xref: /linux/arch/x86/kernel/smpboot.c (revision ff40b5769a50fab654a70575ff0f49853b799b0e)
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 
125 static unsigned int smpboot_warm_reset_vector_count;
126 
127 static inline void smpboot_setup_warm_reset_vector(unsigned long start_eip)
128 {
129 	unsigned long flags;
130 
131 	spin_lock_irqsave(&rtc_lock, flags);
132 	if (!smpboot_warm_reset_vector_count++) {
133 		CMOS_WRITE(0xa, 0xf);
134 		*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_HIGH)) = start_eip >> 4;
135 		*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = start_eip & 0xf;
136 	}
137 	spin_unlock_irqrestore(&rtc_lock, flags);
138 }
139 
140 static inline void smpboot_restore_warm_reset_vector(void)
141 {
142 	unsigned long flags;
143 
144 	/*
145 	 * Paranoid:  Set warm reset code and vector here back
146 	 * to default values.
147 	 */
148 	spin_lock_irqsave(&rtc_lock, flags);
149 	if (!--smpboot_warm_reset_vector_count) {
150 		CMOS_WRITE(0, 0xf);
151 		*((volatile u32 *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = 0;
152 	}
153 	spin_unlock_irqrestore(&rtc_lock, flags);
154 
155 }
156 
157 /*
158  * Report back to the Boot Processor during boot time or to the caller processor
159  * during CPU online.
160  */
161 static void smp_callin(void)
162 {
163 	int cpuid;
164 
165 	/*
166 	 * If waken up by an INIT in an 82489DX configuration
167 	 * cpu_callout_mask guarantees we don't get here before
168 	 * an INIT_deassert IPI reaches our local APIC, so it is
169 	 * now safe to touch our local APIC.
170 	 */
171 	cpuid = smp_processor_id();
172 
173 	/*
174 	 * the boot CPU has finished the init stage and is spinning
175 	 * on callin_map until we finish. We are free to set up this
176 	 * CPU, first the APIC. (this is probably redundant on most
177 	 * boards)
178 	 */
179 	apic_ap_setup();
180 
181 	/*
182 	 * Save our processor parameters. Note: this information
183 	 * is needed for clock calibration.
184 	 */
185 	smp_store_cpu_info(cpuid);
186 
187 	/*
188 	 * The topology information must be up to date before
189 	 * calibrate_delay() and notify_cpu_starting().
190 	 */
191 	set_cpu_sibling_map(raw_smp_processor_id());
192 
193 	ap_init_aperfmperf();
194 
195 	/*
196 	 * Get our bogomips.
197 	 * Update loops_per_jiffy in cpu_data. Previous call to
198 	 * smp_store_cpu_info() stored a value that is close but not as
199 	 * accurate as the value just calculated.
200 	 */
201 	calibrate_delay();
202 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
203 	pr_debug("Stack at about %p\n", &cpuid);
204 
205 	wmb();
206 
207 	notify_cpu_starting(cpuid);
208 
209 	/*
210 	 * Allow the master to continue.
211 	 */
212 	cpumask_set_cpu(cpuid, cpu_callin_mask);
213 }
214 
215 static int cpu0_logical_apicid;
216 static int enable_start_cpu0;
217 /*
218  * Activate a secondary processor.
219  */
220 static void notrace start_secondary(void *unused)
221 {
222 	/*
223 	 * Don't put *anything* except direct CPU state initialization
224 	 * before cpu_init(), SMP booting is too fragile that we want to
225 	 * limit the things done here to the most necessary things.
226 	 */
227 	cr4_init();
228 
229 #ifdef CONFIG_X86_32
230 	/* switch away from the initial page table */
231 	load_cr3(swapper_pg_dir);
232 	__flush_tlb_all();
233 #endif
234 	cpu_init_secondary();
235 	rcu_cpu_starting(raw_smp_processor_id());
236 	x86_cpuinit.early_percpu_clock_init();
237 	smp_callin();
238 
239 	enable_start_cpu0 = 0;
240 
241 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
242 	barrier();
243 	/*
244 	 * Check TSC synchronization with the boot CPU:
245 	 */
246 	check_tsc_sync_target();
247 
248 	speculative_store_bypass_ht_init();
249 
250 	/*
251 	 * Lock vector_lock, set CPU online and bring the vector
252 	 * allocator online. Online must be set with vector_lock held
253 	 * to prevent a concurrent irq setup/teardown from seeing a
254 	 * half valid vector space.
255 	 */
256 	lock_vector_lock();
257 	set_cpu_online(smp_processor_id(), true);
258 	lapic_online();
259 	unlock_vector_lock();
260 	cpu_set_state_online(smp_processor_id());
261 	x86_platform.nmi_init();
262 
263 	/* enable local interrupts */
264 	local_irq_enable();
265 
266 	x86_cpuinit.setup_percpu_clockev();
267 
268 	wmb();
269 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
270 }
271 
272 /**
273  * topology_is_primary_thread - Check whether CPU is the primary SMT thread
274  * @cpu:	CPU to check
275  */
276 bool topology_is_primary_thread(unsigned int cpu)
277 {
278 	return apic_id_is_primary_thread(per_cpu(x86_cpu_to_apicid, cpu));
279 }
280 
281 /**
282  * topology_smt_supported - Check whether SMT is supported by the CPUs
283  */
284 bool topology_smt_supported(void)
285 {
286 	return smp_num_siblings > 1;
287 }
288 
289 /**
290  * topology_phys_to_logical_pkg - Map a physical package id to a logical
291  *
292  * Returns logical package id or -1 if not found
293  */
294 int topology_phys_to_logical_pkg(unsigned int phys_pkg)
295 {
296 	int cpu;
297 
298 	for_each_possible_cpu(cpu) {
299 		struct cpuinfo_x86 *c = &cpu_data(cpu);
300 
301 		if (c->initialized && c->phys_proc_id == phys_pkg)
302 			return c->logical_proc_id;
303 	}
304 	return -1;
305 }
306 EXPORT_SYMBOL(topology_phys_to_logical_pkg);
307 /**
308  * topology_phys_to_logical_die - Map a physical die id to logical
309  *
310  * Returns logical die id or -1 if not found
311  */
312 int topology_phys_to_logical_die(unsigned int die_id, unsigned int cur_cpu)
313 {
314 	int cpu;
315 	int proc_id = cpu_data(cur_cpu).phys_proc_id;
316 
317 	for_each_possible_cpu(cpu) {
318 		struct cpuinfo_x86 *c = &cpu_data(cpu);
319 
320 		if (c->initialized && c->cpu_die_id == die_id &&
321 		    c->phys_proc_id == proc_id)
322 			return c->logical_die_id;
323 	}
324 	return -1;
325 }
326 EXPORT_SYMBOL(topology_phys_to_logical_die);
327 
328 /**
329  * topology_update_package_map - Update the physical to logical package map
330  * @pkg:	The physical package id as retrieved via CPUID
331  * @cpu:	The cpu for which this is updated
332  */
333 int topology_update_package_map(unsigned int pkg, unsigned int cpu)
334 {
335 	int new;
336 
337 	/* Already available somewhere? */
338 	new = topology_phys_to_logical_pkg(pkg);
339 	if (new >= 0)
340 		goto found;
341 
342 	new = logical_packages++;
343 	if (new != pkg) {
344 		pr_info("CPU %u Converting physical %u to logical package %u\n",
345 			cpu, pkg, new);
346 	}
347 found:
348 	cpu_data(cpu).logical_proc_id = new;
349 	return 0;
350 }
351 /**
352  * topology_update_die_map - Update the physical to logical die map
353  * @die:	The die id as retrieved via CPUID
354  * @cpu:	The cpu for which this is updated
355  */
356 int topology_update_die_map(unsigned int die, unsigned int cpu)
357 {
358 	int new;
359 
360 	/* Already available somewhere? */
361 	new = topology_phys_to_logical_die(die, cpu);
362 	if (new >= 0)
363 		goto found;
364 
365 	new = logical_die++;
366 	if (new != die) {
367 		pr_info("CPU %u Converting physical %u to logical die %u\n",
368 			cpu, die, new);
369 	}
370 found:
371 	cpu_data(cpu).logical_die_id = new;
372 	return 0;
373 }
374 
375 void __init smp_store_boot_cpu_info(void)
376 {
377 	int id = 0; /* CPU 0 */
378 	struct cpuinfo_x86 *c = &cpu_data(id);
379 
380 	*c = boot_cpu_data;
381 	c->cpu_index = id;
382 	topology_update_package_map(c->phys_proc_id, id);
383 	topology_update_die_map(c->cpu_die_id, id);
384 	c->initialized = true;
385 }
386 
387 /*
388  * The bootstrap kernel entry code has set these up. Save them for
389  * a given CPU
390  */
391 void smp_store_cpu_info(int id)
392 {
393 	struct cpuinfo_x86 *c = &cpu_data(id);
394 
395 	/* Copy boot_cpu_data only on the first bringup */
396 	if (!c->initialized)
397 		*c = boot_cpu_data;
398 	c->cpu_index = id;
399 	/*
400 	 * During boot time, CPU0 has this setup already. Save the info when
401 	 * bringing up AP or offlined CPU0.
402 	 */
403 	identify_secondary_cpu(c);
404 	c->initialized = true;
405 }
406 
407 static bool
408 topology_same_node(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
409 {
410 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
411 
412 	return (cpu_to_node(cpu1) == cpu_to_node(cpu2));
413 }
414 
415 static bool
416 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
417 {
418 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
419 
420 	return !WARN_ONCE(!topology_same_node(c, o),
421 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
422 		"[node: %d != %d]. Ignoring dependency.\n",
423 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
424 }
425 
426 #define link_mask(mfunc, c1, c2)					\
427 do {									\
428 	cpumask_set_cpu((c1), mfunc(c2));				\
429 	cpumask_set_cpu((c2), mfunc(c1));				\
430 } while (0)
431 
432 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
433 {
434 	if (boot_cpu_has(X86_FEATURE_TOPOEXT)) {
435 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
436 
437 		if (c->phys_proc_id == o->phys_proc_id &&
438 		    c->cpu_die_id == o->cpu_die_id &&
439 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2)) {
440 			if (c->cpu_core_id == o->cpu_core_id)
441 				return topology_sane(c, o, "smt");
442 
443 			if ((c->cu_id != 0xff) &&
444 			    (o->cu_id != 0xff) &&
445 			    (c->cu_id == o->cu_id))
446 				return topology_sane(c, o, "smt");
447 		}
448 
449 	} else if (c->phys_proc_id == o->phys_proc_id &&
450 		   c->cpu_die_id == o->cpu_die_id &&
451 		   c->cpu_core_id == o->cpu_core_id) {
452 		return topology_sane(c, o, "smt");
453 	}
454 
455 	return false;
456 }
457 
458 static bool match_die(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
459 {
460 	if (c->phys_proc_id == o->phys_proc_id &&
461 	    c->cpu_die_id == o->cpu_die_id)
462 		return true;
463 	return false;
464 }
465 
466 static bool match_l2c(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
467 {
468 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
469 
470 	/* If the arch didn't set up l2c_id, fall back to SMT */
471 	if (per_cpu(cpu_l2c_id, cpu1) == BAD_APICID)
472 		return match_smt(c, o);
473 
474 	/* Do not match if L2 cache id does not match: */
475 	if (per_cpu(cpu_l2c_id, cpu1) != per_cpu(cpu_l2c_id, cpu2))
476 		return false;
477 
478 	return topology_sane(c, o, "l2c");
479 }
480 
481 /*
482  * Unlike the other levels, we do not enforce keeping a
483  * multicore group inside a NUMA node.  If this happens, we will
484  * discard the MC level of the topology later.
485  */
486 static bool match_pkg(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
487 {
488 	if (c->phys_proc_id == o->phys_proc_id)
489 		return true;
490 	return false;
491 }
492 
493 /*
494  * Define intel_cod_cpu[] for Intel COD (Cluster-on-Die) CPUs.
495  *
496  * Any Intel CPU that has multiple nodes per package and does not
497  * match intel_cod_cpu[] has the SNC (Sub-NUMA Cluster) topology.
498  *
499  * When in SNC mode, these CPUs enumerate an LLC that is shared
500  * by multiple NUMA nodes. The LLC is shared for off-package data
501  * access but private to the NUMA node (half of the package) for
502  * on-package access. CPUID (the source of the information about
503  * the LLC) can only enumerate the cache as shared or unshared,
504  * but not this particular configuration.
505  */
506 
507 static const struct x86_cpu_id intel_cod_cpu[] = {
508 	X86_MATCH_INTEL_FAM6_MODEL(HASWELL_X, 0),	/* COD */
509 	X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_X, 0),	/* COD */
510 	X86_MATCH_INTEL_FAM6_MODEL(ANY, 1),		/* SNC */
511 	{}
512 };
513 
514 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
515 {
516 	const struct x86_cpu_id *id = x86_match_cpu(intel_cod_cpu);
517 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
518 	bool intel_snc = id && id->driver_data;
519 
520 	/* Do not match if we do not have a valid APICID for cpu: */
521 	if (per_cpu(cpu_llc_id, cpu1) == BAD_APICID)
522 		return false;
523 
524 	/* Do not match if LLC id does not match: */
525 	if (per_cpu(cpu_llc_id, cpu1) != per_cpu(cpu_llc_id, cpu2))
526 		return false;
527 
528 	/*
529 	 * Allow the SNC topology without warning. Return of false
530 	 * means 'c' does not share the LLC of 'o'. This will be
531 	 * reflected to userspace.
532 	 */
533 	if (match_pkg(c, o) && !topology_same_node(c, o) && intel_snc)
534 		return false;
535 
536 	return topology_sane(c, o, "llc");
537 }
538 
539 
540 #if defined(CONFIG_SCHED_SMT) || defined(CONFIG_SCHED_CLUSTER) || defined(CONFIG_SCHED_MC)
541 static inline int x86_sched_itmt_flags(void)
542 {
543 	return sysctl_sched_itmt_enabled ? SD_ASYM_PACKING : 0;
544 }
545 
546 #ifdef CONFIG_SCHED_MC
547 static int x86_core_flags(void)
548 {
549 	return cpu_core_flags() | x86_sched_itmt_flags();
550 }
551 #endif
552 #ifdef CONFIG_SCHED_SMT
553 static int x86_smt_flags(void)
554 {
555 	return cpu_smt_flags() | x86_sched_itmt_flags();
556 }
557 #endif
558 #ifdef CONFIG_SCHED_CLUSTER
559 static int x86_cluster_flags(void)
560 {
561 	return cpu_cluster_flags() | x86_sched_itmt_flags();
562 }
563 #endif
564 #endif
565 
566 static struct sched_domain_topology_level x86_numa_in_package_topology[] = {
567 #ifdef CONFIG_SCHED_SMT
568 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
569 #endif
570 #ifdef CONFIG_SCHED_CLUSTER
571 	{ cpu_clustergroup_mask, x86_cluster_flags, SD_INIT_NAME(CLS) },
572 #endif
573 #ifdef CONFIG_SCHED_MC
574 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
575 #endif
576 	{ NULL, },
577 };
578 
579 static struct sched_domain_topology_level x86_hybrid_topology[] = {
580 #ifdef CONFIG_SCHED_SMT
581 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
582 #endif
583 #ifdef CONFIG_SCHED_MC
584 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
585 #endif
586 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
587 	{ NULL, },
588 };
589 
590 static struct sched_domain_topology_level x86_topology[] = {
591 #ifdef CONFIG_SCHED_SMT
592 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
593 #endif
594 #ifdef CONFIG_SCHED_CLUSTER
595 	{ cpu_clustergroup_mask, x86_cluster_flags, SD_INIT_NAME(CLS) },
596 #endif
597 #ifdef CONFIG_SCHED_MC
598 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
599 #endif
600 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
601 	{ NULL, },
602 };
603 
604 /*
605  * Set if a package/die has multiple NUMA nodes inside.
606  * AMD Magny-Cours, Intel Cluster-on-Die, and Intel
607  * Sub-NUMA Clustering have this.
608  */
609 static bool x86_has_numa_in_package;
610 
611 void set_cpu_sibling_map(int cpu)
612 {
613 	bool has_smt = smp_num_siblings > 1;
614 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
615 	struct cpuinfo_x86 *c = &cpu_data(cpu);
616 	struct cpuinfo_x86 *o;
617 	int i, threads;
618 
619 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
620 
621 	if (!has_mp) {
622 		cpumask_set_cpu(cpu, topology_sibling_cpumask(cpu));
623 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
624 		cpumask_set_cpu(cpu, cpu_l2c_shared_mask(cpu));
625 		cpumask_set_cpu(cpu, topology_core_cpumask(cpu));
626 		cpumask_set_cpu(cpu, topology_die_cpumask(cpu));
627 		c->booted_cores = 1;
628 		return;
629 	}
630 
631 	for_each_cpu(i, cpu_sibling_setup_mask) {
632 		o = &cpu_data(i);
633 
634 		if (match_pkg(c, o) && !topology_same_node(c, o))
635 			x86_has_numa_in_package = true;
636 
637 		if ((i == cpu) || (has_smt && match_smt(c, o)))
638 			link_mask(topology_sibling_cpumask, cpu, i);
639 
640 		if ((i == cpu) || (has_mp && match_llc(c, o)))
641 			link_mask(cpu_llc_shared_mask, cpu, i);
642 
643 		if ((i == cpu) || (has_mp && match_l2c(c, o)))
644 			link_mask(cpu_l2c_shared_mask, cpu, i);
645 
646 		if ((i == cpu) || (has_mp && match_die(c, o)))
647 			link_mask(topology_die_cpumask, cpu, i);
648 	}
649 
650 	threads = cpumask_weight(topology_sibling_cpumask(cpu));
651 	if (threads > __max_smt_threads)
652 		__max_smt_threads = threads;
653 
654 	for_each_cpu(i, topology_sibling_cpumask(cpu))
655 		cpu_data(i).smt_active = threads > 1;
656 
657 	/*
658 	 * This needs a separate iteration over the cpus because we rely on all
659 	 * topology_sibling_cpumask links to be set-up.
660 	 */
661 	for_each_cpu(i, cpu_sibling_setup_mask) {
662 		o = &cpu_data(i);
663 
664 		if ((i == cpu) || (has_mp && match_pkg(c, o))) {
665 			link_mask(topology_core_cpumask, cpu, i);
666 
667 			/*
668 			 *  Does this new cpu bringup a new core?
669 			 */
670 			if (threads == 1) {
671 				/*
672 				 * for each core in package, increment
673 				 * the booted_cores for this new cpu
674 				 */
675 				if (cpumask_first(
676 				    topology_sibling_cpumask(i)) == i)
677 					c->booted_cores++;
678 				/*
679 				 * increment the core count for all
680 				 * the other cpus in this package
681 				 */
682 				if (i != cpu)
683 					cpu_data(i).booted_cores++;
684 			} else if (i != cpu && !c->booted_cores)
685 				c->booted_cores = cpu_data(i).booted_cores;
686 		}
687 	}
688 }
689 
690 /* maps the cpu to the sched domain representing multi-core */
691 const struct cpumask *cpu_coregroup_mask(int cpu)
692 {
693 	return cpu_llc_shared_mask(cpu);
694 }
695 
696 const struct cpumask *cpu_clustergroup_mask(int cpu)
697 {
698 	return cpu_l2c_shared_mask(cpu);
699 }
700 
701 static void impress_friends(void)
702 {
703 	int cpu;
704 	unsigned long bogosum = 0;
705 	/*
706 	 * Allow the user to impress friends.
707 	 */
708 	pr_debug("Before bogomips\n");
709 	for_each_possible_cpu(cpu)
710 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
711 			bogosum += cpu_data(cpu).loops_per_jiffy;
712 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
713 		num_online_cpus(),
714 		bogosum/(500000/HZ),
715 		(bogosum/(5000/HZ))%100);
716 
717 	pr_debug("Before bogocount - setting activated=1\n");
718 }
719 
720 void __inquire_remote_apic(int apicid)
721 {
722 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
723 	const char * const names[] = { "ID", "VERSION", "SPIV" };
724 	int timeout;
725 	u32 status;
726 
727 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
728 
729 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
730 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
731 
732 		/*
733 		 * Wait for idle.
734 		 */
735 		status = safe_apic_wait_icr_idle();
736 		if (status)
737 			pr_cont("a previous APIC delivery may have failed\n");
738 
739 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
740 
741 		timeout = 0;
742 		do {
743 			udelay(100);
744 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
745 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
746 
747 		switch (status) {
748 		case APIC_ICR_RR_VALID:
749 			status = apic_read(APIC_RRR);
750 			pr_cont("%08x\n", status);
751 			break;
752 		default:
753 			pr_cont("failed\n");
754 		}
755 	}
756 }
757 
758 /*
759  * The Multiprocessor Specification 1.4 (1997) example code suggests
760  * that there should be a 10ms delay between the BSP asserting INIT
761  * and de-asserting INIT, when starting a remote processor.
762  * But that slows boot and resume on modern processors, which include
763  * many cores and don't require that delay.
764  *
765  * Cmdline "init_cpu_udelay=" is available to over-ride this delay.
766  * Modern processor families are quirked to remove the delay entirely.
767  */
768 #define UDELAY_10MS_DEFAULT 10000
769 
770 static unsigned int init_udelay = UINT_MAX;
771 
772 static int __init cpu_init_udelay(char *str)
773 {
774 	get_option(&str, &init_udelay);
775 
776 	return 0;
777 }
778 early_param("cpu_init_udelay", cpu_init_udelay);
779 
780 static void __init smp_quirk_init_udelay(void)
781 {
782 	/* if cmdline changed it from default, leave it alone */
783 	if (init_udelay != UINT_MAX)
784 		return;
785 
786 	/* if modern processor, use no delay */
787 	if (((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && (boot_cpu_data.x86 == 6)) ||
788 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) && (boot_cpu_data.x86 >= 0x18)) ||
789 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && (boot_cpu_data.x86 >= 0xF))) {
790 		init_udelay = 0;
791 		return;
792 	}
793 	/* else, use legacy delay */
794 	init_udelay = UDELAY_10MS_DEFAULT;
795 }
796 
797 /*
798  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
799  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
800  * won't ... remember to clear down the APIC, etc later.
801  */
802 int
803 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
804 {
805 	u32 dm = apic->dest_mode_logical ? APIC_DEST_LOGICAL : APIC_DEST_PHYSICAL;
806 	unsigned long send_status, accept_status = 0;
807 	int maxlvt;
808 
809 	/* Target chip */
810 	/* Boot on the stack */
811 	/* Kick the second */
812 	apic_icr_write(APIC_DM_NMI | dm, apicid);
813 
814 	pr_debug("Waiting for send to finish...\n");
815 	send_status = safe_apic_wait_icr_idle();
816 
817 	/*
818 	 * Give the other CPU some time to accept the IPI.
819 	 */
820 	udelay(200);
821 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
822 		maxlvt = lapic_get_maxlvt();
823 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
824 			apic_write(APIC_ESR, 0);
825 		accept_status = (apic_read(APIC_ESR) & 0xEF);
826 	}
827 	pr_debug("NMI sent\n");
828 
829 	if (send_status)
830 		pr_err("APIC never delivered???\n");
831 	if (accept_status)
832 		pr_err("APIC delivery error (%lx)\n", accept_status);
833 
834 	return (send_status | accept_status);
835 }
836 
837 static int
838 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
839 {
840 	unsigned long send_status = 0, accept_status = 0;
841 	int maxlvt, num_starts, j;
842 
843 	maxlvt = lapic_get_maxlvt();
844 
845 	/*
846 	 * Be paranoid about clearing APIC errors.
847 	 */
848 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
849 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
850 			apic_write(APIC_ESR, 0);
851 		apic_read(APIC_ESR);
852 	}
853 
854 	pr_debug("Asserting INIT\n");
855 
856 	/*
857 	 * Turn INIT on target chip
858 	 */
859 	/*
860 	 * Send IPI
861 	 */
862 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
863 		       phys_apicid);
864 
865 	pr_debug("Waiting for send to finish...\n");
866 	send_status = safe_apic_wait_icr_idle();
867 
868 	udelay(init_udelay);
869 
870 	pr_debug("Deasserting INIT\n");
871 
872 	/* Target chip */
873 	/* Send IPI */
874 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
875 
876 	pr_debug("Waiting for send to finish...\n");
877 	send_status = safe_apic_wait_icr_idle();
878 
879 	mb();
880 
881 	/*
882 	 * Should we send STARTUP IPIs ?
883 	 *
884 	 * Determine this based on the APIC version.
885 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
886 	 */
887 	if (APIC_INTEGRATED(boot_cpu_apic_version))
888 		num_starts = 2;
889 	else
890 		num_starts = 0;
891 
892 	/*
893 	 * Run STARTUP IPI loop.
894 	 */
895 	pr_debug("#startup loops: %d\n", num_starts);
896 
897 	for (j = 1; j <= num_starts; j++) {
898 		pr_debug("Sending STARTUP #%d\n", j);
899 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
900 			apic_write(APIC_ESR, 0);
901 		apic_read(APIC_ESR);
902 		pr_debug("After apic_write\n");
903 
904 		/*
905 		 * STARTUP IPI
906 		 */
907 
908 		/* Target chip */
909 		/* Boot on the stack */
910 		/* Kick the second */
911 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
912 			       phys_apicid);
913 
914 		/*
915 		 * Give the other CPU some time to accept the IPI.
916 		 */
917 		if (init_udelay == 0)
918 			udelay(10);
919 		else
920 			udelay(300);
921 
922 		pr_debug("Startup point 1\n");
923 
924 		pr_debug("Waiting for send to finish...\n");
925 		send_status = safe_apic_wait_icr_idle();
926 
927 		/*
928 		 * Give the other CPU some time to accept the IPI.
929 		 */
930 		if (init_udelay == 0)
931 			udelay(10);
932 		else
933 			udelay(200);
934 
935 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
936 			apic_write(APIC_ESR, 0);
937 		accept_status = (apic_read(APIC_ESR) & 0xEF);
938 		if (send_status || accept_status)
939 			break;
940 	}
941 	pr_debug("After Startup\n");
942 
943 	if (send_status)
944 		pr_err("APIC never delivered???\n");
945 	if (accept_status)
946 		pr_err("APIC delivery error (%lx)\n", accept_status);
947 
948 	return (send_status | accept_status);
949 }
950 
951 /* reduce the number of lines printed when booting a large cpu count system */
952 static void announce_cpu(int cpu, int apicid)
953 {
954 	static int current_node = NUMA_NO_NODE;
955 	int node = early_cpu_to_node(cpu);
956 	static int width, node_width;
957 
958 	if (!width)
959 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
960 
961 	if (!node_width)
962 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
963 
964 	if (cpu == 1)
965 		printk(KERN_INFO "x86: Booting SMP configuration:\n");
966 
967 	if (system_state < SYSTEM_RUNNING) {
968 		if (node != current_node) {
969 			if (current_node > (-1))
970 				pr_cont("\n");
971 			current_node = node;
972 
973 			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
974 			       node_width - num_digits(node), " ", node);
975 		}
976 
977 		/* Add padding for the BSP */
978 		if (cpu == 1)
979 			pr_cont("%*s", width + 1, " ");
980 
981 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
982 
983 	} else
984 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
985 			node, cpu, apicid);
986 }
987 
988 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
989 {
990 	int cpu;
991 
992 	cpu = smp_processor_id();
993 	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
994 		return NMI_HANDLED;
995 
996 	return NMI_DONE;
997 }
998 
999 /*
1000  * Wake up AP by INIT, INIT, STARTUP sequence.
1001  *
1002  * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
1003  * boot-strap code which is not a desired behavior for waking up BSP. To
1004  * void the boot-strap code, wake up CPU0 by NMI instead.
1005  *
1006  * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
1007  * (i.e. physically hot removed and then hot added), NMI won't wake it up.
1008  * We'll change this code in the future to wake up hard offlined CPU0 if
1009  * real platform and request are available.
1010  */
1011 static int
1012 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
1013 	       int *cpu0_nmi_registered)
1014 {
1015 	int id;
1016 	int boot_error;
1017 
1018 	preempt_disable();
1019 
1020 	/*
1021 	 * Wake up AP by INIT, INIT, STARTUP sequence.
1022 	 */
1023 	if (cpu) {
1024 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
1025 		goto out;
1026 	}
1027 
1028 	/*
1029 	 * Wake up BSP by nmi.
1030 	 *
1031 	 * Register a NMI handler to help wake up CPU0.
1032 	 */
1033 	boot_error = register_nmi_handler(NMI_LOCAL,
1034 					  wakeup_cpu0_nmi, 0, "wake_cpu0");
1035 
1036 	if (!boot_error) {
1037 		enable_start_cpu0 = 1;
1038 		*cpu0_nmi_registered = 1;
1039 		id = apic->dest_mode_logical ? cpu0_logical_apicid : apicid;
1040 		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
1041 	}
1042 
1043 out:
1044 	preempt_enable();
1045 
1046 	return boot_error;
1047 }
1048 
1049 int common_cpu_up(unsigned int cpu, struct task_struct *idle)
1050 {
1051 	int ret;
1052 
1053 	/* Just in case we booted with a single CPU. */
1054 	alternatives_enable_smp();
1055 
1056 	per_cpu(pcpu_hot.current_task, cpu) = idle;
1057 	cpu_init_stack_canary(cpu, idle);
1058 
1059 	/* Initialize the interrupt stack(s) */
1060 	ret = irq_init_percpu_irqstack(cpu);
1061 	if (ret)
1062 		return ret;
1063 
1064 #ifdef CONFIG_X86_32
1065 	/* Stack for startup_32 can be just as for start_secondary onwards */
1066 	per_cpu(pcpu_hot.top_of_stack, cpu) = task_top_of_stack(idle);
1067 #endif
1068 	return 0;
1069 }
1070 
1071 /*
1072  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
1073  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
1074  * Returns zero if CPU booted OK, else error code from
1075  * ->wakeup_secondary_cpu.
1076  */
1077 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle,
1078 		       int *cpu0_nmi_registered)
1079 {
1080 	/* start_ip had better be page-aligned! */
1081 	unsigned long start_ip = real_mode_header->trampoline_start;
1082 
1083 	unsigned long boot_error = 0;
1084 	unsigned long timeout;
1085 
1086 #ifdef CONFIG_X86_64
1087 	/* If 64-bit wakeup method exists, use the 64-bit mode trampoline IP */
1088 	if (apic->wakeup_secondary_cpu_64)
1089 		start_ip = real_mode_header->trampoline_start64;
1090 #endif
1091 	idle->thread.sp = (unsigned long)task_pt_regs(idle);
1092 	initial_code = (unsigned long)start_secondary;
1093 
1094 	if (IS_ENABLED(CONFIG_X86_32)) {
1095 		early_gdt_descr.address = (unsigned long)get_cpu_gdt_rw(cpu);
1096 		initial_stack  = idle->thread.sp;
1097 	} else {
1098 		smpboot_control = cpu;
1099 	}
1100 
1101 	/* Enable the espfix hack for this CPU */
1102 	init_espfix_ap(cpu);
1103 
1104 	/* So we see what's up */
1105 	announce_cpu(cpu, apicid);
1106 
1107 	/*
1108 	 * This grunge runs the startup process for
1109 	 * the targeted processor.
1110 	 */
1111 
1112 	if (x86_platform.legacy.warm_reset) {
1113 
1114 		pr_debug("Setting warm reset code and vector.\n");
1115 
1116 		smpboot_setup_warm_reset_vector(start_ip);
1117 		/*
1118 		 * Be paranoid about clearing APIC errors.
1119 		*/
1120 		if (APIC_INTEGRATED(boot_cpu_apic_version)) {
1121 			apic_write(APIC_ESR, 0);
1122 			apic_read(APIC_ESR);
1123 		}
1124 	}
1125 
1126 	/*
1127 	 * AP might wait on cpu_callout_mask in cpu_init() with
1128 	 * cpu_initialized_mask set if previous attempt to online
1129 	 * it timed-out. Clear cpu_initialized_mask so that after
1130 	 * INIT/SIPI it could start with a clean state.
1131 	 */
1132 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1133 	smp_mb();
1134 
1135 	/*
1136 	 * Wake up a CPU in difference cases:
1137 	 * - Use a method from the APIC driver if one defined, with wakeup
1138 	 *   straight to 64-bit mode preferred over wakeup to RM.
1139 	 * Otherwise,
1140 	 * - Use an INIT boot APIC message for APs or NMI for BSP.
1141 	 */
1142 	if (apic->wakeup_secondary_cpu_64)
1143 		boot_error = apic->wakeup_secondary_cpu_64(apicid, start_ip);
1144 	else if (apic->wakeup_secondary_cpu)
1145 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
1146 	else
1147 		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
1148 						     cpu0_nmi_registered);
1149 
1150 	if (!boot_error) {
1151 		/*
1152 		 * Wait 10s total for first sign of life from AP
1153 		 */
1154 		boot_error = -1;
1155 		timeout = jiffies + 10*HZ;
1156 		while (time_before(jiffies, timeout)) {
1157 			if (cpumask_test_cpu(cpu, cpu_initialized_mask)) {
1158 				/*
1159 				 * Tell AP to proceed with initialization
1160 				 */
1161 				cpumask_set_cpu(cpu, cpu_callout_mask);
1162 				boot_error = 0;
1163 				break;
1164 			}
1165 			schedule();
1166 		}
1167 	}
1168 
1169 	if (!boot_error) {
1170 		/*
1171 		 * Wait till AP completes initial initialization
1172 		 */
1173 		while (!cpumask_test_cpu(cpu, cpu_callin_mask)) {
1174 			/*
1175 			 * Allow other tasks to run while we wait for the
1176 			 * AP to come online. This also gives a chance
1177 			 * for the MTRR work(triggered by the AP coming online)
1178 			 * to be completed in the stop machine context.
1179 			 */
1180 			schedule();
1181 		}
1182 	}
1183 
1184 	if (x86_platform.legacy.warm_reset) {
1185 		/*
1186 		 * Cleanup possible dangling ends...
1187 		 */
1188 		smpboot_restore_warm_reset_vector();
1189 	}
1190 
1191 	return boot_error;
1192 }
1193 
1194 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
1195 {
1196 	int apicid = apic->cpu_present_to_apicid(cpu);
1197 	int cpu0_nmi_registered = 0;
1198 	unsigned long flags;
1199 	int err, ret = 0;
1200 
1201 	lockdep_assert_irqs_enabled();
1202 
1203 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
1204 
1205 	if (apicid == BAD_APICID ||
1206 	    !physid_isset(apicid, phys_cpu_present_map) ||
1207 	    !apic->apic_id_valid(apicid)) {
1208 		pr_err("%s: bad cpu %d\n", __func__, cpu);
1209 		return -EINVAL;
1210 	}
1211 
1212 	/*
1213 	 * Already booted CPU?
1214 	 */
1215 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
1216 		pr_debug("do_boot_cpu %d Already started\n", cpu);
1217 		return -ENOSYS;
1218 	}
1219 
1220 	/*
1221 	 * Save current MTRR state in case it was changed since early boot
1222 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
1223 	 */
1224 	mtrr_save_state();
1225 
1226 	/* x86 CPUs take themselves offline, so delayed offline is OK. */
1227 	err = cpu_check_up_prepare(cpu);
1228 	if (err && err != -EBUSY)
1229 		return err;
1230 
1231 	/* the FPU context is blank, nobody can own it */
1232 	per_cpu(fpu_fpregs_owner_ctx, cpu) = NULL;
1233 
1234 	err = common_cpu_up(cpu, tidle);
1235 	if (err)
1236 		return err;
1237 
1238 	err = do_boot_cpu(apicid, cpu, tidle, &cpu0_nmi_registered);
1239 	if (err) {
1240 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
1241 		ret = -EIO;
1242 		goto unreg_nmi;
1243 	}
1244 
1245 	/*
1246 	 * Check TSC synchronization with the AP (keep irqs disabled
1247 	 * while doing so):
1248 	 */
1249 	local_irq_save(flags);
1250 	check_tsc_sync_source(cpu);
1251 	local_irq_restore(flags);
1252 
1253 	while (!cpu_online(cpu)) {
1254 		cpu_relax();
1255 		touch_nmi_watchdog();
1256 	}
1257 
1258 unreg_nmi:
1259 	/*
1260 	 * Clean up the nmi handler. Do this after the callin and callout sync
1261 	 * to avoid impact of possible long unregister time.
1262 	 */
1263 	if (cpu0_nmi_registered)
1264 		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
1265 
1266 	return ret;
1267 }
1268 
1269 /**
1270  * arch_disable_smp_support() - disables SMP support for x86 at runtime
1271  */
1272 void arch_disable_smp_support(void)
1273 {
1274 	disable_ioapic_support();
1275 }
1276 
1277 /*
1278  * Fall back to non SMP mode after errors.
1279  *
1280  * RED-PEN audit/test this more. I bet there is more state messed up here.
1281  */
1282 static __init void disable_smp(void)
1283 {
1284 	pr_info("SMP disabled\n");
1285 
1286 	disable_ioapic_support();
1287 
1288 	init_cpu_present(cpumask_of(0));
1289 	init_cpu_possible(cpumask_of(0));
1290 
1291 	if (smp_found_config)
1292 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1293 	else
1294 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1295 	cpumask_set_cpu(0, topology_sibling_cpumask(0));
1296 	cpumask_set_cpu(0, topology_core_cpumask(0));
1297 	cpumask_set_cpu(0, topology_die_cpumask(0));
1298 }
1299 
1300 /*
1301  * Various sanity checks.
1302  */
1303 static void __init smp_sanity_check(void)
1304 {
1305 	preempt_disable();
1306 
1307 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
1308 	if (def_to_bigsmp && nr_cpu_ids > 8) {
1309 		unsigned int cpu;
1310 		unsigned nr;
1311 
1312 		pr_warn("More than 8 CPUs detected - skipping them\n"
1313 			"Use CONFIG_X86_BIGSMP\n");
1314 
1315 		nr = 0;
1316 		for_each_present_cpu(cpu) {
1317 			if (nr >= 8)
1318 				set_cpu_present(cpu, false);
1319 			nr++;
1320 		}
1321 
1322 		nr = 0;
1323 		for_each_possible_cpu(cpu) {
1324 			if (nr >= 8)
1325 				set_cpu_possible(cpu, false);
1326 			nr++;
1327 		}
1328 
1329 		set_nr_cpu_ids(8);
1330 	}
1331 #endif
1332 
1333 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1334 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1335 			hard_smp_processor_id());
1336 
1337 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1338 	}
1339 
1340 	/*
1341 	 * Should not be necessary because the MP table should list the boot
1342 	 * CPU too, but we do it for the sake of robustness anyway.
1343 	 */
1344 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1345 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1346 			  boot_cpu_physical_apicid);
1347 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1348 	}
1349 	preempt_enable();
1350 }
1351 
1352 static void __init smp_cpu_index_default(void)
1353 {
1354 	int i;
1355 	struct cpuinfo_x86 *c;
1356 
1357 	for_each_possible_cpu(i) {
1358 		c = &cpu_data(i);
1359 		/* mark all to hotplug */
1360 		c->cpu_index = nr_cpu_ids;
1361 	}
1362 }
1363 
1364 static void __init smp_get_logical_apicid(void)
1365 {
1366 	if (x2apic_mode)
1367 		cpu0_logical_apicid = apic_read(APIC_LDR);
1368 	else
1369 		cpu0_logical_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1370 }
1371 
1372 void __init smp_prepare_cpus_common(void)
1373 {
1374 	unsigned int i;
1375 
1376 	smp_cpu_index_default();
1377 
1378 	/*
1379 	 * Setup boot CPU information
1380 	 */
1381 	smp_store_boot_cpu_info(); /* Final full version of the data */
1382 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1383 	mb();
1384 
1385 	for_each_possible_cpu(i) {
1386 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1387 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1388 		zalloc_cpumask_var(&per_cpu(cpu_die_map, i), GFP_KERNEL);
1389 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1390 		zalloc_cpumask_var(&per_cpu(cpu_l2c_shared_map, i), GFP_KERNEL);
1391 	}
1392 
1393 	/*
1394 	 * Set 'default' x86 topology, this matches default_topology() in that
1395 	 * it has NUMA nodes as a topology level. See also
1396 	 * native_smp_cpus_done().
1397 	 *
1398 	 * Must be done before set_cpus_sibling_map() is ran.
1399 	 */
1400 	set_sched_topology(x86_topology);
1401 
1402 	set_cpu_sibling_map(0);
1403 }
1404 
1405 /*
1406  * Prepare for SMP bootup.
1407  * @max_cpus: configured maximum number of CPUs, It is a legacy parameter
1408  *            for common interface support.
1409  */
1410 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1411 {
1412 	smp_prepare_cpus_common();
1413 
1414 	smp_sanity_check();
1415 
1416 	switch (apic_intr_mode) {
1417 	case APIC_PIC:
1418 	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1419 		disable_smp();
1420 		return;
1421 	case APIC_SYMMETRIC_IO_NO_ROUTING:
1422 		disable_smp();
1423 		/* Setup local timer */
1424 		x86_init.timers.setup_percpu_clockev();
1425 		return;
1426 	case APIC_VIRTUAL_WIRE:
1427 	case APIC_SYMMETRIC_IO:
1428 		break;
1429 	}
1430 
1431 	/* Setup local timer */
1432 	x86_init.timers.setup_percpu_clockev();
1433 
1434 	smp_get_logical_apicid();
1435 
1436 	pr_info("CPU0: ");
1437 	print_cpu_info(&cpu_data(0));
1438 
1439 	uv_system_init();
1440 
1441 	smp_quirk_init_udelay();
1442 
1443 	speculative_store_bypass_ht_init();
1444 
1445 	snp_set_wakeup_secondary_cpu();
1446 }
1447 
1448 void arch_thaw_secondary_cpus_begin(void)
1449 {
1450 	set_cache_aps_delayed_init(true);
1451 }
1452 
1453 void arch_thaw_secondary_cpus_end(void)
1454 {
1455 	cache_aps_init();
1456 }
1457 
1458 /*
1459  * Early setup to make printk work.
1460  */
1461 void __init native_smp_prepare_boot_cpu(void)
1462 {
1463 	int me = smp_processor_id();
1464 
1465 	/* SMP handles this from setup_per_cpu_areas() */
1466 	if (!IS_ENABLED(CONFIG_SMP))
1467 		switch_gdt_and_percpu_base(me);
1468 
1469 	/* already set me in cpu_online_mask in boot_cpu_init() */
1470 	cpumask_set_cpu(me, cpu_callout_mask);
1471 	cpu_set_state_online(me);
1472 	native_pv_lock_init();
1473 }
1474 
1475 void __init calculate_max_logical_packages(void)
1476 {
1477 	int ncpus;
1478 
1479 	/*
1480 	 * Today neither Intel nor AMD support heterogeneous systems so
1481 	 * extrapolate the boot cpu's data to all packages.
1482 	 */
1483 	ncpus = cpu_data(0).booted_cores * topology_max_smt_threads();
1484 	__max_logical_packages = DIV_ROUND_UP(total_cpus, ncpus);
1485 	pr_info("Max logical packages: %u\n", __max_logical_packages);
1486 }
1487 
1488 void __init native_smp_cpus_done(unsigned int max_cpus)
1489 {
1490 	pr_debug("Boot done\n");
1491 
1492 	calculate_max_logical_packages();
1493 
1494 	/* XXX for now assume numa-in-package and hybrid don't overlap */
1495 	if (x86_has_numa_in_package)
1496 		set_sched_topology(x86_numa_in_package_topology);
1497 	if (cpu_feature_enabled(X86_FEATURE_HYBRID_CPU))
1498 		set_sched_topology(x86_hybrid_topology);
1499 
1500 	nmi_selftest();
1501 	impress_friends();
1502 	cache_aps_init();
1503 }
1504 
1505 static int __initdata setup_possible_cpus = -1;
1506 static int __init _setup_possible_cpus(char *str)
1507 {
1508 	get_option(&str, &setup_possible_cpus);
1509 	return 0;
1510 }
1511 early_param("possible_cpus", _setup_possible_cpus);
1512 
1513 
1514 /*
1515  * cpu_possible_mask should be static, it cannot change as cpu's
1516  * are onlined, or offlined. The reason is per-cpu data-structures
1517  * are allocated by some modules at init time, and don't expect to
1518  * do this dynamically on cpu arrival/departure.
1519  * cpu_present_mask on the other hand can change dynamically.
1520  * In case when cpu_hotplug is not compiled, then we resort to current
1521  * behaviour, which is cpu_possible == cpu_present.
1522  * - Ashok Raj
1523  *
1524  * Three ways to find out the number of additional hotplug CPUs:
1525  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1526  * - The user can overwrite it with possible_cpus=NUM
1527  * - Otherwise don't reserve additional CPUs.
1528  * We do this because additional CPUs waste a lot of memory.
1529  * -AK
1530  */
1531 __init void prefill_possible_map(void)
1532 {
1533 	int i, possible;
1534 
1535 	/* No boot processor was found in mptable or ACPI MADT */
1536 	if (!num_processors) {
1537 		if (boot_cpu_has(X86_FEATURE_APIC)) {
1538 			int apicid = boot_cpu_physical_apicid;
1539 			int cpu = hard_smp_processor_id();
1540 
1541 			pr_warn("Boot CPU (id %d) not listed by BIOS\n", cpu);
1542 
1543 			/* Make sure boot cpu is enumerated */
1544 			if (apic->cpu_present_to_apicid(0) == BAD_APICID &&
1545 			    apic->apic_id_valid(apicid))
1546 				generic_processor_info(apicid, boot_cpu_apic_version);
1547 		}
1548 
1549 		if (!num_processors)
1550 			num_processors = 1;
1551 	}
1552 
1553 	i = setup_max_cpus ?: 1;
1554 	if (setup_possible_cpus == -1) {
1555 		possible = num_processors;
1556 #ifdef CONFIG_HOTPLUG_CPU
1557 		if (setup_max_cpus)
1558 			possible += disabled_cpus;
1559 #else
1560 		if (possible > i)
1561 			possible = i;
1562 #endif
1563 	} else
1564 		possible = setup_possible_cpus;
1565 
1566 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1567 
1568 	/* nr_cpu_ids could be reduced via nr_cpus= */
1569 	if (possible > nr_cpu_ids) {
1570 		pr_warn("%d Processors exceeds NR_CPUS limit of %u\n",
1571 			possible, nr_cpu_ids);
1572 		possible = nr_cpu_ids;
1573 	}
1574 
1575 #ifdef CONFIG_HOTPLUG_CPU
1576 	if (!setup_max_cpus)
1577 #endif
1578 	if (possible > i) {
1579 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1580 			possible, setup_max_cpus);
1581 		possible = i;
1582 	}
1583 
1584 	set_nr_cpu_ids(possible);
1585 
1586 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1587 		possible, max_t(int, possible - num_processors, 0));
1588 
1589 	reset_cpu_possible_mask();
1590 
1591 	for (i = 0; i < possible; i++)
1592 		set_cpu_possible(i, true);
1593 }
1594 
1595 #ifdef CONFIG_HOTPLUG_CPU
1596 
1597 /* Recompute SMT state for all CPUs on offline */
1598 static void recompute_smt_state(void)
1599 {
1600 	int max_threads, cpu;
1601 
1602 	max_threads = 0;
1603 	for_each_online_cpu (cpu) {
1604 		int threads = cpumask_weight(topology_sibling_cpumask(cpu));
1605 
1606 		if (threads > max_threads)
1607 			max_threads = threads;
1608 	}
1609 	__max_smt_threads = max_threads;
1610 }
1611 
1612 static void remove_siblinginfo(int cpu)
1613 {
1614 	int sibling;
1615 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1616 
1617 	for_each_cpu(sibling, topology_core_cpumask(cpu)) {
1618 		cpumask_clear_cpu(cpu, topology_core_cpumask(sibling));
1619 		/*/
1620 		 * last thread sibling in this cpu core going down
1621 		 */
1622 		if (cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
1623 			cpu_data(sibling).booted_cores--;
1624 	}
1625 
1626 	for_each_cpu(sibling, topology_die_cpumask(cpu))
1627 		cpumask_clear_cpu(cpu, topology_die_cpumask(sibling));
1628 
1629 	for_each_cpu(sibling, topology_sibling_cpumask(cpu)) {
1630 		cpumask_clear_cpu(cpu, topology_sibling_cpumask(sibling));
1631 		if (cpumask_weight(topology_sibling_cpumask(sibling)) == 1)
1632 			cpu_data(sibling).smt_active = false;
1633 	}
1634 
1635 	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1636 		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1637 	for_each_cpu(sibling, cpu_l2c_shared_mask(cpu))
1638 		cpumask_clear_cpu(cpu, cpu_l2c_shared_mask(sibling));
1639 	cpumask_clear(cpu_llc_shared_mask(cpu));
1640 	cpumask_clear(cpu_l2c_shared_mask(cpu));
1641 	cpumask_clear(topology_sibling_cpumask(cpu));
1642 	cpumask_clear(topology_core_cpumask(cpu));
1643 	cpumask_clear(topology_die_cpumask(cpu));
1644 	c->cpu_core_id = 0;
1645 	c->booted_cores = 0;
1646 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1647 	recompute_smt_state();
1648 }
1649 
1650 static void remove_cpu_from_maps(int cpu)
1651 {
1652 	set_cpu_online(cpu, false);
1653 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1654 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1655 	/* was set by cpu_init() */
1656 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1657 	numa_remove_cpu(cpu);
1658 }
1659 
1660 void cpu_disable_common(void)
1661 {
1662 	int cpu = smp_processor_id();
1663 
1664 	remove_siblinginfo(cpu);
1665 
1666 	/* It's now safe to remove this processor from the online map */
1667 	lock_vector_lock();
1668 	remove_cpu_from_maps(cpu);
1669 	unlock_vector_lock();
1670 	fixup_irqs();
1671 	lapic_offline();
1672 }
1673 
1674 int native_cpu_disable(void)
1675 {
1676 	int ret;
1677 
1678 	ret = lapic_can_unplug_cpu();
1679 	if (ret)
1680 		return ret;
1681 
1682 	cpu_disable_common();
1683 
1684         /*
1685          * Disable the local APIC. Otherwise IPI broadcasts will reach
1686          * it. It still responds normally to INIT, NMI, SMI, and SIPI
1687          * messages.
1688          *
1689          * Disabling the APIC must happen after cpu_disable_common()
1690          * which invokes fixup_irqs().
1691          *
1692          * Disabling the APIC preserves already set bits in IRR, but
1693          * an interrupt arriving after disabling the local APIC does not
1694          * set the corresponding IRR bit.
1695          *
1696          * fixup_irqs() scans IRR for set bits so it can raise a not
1697          * yet handled interrupt on the new destination CPU via an IPI
1698          * but obviously it can't do so for IRR bits which are not set.
1699          * IOW, interrupts arriving after disabling the local APIC will
1700          * be lost.
1701          */
1702 	apic_soft_disable();
1703 
1704 	return 0;
1705 }
1706 
1707 int common_cpu_die(unsigned int cpu)
1708 {
1709 	int ret = 0;
1710 
1711 	/* We don't do anything here: idle task is faking death itself. */
1712 
1713 	/* They ack this in play_dead() by setting CPU_DEAD */
1714 	if (cpu_wait_death(cpu, 5)) {
1715 		if (system_state == SYSTEM_RUNNING)
1716 			pr_info("CPU %u is now offline\n", cpu);
1717 	} else {
1718 		pr_err("CPU %u didn't die...\n", cpu);
1719 		ret = -1;
1720 	}
1721 
1722 	return ret;
1723 }
1724 
1725 void native_cpu_die(unsigned int cpu)
1726 {
1727 	common_cpu_die(cpu);
1728 }
1729 
1730 void play_dead_common(void)
1731 {
1732 	idle_task_exit();
1733 
1734 	/* Ack it */
1735 	(void)cpu_report_death();
1736 
1737 	/*
1738 	 * With physical CPU hotplug, we should halt the cpu
1739 	 */
1740 	local_irq_disable();
1741 }
1742 
1743 /**
1744  * cond_wakeup_cpu0 - Wake up CPU0 if needed.
1745  *
1746  * If NMI wants to wake up CPU0, start CPU0.
1747  */
1748 void cond_wakeup_cpu0(void)
1749 {
1750 	if (smp_processor_id() == 0 && enable_start_cpu0)
1751 		start_cpu0();
1752 }
1753 EXPORT_SYMBOL_GPL(cond_wakeup_cpu0);
1754 
1755 /*
1756  * We need to flush the caches before going to sleep, lest we have
1757  * dirty data in our caches when we come back up.
1758  */
1759 static inline void mwait_play_dead(void)
1760 {
1761 	unsigned int eax, ebx, ecx, edx;
1762 	unsigned int highest_cstate = 0;
1763 	unsigned int highest_subcstate = 0;
1764 	void *mwait_ptr;
1765 	int i;
1766 
1767 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1768 	    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
1769 		return;
1770 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1771 		return;
1772 	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1773 		return;
1774 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1775 		return;
1776 
1777 	eax = CPUID_MWAIT_LEAF;
1778 	ecx = 0;
1779 	native_cpuid(&eax, &ebx, &ecx, &edx);
1780 
1781 	/*
1782 	 * eax will be 0 if EDX enumeration is not valid.
1783 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1784 	 */
1785 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1786 		eax = 0;
1787 	} else {
1788 		edx >>= MWAIT_SUBSTATE_SIZE;
1789 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1790 			if (edx & MWAIT_SUBSTATE_MASK) {
1791 				highest_cstate = i;
1792 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1793 			}
1794 		}
1795 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1796 			(highest_subcstate - 1);
1797 	}
1798 
1799 	/*
1800 	 * This should be a memory location in a cache line which is
1801 	 * unlikely to be touched by other processors.  The actual
1802 	 * content is immaterial as it is not actually modified in any way.
1803 	 */
1804 	mwait_ptr = &current_thread_info()->flags;
1805 
1806 	wbinvd();
1807 
1808 	while (1) {
1809 		/*
1810 		 * The CLFLUSH is a workaround for erratum AAI65 for
1811 		 * the Xeon 7400 series.  It's not clear it is actually
1812 		 * needed, but it should be harmless in either case.
1813 		 * The WBINVD is insufficient due to the spurious-wakeup
1814 		 * case where we return around the loop.
1815 		 */
1816 		mb();
1817 		clflush(mwait_ptr);
1818 		mb();
1819 		__monitor(mwait_ptr, 0, 0);
1820 		mb();
1821 		__mwait(eax, 0);
1822 
1823 		cond_wakeup_cpu0();
1824 	}
1825 }
1826 
1827 void __noreturn hlt_play_dead(void)
1828 {
1829 	if (__this_cpu_read(cpu_info.x86) >= 4)
1830 		wbinvd();
1831 
1832 	while (1) {
1833 		native_halt();
1834 
1835 		cond_wakeup_cpu0();
1836 	}
1837 }
1838 
1839 void native_play_dead(void)
1840 {
1841 	play_dead_common();
1842 	tboot_shutdown(TB_SHUTDOWN_WFS);
1843 
1844 	mwait_play_dead();
1845 	if (cpuidle_play_dead())
1846 		hlt_play_dead();
1847 }
1848 
1849 #else /* ... !CONFIG_HOTPLUG_CPU */
1850 int native_cpu_disable(void)
1851 {
1852 	return -ENOSYS;
1853 }
1854 
1855 void native_cpu_die(unsigned int cpu)
1856 {
1857 	/* We said "no" in __cpu_disable */
1858 	BUG();
1859 }
1860 
1861 void native_play_dead(void)
1862 {
1863 	BUG();
1864 }
1865 
1866 #endif
1867