xref: /linux/arch/x86/kernel/smpboot.c (revision d4f28f07)
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 /* All of these masks are initialized in setup_cpu_local_masks() */
105 static cpumask_var_t cpu_initialized_mask;
106 static cpumask_var_t cpu_callout_mask;
107 static cpumask_var_t cpu_callin_mask;
108 /* Representing CPUs for which sibling maps can be computed */
109 static cpumask_var_t cpu_sibling_setup_mask;
110 
111 /* Logical package management. We might want to allocate that dynamically */
112 unsigned int __max_logical_packages __read_mostly;
113 EXPORT_SYMBOL(__max_logical_packages);
114 static unsigned int logical_packages __read_mostly;
115 static unsigned int logical_die __read_mostly;
116 
117 /* Maximum number of SMT threads on any online core */
118 int __read_mostly __max_smt_threads = 1;
119 
120 /* Flag to indicate if a complete sched domain rebuild is required */
121 bool x86_topology_update;
122 
123 int arch_update_cpu_topology(void)
124 {
125 	int retval = x86_topology_update;
126 
127 	x86_topology_update = false;
128 	return retval;
129 }
130 
131 
132 static unsigned int smpboot_warm_reset_vector_count;
133 
134 static inline void smpboot_setup_warm_reset_vector(unsigned long start_eip)
135 {
136 	unsigned long flags;
137 
138 	spin_lock_irqsave(&rtc_lock, flags);
139 	if (!smpboot_warm_reset_vector_count++) {
140 		CMOS_WRITE(0xa, 0xf);
141 		*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_HIGH)) = start_eip >> 4;
142 		*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = start_eip & 0xf;
143 	}
144 	spin_unlock_irqrestore(&rtc_lock, flags);
145 }
146 
147 static inline void smpboot_restore_warm_reset_vector(void)
148 {
149 	unsigned long flags;
150 
151 	/*
152 	 * Paranoid:  Set warm reset code and vector here back
153 	 * to default values.
154 	 */
155 	spin_lock_irqsave(&rtc_lock, flags);
156 	if (!--smpboot_warm_reset_vector_count) {
157 		CMOS_WRITE(0, 0xf);
158 		*((volatile u32 *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = 0;
159 	}
160 	spin_unlock_irqrestore(&rtc_lock, flags);
161 
162 }
163 
164 /*
165  * Report back to the Boot Processor during boot time or to the caller processor
166  * during CPU online.
167  */
168 static void smp_callin(void)
169 {
170 	int cpuid;
171 
172 	/*
173 	 * If waken up by an INIT in an 82489DX configuration
174 	 * cpu_callout_mask guarantees we don't get here before
175 	 * an INIT_deassert IPI reaches our local APIC, so it is
176 	 * now safe to touch our local APIC.
177 	 */
178 	cpuid = smp_processor_id();
179 
180 	/*
181 	 * the boot CPU has finished the init stage and is spinning
182 	 * on callin_map until we finish. We are free to set up this
183 	 * CPU, first the APIC. (this is probably redundant on most
184 	 * boards)
185 	 */
186 	apic_ap_setup();
187 
188 	/* Save our processor parameters. */
189 	smp_store_cpu_info(cpuid);
190 
191 	/*
192 	 * The topology information must be up to date before
193 	 * notify_cpu_starting().
194 	 */
195 	set_cpu_sibling_map(raw_smp_processor_id());
196 
197 	ap_init_aperfmperf();
198 
199 	pr_debug("Stack at about %p\n", &cpuid);
200 
201 	wmb();
202 
203 	/*
204 	 * This runs the AP through all the cpuhp states to its target
205 	 * state CPUHP_ONLINE.
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 void ap_calibrate_delay(void)
216 {
217 	/*
218 	 * Calibrate the delay loop and update loops_per_jiffy in cpu_data.
219 	 * smp_store_cpu_info() stored a value that is close but not as
220 	 * accurate as the value just calculated.
221 	 *
222 	 * As this is invoked after the TSC synchronization check,
223 	 * calibrate_delay_is_known() will skip the calibration routine
224 	 * when TSC is synchronized across sockets.
225 	 */
226 	calibrate_delay();
227 	cpu_data(smp_processor_id()).loops_per_jiffy = loops_per_jiffy;
228 }
229 
230 static void wait_for_master_cpu(int cpu)
231 {
232 	/*
233 	 * Wait for release by control CPU before continuing with AP
234 	 * initialization.
235 	 */
236 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
237 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
238 		cpu_relax();
239 }
240 
241 /*
242  * Activate a secondary processor.
243  */
244 static void notrace start_secondary(void *unused)
245 {
246 	/*
247 	 * Don't put *anything* except direct CPU state initialization
248 	 * before cpu_init(), SMP booting is too fragile that we want to
249 	 * limit the things done here to the most necessary things.
250 	 */
251 	cr4_init();
252 
253 #ifdef CONFIG_X86_32
254 	/* switch away from the initial page table */
255 	load_cr3(swapper_pg_dir);
256 	__flush_tlb_all();
257 #endif
258 	cpu_init_exception_handling();
259 
260 	/*
261 	 * Sync point with wait_cpu_initialized(). Sets AP in
262 	 * cpu_initialized_mask and then waits for the control CPU
263 	 * to release it.
264 	 */
265 	wait_for_master_cpu(raw_smp_processor_id());
266 
267 	cpu_init();
268 	rcu_cpu_starting(raw_smp_processor_id());
269 	x86_cpuinit.early_percpu_clock_init();
270 
271 	/*
272 	 * Sync point with wait_cpu_callin(). The AP doesn't wait here
273 	 * but just sets the bit to let the controlling CPU (BSP) know that
274 	 * it's got this far.
275 	 */
276 	smp_callin();
277 
278 	/*
279 	 * Check TSC synchronization with the control CPU, which will do
280 	 * its part of this from wait_cpu_online(), making it an implicit
281 	 * synchronization point.
282 	 */
283 	check_tsc_sync_target();
284 
285 	/*
286 	 * Calibrate the delay loop after the TSC synchronization check.
287 	 * This allows to skip the calibration when TSC is synchronized
288 	 * across sockets.
289 	 */
290 	ap_calibrate_delay();
291 
292 	speculative_store_bypass_ht_init();
293 
294 	/*
295 	 * Lock vector_lock, set CPU online and bring the vector
296 	 * allocator online. Online must be set with vector_lock held
297 	 * to prevent a concurrent irq setup/teardown from seeing a
298 	 * half valid vector space.
299 	 */
300 	lock_vector_lock();
301 	/* Sync point with do_wait_cpu_online() */
302 	set_cpu_online(smp_processor_id(), true);
303 	lapic_online();
304 	unlock_vector_lock();
305 	cpu_set_state_online(smp_processor_id());
306 	x86_platform.nmi_init();
307 
308 	/* enable local interrupts */
309 	local_irq_enable();
310 
311 	x86_cpuinit.setup_percpu_clockev();
312 
313 	wmb();
314 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
315 }
316 
317 /**
318  * topology_is_primary_thread - Check whether CPU is the primary SMT thread
319  * @cpu:	CPU to check
320  */
321 bool topology_is_primary_thread(unsigned int cpu)
322 {
323 	return apic_id_is_primary_thread(per_cpu(x86_cpu_to_apicid, cpu));
324 }
325 
326 /**
327  * topology_smt_supported - Check whether SMT is supported by the CPUs
328  */
329 bool topology_smt_supported(void)
330 {
331 	return smp_num_siblings > 1;
332 }
333 
334 /**
335  * topology_phys_to_logical_pkg - Map a physical package id to a logical
336  * @phys_pkg:	The physical package id to map
337  *
338  * Returns logical package id or -1 if not found
339  */
340 int topology_phys_to_logical_pkg(unsigned int phys_pkg)
341 {
342 	int cpu;
343 
344 	for_each_possible_cpu(cpu) {
345 		struct cpuinfo_x86 *c = &cpu_data(cpu);
346 
347 		if (c->initialized && c->phys_proc_id == phys_pkg)
348 			return c->logical_proc_id;
349 	}
350 	return -1;
351 }
352 EXPORT_SYMBOL(topology_phys_to_logical_pkg);
353 
354 /**
355  * topology_phys_to_logical_die - Map a physical die id to logical
356  * @die_id:	The physical die id to map
357  * @cur_cpu:	The CPU for which the mapping is done
358  *
359  * Returns logical die id or -1 if not found
360  */
361 static int topology_phys_to_logical_die(unsigned int die_id, unsigned int cur_cpu)
362 {
363 	int cpu, proc_id = cpu_data(cur_cpu).phys_proc_id;
364 
365 	for_each_possible_cpu(cpu) {
366 		struct cpuinfo_x86 *c = &cpu_data(cpu);
367 
368 		if (c->initialized && c->cpu_die_id == die_id &&
369 		    c->phys_proc_id == proc_id)
370 			return c->logical_die_id;
371 	}
372 	return -1;
373 }
374 
375 /**
376  * topology_update_package_map - Update the physical to logical package map
377  * @pkg:	The physical package id as retrieved via CPUID
378  * @cpu:	The cpu for which this is updated
379  */
380 int topology_update_package_map(unsigned int pkg, unsigned int cpu)
381 {
382 	int new;
383 
384 	/* Already available somewhere? */
385 	new = topology_phys_to_logical_pkg(pkg);
386 	if (new >= 0)
387 		goto found;
388 
389 	new = logical_packages++;
390 	if (new != pkg) {
391 		pr_info("CPU %u Converting physical %u to logical package %u\n",
392 			cpu, pkg, new);
393 	}
394 found:
395 	cpu_data(cpu).logical_proc_id = new;
396 	return 0;
397 }
398 /**
399  * topology_update_die_map - Update the physical to logical die map
400  * @die:	The die id as retrieved via CPUID
401  * @cpu:	The cpu for which this is updated
402  */
403 int topology_update_die_map(unsigned int die, unsigned int cpu)
404 {
405 	int new;
406 
407 	/* Already available somewhere? */
408 	new = topology_phys_to_logical_die(die, cpu);
409 	if (new >= 0)
410 		goto found;
411 
412 	new = logical_die++;
413 	if (new != die) {
414 		pr_info("CPU %u Converting physical %u to logical die %u\n",
415 			cpu, die, new);
416 	}
417 found:
418 	cpu_data(cpu).logical_die_id = new;
419 	return 0;
420 }
421 
422 void __init smp_store_boot_cpu_info(void)
423 {
424 	int id = 0; /* CPU 0 */
425 	struct cpuinfo_x86 *c = &cpu_data(id);
426 
427 	*c = boot_cpu_data;
428 	c->cpu_index = id;
429 	topology_update_package_map(c->phys_proc_id, id);
430 	topology_update_die_map(c->cpu_die_id, id);
431 	c->initialized = true;
432 }
433 
434 /*
435  * The bootstrap kernel entry code has set these up. Save them for
436  * a given CPU
437  */
438 void smp_store_cpu_info(int id)
439 {
440 	struct cpuinfo_x86 *c = &cpu_data(id);
441 
442 	/* Copy boot_cpu_data only on the first bringup */
443 	if (!c->initialized)
444 		*c = boot_cpu_data;
445 	c->cpu_index = id;
446 	/*
447 	 * During boot time, CPU0 has this setup already. Save the info when
448 	 * bringing up an AP.
449 	 */
450 	identify_secondary_cpu(c);
451 	c->initialized = true;
452 }
453 
454 static bool
455 topology_same_node(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
456 {
457 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
458 
459 	return (cpu_to_node(cpu1) == cpu_to_node(cpu2));
460 }
461 
462 static bool
463 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
464 {
465 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
466 
467 	return !WARN_ONCE(!topology_same_node(c, o),
468 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
469 		"[node: %d != %d]. Ignoring dependency.\n",
470 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
471 }
472 
473 #define link_mask(mfunc, c1, c2)					\
474 do {									\
475 	cpumask_set_cpu((c1), mfunc(c2));				\
476 	cpumask_set_cpu((c2), mfunc(c1));				\
477 } while (0)
478 
479 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
480 {
481 	if (boot_cpu_has(X86_FEATURE_TOPOEXT)) {
482 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
483 
484 		if (c->phys_proc_id == o->phys_proc_id &&
485 		    c->cpu_die_id == o->cpu_die_id &&
486 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2)) {
487 			if (c->cpu_core_id == o->cpu_core_id)
488 				return topology_sane(c, o, "smt");
489 
490 			if ((c->cu_id != 0xff) &&
491 			    (o->cu_id != 0xff) &&
492 			    (c->cu_id == o->cu_id))
493 				return topology_sane(c, o, "smt");
494 		}
495 
496 	} else if (c->phys_proc_id == o->phys_proc_id &&
497 		   c->cpu_die_id == o->cpu_die_id &&
498 		   c->cpu_core_id == o->cpu_core_id) {
499 		return topology_sane(c, o, "smt");
500 	}
501 
502 	return false;
503 }
504 
505 static bool match_die(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
506 {
507 	if (c->phys_proc_id == o->phys_proc_id &&
508 	    c->cpu_die_id == o->cpu_die_id)
509 		return true;
510 	return false;
511 }
512 
513 static bool match_l2c(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
514 {
515 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
516 
517 	/* If the arch didn't set up l2c_id, fall back to SMT */
518 	if (per_cpu(cpu_l2c_id, cpu1) == BAD_APICID)
519 		return match_smt(c, o);
520 
521 	/* Do not match if L2 cache id does not match: */
522 	if (per_cpu(cpu_l2c_id, cpu1) != per_cpu(cpu_l2c_id, cpu2))
523 		return false;
524 
525 	return topology_sane(c, o, "l2c");
526 }
527 
528 /*
529  * Unlike the other levels, we do not enforce keeping a
530  * multicore group inside a NUMA node.  If this happens, we will
531  * discard the MC level of the topology later.
532  */
533 static bool match_pkg(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
534 {
535 	if (c->phys_proc_id == o->phys_proc_id)
536 		return true;
537 	return false;
538 }
539 
540 /*
541  * Define intel_cod_cpu[] for Intel COD (Cluster-on-Die) CPUs.
542  *
543  * Any Intel CPU that has multiple nodes per package and does not
544  * match intel_cod_cpu[] has the SNC (Sub-NUMA Cluster) topology.
545  *
546  * When in SNC mode, these CPUs enumerate an LLC that is shared
547  * by multiple NUMA nodes. The LLC is shared for off-package data
548  * access but private to the NUMA node (half of the package) for
549  * on-package access. CPUID (the source of the information about
550  * the LLC) can only enumerate the cache as shared or unshared,
551  * but not this particular configuration.
552  */
553 
554 static const struct x86_cpu_id intel_cod_cpu[] = {
555 	X86_MATCH_INTEL_FAM6_MODEL(HASWELL_X, 0),	/* COD */
556 	X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_X, 0),	/* COD */
557 	X86_MATCH_INTEL_FAM6_MODEL(ANY, 1),		/* SNC */
558 	{}
559 };
560 
561 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
562 {
563 	const struct x86_cpu_id *id = x86_match_cpu(intel_cod_cpu);
564 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
565 	bool intel_snc = id && id->driver_data;
566 
567 	/* Do not match if we do not have a valid APICID for cpu: */
568 	if (per_cpu(cpu_llc_id, cpu1) == BAD_APICID)
569 		return false;
570 
571 	/* Do not match if LLC id does not match: */
572 	if (per_cpu(cpu_llc_id, cpu1) != per_cpu(cpu_llc_id, cpu2))
573 		return false;
574 
575 	/*
576 	 * Allow the SNC topology without warning. Return of false
577 	 * means 'c' does not share the LLC of 'o'. This will be
578 	 * reflected to userspace.
579 	 */
580 	if (match_pkg(c, o) && !topology_same_node(c, o) && intel_snc)
581 		return false;
582 
583 	return topology_sane(c, o, "llc");
584 }
585 
586 
587 #if defined(CONFIG_SCHED_SMT) || defined(CONFIG_SCHED_CLUSTER) || defined(CONFIG_SCHED_MC)
588 static inline int x86_sched_itmt_flags(void)
589 {
590 	return sysctl_sched_itmt_enabled ? SD_ASYM_PACKING : 0;
591 }
592 
593 #ifdef CONFIG_SCHED_MC
594 static int x86_core_flags(void)
595 {
596 	return cpu_core_flags() | x86_sched_itmt_flags();
597 }
598 #endif
599 #ifdef CONFIG_SCHED_SMT
600 static int x86_smt_flags(void)
601 {
602 	return cpu_smt_flags() | x86_sched_itmt_flags();
603 }
604 #endif
605 #ifdef CONFIG_SCHED_CLUSTER
606 static int x86_cluster_flags(void)
607 {
608 	return cpu_cluster_flags() | x86_sched_itmt_flags();
609 }
610 #endif
611 #endif
612 
613 static struct sched_domain_topology_level x86_numa_in_package_topology[] = {
614 #ifdef CONFIG_SCHED_SMT
615 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
616 #endif
617 #ifdef CONFIG_SCHED_CLUSTER
618 	{ cpu_clustergroup_mask, x86_cluster_flags, SD_INIT_NAME(CLS) },
619 #endif
620 #ifdef CONFIG_SCHED_MC
621 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
622 #endif
623 	{ NULL, },
624 };
625 
626 static struct sched_domain_topology_level x86_hybrid_topology[] = {
627 #ifdef CONFIG_SCHED_SMT
628 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
629 #endif
630 #ifdef CONFIG_SCHED_MC
631 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
632 #endif
633 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
634 	{ NULL, },
635 };
636 
637 static struct sched_domain_topology_level x86_topology[] = {
638 #ifdef CONFIG_SCHED_SMT
639 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
640 #endif
641 #ifdef CONFIG_SCHED_CLUSTER
642 	{ cpu_clustergroup_mask, x86_cluster_flags, SD_INIT_NAME(CLS) },
643 #endif
644 #ifdef CONFIG_SCHED_MC
645 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
646 #endif
647 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
648 	{ NULL, },
649 };
650 
651 /*
652  * Set if a package/die has multiple NUMA nodes inside.
653  * AMD Magny-Cours, Intel Cluster-on-Die, and Intel
654  * Sub-NUMA Clustering have this.
655  */
656 static bool x86_has_numa_in_package;
657 
658 void set_cpu_sibling_map(int cpu)
659 {
660 	bool has_smt = smp_num_siblings > 1;
661 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
662 	struct cpuinfo_x86 *c = &cpu_data(cpu);
663 	struct cpuinfo_x86 *o;
664 	int i, threads;
665 
666 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
667 
668 	if (!has_mp) {
669 		cpumask_set_cpu(cpu, topology_sibling_cpumask(cpu));
670 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
671 		cpumask_set_cpu(cpu, cpu_l2c_shared_mask(cpu));
672 		cpumask_set_cpu(cpu, topology_core_cpumask(cpu));
673 		cpumask_set_cpu(cpu, topology_die_cpumask(cpu));
674 		c->booted_cores = 1;
675 		return;
676 	}
677 
678 	for_each_cpu(i, cpu_sibling_setup_mask) {
679 		o = &cpu_data(i);
680 
681 		if (match_pkg(c, o) && !topology_same_node(c, o))
682 			x86_has_numa_in_package = true;
683 
684 		if ((i == cpu) || (has_smt && match_smt(c, o)))
685 			link_mask(topology_sibling_cpumask, cpu, i);
686 
687 		if ((i == cpu) || (has_mp && match_llc(c, o)))
688 			link_mask(cpu_llc_shared_mask, cpu, i);
689 
690 		if ((i == cpu) || (has_mp && match_l2c(c, o)))
691 			link_mask(cpu_l2c_shared_mask, cpu, i);
692 
693 		if ((i == cpu) || (has_mp && match_die(c, o)))
694 			link_mask(topology_die_cpumask, cpu, i);
695 	}
696 
697 	threads = cpumask_weight(topology_sibling_cpumask(cpu));
698 	if (threads > __max_smt_threads)
699 		__max_smt_threads = threads;
700 
701 	for_each_cpu(i, topology_sibling_cpumask(cpu))
702 		cpu_data(i).smt_active = threads > 1;
703 
704 	/*
705 	 * This needs a separate iteration over the cpus because we rely on all
706 	 * topology_sibling_cpumask links to be set-up.
707 	 */
708 	for_each_cpu(i, cpu_sibling_setup_mask) {
709 		o = &cpu_data(i);
710 
711 		if ((i == cpu) || (has_mp && match_pkg(c, o))) {
712 			link_mask(topology_core_cpumask, cpu, i);
713 
714 			/*
715 			 *  Does this new cpu bringup a new core?
716 			 */
717 			if (threads == 1) {
718 				/*
719 				 * for each core in package, increment
720 				 * the booted_cores for this new cpu
721 				 */
722 				if (cpumask_first(
723 				    topology_sibling_cpumask(i)) == i)
724 					c->booted_cores++;
725 				/*
726 				 * increment the core count for all
727 				 * the other cpus in this package
728 				 */
729 				if (i != cpu)
730 					cpu_data(i).booted_cores++;
731 			} else if (i != cpu && !c->booted_cores)
732 				c->booted_cores = cpu_data(i).booted_cores;
733 		}
734 	}
735 }
736 
737 /* maps the cpu to the sched domain representing multi-core */
738 const struct cpumask *cpu_coregroup_mask(int cpu)
739 {
740 	return cpu_llc_shared_mask(cpu);
741 }
742 
743 const struct cpumask *cpu_clustergroup_mask(int cpu)
744 {
745 	return cpu_l2c_shared_mask(cpu);
746 }
747 
748 static void impress_friends(void)
749 {
750 	int cpu;
751 	unsigned long bogosum = 0;
752 	/*
753 	 * Allow the user to impress friends.
754 	 */
755 	pr_debug("Before bogomips\n");
756 	for_each_possible_cpu(cpu)
757 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
758 			bogosum += cpu_data(cpu).loops_per_jiffy;
759 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
760 		num_online_cpus(),
761 		bogosum/(500000/HZ),
762 		(bogosum/(5000/HZ))%100);
763 
764 	pr_debug("Before bogocount - setting activated=1\n");
765 }
766 
767 void __inquire_remote_apic(int apicid)
768 {
769 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
770 	const char * const names[] = { "ID", "VERSION", "SPIV" };
771 	int timeout;
772 	u32 status;
773 
774 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
775 
776 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
777 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
778 
779 		/*
780 		 * Wait for idle.
781 		 */
782 		status = safe_apic_wait_icr_idle();
783 		if (status)
784 			pr_cont("a previous APIC delivery may have failed\n");
785 
786 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
787 
788 		timeout = 0;
789 		do {
790 			udelay(100);
791 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
792 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
793 
794 		switch (status) {
795 		case APIC_ICR_RR_VALID:
796 			status = apic_read(APIC_RRR);
797 			pr_cont("%08x\n", status);
798 			break;
799 		default:
800 			pr_cont("failed\n");
801 		}
802 	}
803 }
804 
805 /*
806  * The Multiprocessor Specification 1.4 (1997) example code suggests
807  * that there should be a 10ms delay between the BSP asserting INIT
808  * and de-asserting INIT, when starting a remote processor.
809  * But that slows boot and resume on modern processors, which include
810  * many cores and don't require that delay.
811  *
812  * Cmdline "init_cpu_udelay=" is available to over-ride this delay.
813  * Modern processor families are quirked to remove the delay entirely.
814  */
815 #define UDELAY_10MS_DEFAULT 10000
816 
817 static unsigned int init_udelay = UINT_MAX;
818 
819 static int __init cpu_init_udelay(char *str)
820 {
821 	get_option(&str, &init_udelay);
822 
823 	return 0;
824 }
825 early_param("cpu_init_udelay", cpu_init_udelay);
826 
827 static void __init smp_quirk_init_udelay(void)
828 {
829 	/* if cmdline changed it from default, leave it alone */
830 	if (init_udelay != UINT_MAX)
831 		return;
832 
833 	/* if modern processor, use no delay */
834 	if (((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && (boot_cpu_data.x86 == 6)) ||
835 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) && (boot_cpu_data.x86 >= 0x18)) ||
836 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && (boot_cpu_data.x86 >= 0xF))) {
837 		init_udelay = 0;
838 		return;
839 	}
840 	/* else, use legacy delay */
841 	init_udelay = UDELAY_10MS_DEFAULT;
842 }
843 
844 /*
845  * Wake up AP by INIT, INIT, STARTUP sequence.
846  */
847 static int wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
848 {
849 	unsigned long send_status = 0, accept_status = 0;
850 	int maxlvt, num_starts, j;
851 
852 	preempt_disable();
853 	maxlvt = lapic_get_maxlvt();
854 
855 	/*
856 	 * Be paranoid about clearing APIC errors.
857 	 */
858 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
859 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
860 			apic_write(APIC_ESR, 0);
861 		apic_read(APIC_ESR);
862 	}
863 
864 	pr_debug("Asserting INIT\n");
865 
866 	/*
867 	 * Turn INIT on target chip
868 	 */
869 	/*
870 	 * Send IPI
871 	 */
872 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
873 		       phys_apicid);
874 
875 	pr_debug("Waiting for send to finish...\n");
876 	send_status = safe_apic_wait_icr_idle();
877 
878 	udelay(init_udelay);
879 
880 	pr_debug("Deasserting INIT\n");
881 
882 	/* Target chip */
883 	/* Send IPI */
884 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
885 
886 	pr_debug("Waiting for send to finish...\n");
887 	send_status = safe_apic_wait_icr_idle();
888 
889 	mb();
890 
891 	/*
892 	 * Should we send STARTUP IPIs ?
893 	 *
894 	 * Determine this based on the APIC version.
895 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
896 	 */
897 	if (APIC_INTEGRATED(boot_cpu_apic_version))
898 		num_starts = 2;
899 	else
900 		num_starts = 0;
901 
902 	/*
903 	 * Run STARTUP IPI loop.
904 	 */
905 	pr_debug("#startup loops: %d\n", num_starts);
906 
907 	for (j = 1; j <= num_starts; j++) {
908 		pr_debug("Sending STARTUP #%d\n", j);
909 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
910 			apic_write(APIC_ESR, 0);
911 		apic_read(APIC_ESR);
912 		pr_debug("After apic_write\n");
913 
914 		/*
915 		 * STARTUP IPI
916 		 */
917 
918 		/* Target chip */
919 		/* Boot on the stack */
920 		/* Kick the second */
921 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
922 			       phys_apicid);
923 
924 		/*
925 		 * Give the other CPU some time to accept the IPI.
926 		 */
927 		if (init_udelay == 0)
928 			udelay(10);
929 		else
930 			udelay(300);
931 
932 		pr_debug("Startup point 1\n");
933 
934 		pr_debug("Waiting for send to finish...\n");
935 		send_status = safe_apic_wait_icr_idle();
936 
937 		/*
938 		 * Give the other CPU some time to accept the IPI.
939 		 */
940 		if (init_udelay == 0)
941 			udelay(10);
942 		else
943 			udelay(200);
944 
945 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
946 			apic_write(APIC_ESR, 0);
947 		accept_status = (apic_read(APIC_ESR) & 0xEF);
948 		if (send_status || accept_status)
949 			break;
950 	}
951 	pr_debug("After Startup\n");
952 
953 	if (send_status)
954 		pr_err("APIC never delivered???\n");
955 	if (accept_status)
956 		pr_err("APIC delivery error (%lx)\n", accept_status);
957 
958 	preempt_enable();
959 	return (send_status | accept_status);
960 }
961 
962 /* reduce the number of lines printed when booting a large cpu count system */
963 static void announce_cpu(int cpu, int apicid)
964 {
965 	static int current_node = NUMA_NO_NODE;
966 	int node = early_cpu_to_node(cpu);
967 	static int width, node_width;
968 
969 	if (!width)
970 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
971 
972 	if (!node_width)
973 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
974 
975 	if (cpu == 1)
976 		printk(KERN_INFO "x86: Booting SMP configuration:\n");
977 
978 	if (system_state < SYSTEM_RUNNING) {
979 		if (node != current_node) {
980 			if (current_node > (-1))
981 				pr_cont("\n");
982 			current_node = node;
983 
984 			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
985 			       node_width - num_digits(node), " ", node);
986 		}
987 
988 		/* Add padding for the BSP */
989 		if (cpu == 1)
990 			pr_cont("%*s", width + 1, " ");
991 
992 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
993 
994 	} else
995 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
996 			node, cpu, apicid);
997 }
998 
999 int common_cpu_up(unsigned int cpu, struct task_struct *idle)
1000 {
1001 	int ret;
1002 
1003 	/* Just in case we booted with a single CPU. */
1004 	alternatives_enable_smp();
1005 
1006 	per_cpu(pcpu_hot.current_task, cpu) = idle;
1007 	cpu_init_stack_canary(cpu, idle);
1008 
1009 	/* Initialize the interrupt stack(s) */
1010 	ret = irq_init_percpu_irqstack(cpu);
1011 	if (ret)
1012 		return ret;
1013 
1014 #ifdef CONFIG_X86_32
1015 	/* Stack for startup_32 can be just as for start_secondary onwards */
1016 	per_cpu(pcpu_hot.top_of_stack, cpu) = task_top_of_stack(idle);
1017 #endif
1018 	return 0;
1019 }
1020 
1021 /*
1022  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
1023  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
1024  * Returns zero if startup was successfully sent, else error code from
1025  * ->wakeup_secondary_cpu.
1026  */
1027 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
1028 {
1029 	unsigned long start_ip = real_mode_header->trampoline_start;
1030 
1031 #ifdef CONFIG_X86_64
1032 	/* If 64-bit wakeup method exists, use the 64-bit mode trampoline IP */
1033 	if (apic->wakeup_secondary_cpu_64)
1034 		start_ip = real_mode_header->trampoline_start64;
1035 #endif
1036 	idle->thread.sp = (unsigned long)task_pt_regs(idle);
1037 	initial_code = (unsigned long)start_secondary;
1038 
1039 	if (IS_ENABLED(CONFIG_X86_32)) {
1040 		early_gdt_descr.address = (unsigned long)get_cpu_gdt_rw(cpu);
1041 		initial_stack  = idle->thread.sp;
1042 	} else {
1043 		smpboot_control = cpu;
1044 	}
1045 
1046 	/* Enable the espfix hack for this CPU */
1047 	init_espfix_ap(cpu);
1048 
1049 	/* So we see what's up */
1050 	announce_cpu(cpu, apicid);
1051 
1052 	/*
1053 	 * This grunge runs the startup process for
1054 	 * the targeted processor.
1055 	 */
1056 	if (x86_platform.legacy.warm_reset) {
1057 
1058 		pr_debug("Setting warm reset code and vector.\n");
1059 
1060 		smpboot_setup_warm_reset_vector(start_ip);
1061 		/*
1062 		 * Be paranoid about clearing APIC errors.
1063 		*/
1064 		if (APIC_INTEGRATED(boot_cpu_apic_version)) {
1065 			apic_write(APIC_ESR, 0);
1066 			apic_read(APIC_ESR);
1067 		}
1068 	}
1069 
1070 	/*
1071 	 * AP might wait on cpu_callout_mask in cpu_init() with
1072 	 * cpu_initialized_mask set if previous attempt to online
1073 	 * it timed-out. Clear cpu_initialized_mask so that after
1074 	 * INIT/SIPI it could start with a clean state.
1075 	 */
1076 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1077 	smp_mb();
1078 
1079 	/*
1080 	 * Wake up a CPU in difference cases:
1081 	 * - Use a method from the APIC driver if one defined, with wakeup
1082 	 *   straight to 64-bit mode preferred over wakeup to RM.
1083 	 * Otherwise,
1084 	 * - Use an INIT boot APIC message
1085 	 */
1086 	if (apic->wakeup_secondary_cpu_64)
1087 		return apic->wakeup_secondary_cpu_64(apicid, start_ip);
1088 	else if (apic->wakeup_secondary_cpu)
1089 		return apic->wakeup_secondary_cpu(apicid, start_ip);
1090 
1091 	return wakeup_secondary_cpu_via_init(apicid, start_ip);
1092 }
1093 
1094 static int wait_cpu_cpumask(unsigned int cpu, const struct cpumask *mask)
1095 {
1096 	unsigned long timeout;
1097 
1098 	/*
1099 	 * Wait up to 10s for the CPU to report in.
1100 	 */
1101 	timeout = jiffies + 10*HZ;
1102 	while (time_before(jiffies, timeout)) {
1103 		if (cpumask_test_cpu(cpu, mask))
1104 			return 0;
1105 
1106 		schedule();
1107 	}
1108 	return -1;
1109 }
1110 
1111 /*
1112  * Bringup step two: Wait for the target AP to reach cpu_init_secondary()
1113  * and thus wait_for_master_cpu(), then set cpu_callout_mask to allow it
1114  * to proceed.  The AP will then proceed past setting its 'callin' bit
1115  * and end up waiting in check_tsc_sync_target() until we reach
1116  * do_wait_cpu_online() to tend to it.
1117  */
1118 static int wait_cpu_initialized(unsigned int cpu)
1119 {
1120 	/*
1121 	 * Wait for first sign of life from AP.
1122 	 */
1123 	if (wait_cpu_cpumask(cpu, cpu_initialized_mask))
1124 		return -1;
1125 
1126 	cpumask_set_cpu(cpu, cpu_callout_mask);
1127 	return 0;
1128 }
1129 
1130 /*
1131  * Bringup step three: Wait for the target AP to reach smp_callin().
1132  * The AP is not waiting for us here so we don't need to parallelise
1133  * this step. Not entirely clear why we care about this, since we just
1134  * proceed directly to TSC synchronization which is the next sync
1135  * point with the AP anyway.
1136  */
1137 static void wait_cpu_callin(unsigned int cpu)
1138 {
1139 	while (!cpumask_test_cpu(cpu, cpu_callin_mask))
1140 		schedule();
1141 }
1142 
1143 /*
1144  * Bringup step four: Synchronize the TSC and wait for the target AP
1145  * to reach set_cpu_online() in start_secondary().
1146  */
1147 static void wait_cpu_online(unsigned int cpu)
1148 {
1149 	unsigned long flags;
1150 
1151 	/*
1152 	 * Check TSC synchronization with the AP (keep irqs disabled
1153 	 * while doing so):
1154 	 */
1155 	local_irq_save(flags);
1156 	check_tsc_sync_source(cpu);
1157 	local_irq_restore(flags);
1158 
1159 	/*
1160 	 * Wait for the AP to mark itself online, so the core caller
1161 	 * can drop sparse_irq_lock.
1162 	 */
1163 	while (!cpu_online(cpu))
1164 		schedule();
1165 }
1166 
1167 static int native_kick_ap(unsigned int cpu, struct task_struct *tidle)
1168 {
1169 	int apicid = apic->cpu_present_to_apicid(cpu);
1170 	int err;
1171 
1172 	lockdep_assert_irqs_enabled();
1173 
1174 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
1175 
1176 	if (apicid == BAD_APICID ||
1177 	    !physid_isset(apicid, phys_cpu_present_map) ||
1178 	    !apic->apic_id_valid(apicid)) {
1179 		pr_err("%s: bad cpu %d\n", __func__, cpu);
1180 		return -EINVAL;
1181 	}
1182 
1183 	/*
1184 	 * Already booted CPU?
1185 	 */
1186 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
1187 		pr_debug("do_boot_cpu %d Already started\n", cpu);
1188 		return -ENOSYS;
1189 	}
1190 
1191 	/*
1192 	 * Save current MTRR state in case it was changed since early boot
1193 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
1194 	 */
1195 	mtrr_save_state();
1196 
1197 	/* x86 CPUs take themselves offline, so delayed offline is OK. */
1198 	err = cpu_check_up_prepare(cpu);
1199 	if (err && err != -EBUSY)
1200 		return err;
1201 
1202 	/* the FPU context is blank, nobody can own it */
1203 	per_cpu(fpu_fpregs_owner_ctx, cpu) = NULL;
1204 
1205 	err = common_cpu_up(cpu, tidle);
1206 	if (err)
1207 		return err;
1208 
1209 	err = do_boot_cpu(apicid, cpu, tidle);
1210 	if (err)
1211 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
1212 
1213 	return err;
1214 }
1215 
1216 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
1217 {
1218 	int ret;
1219 
1220 	ret = native_kick_ap(cpu, tidle);
1221 	if (ret)
1222 		goto out;
1223 
1224 	ret = wait_cpu_initialized(cpu);
1225 	if (ret)
1226 		goto out;
1227 
1228 	wait_cpu_callin(cpu);
1229 	wait_cpu_online(cpu);
1230 
1231 out:
1232 	/* Cleanup possible dangling ends... */
1233 	if (x86_platform.legacy.warm_reset)
1234 		smpboot_restore_warm_reset_vector();
1235 
1236 	return ret;
1237 }
1238 
1239 /**
1240  * arch_disable_smp_support() - Disables SMP support for x86 at boottime
1241  */
1242 void __init arch_disable_smp_support(void)
1243 {
1244 	disable_ioapic_support();
1245 }
1246 
1247 /*
1248  * Fall back to non SMP mode after errors.
1249  *
1250  * RED-PEN audit/test this more. I bet there is more state messed up here.
1251  */
1252 static __init void disable_smp(void)
1253 {
1254 	pr_info("SMP disabled\n");
1255 
1256 	disable_ioapic_support();
1257 
1258 	init_cpu_present(cpumask_of(0));
1259 	init_cpu_possible(cpumask_of(0));
1260 
1261 	if (smp_found_config)
1262 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1263 	else
1264 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1265 	cpumask_set_cpu(0, topology_sibling_cpumask(0));
1266 	cpumask_set_cpu(0, topology_core_cpumask(0));
1267 	cpumask_set_cpu(0, topology_die_cpumask(0));
1268 }
1269 
1270 /*
1271  * Various sanity checks.
1272  */
1273 static void __init smp_sanity_check(void)
1274 {
1275 	preempt_disable();
1276 
1277 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
1278 	if (def_to_bigsmp && nr_cpu_ids > 8) {
1279 		unsigned int cpu;
1280 		unsigned nr;
1281 
1282 		pr_warn("More than 8 CPUs detected - skipping them\n"
1283 			"Use CONFIG_X86_BIGSMP\n");
1284 
1285 		nr = 0;
1286 		for_each_present_cpu(cpu) {
1287 			if (nr >= 8)
1288 				set_cpu_present(cpu, false);
1289 			nr++;
1290 		}
1291 
1292 		nr = 0;
1293 		for_each_possible_cpu(cpu) {
1294 			if (nr >= 8)
1295 				set_cpu_possible(cpu, false);
1296 			nr++;
1297 		}
1298 
1299 		set_nr_cpu_ids(8);
1300 	}
1301 #endif
1302 
1303 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1304 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1305 			hard_smp_processor_id());
1306 
1307 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1308 	}
1309 
1310 	/*
1311 	 * Should not be necessary because the MP table should list the boot
1312 	 * CPU too, but we do it for the sake of robustness anyway.
1313 	 */
1314 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1315 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1316 			  boot_cpu_physical_apicid);
1317 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1318 	}
1319 	preempt_enable();
1320 }
1321 
1322 static void __init smp_cpu_index_default(void)
1323 {
1324 	int i;
1325 	struct cpuinfo_x86 *c;
1326 
1327 	for_each_possible_cpu(i) {
1328 		c = &cpu_data(i);
1329 		/* mark all to hotplug */
1330 		c->cpu_index = nr_cpu_ids;
1331 	}
1332 }
1333 
1334 void __init smp_prepare_cpus_common(void)
1335 {
1336 	unsigned int i;
1337 
1338 	smp_cpu_index_default();
1339 
1340 	/*
1341 	 * Setup boot CPU information
1342 	 */
1343 	smp_store_boot_cpu_info(); /* Final full version of the data */
1344 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1345 	mb();
1346 
1347 	for_each_possible_cpu(i) {
1348 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1349 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1350 		zalloc_cpumask_var(&per_cpu(cpu_die_map, i), GFP_KERNEL);
1351 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1352 		zalloc_cpumask_var(&per_cpu(cpu_l2c_shared_map, i), GFP_KERNEL);
1353 	}
1354 
1355 	/*
1356 	 * Set 'default' x86 topology, this matches default_topology() in that
1357 	 * it has NUMA nodes as a topology level. See also
1358 	 * native_smp_cpus_done().
1359 	 *
1360 	 * Must be done before set_cpus_sibling_map() is ran.
1361 	 */
1362 	set_sched_topology(x86_topology);
1363 
1364 	set_cpu_sibling_map(0);
1365 }
1366 
1367 /*
1368  * Prepare for SMP bootup.
1369  * @max_cpus: configured maximum number of CPUs, It is a legacy parameter
1370  *            for common interface support.
1371  */
1372 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1373 {
1374 	smp_prepare_cpus_common();
1375 
1376 	smp_sanity_check();
1377 
1378 	switch (apic_intr_mode) {
1379 	case APIC_PIC:
1380 	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1381 		disable_smp();
1382 		return;
1383 	case APIC_SYMMETRIC_IO_NO_ROUTING:
1384 		disable_smp();
1385 		/* Setup local timer */
1386 		x86_init.timers.setup_percpu_clockev();
1387 		return;
1388 	case APIC_VIRTUAL_WIRE:
1389 	case APIC_SYMMETRIC_IO:
1390 		break;
1391 	}
1392 
1393 	/* Setup local timer */
1394 	x86_init.timers.setup_percpu_clockev();
1395 
1396 	pr_info("CPU0: ");
1397 	print_cpu_info(&cpu_data(0));
1398 
1399 	uv_system_init();
1400 
1401 	smp_quirk_init_udelay();
1402 
1403 	speculative_store_bypass_ht_init();
1404 
1405 	snp_set_wakeup_secondary_cpu();
1406 }
1407 
1408 void arch_thaw_secondary_cpus_begin(void)
1409 {
1410 	set_cache_aps_delayed_init(true);
1411 }
1412 
1413 void arch_thaw_secondary_cpus_end(void)
1414 {
1415 	cache_aps_init();
1416 }
1417 
1418 /*
1419  * Early setup to make printk work.
1420  */
1421 void __init native_smp_prepare_boot_cpu(void)
1422 {
1423 	int me = smp_processor_id();
1424 
1425 	/* SMP handles this from setup_per_cpu_areas() */
1426 	if (!IS_ENABLED(CONFIG_SMP))
1427 		switch_gdt_and_percpu_base(me);
1428 
1429 	/* already set me in cpu_online_mask in boot_cpu_init() */
1430 	cpumask_set_cpu(me, cpu_callout_mask);
1431 	cpu_set_state_online(me);
1432 	native_pv_lock_init();
1433 }
1434 
1435 void __init calculate_max_logical_packages(void)
1436 {
1437 	int ncpus;
1438 
1439 	/*
1440 	 * Today neither Intel nor AMD support heterogeneous systems so
1441 	 * extrapolate the boot cpu's data to all packages.
1442 	 */
1443 	ncpus = cpu_data(0).booted_cores * topology_max_smt_threads();
1444 	__max_logical_packages = DIV_ROUND_UP(total_cpus, ncpus);
1445 	pr_info("Max logical packages: %u\n", __max_logical_packages);
1446 }
1447 
1448 void __init native_smp_cpus_done(unsigned int max_cpus)
1449 {
1450 	pr_debug("Boot done\n");
1451 
1452 	calculate_max_logical_packages();
1453 
1454 	/* XXX for now assume numa-in-package and hybrid don't overlap */
1455 	if (x86_has_numa_in_package)
1456 		set_sched_topology(x86_numa_in_package_topology);
1457 	if (cpu_feature_enabled(X86_FEATURE_HYBRID_CPU))
1458 		set_sched_topology(x86_hybrid_topology);
1459 
1460 	nmi_selftest();
1461 	impress_friends();
1462 	cache_aps_init();
1463 }
1464 
1465 static int __initdata setup_possible_cpus = -1;
1466 static int __init _setup_possible_cpus(char *str)
1467 {
1468 	get_option(&str, &setup_possible_cpus);
1469 	return 0;
1470 }
1471 early_param("possible_cpus", _setup_possible_cpus);
1472 
1473 
1474 /*
1475  * cpu_possible_mask should be static, it cannot change as cpu's
1476  * are onlined, or offlined. The reason is per-cpu data-structures
1477  * are allocated by some modules at init time, and don't expect to
1478  * do this dynamically on cpu arrival/departure.
1479  * cpu_present_mask on the other hand can change dynamically.
1480  * In case when cpu_hotplug is not compiled, then we resort to current
1481  * behaviour, which is cpu_possible == cpu_present.
1482  * - Ashok Raj
1483  *
1484  * Three ways to find out the number of additional hotplug CPUs:
1485  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1486  * - The user can overwrite it with possible_cpus=NUM
1487  * - Otherwise don't reserve additional CPUs.
1488  * We do this because additional CPUs waste a lot of memory.
1489  * -AK
1490  */
1491 __init void prefill_possible_map(void)
1492 {
1493 	int i, possible;
1494 
1495 	/* No boot processor was found in mptable or ACPI MADT */
1496 	if (!num_processors) {
1497 		if (boot_cpu_has(X86_FEATURE_APIC)) {
1498 			int apicid = boot_cpu_physical_apicid;
1499 			int cpu = hard_smp_processor_id();
1500 
1501 			pr_warn("Boot CPU (id %d) not listed by BIOS\n", cpu);
1502 
1503 			/* Make sure boot cpu is enumerated */
1504 			if (apic->cpu_present_to_apicid(0) == BAD_APICID &&
1505 			    apic->apic_id_valid(apicid))
1506 				generic_processor_info(apicid, boot_cpu_apic_version);
1507 		}
1508 
1509 		if (!num_processors)
1510 			num_processors = 1;
1511 	}
1512 
1513 	i = setup_max_cpus ?: 1;
1514 	if (setup_possible_cpus == -1) {
1515 		possible = num_processors;
1516 #ifdef CONFIG_HOTPLUG_CPU
1517 		if (setup_max_cpus)
1518 			possible += disabled_cpus;
1519 #else
1520 		if (possible > i)
1521 			possible = i;
1522 #endif
1523 	} else
1524 		possible = setup_possible_cpus;
1525 
1526 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1527 
1528 	/* nr_cpu_ids could be reduced via nr_cpus= */
1529 	if (possible > nr_cpu_ids) {
1530 		pr_warn("%d Processors exceeds NR_CPUS limit of %u\n",
1531 			possible, nr_cpu_ids);
1532 		possible = nr_cpu_ids;
1533 	}
1534 
1535 #ifdef CONFIG_HOTPLUG_CPU
1536 	if (!setup_max_cpus)
1537 #endif
1538 	if (possible > i) {
1539 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1540 			possible, setup_max_cpus);
1541 		possible = i;
1542 	}
1543 
1544 	set_nr_cpu_ids(possible);
1545 
1546 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1547 		possible, max_t(int, possible - num_processors, 0));
1548 
1549 	reset_cpu_possible_mask();
1550 
1551 	for (i = 0; i < possible; i++)
1552 		set_cpu_possible(i, true);
1553 }
1554 
1555 /* correctly size the local cpu masks */
1556 void __init setup_cpu_local_masks(void)
1557 {
1558 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
1559 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
1560 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
1561 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
1562 }
1563 
1564 #ifdef CONFIG_HOTPLUG_CPU
1565 
1566 /* Recompute SMT state for all CPUs on offline */
1567 static void recompute_smt_state(void)
1568 {
1569 	int max_threads, cpu;
1570 
1571 	max_threads = 0;
1572 	for_each_online_cpu (cpu) {
1573 		int threads = cpumask_weight(topology_sibling_cpumask(cpu));
1574 
1575 		if (threads > max_threads)
1576 			max_threads = threads;
1577 	}
1578 	__max_smt_threads = max_threads;
1579 }
1580 
1581 static void remove_siblinginfo(int cpu)
1582 {
1583 	int sibling;
1584 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1585 
1586 	for_each_cpu(sibling, topology_core_cpumask(cpu)) {
1587 		cpumask_clear_cpu(cpu, topology_core_cpumask(sibling));
1588 		/*/
1589 		 * last thread sibling in this cpu core going down
1590 		 */
1591 		if (cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
1592 			cpu_data(sibling).booted_cores--;
1593 	}
1594 
1595 	for_each_cpu(sibling, topology_die_cpumask(cpu))
1596 		cpumask_clear_cpu(cpu, topology_die_cpumask(sibling));
1597 
1598 	for_each_cpu(sibling, topology_sibling_cpumask(cpu)) {
1599 		cpumask_clear_cpu(cpu, topology_sibling_cpumask(sibling));
1600 		if (cpumask_weight(topology_sibling_cpumask(sibling)) == 1)
1601 			cpu_data(sibling).smt_active = false;
1602 	}
1603 
1604 	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1605 		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1606 	for_each_cpu(sibling, cpu_l2c_shared_mask(cpu))
1607 		cpumask_clear_cpu(cpu, cpu_l2c_shared_mask(sibling));
1608 	cpumask_clear(cpu_llc_shared_mask(cpu));
1609 	cpumask_clear(cpu_l2c_shared_mask(cpu));
1610 	cpumask_clear(topology_sibling_cpumask(cpu));
1611 	cpumask_clear(topology_core_cpumask(cpu));
1612 	cpumask_clear(topology_die_cpumask(cpu));
1613 	c->cpu_core_id = 0;
1614 	c->booted_cores = 0;
1615 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1616 	recompute_smt_state();
1617 }
1618 
1619 static void remove_cpu_from_maps(int cpu)
1620 {
1621 	set_cpu_online(cpu, false);
1622 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1623 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1624 	/* was set by cpu_init() */
1625 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1626 	numa_remove_cpu(cpu);
1627 }
1628 
1629 void cpu_disable_common(void)
1630 {
1631 	int cpu = smp_processor_id();
1632 
1633 	remove_siblinginfo(cpu);
1634 
1635 	/* It's now safe to remove this processor from the online map */
1636 	lock_vector_lock();
1637 	remove_cpu_from_maps(cpu);
1638 	unlock_vector_lock();
1639 	fixup_irqs();
1640 	lapic_offline();
1641 }
1642 
1643 int native_cpu_disable(void)
1644 {
1645 	int ret;
1646 
1647 	ret = lapic_can_unplug_cpu();
1648 	if (ret)
1649 		return ret;
1650 
1651 	cpu_disable_common();
1652 
1653         /*
1654          * Disable the local APIC. Otherwise IPI broadcasts will reach
1655          * it. It still responds normally to INIT, NMI, SMI, and SIPI
1656          * messages.
1657          *
1658          * Disabling the APIC must happen after cpu_disable_common()
1659          * which invokes fixup_irqs().
1660          *
1661          * Disabling the APIC preserves already set bits in IRR, but
1662          * an interrupt arriving after disabling the local APIC does not
1663          * set the corresponding IRR bit.
1664          *
1665          * fixup_irqs() scans IRR for set bits so it can raise a not
1666          * yet handled interrupt on the new destination CPU via an IPI
1667          * but obviously it can't do so for IRR bits which are not set.
1668          * IOW, interrupts arriving after disabling the local APIC will
1669          * be lost.
1670          */
1671 	apic_soft_disable();
1672 
1673 	return 0;
1674 }
1675 
1676 int common_cpu_die(unsigned int cpu)
1677 {
1678 	int ret = 0;
1679 
1680 	/* We don't do anything here: idle task is faking death itself. */
1681 
1682 	/* They ack this in play_dead() by setting CPU_DEAD */
1683 	if (cpu_wait_death(cpu, 5)) {
1684 		if (system_state == SYSTEM_RUNNING)
1685 			pr_info("CPU %u is now offline\n", cpu);
1686 	} else {
1687 		pr_err("CPU %u didn't die...\n", cpu);
1688 		ret = -1;
1689 	}
1690 
1691 	return ret;
1692 }
1693 
1694 void native_cpu_die(unsigned int cpu)
1695 {
1696 	common_cpu_die(cpu);
1697 }
1698 
1699 void play_dead_common(void)
1700 {
1701 	idle_task_exit();
1702 
1703 	/* Ack it */
1704 	(void)cpu_report_death();
1705 
1706 	/*
1707 	 * With physical CPU hotplug, we should halt the cpu
1708 	 */
1709 	local_irq_disable();
1710 }
1711 
1712 /*
1713  * We need to flush the caches before going to sleep, lest we have
1714  * dirty data in our caches when we come back up.
1715  */
1716 static inline void mwait_play_dead(void)
1717 {
1718 	unsigned int eax, ebx, ecx, edx;
1719 	unsigned int highest_cstate = 0;
1720 	unsigned int highest_subcstate = 0;
1721 	void *mwait_ptr;
1722 	int i;
1723 
1724 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1725 	    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
1726 		return;
1727 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1728 		return;
1729 	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1730 		return;
1731 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1732 		return;
1733 
1734 	eax = CPUID_MWAIT_LEAF;
1735 	ecx = 0;
1736 	native_cpuid(&eax, &ebx, &ecx, &edx);
1737 
1738 	/*
1739 	 * eax will be 0 if EDX enumeration is not valid.
1740 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1741 	 */
1742 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1743 		eax = 0;
1744 	} else {
1745 		edx >>= MWAIT_SUBSTATE_SIZE;
1746 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1747 			if (edx & MWAIT_SUBSTATE_MASK) {
1748 				highest_cstate = i;
1749 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1750 			}
1751 		}
1752 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1753 			(highest_subcstate - 1);
1754 	}
1755 
1756 	/*
1757 	 * This should be a memory location in a cache line which is
1758 	 * unlikely to be touched by other processors.  The actual
1759 	 * content is immaterial as it is not actually modified in any way.
1760 	 */
1761 	mwait_ptr = &current_thread_info()->flags;
1762 
1763 	wbinvd();
1764 
1765 	while (1) {
1766 		/*
1767 		 * The CLFLUSH is a workaround for erratum AAI65 for
1768 		 * the Xeon 7400 series.  It's not clear it is actually
1769 		 * needed, but it should be harmless in either case.
1770 		 * The WBINVD is insufficient due to the spurious-wakeup
1771 		 * case where we return around the loop.
1772 		 */
1773 		mb();
1774 		clflush(mwait_ptr);
1775 		mb();
1776 		__monitor(mwait_ptr, 0, 0);
1777 		mb();
1778 		__mwait(eax, 0);
1779 	}
1780 }
1781 
1782 void __noreturn hlt_play_dead(void)
1783 {
1784 	if (__this_cpu_read(cpu_info.x86) >= 4)
1785 		wbinvd();
1786 
1787 	while (1)
1788 		native_halt();
1789 }
1790 
1791 void native_play_dead(void)
1792 {
1793 	play_dead_common();
1794 	tboot_shutdown(TB_SHUTDOWN_WFS);
1795 
1796 	mwait_play_dead();
1797 	if (cpuidle_play_dead())
1798 		hlt_play_dead();
1799 }
1800 
1801 #else /* ... !CONFIG_HOTPLUG_CPU */
1802 int native_cpu_disable(void)
1803 {
1804 	return -ENOSYS;
1805 }
1806 
1807 void native_cpu_die(unsigned int cpu)
1808 {
1809 	/* We said "no" in __cpu_disable */
1810 	BUG();
1811 }
1812 
1813 void native_play_dead(void)
1814 {
1815 	BUG();
1816 }
1817 
1818 #endif
1819