xref: /openbmc/linux/arch/x86/kernel/smpboot.c (revision a72b9869)
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/cacheinfo.h>
63 #include <asm/desc.h>
64 #include <asm/nmi.h>
65 #include <asm/irq.h>
66 #include <asm/realmode.h>
67 #include <asm/cpu.h>
68 #include <asm/numa.h>
69 #include <asm/tlbflush.h>
70 #include <asm/mtrr.h>
71 #include <asm/mwait.h>
72 #include <asm/apic.h>
73 #include <asm/io_apic.h>
74 #include <asm/fpu/api.h>
75 #include <asm/setup.h>
76 #include <asm/uv/uv.h>
77 #include <linux/mc146818rtc.h>
78 #include <asm/i8259.h>
79 #include <asm/misc.h>
80 #include <asm/qspinlock.h>
81 #include <asm/intel-family.h>
82 #include <asm/cpu_device_id.h>
83 #include <asm/spec-ctrl.h>
84 #include <asm/hw_irq.h>
85 #include <asm/stackprotector.h>
86 #include <asm/sev.h>
87 
88 /* representing HT siblings of each logical CPU */
89 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
90 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
91 
92 /* representing HT and core siblings of each logical CPU */
93 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_map);
94 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
95 
96 /* representing HT, core, and die siblings of each logical CPU */
97 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_die_map);
98 EXPORT_PER_CPU_SYMBOL(cpu_die_map);
99 
100 /* Per CPU bogomips and other parameters */
101 DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
102 EXPORT_PER_CPU_SYMBOL(cpu_info);
103 
104 /* Logical package management. We might want to allocate that dynamically */
105 unsigned int __max_logical_packages __read_mostly;
106 EXPORT_SYMBOL(__max_logical_packages);
107 static unsigned int logical_packages __read_mostly;
108 static unsigned int logical_die __read_mostly;
109 
110 /* Maximum number of SMT threads on any online core */
111 int __read_mostly __max_smt_threads = 1;
112 
113 /* Flag to indicate if a complete sched domain rebuild is required */
114 bool x86_topology_update;
115 
116 int arch_update_cpu_topology(void)
117 {
118 	int retval = x86_topology_update;
119 
120 	x86_topology_update = false;
121 	return retval;
122 }
123 
124 static inline void smpboot_setup_warm_reset_vector(unsigned long start_eip)
125 {
126 	unsigned long flags;
127 
128 	spin_lock_irqsave(&rtc_lock, flags);
129 	CMOS_WRITE(0xa, 0xf);
130 	spin_unlock_irqrestore(&rtc_lock, flags);
131 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_HIGH)) =
132 							start_eip >> 4;
133 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) =
134 							start_eip & 0xf;
135 }
136 
137 static inline void smpboot_restore_warm_reset_vector(void)
138 {
139 	unsigned long flags;
140 
141 	/*
142 	 * Paranoid:  Set warm reset code and vector here back
143 	 * to default values.
144 	 */
145 	spin_lock_irqsave(&rtc_lock, flags);
146 	CMOS_WRITE(0, 0xf);
147 	spin_unlock_irqrestore(&rtc_lock, flags);
148 
149 	*((volatile u32 *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = 0;
150 }
151 
152 /*
153  * Report back to the Boot Processor during boot time or to the caller processor
154  * during CPU online.
155  */
156 static void smp_callin(void)
157 {
158 	int cpuid;
159 
160 	/*
161 	 * If waken up by an INIT in an 82489DX configuration
162 	 * cpu_callout_mask guarantees we don't get here before
163 	 * an INIT_deassert IPI reaches our local APIC, so it is
164 	 * now safe to touch our local APIC.
165 	 */
166 	cpuid = smp_processor_id();
167 
168 	/*
169 	 * the boot CPU has finished the init stage and is spinning
170 	 * on callin_map until we finish. We are free to set up this
171 	 * CPU, first the APIC. (this is probably redundant on most
172 	 * boards)
173 	 */
174 	apic_ap_setup();
175 
176 	/*
177 	 * Save our processor parameters. Note: this information
178 	 * is needed for clock calibration.
179 	 */
180 	smp_store_cpu_info(cpuid);
181 
182 	/*
183 	 * The topology information must be up to date before
184 	 * calibrate_delay() and notify_cpu_starting().
185 	 */
186 	set_cpu_sibling_map(raw_smp_processor_id());
187 
188 	ap_init_aperfmperf();
189 
190 	/*
191 	 * Get our bogomips.
192 	 * Update loops_per_jiffy in cpu_data. Previous call to
193 	 * smp_store_cpu_info() stored a value that is close but not as
194 	 * accurate as the value just calculated.
195 	 */
196 	calibrate_delay();
197 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
198 	pr_debug("Stack at about %p\n", &cpuid);
199 
200 	wmb();
201 
202 	notify_cpu_starting(cpuid);
203 
204 	/*
205 	 * Allow the master to continue.
206 	 */
207 	cpumask_set_cpu(cpuid, cpu_callin_mask);
208 }
209 
210 static int cpu0_logical_apicid;
211 static int enable_start_cpu0;
212 /*
213  * Activate a secondary processor.
214  */
215 static void notrace start_secondary(void *unused)
216 {
217 	/*
218 	 * Don't put *anything* except direct CPU state initialization
219 	 * before cpu_init(), SMP booting is too fragile that we want to
220 	 * limit the things done here to the most necessary things.
221 	 */
222 	cr4_init();
223 
224 #ifdef CONFIG_X86_32
225 	/* switch away from the initial page table */
226 	load_cr3(swapper_pg_dir);
227 	__flush_tlb_all();
228 #endif
229 	cpu_init_secondary();
230 	rcu_cpu_starting(raw_smp_processor_id());
231 	x86_cpuinit.early_percpu_clock_init();
232 	smp_callin();
233 
234 	enable_start_cpu0 = 0;
235 
236 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
237 	barrier();
238 	/*
239 	 * Check TSC synchronization with the boot CPU:
240 	 */
241 	check_tsc_sync_target();
242 
243 	speculative_store_bypass_ht_init();
244 
245 	/*
246 	 * Lock vector_lock, set CPU online and bring the vector
247 	 * allocator online. Online must be set with vector_lock held
248 	 * to prevent a concurrent irq setup/teardown from seeing a
249 	 * half valid vector space.
250 	 */
251 	lock_vector_lock();
252 	set_cpu_online(smp_processor_id(), true);
253 	lapic_online();
254 	unlock_vector_lock();
255 	cpu_set_state_online(smp_processor_id());
256 	x86_platform.nmi_init();
257 
258 	/* enable local interrupts */
259 	local_irq_enable();
260 
261 	x86_cpuinit.setup_percpu_clockev();
262 
263 	wmb();
264 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
265 }
266 
267 /**
268  * topology_is_primary_thread - Check whether CPU is the primary SMT thread
269  * @cpu:	CPU to check
270  */
271 bool topology_is_primary_thread(unsigned int cpu)
272 {
273 	return apic_id_is_primary_thread(per_cpu(x86_cpu_to_apicid, cpu));
274 }
275 
276 /**
277  * topology_smt_supported - Check whether SMT is supported by the CPUs
278  */
279 bool topology_smt_supported(void)
280 {
281 	return smp_num_siblings > 1;
282 }
283 
284 /**
285  * topology_phys_to_logical_pkg - Map a physical package id to a logical
286  *
287  * Returns logical package id or -1 if not found
288  */
289 int topology_phys_to_logical_pkg(unsigned int phys_pkg)
290 {
291 	int cpu;
292 
293 	for_each_possible_cpu(cpu) {
294 		struct cpuinfo_x86 *c = &cpu_data(cpu);
295 
296 		if (c->initialized && c->phys_proc_id == phys_pkg)
297 			return c->logical_proc_id;
298 	}
299 	return -1;
300 }
301 EXPORT_SYMBOL(topology_phys_to_logical_pkg);
302 /**
303  * topology_phys_to_logical_die - Map a physical die id to logical
304  *
305  * Returns logical die id or -1 if not found
306  */
307 int topology_phys_to_logical_die(unsigned int die_id, unsigned int cur_cpu)
308 {
309 	int cpu;
310 	int proc_id = cpu_data(cur_cpu).phys_proc_id;
311 
312 	for_each_possible_cpu(cpu) {
313 		struct cpuinfo_x86 *c = &cpu_data(cpu);
314 
315 		if (c->initialized && c->cpu_die_id == die_id &&
316 		    c->phys_proc_id == proc_id)
317 			return c->logical_die_id;
318 	}
319 	return -1;
320 }
321 EXPORT_SYMBOL(topology_phys_to_logical_die);
322 
323 /**
324  * topology_update_package_map - Update the physical to logical package map
325  * @pkg:	The physical package id as retrieved via CPUID
326  * @cpu:	The cpu for which this is updated
327  */
328 int topology_update_package_map(unsigned int pkg, unsigned int cpu)
329 {
330 	int new;
331 
332 	/* Already available somewhere? */
333 	new = topology_phys_to_logical_pkg(pkg);
334 	if (new >= 0)
335 		goto found;
336 
337 	new = logical_packages++;
338 	if (new != pkg) {
339 		pr_info("CPU %u Converting physical %u to logical package %u\n",
340 			cpu, pkg, new);
341 	}
342 found:
343 	cpu_data(cpu).logical_proc_id = new;
344 	return 0;
345 }
346 /**
347  * topology_update_die_map - Update the physical to logical die map
348  * @die:	The die id as retrieved via CPUID
349  * @cpu:	The cpu for which this is updated
350  */
351 int topology_update_die_map(unsigned int die, unsigned int cpu)
352 {
353 	int new;
354 
355 	/* Already available somewhere? */
356 	new = topology_phys_to_logical_die(die, cpu);
357 	if (new >= 0)
358 		goto found;
359 
360 	new = logical_die++;
361 	if (new != die) {
362 		pr_info("CPU %u Converting physical %u to logical die %u\n",
363 			cpu, die, new);
364 	}
365 found:
366 	cpu_data(cpu).logical_die_id = new;
367 	return 0;
368 }
369 
370 void __init smp_store_boot_cpu_info(void)
371 {
372 	int id = 0; /* CPU 0 */
373 	struct cpuinfo_x86 *c = &cpu_data(id);
374 
375 	*c = boot_cpu_data;
376 	c->cpu_index = id;
377 	topology_update_package_map(c->phys_proc_id, id);
378 	topology_update_die_map(c->cpu_die_id, id);
379 	c->initialized = true;
380 }
381 
382 /*
383  * The bootstrap kernel entry code has set these up. Save them for
384  * a given CPU
385  */
386 void smp_store_cpu_info(int id)
387 {
388 	struct cpuinfo_x86 *c = &cpu_data(id);
389 
390 	/* Copy boot_cpu_data only on the first bringup */
391 	if (!c->initialized)
392 		*c = boot_cpu_data;
393 	c->cpu_index = id;
394 	/*
395 	 * During boot time, CPU0 has this setup already. Save the info when
396 	 * bringing up AP or offlined CPU0.
397 	 */
398 	identify_secondary_cpu(c);
399 	c->initialized = true;
400 }
401 
402 static bool
403 topology_same_node(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
404 {
405 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
406 
407 	return (cpu_to_node(cpu1) == cpu_to_node(cpu2));
408 }
409 
410 static bool
411 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
412 {
413 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
414 
415 	return !WARN_ONCE(!topology_same_node(c, o),
416 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
417 		"[node: %d != %d]. Ignoring dependency.\n",
418 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
419 }
420 
421 #define link_mask(mfunc, c1, c2)					\
422 do {									\
423 	cpumask_set_cpu((c1), mfunc(c2));				\
424 	cpumask_set_cpu((c2), mfunc(c1));				\
425 } while (0)
426 
427 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
428 {
429 	if (boot_cpu_has(X86_FEATURE_TOPOEXT)) {
430 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
431 
432 		if (c->phys_proc_id == o->phys_proc_id &&
433 		    c->cpu_die_id == o->cpu_die_id &&
434 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2)) {
435 			if (c->cpu_core_id == o->cpu_core_id)
436 				return topology_sane(c, o, "smt");
437 
438 			if ((c->cu_id != 0xff) &&
439 			    (o->cu_id != 0xff) &&
440 			    (c->cu_id == o->cu_id))
441 				return topology_sane(c, o, "smt");
442 		}
443 
444 	} else if (c->phys_proc_id == o->phys_proc_id &&
445 		   c->cpu_die_id == o->cpu_die_id &&
446 		   c->cpu_core_id == o->cpu_core_id) {
447 		return topology_sane(c, o, "smt");
448 	}
449 
450 	return false;
451 }
452 
453 static bool match_die(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
454 {
455 	if (c->phys_proc_id == o->phys_proc_id &&
456 	    c->cpu_die_id == o->cpu_die_id)
457 		return true;
458 	return false;
459 }
460 
461 static bool match_l2c(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
462 {
463 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
464 
465 	/* If the arch didn't set up l2c_id, fall back to SMT */
466 	if (per_cpu(cpu_l2c_id, cpu1) == BAD_APICID)
467 		return match_smt(c, o);
468 
469 	/* Do not match if L2 cache id does not match: */
470 	if (per_cpu(cpu_l2c_id, cpu1) != per_cpu(cpu_l2c_id, cpu2))
471 		return false;
472 
473 	return topology_sane(c, o, "l2c");
474 }
475 
476 /*
477  * Unlike the other levels, we do not enforce keeping a
478  * multicore group inside a NUMA node.  If this happens, we will
479  * discard the MC level of the topology later.
480  */
481 static bool match_pkg(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
482 {
483 	if (c->phys_proc_id == o->phys_proc_id)
484 		return true;
485 	return false;
486 }
487 
488 /*
489  * Define intel_cod_cpu[] for Intel COD (Cluster-on-Die) CPUs.
490  *
491  * Any Intel CPU that has multiple nodes per package and does not
492  * match intel_cod_cpu[] has the SNC (Sub-NUMA Cluster) topology.
493  *
494  * When in SNC mode, these CPUs enumerate an LLC that is shared
495  * by multiple NUMA nodes. The LLC is shared for off-package data
496  * access but private to the NUMA node (half of the package) for
497  * on-package access. CPUID (the source of the information about
498  * the LLC) can only enumerate the cache as shared or unshared,
499  * but not this particular configuration.
500  */
501 
502 static const struct x86_cpu_id intel_cod_cpu[] = {
503 	X86_MATCH_INTEL_FAM6_MODEL(HASWELL_X, 0),	/* COD */
504 	X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_X, 0),	/* COD */
505 	X86_MATCH_INTEL_FAM6_MODEL(ANY, 1),		/* SNC */
506 	{}
507 };
508 
509 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
510 {
511 	const struct x86_cpu_id *id = x86_match_cpu(intel_cod_cpu);
512 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
513 	bool intel_snc = id && id->driver_data;
514 
515 	/* Do not match if we do not have a valid APICID for cpu: */
516 	if (per_cpu(cpu_llc_id, cpu1) == BAD_APICID)
517 		return false;
518 
519 	/* Do not match if LLC id does not match: */
520 	if (per_cpu(cpu_llc_id, cpu1) != per_cpu(cpu_llc_id, cpu2))
521 		return false;
522 
523 	/*
524 	 * Allow the SNC topology without warning. Return of false
525 	 * means 'c' does not share the LLC of 'o'. This will be
526 	 * reflected to userspace.
527 	 */
528 	if (match_pkg(c, o) && !topology_same_node(c, o) && intel_snc)
529 		return false;
530 
531 	return topology_sane(c, o, "llc");
532 }
533 
534 
535 #if defined(CONFIG_SCHED_SMT) || defined(CONFIG_SCHED_CLUSTER) || defined(CONFIG_SCHED_MC)
536 static inline int x86_sched_itmt_flags(void)
537 {
538 	return sysctl_sched_itmt_enabled ? SD_ASYM_PACKING : 0;
539 }
540 
541 #ifdef CONFIG_SCHED_MC
542 static int x86_core_flags(void)
543 {
544 	return cpu_core_flags() | x86_sched_itmt_flags();
545 }
546 #endif
547 #ifdef CONFIG_SCHED_SMT
548 static int x86_smt_flags(void)
549 {
550 	return cpu_smt_flags() | x86_sched_itmt_flags();
551 }
552 #endif
553 #ifdef CONFIG_SCHED_CLUSTER
554 static int x86_cluster_flags(void)
555 {
556 	return cpu_cluster_flags() | x86_sched_itmt_flags();
557 }
558 #endif
559 #endif
560 
561 static struct sched_domain_topology_level x86_numa_in_package_topology[] = {
562 #ifdef CONFIG_SCHED_SMT
563 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
564 #endif
565 #ifdef CONFIG_SCHED_CLUSTER
566 	{ cpu_clustergroup_mask, x86_cluster_flags, SD_INIT_NAME(CLS) },
567 #endif
568 #ifdef CONFIG_SCHED_MC
569 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
570 #endif
571 	{ NULL, },
572 };
573 
574 static struct sched_domain_topology_level x86_hybrid_topology[] = {
575 #ifdef CONFIG_SCHED_SMT
576 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
577 #endif
578 #ifdef CONFIG_SCHED_MC
579 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
580 #endif
581 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
582 	{ NULL, },
583 };
584 
585 static struct sched_domain_topology_level x86_topology[] = {
586 #ifdef CONFIG_SCHED_SMT
587 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
588 #endif
589 #ifdef CONFIG_SCHED_CLUSTER
590 	{ cpu_clustergroup_mask, x86_cluster_flags, SD_INIT_NAME(CLS) },
591 #endif
592 #ifdef CONFIG_SCHED_MC
593 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
594 #endif
595 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
596 	{ NULL, },
597 };
598 
599 /*
600  * Set if a package/die has multiple NUMA nodes inside.
601  * AMD Magny-Cours, Intel Cluster-on-Die, and Intel
602  * Sub-NUMA Clustering have this.
603  */
604 static bool x86_has_numa_in_package;
605 
606 void set_cpu_sibling_map(int cpu)
607 {
608 	bool has_smt = smp_num_siblings > 1;
609 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
610 	struct cpuinfo_x86 *c = &cpu_data(cpu);
611 	struct cpuinfo_x86 *o;
612 	int i, threads;
613 
614 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
615 
616 	if (!has_mp) {
617 		cpumask_set_cpu(cpu, topology_sibling_cpumask(cpu));
618 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
619 		cpumask_set_cpu(cpu, cpu_l2c_shared_mask(cpu));
620 		cpumask_set_cpu(cpu, topology_core_cpumask(cpu));
621 		cpumask_set_cpu(cpu, topology_die_cpumask(cpu));
622 		c->booted_cores = 1;
623 		return;
624 	}
625 
626 	for_each_cpu(i, cpu_sibling_setup_mask) {
627 		o = &cpu_data(i);
628 
629 		if (match_pkg(c, o) && !topology_same_node(c, o))
630 			x86_has_numa_in_package = true;
631 
632 		if ((i == cpu) || (has_smt && match_smt(c, o)))
633 			link_mask(topology_sibling_cpumask, cpu, i);
634 
635 		if ((i == cpu) || (has_mp && match_llc(c, o)))
636 			link_mask(cpu_llc_shared_mask, cpu, i);
637 
638 		if ((i == cpu) || (has_mp && match_l2c(c, o)))
639 			link_mask(cpu_l2c_shared_mask, cpu, i);
640 
641 		if ((i == cpu) || (has_mp && match_die(c, o)))
642 			link_mask(topology_die_cpumask, cpu, i);
643 	}
644 
645 	threads = cpumask_weight(topology_sibling_cpumask(cpu));
646 	if (threads > __max_smt_threads)
647 		__max_smt_threads = threads;
648 
649 	for_each_cpu(i, topology_sibling_cpumask(cpu))
650 		cpu_data(i).smt_active = threads > 1;
651 
652 	/*
653 	 * This needs a separate iteration over the cpus because we rely on all
654 	 * topology_sibling_cpumask links to be set-up.
655 	 */
656 	for_each_cpu(i, cpu_sibling_setup_mask) {
657 		o = &cpu_data(i);
658 
659 		if ((i == cpu) || (has_mp && match_pkg(c, o))) {
660 			link_mask(topology_core_cpumask, cpu, i);
661 
662 			/*
663 			 *  Does this new cpu bringup a new core?
664 			 */
665 			if (threads == 1) {
666 				/*
667 				 * for each core in package, increment
668 				 * the booted_cores for this new cpu
669 				 */
670 				if (cpumask_first(
671 				    topology_sibling_cpumask(i)) == i)
672 					c->booted_cores++;
673 				/*
674 				 * increment the core count for all
675 				 * the other cpus in this package
676 				 */
677 				if (i != cpu)
678 					cpu_data(i).booted_cores++;
679 			} else if (i != cpu && !c->booted_cores)
680 				c->booted_cores = cpu_data(i).booted_cores;
681 		}
682 	}
683 }
684 
685 /* maps the cpu to the sched domain representing multi-core */
686 const struct cpumask *cpu_coregroup_mask(int cpu)
687 {
688 	return cpu_llc_shared_mask(cpu);
689 }
690 
691 const struct cpumask *cpu_clustergroup_mask(int cpu)
692 {
693 	return cpu_l2c_shared_mask(cpu);
694 }
695 
696 static void impress_friends(void)
697 {
698 	int cpu;
699 	unsigned long bogosum = 0;
700 	/*
701 	 * Allow the user to impress friends.
702 	 */
703 	pr_debug("Before bogomips\n");
704 	for_each_possible_cpu(cpu)
705 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
706 			bogosum += cpu_data(cpu).loops_per_jiffy;
707 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
708 		num_online_cpus(),
709 		bogosum/(500000/HZ),
710 		(bogosum/(5000/HZ))%100);
711 
712 	pr_debug("Before bogocount - setting activated=1\n");
713 }
714 
715 void __inquire_remote_apic(int apicid)
716 {
717 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
718 	const char * const names[] = { "ID", "VERSION", "SPIV" };
719 	int timeout;
720 	u32 status;
721 
722 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
723 
724 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
725 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
726 
727 		/*
728 		 * Wait for idle.
729 		 */
730 		status = safe_apic_wait_icr_idle();
731 		if (status)
732 			pr_cont("a previous APIC delivery may have failed\n");
733 
734 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
735 
736 		timeout = 0;
737 		do {
738 			udelay(100);
739 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
740 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
741 
742 		switch (status) {
743 		case APIC_ICR_RR_VALID:
744 			status = apic_read(APIC_RRR);
745 			pr_cont("%08x\n", status);
746 			break;
747 		default:
748 			pr_cont("failed\n");
749 		}
750 	}
751 }
752 
753 /*
754  * The Multiprocessor Specification 1.4 (1997) example code suggests
755  * that there should be a 10ms delay between the BSP asserting INIT
756  * and de-asserting INIT, when starting a remote processor.
757  * But that slows boot and resume on modern processors, which include
758  * many cores and don't require that delay.
759  *
760  * Cmdline "init_cpu_udelay=" is available to over-ride this delay.
761  * Modern processor families are quirked to remove the delay entirely.
762  */
763 #define UDELAY_10MS_DEFAULT 10000
764 
765 static unsigned int init_udelay = UINT_MAX;
766 
767 static int __init cpu_init_udelay(char *str)
768 {
769 	get_option(&str, &init_udelay);
770 
771 	return 0;
772 }
773 early_param("cpu_init_udelay", cpu_init_udelay);
774 
775 static void __init smp_quirk_init_udelay(void)
776 {
777 	/* if cmdline changed it from default, leave it alone */
778 	if (init_udelay != UINT_MAX)
779 		return;
780 
781 	/* if modern processor, use no delay */
782 	if (((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && (boot_cpu_data.x86 == 6)) ||
783 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) && (boot_cpu_data.x86 >= 0x18)) ||
784 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && (boot_cpu_data.x86 >= 0xF))) {
785 		init_udelay = 0;
786 		return;
787 	}
788 	/* else, use legacy delay */
789 	init_udelay = UDELAY_10MS_DEFAULT;
790 }
791 
792 /*
793  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
794  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
795  * won't ... remember to clear down the APIC, etc later.
796  */
797 int
798 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
799 {
800 	u32 dm = apic->dest_mode_logical ? APIC_DEST_LOGICAL : APIC_DEST_PHYSICAL;
801 	unsigned long send_status, accept_status = 0;
802 	int maxlvt;
803 
804 	/* Target chip */
805 	/* Boot on the stack */
806 	/* Kick the second */
807 	apic_icr_write(APIC_DM_NMI | dm, apicid);
808 
809 	pr_debug("Waiting for send to finish...\n");
810 	send_status = safe_apic_wait_icr_idle();
811 
812 	/*
813 	 * Give the other CPU some time to accept the IPI.
814 	 */
815 	udelay(200);
816 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
817 		maxlvt = lapic_get_maxlvt();
818 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
819 			apic_write(APIC_ESR, 0);
820 		accept_status = (apic_read(APIC_ESR) & 0xEF);
821 	}
822 	pr_debug("NMI sent\n");
823 
824 	if (send_status)
825 		pr_err("APIC never delivered???\n");
826 	if (accept_status)
827 		pr_err("APIC delivery error (%lx)\n", accept_status);
828 
829 	return (send_status | accept_status);
830 }
831 
832 static int
833 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
834 {
835 	unsigned long send_status = 0, accept_status = 0;
836 	int maxlvt, num_starts, j;
837 
838 	maxlvt = lapic_get_maxlvt();
839 
840 	/*
841 	 * Be paranoid about clearing APIC errors.
842 	 */
843 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
844 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
845 			apic_write(APIC_ESR, 0);
846 		apic_read(APIC_ESR);
847 	}
848 
849 	pr_debug("Asserting INIT\n");
850 
851 	/*
852 	 * Turn INIT on target chip
853 	 */
854 	/*
855 	 * Send IPI
856 	 */
857 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
858 		       phys_apicid);
859 
860 	pr_debug("Waiting for send to finish...\n");
861 	send_status = safe_apic_wait_icr_idle();
862 
863 	udelay(init_udelay);
864 
865 	pr_debug("Deasserting INIT\n");
866 
867 	/* Target chip */
868 	/* Send IPI */
869 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
870 
871 	pr_debug("Waiting for send to finish...\n");
872 	send_status = safe_apic_wait_icr_idle();
873 
874 	mb();
875 
876 	/*
877 	 * Should we send STARTUP IPIs ?
878 	 *
879 	 * Determine this based on the APIC version.
880 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
881 	 */
882 	if (APIC_INTEGRATED(boot_cpu_apic_version))
883 		num_starts = 2;
884 	else
885 		num_starts = 0;
886 
887 	/*
888 	 * Run STARTUP IPI loop.
889 	 */
890 	pr_debug("#startup loops: %d\n", num_starts);
891 
892 	for (j = 1; j <= num_starts; j++) {
893 		pr_debug("Sending STARTUP #%d\n", j);
894 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
895 			apic_write(APIC_ESR, 0);
896 		apic_read(APIC_ESR);
897 		pr_debug("After apic_write\n");
898 
899 		/*
900 		 * STARTUP IPI
901 		 */
902 
903 		/* Target chip */
904 		/* Boot on the stack */
905 		/* Kick the second */
906 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
907 			       phys_apicid);
908 
909 		/*
910 		 * Give the other CPU some time to accept the IPI.
911 		 */
912 		if (init_udelay == 0)
913 			udelay(10);
914 		else
915 			udelay(300);
916 
917 		pr_debug("Startup point 1\n");
918 
919 		pr_debug("Waiting for send to finish...\n");
920 		send_status = safe_apic_wait_icr_idle();
921 
922 		/*
923 		 * Give the other CPU some time to accept the IPI.
924 		 */
925 		if (init_udelay == 0)
926 			udelay(10);
927 		else
928 			udelay(200);
929 
930 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
931 			apic_write(APIC_ESR, 0);
932 		accept_status = (apic_read(APIC_ESR) & 0xEF);
933 		if (send_status || accept_status)
934 			break;
935 	}
936 	pr_debug("After Startup\n");
937 
938 	if (send_status)
939 		pr_err("APIC never delivered???\n");
940 	if (accept_status)
941 		pr_err("APIC delivery error (%lx)\n", accept_status);
942 
943 	return (send_status | accept_status);
944 }
945 
946 /* reduce the number of lines printed when booting a large cpu count system */
947 static void announce_cpu(int cpu, int apicid)
948 {
949 	static int current_node = NUMA_NO_NODE;
950 	int node = early_cpu_to_node(cpu);
951 	static int width, node_width;
952 
953 	if (!width)
954 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
955 
956 	if (!node_width)
957 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
958 
959 	if (cpu == 1)
960 		printk(KERN_INFO "x86: Booting SMP configuration:\n");
961 
962 	if (system_state < SYSTEM_RUNNING) {
963 		if (node != current_node) {
964 			if (current_node > (-1))
965 				pr_cont("\n");
966 			current_node = node;
967 
968 			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
969 			       node_width - num_digits(node), " ", node);
970 		}
971 
972 		/* Add padding for the BSP */
973 		if (cpu == 1)
974 			pr_cont("%*s", width + 1, " ");
975 
976 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
977 
978 	} else
979 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
980 			node, cpu, apicid);
981 }
982 
983 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
984 {
985 	int cpu;
986 
987 	cpu = smp_processor_id();
988 	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
989 		return NMI_HANDLED;
990 
991 	return NMI_DONE;
992 }
993 
994 /*
995  * Wake up AP by INIT, INIT, STARTUP sequence.
996  *
997  * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
998  * boot-strap code which is not a desired behavior for waking up BSP. To
999  * void the boot-strap code, wake up CPU0 by NMI instead.
1000  *
1001  * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
1002  * (i.e. physically hot removed and then hot added), NMI won't wake it up.
1003  * We'll change this code in the future to wake up hard offlined CPU0 if
1004  * real platform and request are available.
1005  */
1006 static int
1007 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
1008 	       int *cpu0_nmi_registered)
1009 {
1010 	int id;
1011 	int boot_error;
1012 
1013 	preempt_disable();
1014 
1015 	/*
1016 	 * Wake up AP by INIT, INIT, STARTUP sequence.
1017 	 */
1018 	if (cpu) {
1019 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
1020 		goto out;
1021 	}
1022 
1023 	/*
1024 	 * Wake up BSP by nmi.
1025 	 *
1026 	 * Register a NMI handler to help wake up CPU0.
1027 	 */
1028 	boot_error = register_nmi_handler(NMI_LOCAL,
1029 					  wakeup_cpu0_nmi, 0, "wake_cpu0");
1030 
1031 	if (!boot_error) {
1032 		enable_start_cpu0 = 1;
1033 		*cpu0_nmi_registered = 1;
1034 		id = apic->dest_mode_logical ? cpu0_logical_apicid : apicid;
1035 		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
1036 	}
1037 
1038 out:
1039 	preempt_enable();
1040 
1041 	return boot_error;
1042 }
1043 
1044 int common_cpu_up(unsigned int cpu, struct task_struct *idle)
1045 {
1046 	int ret;
1047 
1048 	/* Just in case we booted with a single CPU. */
1049 	alternatives_enable_smp();
1050 
1051 	per_cpu(pcpu_hot.current_task, cpu) = idle;
1052 	cpu_init_stack_canary(cpu, idle);
1053 
1054 	/* Initialize the interrupt stack(s) */
1055 	ret = irq_init_percpu_irqstack(cpu);
1056 	if (ret)
1057 		return ret;
1058 
1059 #ifdef CONFIG_X86_32
1060 	/* Stack for startup_32 can be just as for start_secondary onwards */
1061 	per_cpu(pcpu_hot.top_of_stack, cpu) = task_top_of_stack(idle);
1062 #else
1063 	initial_gs = per_cpu_offset(cpu);
1064 #endif
1065 	return 0;
1066 }
1067 
1068 /*
1069  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
1070  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
1071  * Returns zero if CPU booted OK, else error code from
1072  * ->wakeup_secondary_cpu.
1073  */
1074 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle,
1075 		       int *cpu0_nmi_registered)
1076 {
1077 	/* start_ip had better be page-aligned! */
1078 	unsigned long start_ip = real_mode_header->trampoline_start;
1079 
1080 	unsigned long boot_error = 0;
1081 	unsigned long timeout;
1082 
1083 #ifdef CONFIG_X86_64
1084 	/* If 64-bit wakeup method exists, use the 64-bit mode trampoline IP */
1085 	if (apic->wakeup_secondary_cpu_64)
1086 		start_ip = real_mode_header->trampoline_start64;
1087 #endif
1088 	idle->thread.sp = (unsigned long)task_pt_regs(idle);
1089 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_rw(cpu);
1090 	initial_code = (unsigned long)start_secondary;
1091 	initial_stack  = idle->thread.sp;
1092 
1093 	/* Enable the espfix hack for this CPU */
1094 	init_espfix_ap(cpu);
1095 
1096 	/* So we see what's up */
1097 	announce_cpu(cpu, apicid);
1098 
1099 	/*
1100 	 * This grunge runs the startup process for
1101 	 * the targeted processor.
1102 	 */
1103 
1104 	if (x86_platform.legacy.warm_reset) {
1105 
1106 		pr_debug("Setting warm reset code and vector.\n");
1107 
1108 		smpboot_setup_warm_reset_vector(start_ip);
1109 		/*
1110 		 * Be paranoid about clearing APIC errors.
1111 		*/
1112 		if (APIC_INTEGRATED(boot_cpu_apic_version)) {
1113 			apic_write(APIC_ESR, 0);
1114 			apic_read(APIC_ESR);
1115 		}
1116 	}
1117 
1118 	/*
1119 	 * AP might wait on cpu_callout_mask in cpu_init() with
1120 	 * cpu_initialized_mask set if previous attempt to online
1121 	 * it timed-out. Clear cpu_initialized_mask so that after
1122 	 * INIT/SIPI it could start with a clean state.
1123 	 */
1124 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1125 	smp_mb();
1126 
1127 	/*
1128 	 * Wake up a CPU in difference cases:
1129 	 * - Use a method from the APIC driver if one defined, with wakeup
1130 	 *   straight to 64-bit mode preferred over wakeup to RM.
1131 	 * Otherwise,
1132 	 * - Use an INIT boot APIC message for APs or NMI for BSP.
1133 	 */
1134 	if (apic->wakeup_secondary_cpu_64)
1135 		boot_error = apic->wakeup_secondary_cpu_64(apicid, start_ip);
1136 	else if (apic->wakeup_secondary_cpu)
1137 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
1138 	else
1139 		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
1140 						     cpu0_nmi_registered);
1141 
1142 	if (!boot_error) {
1143 		/*
1144 		 * Wait 10s total for first sign of life from AP
1145 		 */
1146 		boot_error = -1;
1147 		timeout = jiffies + 10*HZ;
1148 		while (time_before(jiffies, timeout)) {
1149 			if (cpumask_test_cpu(cpu, cpu_initialized_mask)) {
1150 				/*
1151 				 * Tell AP to proceed with initialization
1152 				 */
1153 				cpumask_set_cpu(cpu, cpu_callout_mask);
1154 				boot_error = 0;
1155 				break;
1156 			}
1157 			schedule();
1158 		}
1159 	}
1160 
1161 	if (!boot_error) {
1162 		/*
1163 		 * Wait till AP completes initial initialization
1164 		 */
1165 		while (!cpumask_test_cpu(cpu, cpu_callin_mask)) {
1166 			/*
1167 			 * Allow other tasks to run while we wait for the
1168 			 * AP to come online. This also gives a chance
1169 			 * for the MTRR work(triggered by the AP coming online)
1170 			 * to be completed in the stop machine context.
1171 			 */
1172 			schedule();
1173 		}
1174 	}
1175 
1176 	if (x86_platform.legacy.warm_reset) {
1177 		/*
1178 		 * Cleanup possible dangling ends...
1179 		 */
1180 		smpboot_restore_warm_reset_vector();
1181 	}
1182 
1183 	return boot_error;
1184 }
1185 
1186 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
1187 {
1188 	int apicid = apic->cpu_present_to_apicid(cpu);
1189 	int cpu0_nmi_registered = 0;
1190 	unsigned long flags;
1191 	int err, ret = 0;
1192 
1193 	lockdep_assert_irqs_enabled();
1194 
1195 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
1196 
1197 	if (apicid == BAD_APICID ||
1198 	    !physid_isset(apicid, phys_cpu_present_map) ||
1199 	    !apic->apic_id_valid(apicid)) {
1200 		pr_err("%s: bad cpu %d\n", __func__, cpu);
1201 		return -EINVAL;
1202 	}
1203 
1204 	/*
1205 	 * Already booted CPU?
1206 	 */
1207 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
1208 		pr_debug("do_boot_cpu %d Already started\n", cpu);
1209 		return -ENOSYS;
1210 	}
1211 
1212 	/*
1213 	 * Save current MTRR state in case it was changed since early boot
1214 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
1215 	 */
1216 	mtrr_save_state();
1217 
1218 	/* x86 CPUs take themselves offline, so delayed offline is OK. */
1219 	err = cpu_check_up_prepare(cpu);
1220 	if (err && err != -EBUSY)
1221 		return err;
1222 
1223 	/* the FPU context is blank, nobody can own it */
1224 	per_cpu(fpu_fpregs_owner_ctx, cpu) = NULL;
1225 
1226 	err = common_cpu_up(cpu, tidle);
1227 	if (err)
1228 		return err;
1229 
1230 	err = do_boot_cpu(apicid, cpu, tidle, &cpu0_nmi_registered);
1231 	if (err) {
1232 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
1233 		ret = -EIO;
1234 		goto unreg_nmi;
1235 	}
1236 
1237 	/*
1238 	 * Check TSC synchronization with the AP (keep irqs disabled
1239 	 * while doing so):
1240 	 */
1241 	local_irq_save(flags);
1242 	check_tsc_sync_source(cpu);
1243 	local_irq_restore(flags);
1244 
1245 	while (!cpu_online(cpu)) {
1246 		cpu_relax();
1247 		touch_nmi_watchdog();
1248 	}
1249 
1250 unreg_nmi:
1251 	/*
1252 	 * Clean up the nmi handler. Do this after the callin and callout sync
1253 	 * to avoid impact of possible long unregister time.
1254 	 */
1255 	if (cpu0_nmi_registered)
1256 		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
1257 
1258 	return ret;
1259 }
1260 
1261 /**
1262  * arch_disable_smp_support() - disables SMP support for x86 at runtime
1263  */
1264 void arch_disable_smp_support(void)
1265 {
1266 	disable_ioapic_support();
1267 }
1268 
1269 /*
1270  * Fall back to non SMP mode after errors.
1271  *
1272  * RED-PEN audit/test this more. I bet there is more state messed up here.
1273  */
1274 static __init void disable_smp(void)
1275 {
1276 	pr_info("SMP disabled\n");
1277 
1278 	disable_ioapic_support();
1279 
1280 	init_cpu_present(cpumask_of(0));
1281 	init_cpu_possible(cpumask_of(0));
1282 
1283 	if (smp_found_config)
1284 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1285 	else
1286 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1287 	cpumask_set_cpu(0, topology_sibling_cpumask(0));
1288 	cpumask_set_cpu(0, topology_core_cpumask(0));
1289 	cpumask_set_cpu(0, topology_die_cpumask(0));
1290 }
1291 
1292 /*
1293  * Various sanity checks.
1294  */
1295 static void __init smp_sanity_check(void)
1296 {
1297 	preempt_disable();
1298 
1299 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
1300 	if (def_to_bigsmp && nr_cpu_ids > 8) {
1301 		unsigned int cpu;
1302 		unsigned nr;
1303 
1304 		pr_warn("More than 8 CPUs detected - skipping them\n"
1305 			"Use CONFIG_X86_BIGSMP\n");
1306 
1307 		nr = 0;
1308 		for_each_present_cpu(cpu) {
1309 			if (nr >= 8)
1310 				set_cpu_present(cpu, false);
1311 			nr++;
1312 		}
1313 
1314 		nr = 0;
1315 		for_each_possible_cpu(cpu) {
1316 			if (nr >= 8)
1317 				set_cpu_possible(cpu, false);
1318 			nr++;
1319 		}
1320 
1321 		set_nr_cpu_ids(8);
1322 	}
1323 #endif
1324 
1325 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1326 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1327 			hard_smp_processor_id());
1328 
1329 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1330 	}
1331 
1332 	/*
1333 	 * Should not be necessary because the MP table should list the boot
1334 	 * CPU too, but we do it for the sake of robustness anyway.
1335 	 */
1336 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1337 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1338 			  boot_cpu_physical_apicid);
1339 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1340 	}
1341 	preempt_enable();
1342 }
1343 
1344 static void __init smp_cpu_index_default(void)
1345 {
1346 	int i;
1347 	struct cpuinfo_x86 *c;
1348 
1349 	for_each_possible_cpu(i) {
1350 		c = &cpu_data(i);
1351 		/* mark all to hotplug */
1352 		c->cpu_index = nr_cpu_ids;
1353 	}
1354 }
1355 
1356 static void __init smp_get_logical_apicid(void)
1357 {
1358 	if (x2apic_mode)
1359 		cpu0_logical_apicid = apic_read(APIC_LDR);
1360 	else
1361 		cpu0_logical_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1362 }
1363 
1364 void __init smp_prepare_cpus_common(void)
1365 {
1366 	unsigned int i;
1367 
1368 	smp_cpu_index_default();
1369 
1370 	/*
1371 	 * Setup boot CPU information
1372 	 */
1373 	smp_store_boot_cpu_info(); /* Final full version of the data */
1374 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1375 	mb();
1376 
1377 	for_each_possible_cpu(i) {
1378 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1379 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1380 		zalloc_cpumask_var(&per_cpu(cpu_die_map, i), GFP_KERNEL);
1381 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1382 		zalloc_cpumask_var(&per_cpu(cpu_l2c_shared_map, i), GFP_KERNEL);
1383 	}
1384 
1385 	/*
1386 	 * Set 'default' x86 topology, this matches default_topology() in that
1387 	 * it has NUMA nodes as a topology level. See also
1388 	 * native_smp_cpus_done().
1389 	 *
1390 	 * Must be done before set_cpus_sibling_map() is ran.
1391 	 */
1392 	set_sched_topology(x86_topology);
1393 
1394 	set_cpu_sibling_map(0);
1395 }
1396 
1397 /*
1398  * Prepare for SMP bootup.
1399  * @max_cpus: configured maximum number of CPUs, It is a legacy parameter
1400  *            for common interface support.
1401  */
1402 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1403 {
1404 	smp_prepare_cpus_common();
1405 
1406 	smp_sanity_check();
1407 
1408 	switch (apic_intr_mode) {
1409 	case APIC_PIC:
1410 	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1411 		disable_smp();
1412 		return;
1413 	case APIC_SYMMETRIC_IO_NO_ROUTING:
1414 		disable_smp();
1415 		/* Setup local timer */
1416 		x86_init.timers.setup_percpu_clockev();
1417 		return;
1418 	case APIC_VIRTUAL_WIRE:
1419 	case APIC_SYMMETRIC_IO:
1420 		break;
1421 	}
1422 
1423 	/* Setup local timer */
1424 	x86_init.timers.setup_percpu_clockev();
1425 
1426 	smp_get_logical_apicid();
1427 
1428 	pr_info("CPU0: ");
1429 	print_cpu_info(&cpu_data(0));
1430 
1431 	uv_system_init();
1432 
1433 	smp_quirk_init_udelay();
1434 
1435 	speculative_store_bypass_ht_init();
1436 
1437 	snp_set_wakeup_secondary_cpu();
1438 }
1439 
1440 void arch_thaw_secondary_cpus_begin(void)
1441 {
1442 	set_cache_aps_delayed_init(true);
1443 }
1444 
1445 void arch_thaw_secondary_cpus_end(void)
1446 {
1447 	cache_aps_init();
1448 }
1449 
1450 /*
1451  * Early setup to make printk work.
1452  */
1453 void __init native_smp_prepare_boot_cpu(void)
1454 {
1455 	int me = smp_processor_id();
1456 
1457 	/* SMP handles this from setup_per_cpu_areas() */
1458 	if (!IS_ENABLED(CONFIG_SMP))
1459 		switch_gdt_and_percpu_base(me);
1460 
1461 	/* already set me in cpu_online_mask in boot_cpu_init() */
1462 	cpumask_set_cpu(me, cpu_callout_mask);
1463 	cpu_set_state_online(me);
1464 	native_pv_lock_init();
1465 }
1466 
1467 void __init calculate_max_logical_packages(void)
1468 {
1469 	int ncpus;
1470 
1471 	/*
1472 	 * Today neither Intel nor AMD support heterogeneous systems so
1473 	 * extrapolate the boot cpu's data to all packages.
1474 	 */
1475 	ncpus = cpu_data(0).booted_cores * topology_max_smt_threads();
1476 	__max_logical_packages = DIV_ROUND_UP(total_cpus, ncpus);
1477 	pr_info("Max logical packages: %u\n", __max_logical_packages);
1478 }
1479 
1480 void __init native_smp_cpus_done(unsigned int max_cpus)
1481 {
1482 	pr_debug("Boot done\n");
1483 
1484 	calculate_max_logical_packages();
1485 
1486 	/* XXX for now assume numa-in-package and hybrid don't overlap */
1487 	if (x86_has_numa_in_package)
1488 		set_sched_topology(x86_numa_in_package_topology);
1489 	if (cpu_feature_enabled(X86_FEATURE_HYBRID_CPU))
1490 		set_sched_topology(x86_hybrid_topology);
1491 
1492 	nmi_selftest();
1493 	impress_friends();
1494 	cache_aps_init();
1495 }
1496 
1497 static int __initdata setup_possible_cpus = -1;
1498 static int __init _setup_possible_cpus(char *str)
1499 {
1500 	get_option(&str, &setup_possible_cpus);
1501 	return 0;
1502 }
1503 early_param("possible_cpus", _setup_possible_cpus);
1504 
1505 
1506 /*
1507  * cpu_possible_mask should be static, it cannot change as cpu's
1508  * are onlined, or offlined. The reason is per-cpu data-structures
1509  * are allocated by some modules at init time, and don't expect to
1510  * do this dynamically on cpu arrival/departure.
1511  * cpu_present_mask on the other hand can change dynamically.
1512  * In case when cpu_hotplug is not compiled, then we resort to current
1513  * behaviour, which is cpu_possible == cpu_present.
1514  * - Ashok Raj
1515  *
1516  * Three ways to find out the number of additional hotplug CPUs:
1517  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1518  * - The user can overwrite it with possible_cpus=NUM
1519  * - Otherwise don't reserve additional CPUs.
1520  * We do this because additional CPUs waste a lot of memory.
1521  * -AK
1522  */
1523 __init void prefill_possible_map(void)
1524 {
1525 	int i, possible;
1526 
1527 	/* No boot processor was found in mptable or ACPI MADT */
1528 	if (!num_processors) {
1529 		if (boot_cpu_has(X86_FEATURE_APIC)) {
1530 			int apicid = boot_cpu_physical_apicid;
1531 			int cpu = hard_smp_processor_id();
1532 
1533 			pr_warn("Boot CPU (id %d) not listed by BIOS\n", cpu);
1534 
1535 			/* Make sure boot cpu is enumerated */
1536 			if (apic->cpu_present_to_apicid(0) == BAD_APICID &&
1537 			    apic->apic_id_valid(apicid))
1538 				generic_processor_info(apicid, boot_cpu_apic_version);
1539 		}
1540 
1541 		if (!num_processors)
1542 			num_processors = 1;
1543 	}
1544 
1545 	i = setup_max_cpus ?: 1;
1546 	if (setup_possible_cpus == -1) {
1547 		possible = num_processors;
1548 #ifdef CONFIG_HOTPLUG_CPU
1549 		if (setup_max_cpus)
1550 			possible += disabled_cpus;
1551 #else
1552 		if (possible > i)
1553 			possible = i;
1554 #endif
1555 	} else
1556 		possible = setup_possible_cpus;
1557 
1558 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1559 
1560 	/* nr_cpu_ids could be reduced via nr_cpus= */
1561 	if (possible > nr_cpu_ids) {
1562 		pr_warn("%d Processors exceeds NR_CPUS limit of %u\n",
1563 			possible, nr_cpu_ids);
1564 		possible = nr_cpu_ids;
1565 	}
1566 
1567 #ifdef CONFIG_HOTPLUG_CPU
1568 	if (!setup_max_cpus)
1569 #endif
1570 	if (possible > i) {
1571 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1572 			possible, setup_max_cpus);
1573 		possible = i;
1574 	}
1575 
1576 	set_nr_cpu_ids(possible);
1577 
1578 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1579 		possible, max_t(int, possible - num_processors, 0));
1580 
1581 	reset_cpu_possible_mask();
1582 
1583 	for (i = 0; i < possible; i++)
1584 		set_cpu_possible(i, true);
1585 }
1586 
1587 #ifdef CONFIG_HOTPLUG_CPU
1588 
1589 /* Recompute SMT state for all CPUs on offline */
1590 static void recompute_smt_state(void)
1591 {
1592 	int max_threads, cpu;
1593 
1594 	max_threads = 0;
1595 	for_each_online_cpu (cpu) {
1596 		int threads = cpumask_weight(topology_sibling_cpumask(cpu));
1597 
1598 		if (threads > max_threads)
1599 			max_threads = threads;
1600 	}
1601 	__max_smt_threads = max_threads;
1602 }
1603 
1604 static void remove_siblinginfo(int cpu)
1605 {
1606 	int sibling;
1607 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1608 
1609 	for_each_cpu(sibling, topology_core_cpumask(cpu)) {
1610 		cpumask_clear_cpu(cpu, topology_core_cpumask(sibling));
1611 		/*/
1612 		 * last thread sibling in this cpu core going down
1613 		 */
1614 		if (cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
1615 			cpu_data(sibling).booted_cores--;
1616 	}
1617 
1618 	for_each_cpu(sibling, topology_die_cpumask(cpu))
1619 		cpumask_clear_cpu(cpu, topology_die_cpumask(sibling));
1620 
1621 	for_each_cpu(sibling, topology_sibling_cpumask(cpu)) {
1622 		cpumask_clear_cpu(cpu, topology_sibling_cpumask(sibling));
1623 		if (cpumask_weight(topology_sibling_cpumask(sibling)) == 1)
1624 			cpu_data(sibling).smt_active = false;
1625 	}
1626 
1627 	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1628 		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1629 	for_each_cpu(sibling, cpu_l2c_shared_mask(cpu))
1630 		cpumask_clear_cpu(cpu, cpu_l2c_shared_mask(sibling));
1631 	cpumask_clear(cpu_llc_shared_mask(cpu));
1632 	cpumask_clear(cpu_l2c_shared_mask(cpu));
1633 	cpumask_clear(topology_sibling_cpumask(cpu));
1634 	cpumask_clear(topology_core_cpumask(cpu));
1635 	cpumask_clear(topology_die_cpumask(cpu));
1636 	c->cpu_core_id = 0;
1637 	c->booted_cores = 0;
1638 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1639 	recompute_smt_state();
1640 }
1641 
1642 static void remove_cpu_from_maps(int cpu)
1643 {
1644 	set_cpu_online(cpu, false);
1645 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1646 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1647 	/* was set by cpu_init() */
1648 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1649 	numa_remove_cpu(cpu);
1650 }
1651 
1652 void cpu_disable_common(void)
1653 {
1654 	int cpu = smp_processor_id();
1655 
1656 	remove_siblinginfo(cpu);
1657 
1658 	/* It's now safe to remove this processor from the online map */
1659 	lock_vector_lock();
1660 	remove_cpu_from_maps(cpu);
1661 	unlock_vector_lock();
1662 	fixup_irqs();
1663 	lapic_offline();
1664 }
1665 
1666 int native_cpu_disable(void)
1667 {
1668 	int ret;
1669 
1670 	ret = lapic_can_unplug_cpu();
1671 	if (ret)
1672 		return ret;
1673 
1674 	cpu_disable_common();
1675 
1676         /*
1677          * Disable the local APIC. Otherwise IPI broadcasts will reach
1678          * it. It still responds normally to INIT, NMI, SMI, and SIPI
1679          * messages.
1680          *
1681          * Disabling the APIC must happen after cpu_disable_common()
1682          * which invokes fixup_irqs().
1683          *
1684          * Disabling the APIC preserves already set bits in IRR, but
1685          * an interrupt arriving after disabling the local APIC does not
1686          * set the corresponding IRR bit.
1687          *
1688          * fixup_irqs() scans IRR for set bits so it can raise a not
1689          * yet handled interrupt on the new destination CPU via an IPI
1690          * but obviously it can't do so for IRR bits which are not set.
1691          * IOW, interrupts arriving after disabling the local APIC will
1692          * be lost.
1693          */
1694 	apic_soft_disable();
1695 
1696 	return 0;
1697 }
1698 
1699 int common_cpu_die(unsigned int cpu)
1700 {
1701 	int ret = 0;
1702 
1703 	/* We don't do anything here: idle task is faking death itself. */
1704 
1705 	/* They ack this in play_dead() by setting CPU_DEAD */
1706 	if (cpu_wait_death(cpu, 5)) {
1707 		if (system_state == SYSTEM_RUNNING)
1708 			pr_info("CPU %u is now offline\n", cpu);
1709 	} else {
1710 		pr_err("CPU %u didn't die...\n", cpu);
1711 		ret = -1;
1712 	}
1713 
1714 	return ret;
1715 }
1716 
1717 void native_cpu_die(unsigned int cpu)
1718 {
1719 	common_cpu_die(cpu);
1720 }
1721 
1722 void play_dead_common(void)
1723 {
1724 	idle_task_exit();
1725 
1726 	/* Ack it */
1727 	(void)cpu_report_death();
1728 
1729 	/*
1730 	 * With physical CPU hotplug, we should halt the cpu
1731 	 */
1732 	local_irq_disable();
1733 }
1734 
1735 /**
1736  * cond_wakeup_cpu0 - Wake up CPU0 if needed.
1737  *
1738  * If NMI wants to wake up CPU0, start CPU0.
1739  */
1740 void cond_wakeup_cpu0(void)
1741 {
1742 	if (smp_processor_id() == 0 && enable_start_cpu0)
1743 		start_cpu0();
1744 }
1745 EXPORT_SYMBOL_GPL(cond_wakeup_cpu0);
1746 
1747 /*
1748  * We need to flush the caches before going to sleep, lest we have
1749  * dirty data in our caches when we come back up.
1750  */
1751 static inline void mwait_play_dead(void)
1752 {
1753 	unsigned int eax, ebx, ecx, edx;
1754 	unsigned int highest_cstate = 0;
1755 	unsigned int highest_subcstate = 0;
1756 	void *mwait_ptr;
1757 	int i;
1758 
1759 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1760 	    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
1761 		return;
1762 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1763 		return;
1764 	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1765 		return;
1766 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1767 		return;
1768 
1769 	eax = CPUID_MWAIT_LEAF;
1770 	ecx = 0;
1771 	native_cpuid(&eax, &ebx, &ecx, &edx);
1772 
1773 	/*
1774 	 * eax will be 0 if EDX enumeration is not valid.
1775 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1776 	 */
1777 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1778 		eax = 0;
1779 	} else {
1780 		edx >>= MWAIT_SUBSTATE_SIZE;
1781 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1782 			if (edx & MWAIT_SUBSTATE_MASK) {
1783 				highest_cstate = i;
1784 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1785 			}
1786 		}
1787 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1788 			(highest_subcstate - 1);
1789 	}
1790 
1791 	/*
1792 	 * This should be a memory location in a cache line which is
1793 	 * unlikely to be touched by other processors.  The actual
1794 	 * content is immaterial as it is not actually modified in any way.
1795 	 */
1796 	mwait_ptr = &current_thread_info()->flags;
1797 
1798 	wbinvd();
1799 
1800 	while (1) {
1801 		/*
1802 		 * The CLFLUSH is a workaround for erratum AAI65 for
1803 		 * the Xeon 7400 series.  It's not clear it is actually
1804 		 * needed, but it should be harmless in either case.
1805 		 * The WBINVD is insufficient due to the spurious-wakeup
1806 		 * case where we return around the loop.
1807 		 */
1808 		mb();
1809 		clflush(mwait_ptr);
1810 		mb();
1811 		__monitor(mwait_ptr, 0, 0);
1812 		mb();
1813 		__mwait(eax, 0);
1814 
1815 		cond_wakeup_cpu0();
1816 	}
1817 }
1818 
1819 void hlt_play_dead(void)
1820 {
1821 	if (__this_cpu_read(cpu_info.x86) >= 4)
1822 		wbinvd();
1823 
1824 	while (1) {
1825 		native_halt();
1826 
1827 		cond_wakeup_cpu0();
1828 	}
1829 }
1830 
1831 void native_play_dead(void)
1832 {
1833 	play_dead_common();
1834 	tboot_shutdown(TB_SHUTDOWN_WFS);
1835 
1836 	mwait_play_dead();
1837 	if (cpuidle_play_dead())
1838 		hlt_play_dead();
1839 }
1840 
1841 #else /* ... !CONFIG_HOTPLUG_CPU */
1842 int native_cpu_disable(void)
1843 {
1844 	return -ENOSYS;
1845 }
1846 
1847 void native_cpu_die(unsigned int cpu)
1848 {
1849 	/* We said "no" in __cpu_disable */
1850 	BUG();
1851 }
1852 
1853 void native_play_dead(void)
1854 {
1855 	BUG();
1856 }
1857 
1858 #endif
1859