif
1552 	} else
1553 		possible = setup_possible_cpus;
1554 
1555 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1556 
1557 	/* nr_cpu_ids could be reduced via nr_cpus= */
1558 	if (possible > nr_cpu_ids) {
1559 		pr_warn("%d Processors exceeds NR_CPUS limit of %u\n",
1560 			possible, nr_cpu_ids);
1561 		possible = nr_cpu_ids;
1562 	}
1563 
1564 #ifdef CONFIG_HOTPLUG_CPU
1565 	if (!setup_max_cpus)
1566 #endif
1567 	if (possible > i) {
1568 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1569 			possible, setup_max_cpus);
1570 		possible = i;
1571 	}
1572 
1573 	set_nr_cpu_ids(possible);
1574 
1575 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1576 		possible, max_t(int, possible - num_processors, 0));
1577 
1578 	reset_cpu_possible_mask();
1579 
1580 	for (i = 0; i < possible; i++)
1581 		set_cpu_possible(i, true);
1582 }
1583 
1584 #ifdef CONFIG_HOTPLUG_CPU
1585 
1586 /* Recompute SMT state for all CPUs on offline */
1587 static void recompute_smt_state(void)
1588 {
1589 	int max_threads, cpu;
1590 
1591 	max_threads = 0;
1592 	for_each_online_cpu (cpu) {
1593 		int threads = cpumask_weight(topology_sibling_cpumask(cpu));
1594 
1595 		if (threads > max_threads)
1596 			max_threads = threads;
1597 	}
1598 	__max_smt_threads = max_threads;
1599 }
1600 
1601 static void remove_siblinginfo(int cpu)
1602 {
1603 	int sibling;
1604 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1605 
1606 	for_each_cpu(sibling, topology_core_cpumask(cpu)) {
1607 		cpumask_clear_cpu(cpu, topology_core_cpumask(sibling));
1608 		/*/
1609 		 * last thread sibling in this cpu core going down
1610 		 */
1611 		if (cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
1612 			cpu_data(sibling).booted_cores--;
1613 	}
1614 
1615 	for_each_cpu(sibling, topology_die_cpumask(cpu))
1616 		cpumask_clear_cpu(cpu, topology_die_cpumask(sibling));
1617 
1618 	for_each_cpu(sibling, topology_sibling_cpumask(cpu)) {
1619 		cpumask_clear_cpu(cpu, topology_sibling_cpumask(sibling));
1620 		if (cpumask_weight(topology_sibling_cpumask(sibling)) == 1)
1621 			cpu_data(sibling).smt_active = false;
1622 	}
1623 
1624 	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1625 		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1626 	for_each_cpu(sibling, cpu_l2c_shared_mask(cpu))
1627 		cpumask_clear_cpu(cpu, cpu_l2c_shared_mask(sibling));
1628 	cpumask_clear(cpu_llc_shared_mask(cpu));
1629 	cpumask_clear(cpu_l2c_shared_mask(cpu));
1630 	cpumask_clear(topology_sibling_cpumask(cpu));
1631 	cpumask_clear(topology_core_cpumask(cpu));
1632 	cpumask_clear(topology_die_cpumask(cpu));
1633 	c->cpu_core_id = 0;
1634 	c->booted_cores = 0;
1635 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1636 	recompute_smt_state();
1637 }
1638 
1639 static void remove_cpu_from_maps(int cpu)
1640 {
1641 	set_cpu_online(cpu, false);
1642 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1643 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1644 	/* was set by cpu_init() */
1645 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1646 	numa_remove_cpu(cpu);
1647 }
1648 
1649 void cpu_disable_common(void)
1650 {
1651 	int cpu = smp_processor_id();
1652 
1653 	remove_siblinginfo(cpu);
1654 
1655 	/* It's now safe to remove this processor from the online map */
1656 	lock_vector_lock();
1657 	remove_cpu_from_maps(cpu);
1658 	unlock_vector_lock();
1659 	fixup_irqs();
1660 	lapic_offline();
1661 }
1662 
1663 int native_cpu_disable(void)
1664 {
1665 	int ret;
1666 
1667 	ret = lapic_can_unplug_cpu();
1668 	if (ret)
1669 		return ret;
1670 
1671 	cpu_disable_common();
1672 
1673         /*
1674          * Disable the local APIC. Otherwise IPI broadcasts will reach
1675          * it. It still responds normally to INIT, NMI, SMI, and SIPI
1676          * messages.
1677          *
1678          * Disabling the APIC must happen after cpu_disable_common()
1679          * which invokes fixup_irqs().
1680          *
1681          * Disabling the APIC preserves already set bits in IRR, but
1682          * an interrupt arriving after disabling the local APIC does not
1683          * set the corresponding IRR bit.
1684          *
1685          * fixup_irqs() scans IRR for set bits so it can raise a not
1686          * yet handled interrupt on the new destination CPU via an IPI
1687          * but obviously it can't do so for IRR bits which are not set.
1688          * IOW, interrupts arriving after disabling the local APIC will
1689          * be lost.
1690          */
1691 	apic_soft_disable();
1692 
1693 	return 0;
1694 }
1695 
1696 int common_cpu_die(unsigned int cpu)
1697 {
1698 	int ret = 0;
1699 
1700 	/* We don't do anything here: idle task is faking death itself. */
1701 
1702 	/* They ack this in play_dead() by setting CPU_DEAD */
1703 	if (cpu_wait_death(cpu, 5)) {
1704 		if (system_state == SYSTEM_RUNNING)
1705 			pr_info("CPU %u is now offline\n", cpu);
1706 	} else {
1707 		pr_err("CPU %u didn't die...\n", cpu);
1708 		ret = -1;
1709 	}
1710 
1711 	return ret;
1712 }
1713 
1714 void native_cpu_die(unsigned int cpu)
1715 {
1716 	common_cpu_die(cpu);
1717 }
1718 
1719 void play_dead_common(void)
1720 {
1721 	idle_task_exit();
1722 
1723 	/* Ack it */
1724 	(void)cpu_report_death();
1725 
1726 	/*
1727 	 * With physical CPU hotplug, we should halt the cpu
1728 	 */
1729 	local_irq_disable();
1730 }
1731 
1732 /**
1733  * cond_wakeup_cpu0 - Wake up CPU0 if needed.
1734  *
1735  * If NMI wants to wake up CPU0, start CPU0.
1736  */
1737 void cond_wakeup_cpu0(void)
1738 {
1739 	if (smp_processor_id() == 0 && enable_start_cpu0)
1740 		start_cpu0();
1741 }
1742 EXPORT_SYMBOL_GPL(cond_wakeup_cpu0);
1743 
1744 /*
1745  * We need to flush the caches before going to sleep, lest we have
1746  * dirty data in our caches when we come back up.
1747  */
1748 static inline void mwait_play_dead(void)
1749 {
1750 	unsigned int eax, ebx, ecx, edx;
1751 	unsigned int highest_cstate = 0;
1752 	unsigned int highest_subcstate = 0;
1753 	void *mwait_ptr;
1754 	int i;
1755 
1756 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1757 	    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
1758 		return;
1759 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1760 		return;
1761 	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1762 		return;
1763 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1764 		return;
1765 
1766 	eax = CPUID_MWAIT_LEAF;
1767 	ecx = 0;
1768 	native_cpuid(&eax, &ebx, &ecx, &edx);
1769 
1770 	/*
1771 	 * eax will be 0 if EDX enumeration is not valid.
1772 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1773 	 */
1774 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1775 		eax = 0;
1776 	} else {
1777 		edx >>= MWAIT_SUBSTATE_SIZE;
1778 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1779 			if (edx & MWAIT_SUBSTATE_MASK) {
1780 				highest_cstate = i;
1781 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1782 			}
1783 		}
1784 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1785 			(highest_subcstate - 1);
1786 	}
1787 
1788 	/*
1789 	 * This should be a memory location in a cache line which is
1790 	 * unlikely to be touched by other processors.  The actual
1791 	 * content is immaterial as it is not actually modified in any way.
1792 	 */
1793 	mwait_ptr = &current_thread_info()->flags;
1794 
1795 	wbinvd();
1796 
1797 	while (1) {
1798 		/*
1799 		 * The CLFLUSH is a workaround for erratum AAI65 for
1800 		 * the Xeon 7400 series.  It's not clear it is actually
1801 		 * needed, but it should be harmless in either case.
1802 		 * The WBINVD is insufficient due to the spurious-wakeup
1803 		 * case where we return around the loop.
1804 		 */
1805 		mb();
1806 		clflush(mwait_ptr);
1807 		mb();
1808 		__monitor(mwait_ptr, 0, 0);
1809 		mb();
1810 		__mwait(eax, 0);
1811 
1812 		cond_wakeup_cpu0();
1813 	}
1814 }
1815 
1816 void hlt_play_dead(void)
1817 {
1818 	if (__this_cpu_read(cpu_info.x86) >= 4)
1819 		wbinvd();
1820 
1821 	while (1) {
1822 		native_halt();
1823 
1824 		cond_wakeup_cpu0();
1825 	}
1826 }
1827 
1828 void native_play_dead(void)
1829 {
1830 	play_dead_common();
1831 	tboot_shutdown(TB_SHUTDOWN_WFS);
1832 
1833 	mwait_play_dead();	/* Only returns on failure */
1834 	if (cpuidle_play_dead())
1835 		hlt_play_dead();
1836 }
1837 
1838 #else /* ... !CONFIG_HOTPLUG_CPU */
1839 int native_cpu_disable(void)
1840 {
1841 	return -ENOSYS;
1842 }
1843 
1844 void native_cpu_die(unsigned int cpu)
1845 {
1846 	/* We said "no" in __cpu_disable */
1847 	BUG();
1848 }
1849 
1850 void native_play_dead(void)
1851 {
1852 	BUG();
1853 }
1854 
1855 #endif
1856