xref: /openbmc/linux/arch/x86/kernel/smpboot.c (revision 8b030a57)
1  /*
2  *	x86 SMP booting functions
3  *
4  *	(c) 1995 Alan Cox, Building #3 <alan@lxorguk.ukuu.org.uk>
5  *	(c) 1998, 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
6  *	Copyright 2001 Andi Kleen, SuSE Labs.
7  *
8  *	Much of the core SMP work is based on previous work by Thomas Radke, to
9  *	whom a great many thanks are extended.
10  *
11  *	Thanks to Intel for making available several different Pentium,
12  *	Pentium Pro and Pentium-II/Xeon MP machines.
13  *	Original development of Linux SMP code supported by Caldera.
14  *
15  *	This code is released under the GNU General Public License version 2 or
16  *	later.
17  *
18  *	Fixes
19  *		Felix Koop	:	NR_CPUS used properly
20  *		Jose Renau	:	Handle single CPU case.
21  *		Alan Cox	:	By repeated request 8) - Total BogoMIPS report.
22  *		Greg Wright	:	Fix for kernel stacks panic.
23  *		Erich Boleyn	:	MP v1.4 and additional changes.
24  *	Matthias Sattler	:	Changes for 2.1 kernel map.
25  *	Michel Lespinasse	:	Changes for 2.1 kernel map.
26  *	Michael Chastain	:	Change trampoline.S to gnu as.
27  *		Alan Cox	:	Dumb bug: 'B' step PPro's are fine
28  *		Ingo Molnar	:	Added APIC timers, based on code
29  *					from Jose Renau
30  *		Ingo Molnar	:	various cleanups and rewrites
31  *		Tigran Aivazian	:	fixed "0.00 in /proc/uptime on SMP" bug.
32  *	Maciej W. Rozycki	:	Bits for genuine 82489DX APICs
33  *	Andi Kleen		:	Changed for SMP boot into long mode.
34  *		Martin J. Bligh	: 	Added support for multi-quad systems
35  *		Dave Jones	:	Report invalid combinations of Athlon CPUs.
36  *		Rusty Russell	:	Hacked into shape for new "hotplug" boot process.
37  *      Andi Kleen              :       Converted to new state machine.
38  *	Ashok Raj		: 	CPU hotplug support
39  *	Glauber Costa		:	i386 and x86_64 integration
40  */
41 
42 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
43 
44 #include <linux/init.h>
45 #include <linux/smp.h>
46 #include <linux/export.h>
47 #include <linux/sched.h>
48 #include <linux/sched/topology.h>
49 #include <linux/sched/hotplug.h>
50 #include <linux/sched/task_stack.h>
51 #include <linux/percpu.h>
52 #include <linux/memblock.h>
53 #include <linux/err.h>
54 #include <linux/nmi.h>
55 #include <linux/tboot.h>
56 #include <linux/stackprotector.h>
57 #include <linux/gfp.h>
58 #include <linux/cpuidle.h>
59 
60 #include <asm/acpi.h>
61 #include <asm/desc.h>
62 #include <asm/nmi.h>
63 #include <asm/irq.h>
64 #include <asm/realmode.h>
65 #include <asm/cpu.h>
66 #include <asm/numa.h>
67 #include <asm/pgtable.h>
68 #include <asm/tlbflush.h>
69 #include <asm/mtrr.h>
70 #include <asm/mwait.h>
71 #include <asm/apic.h>
72 #include <asm/io_apic.h>
73 #include <asm/fpu/internal.h>
74 #include <asm/setup.h>
75 #include <asm/uv/uv.h>
76 #include <linux/mc146818rtc.h>
77 #include <asm/i8259.h>
78 #include <asm/misc.h>
79 #include <asm/qspinlock.h>
80 #include <asm/intel-family.h>
81 #include <asm/cpu_device_id.h>
82 #include <asm/spec-ctrl.h>
83 #include <asm/hw_irq.h>
84 
85 /* representing HT siblings of each logical CPU */
86 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
87 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
88 
89 /* representing HT and core siblings of each logical CPU */
90 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_map);
91 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
92 
93 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_llc_shared_map);
94 
95 /* Per CPU bogomips and other parameters */
96 DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
97 EXPORT_PER_CPU_SYMBOL(cpu_info);
98 
99 /* Logical package management. We might want to allocate that dynamically */
100 unsigned int __max_logical_packages __read_mostly;
101 EXPORT_SYMBOL(__max_logical_packages);
102 static unsigned int logical_packages __read_mostly;
103 
104 /* Maximum number of SMT threads on any online core */
105 int __read_mostly __max_smt_threads = 1;
106 
107 /* Flag to indicate if a complete sched domain rebuild is required */
108 bool x86_topology_update;
109 
110 int arch_update_cpu_topology(void)
111 {
112 	int retval = x86_topology_update;
113 
114 	x86_topology_update = false;
115 	return retval;
116 }
117 
118 static inline void smpboot_setup_warm_reset_vector(unsigned long start_eip)
119 {
120 	unsigned long flags;
121 
122 	spin_lock_irqsave(&rtc_lock, flags);
123 	CMOS_WRITE(0xa, 0xf);
124 	spin_unlock_irqrestore(&rtc_lock, flags);
125 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_HIGH)) =
126 							start_eip >> 4;
127 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) =
128 							start_eip & 0xf;
129 }
130 
131 static inline void smpboot_restore_warm_reset_vector(void)
132 {
133 	unsigned long flags;
134 
135 	/*
136 	 * Paranoid:  Set warm reset code and vector here back
137 	 * to default values.
138 	 */
139 	spin_lock_irqsave(&rtc_lock, flags);
140 	CMOS_WRITE(0, 0xf);
141 	spin_unlock_irqrestore(&rtc_lock, flags);
142 
143 	*((volatile u32 *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = 0;
144 }
145 
146 /*
147  * Report back to the Boot Processor during boot time or to the caller processor
148  * during CPU online.
149  */
150 static void smp_callin(void)
151 {
152 	int cpuid, phys_id;
153 
154 	/*
155 	 * If waken up by an INIT in an 82489DX configuration
156 	 * cpu_callout_mask guarantees we don't get here before
157 	 * an INIT_deassert IPI reaches our local APIC, so it is
158 	 * now safe to touch our local APIC.
159 	 */
160 	cpuid = smp_processor_id();
161 
162 	/*
163 	 * (This works even if the APIC is not enabled.)
164 	 */
165 	phys_id = read_apic_id();
166 
167 	/*
168 	 * the boot CPU has finished the init stage and is spinning
169 	 * on callin_map until we finish. We are free to set up this
170 	 * CPU, first the APIC. (this is probably redundant on most
171 	 * boards)
172 	 */
173 	apic_ap_setup();
174 
175 	/*
176 	 * Save our processor parameters. Note: this information
177 	 * is needed for clock calibration.
178 	 */
179 	smp_store_cpu_info(cpuid);
180 
181 	/*
182 	 * The topology information must be up to date before
183 	 * calibrate_delay() and notify_cpu_starting().
184 	 */
185 	set_cpu_sibling_map(raw_smp_processor_id());
186 
187 	/*
188 	 * Get our bogomips.
189 	 * Update loops_per_jiffy in cpu_data. Previous call to
190 	 * smp_store_cpu_info() stored a value that is close but not as
191 	 * accurate as the value just calculated.
192 	 */
193 	calibrate_delay();
194 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
195 	pr_debug("Stack at about %p\n", &cpuid);
196 
197 	wmb();
198 
199 	notify_cpu_starting(cpuid);
200 
201 	/*
202 	 * Allow the master to continue.
203 	 */
204 	cpumask_set_cpu(cpuid, cpu_callin_mask);
205 }
206 
207 static int cpu0_logical_apicid;
208 static int enable_start_cpu0;
209 /*
210  * Activate a secondary processor.
211  */
212 static void notrace start_secondary(void *unused)
213 {
214 	/*
215 	 * Don't put *anything* except direct CPU state initialization
216 	 * before cpu_init(), SMP booting is too fragile that we want to
217 	 * limit the things done here to the most necessary things.
218 	 */
219 	if (boot_cpu_has(X86_FEATURE_PCID))
220 		__write_cr4(__read_cr4() | X86_CR4_PCIDE);
221 
222 #ifdef CONFIG_X86_32
223 	/* switch away from the initial page table */
224 	load_cr3(swapper_pg_dir);
225 	/*
226 	 * Initialize the CR4 shadow before doing anything that could
227 	 * try to read it.
228 	 */
229 	cr4_init_shadow();
230 	__flush_tlb_all();
231 #endif
232 	load_current_idt();
233 	cpu_init();
234 	x86_cpuinit.early_percpu_clock_init();
235 	preempt_disable();
236 	smp_callin();
237 
238 	enable_start_cpu0 = 0;
239 
240 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
241 	barrier();
242 	/*
243 	 * Check TSC synchronization with the boot CPU:
244 	 */
245 	check_tsc_sync_target();
246 
247 	speculative_store_bypass_ht_init();
248 
249 	/*
250 	 * Lock vector_lock, set CPU online and bring the vector
251 	 * allocator online. Online must be set with vector_lock held
252 	 * to prevent a concurrent irq setup/teardown from seeing a
253 	 * half valid vector space.
254 	 */
255 	lock_vector_lock();
256 	set_cpu_online(smp_processor_id(), true);
257 	lapic_online();
258 	unlock_vector_lock();
259 	cpu_set_state_online(smp_processor_id());
260 	x86_platform.nmi_init();
261 
262 	/* enable local interrupts */
263 	local_irq_enable();
264 
265 	/* to prevent fake stack check failure in clock setup */
266 	boot_init_stack_canary();
267 
268 	x86_cpuinit.setup_percpu_clockev();
269 
270 	wmb();
271 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
272 }
273 
274 /**
275  * topology_is_primary_thread - Check whether CPU is the primary SMT thread
276  * @cpu:	CPU to check
277  */
278 bool topology_is_primary_thread(unsigned int cpu)
279 {
280 	return apic_id_is_primary_thread(per_cpu(x86_cpu_to_apicid, cpu));
281 }
282 
283 /**
284  * topology_smt_supported - Check whether SMT is supported by the CPUs
285  */
286 bool topology_smt_supported(void)
287 {
288 	return smp_num_siblings > 1;
289 }
290 
291 /**
292  * topology_phys_to_logical_pkg - Map a physical package id to a logical
293  *
294  * Returns logical package id or -1 if not found
295  */
296 int topology_phys_to_logical_pkg(unsigned int phys_pkg)
297 {
298 	int cpu;
299 
300 	for_each_possible_cpu(cpu) {
301 		struct cpuinfo_x86 *c = &cpu_data(cpu);
302 
303 		if (c->initialized && c->phys_proc_id == phys_pkg)
304 			return c->logical_proc_id;
305 	}
306 	return -1;
307 }
308 EXPORT_SYMBOL(topology_phys_to_logical_pkg);
309 
310 /**
311  * topology_update_package_map - Update the physical to logical package map
312  * @pkg:	The physical package id as retrieved via CPUID
313  * @cpu:	The cpu for which this is updated
314  */
315 int topology_update_package_map(unsigned int pkg, unsigned int cpu)
316 {
317 	int new;
318 
319 	/* Already available somewhere? */
320 	new = topology_phys_to_logical_pkg(pkg);
321 	if (new >= 0)
322 		goto found;
323 
324 	new = logical_packages++;
325 	if (new != pkg) {
326 		pr_info("CPU %u Converting physical %u to logical package %u\n",
327 			cpu, pkg, new);
328 	}
329 found:
330 	cpu_data(cpu).logical_proc_id = new;
331 	return 0;
332 }
333 
334 void __init smp_store_boot_cpu_info(void)
335 {
336 	int id = 0; /* CPU 0 */
337 	struct cpuinfo_x86 *c = &cpu_data(id);
338 
339 	*c = boot_cpu_data;
340 	c->cpu_index = id;
341 	topology_update_package_map(c->phys_proc_id, id);
342 	c->initialized = true;
343 }
344 
345 /*
346  * The bootstrap kernel entry code has set these up. Save them for
347  * a given CPU
348  */
349 void smp_store_cpu_info(int id)
350 {
351 	struct cpuinfo_x86 *c = &cpu_data(id);
352 
353 	/* Copy boot_cpu_data only on the first bringup */
354 	if (!c->initialized)
355 		*c = boot_cpu_data;
356 	c->cpu_index = id;
357 	/*
358 	 * During boot time, CPU0 has this setup already. Save the info when
359 	 * bringing up AP or offlined CPU0.
360 	 */
361 	identify_secondary_cpu(c);
362 	c->initialized = true;
363 }
364 
365 static bool
366 topology_same_node(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
367 {
368 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
369 
370 	return (cpu_to_node(cpu1) == cpu_to_node(cpu2));
371 }
372 
373 static bool
374 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
375 {
376 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
377 
378 	return !WARN_ONCE(!topology_same_node(c, o),
379 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
380 		"[node: %d != %d]. Ignoring dependency.\n",
381 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
382 }
383 
384 #define link_mask(mfunc, c1, c2)					\
385 do {									\
386 	cpumask_set_cpu((c1), mfunc(c2));				\
387 	cpumask_set_cpu((c2), mfunc(c1));				\
388 } while (0)
389 
390 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
391 {
392 	if (boot_cpu_has(X86_FEATURE_TOPOEXT)) {
393 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
394 
395 		if (c->phys_proc_id == o->phys_proc_id &&
396 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2)) {
397 			if (c->cpu_core_id == o->cpu_core_id)
398 				return topology_sane(c, o, "smt");
399 
400 			if ((c->cu_id != 0xff) &&
401 			    (o->cu_id != 0xff) &&
402 			    (c->cu_id == o->cu_id))
403 				return topology_sane(c, o, "smt");
404 		}
405 
406 	} else if (c->phys_proc_id == o->phys_proc_id &&
407 		   c->cpu_core_id == o->cpu_core_id) {
408 		return topology_sane(c, o, "smt");
409 	}
410 
411 	return false;
412 }
413 
414 /*
415  * Define snc_cpu[] for SNC (Sub-NUMA Cluster) CPUs.
416  *
417  * These are Intel CPUs that enumerate an LLC that is shared by
418  * multiple NUMA nodes. The LLC on these systems is shared for
419  * off-package data access but private to the NUMA node (half
420  * of the package) for on-package access.
421  *
422  * CPUID (the source of the information about the LLC) can only
423  * enumerate the cache as being shared *or* unshared, but not
424  * this particular configuration. The CPU in this case enumerates
425  * the cache to be shared across the entire package (spanning both
426  * NUMA nodes).
427  */
428 
429 static const struct x86_cpu_id snc_cpu[] = {
430 	{ X86_VENDOR_INTEL, 6, INTEL_FAM6_SKYLAKE_X },
431 	{}
432 };
433 
434 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
435 {
436 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
437 
438 	/* Do not match if we do not have a valid APICID for cpu: */
439 	if (per_cpu(cpu_llc_id, cpu1) == BAD_APICID)
440 		return false;
441 
442 	/* Do not match if LLC id does not match: */
443 	if (per_cpu(cpu_llc_id, cpu1) != per_cpu(cpu_llc_id, cpu2))
444 		return false;
445 
446 	/*
447 	 * Allow the SNC topology without warning. Return of false
448 	 * means 'c' does not share the LLC of 'o'. This will be
449 	 * reflected to userspace.
450 	 */
451 	if (!topology_same_node(c, o) && x86_match_cpu(snc_cpu))
452 		return false;
453 
454 	return topology_sane(c, o, "llc");
455 }
456 
457 /*
458  * Unlike the other levels, we do not enforce keeping a
459  * multicore group inside a NUMA node.  If this happens, we will
460  * discard the MC level of the topology later.
461  */
462 static bool match_die(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
463 {
464 	if (c->phys_proc_id == o->phys_proc_id)
465 		return true;
466 	return false;
467 }
468 
469 #if defined(CONFIG_SCHED_SMT) || defined(CONFIG_SCHED_MC)
470 static inline int x86_sched_itmt_flags(void)
471 {
472 	return sysctl_sched_itmt_enabled ? SD_ASYM_PACKING : 0;
473 }
474 
475 #ifdef CONFIG_SCHED_MC
476 static int x86_core_flags(void)
477 {
478 	return cpu_core_flags() | x86_sched_itmt_flags();
479 }
480 #endif
481 #ifdef CONFIG_SCHED_SMT
482 static int x86_smt_flags(void)
483 {
484 	return cpu_smt_flags() | x86_sched_itmt_flags();
485 }
486 #endif
487 #endif
488 
489 static struct sched_domain_topology_level x86_numa_in_package_topology[] = {
490 #ifdef CONFIG_SCHED_SMT
491 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
492 #endif
493 #ifdef CONFIG_SCHED_MC
494 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
495 #endif
496 	{ NULL, },
497 };
498 
499 static struct sched_domain_topology_level x86_topology[] = {
500 #ifdef CONFIG_SCHED_SMT
501 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
502 #endif
503 #ifdef CONFIG_SCHED_MC
504 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
505 #endif
506 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
507 	{ NULL, },
508 };
509 
510 /*
511  * Set if a package/die has multiple NUMA nodes inside.
512  * AMD Magny-Cours, Intel Cluster-on-Die, and Intel
513  * Sub-NUMA Clustering have this.
514  */
515 static bool x86_has_numa_in_package;
516 
517 void set_cpu_sibling_map(int cpu)
518 {
519 	bool has_smt = smp_num_siblings > 1;
520 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
521 	struct cpuinfo_x86 *c = &cpu_data(cpu);
522 	struct cpuinfo_x86 *o;
523 	int i, threads;
524 
525 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
526 
527 	if (!has_mp) {
528 		cpumask_set_cpu(cpu, topology_sibling_cpumask(cpu));
529 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
530 		cpumask_set_cpu(cpu, topology_core_cpumask(cpu));
531 		c->booted_cores = 1;
532 		return;
533 	}
534 
535 	for_each_cpu(i, cpu_sibling_setup_mask) {
536 		o = &cpu_data(i);
537 
538 		if ((i == cpu) || (has_smt && match_smt(c, o)))
539 			link_mask(topology_sibling_cpumask, cpu, i);
540 
541 		if ((i == cpu) || (has_mp && match_llc(c, o)))
542 			link_mask(cpu_llc_shared_mask, cpu, i);
543 
544 	}
545 
546 	/*
547 	 * This needs a separate iteration over the cpus because we rely on all
548 	 * topology_sibling_cpumask links to be set-up.
549 	 */
550 	for_each_cpu(i, cpu_sibling_setup_mask) {
551 		o = &cpu_data(i);
552 
553 		if ((i == cpu) || (has_mp && match_die(c, o))) {
554 			link_mask(topology_core_cpumask, cpu, i);
555 
556 			/*
557 			 *  Does this new cpu bringup a new core?
558 			 */
559 			if (cpumask_weight(
560 			    topology_sibling_cpumask(cpu)) == 1) {
561 				/*
562 				 * for each core in package, increment
563 				 * the booted_cores for this new cpu
564 				 */
565 				if (cpumask_first(
566 				    topology_sibling_cpumask(i)) == i)
567 					c->booted_cores++;
568 				/*
569 				 * increment the core count for all
570 				 * the other cpus in this package
571 				 */
572 				if (i != cpu)
573 					cpu_data(i).booted_cores++;
574 			} else if (i != cpu && !c->booted_cores)
575 				c->booted_cores = cpu_data(i).booted_cores;
576 		}
577 		if (match_die(c, o) && !topology_same_node(c, o))
578 			x86_has_numa_in_package = true;
579 	}
580 
581 	threads = cpumask_weight(topology_sibling_cpumask(cpu));
582 	if (threads > __max_smt_threads)
583 		__max_smt_threads = threads;
584 }
585 
586 /* maps the cpu to the sched domain representing multi-core */
587 const struct cpumask *cpu_coregroup_mask(int cpu)
588 {
589 	return cpu_llc_shared_mask(cpu);
590 }
591 
592 static void impress_friends(void)
593 {
594 	int cpu;
595 	unsigned long bogosum = 0;
596 	/*
597 	 * Allow the user to impress friends.
598 	 */
599 	pr_debug("Before bogomips\n");
600 	for_each_possible_cpu(cpu)
601 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
602 			bogosum += cpu_data(cpu).loops_per_jiffy;
603 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
604 		num_online_cpus(),
605 		bogosum/(500000/HZ),
606 		(bogosum/(5000/HZ))%100);
607 
608 	pr_debug("Before bogocount - setting activated=1\n");
609 }
610 
611 void __inquire_remote_apic(int apicid)
612 {
613 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
614 	const char * const names[] = { "ID", "VERSION", "SPIV" };
615 	int timeout;
616 	u32 status;
617 
618 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
619 
620 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
621 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
622 
623 		/*
624 		 * Wait for idle.
625 		 */
626 		status = safe_apic_wait_icr_idle();
627 		if (status)
628 			pr_cont("a previous APIC delivery may have failed\n");
629 
630 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
631 
632 		timeout = 0;
633 		do {
634 			udelay(100);
635 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
636 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
637 
638 		switch (status) {
639 		case APIC_ICR_RR_VALID:
640 			status = apic_read(APIC_RRR);
641 			pr_cont("%08x\n", status);
642 			break;
643 		default:
644 			pr_cont("failed\n");
645 		}
646 	}
647 }
648 
649 /*
650  * The Multiprocessor Specification 1.4 (1997) example code suggests
651  * that there should be a 10ms delay between the BSP asserting INIT
652  * and de-asserting INIT, when starting a remote processor.
653  * But that slows boot and resume on modern processors, which include
654  * many cores and don't require that delay.
655  *
656  * Cmdline "init_cpu_udelay=" is available to over-ride this delay.
657  * Modern processor families are quirked to remove the delay entirely.
658  */
659 #define UDELAY_10MS_DEFAULT 10000
660 
661 static unsigned int init_udelay = UINT_MAX;
662 
663 static int __init cpu_init_udelay(char *str)
664 {
665 	get_option(&str, &init_udelay);
666 
667 	return 0;
668 }
669 early_param("cpu_init_udelay", cpu_init_udelay);
670 
671 static void __init smp_quirk_init_udelay(void)
672 {
673 	/* if cmdline changed it from default, leave it alone */
674 	if (init_udelay != UINT_MAX)
675 		return;
676 
677 	/* if modern processor, use no delay */
678 	if (((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && (boot_cpu_data.x86 == 6)) ||
679 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) && (boot_cpu_data.x86 >= 0x18)) ||
680 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && (boot_cpu_data.x86 >= 0xF))) {
681 		init_udelay = 0;
682 		return;
683 	}
684 	/* else, use legacy delay */
685 	init_udelay = UDELAY_10MS_DEFAULT;
686 }
687 
688 /*
689  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
690  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
691  * won't ... remember to clear down the APIC, etc later.
692  */
693 int
694 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
695 {
696 	unsigned long send_status, accept_status = 0;
697 	int maxlvt;
698 
699 	/* Target chip */
700 	/* Boot on the stack */
701 	/* Kick the second */
702 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, apicid);
703 
704 	pr_debug("Waiting for send to finish...\n");
705 	send_status = safe_apic_wait_icr_idle();
706 
707 	/*
708 	 * Give the other CPU some time to accept the IPI.
709 	 */
710 	udelay(200);
711 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
712 		maxlvt = lapic_get_maxlvt();
713 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
714 			apic_write(APIC_ESR, 0);
715 		accept_status = (apic_read(APIC_ESR) & 0xEF);
716 	}
717 	pr_debug("NMI sent\n");
718 
719 	if (send_status)
720 		pr_err("APIC never delivered???\n");
721 	if (accept_status)
722 		pr_err("APIC delivery error (%lx)\n", accept_status);
723 
724 	return (send_status | accept_status);
725 }
726 
727 static int
728 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
729 {
730 	unsigned long send_status = 0, accept_status = 0;
731 	int maxlvt, num_starts, j;
732 
733 	maxlvt = lapic_get_maxlvt();
734 
735 	/*
736 	 * Be paranoid about clearing APIC errors.
737 	 */
738 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
739 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
740 			apic_write(APIC_ESR, 0);
741 		apic_read(APIC_ESR);
742 	}
743 
744 	pr_debug("Asserting INIT\n");
745 
746 	/*
747 	 * Turn INIT on target chip
748 	 */
749 	/*
750 	 * Send IPI
751 	 */
752 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
753 		       phys_apicid);
754 
755 	pr_debug("Waiting for send to finish...\n");
756 	send_status = safe_apic_wait_icr_idle();
757 
758 	udelay(init_udelay);
759 
760 	pr_debug("Deasserting INIT\n");
761 
762 	/* Target chip */
763 	/* Send IPI */
764 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
765 
766 	pr_debug("Waiting for send to finish...\n");
767 	send_status = safe_apic_wait_icr_idle();
768 
769 	mb();
770 
771 	/*
772 	 * Should we send STARTUP IPIs ?
773 	 *
774 	 * Determine this based on the APIC version.
775 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
776 	 */
777 	if (APIC_INTEGRATED(boot_cpu_apic_version))
778 		num_starts = 2;
779 	else
780 		num_starts = 0;
781 
782 	/*
783 	 * Run STARTUP IPI loop.
784 	 */
785 	pr_debug("#startup loops: %d\n", num_starts);
786 
787 	for (j = 1; j <= num_starts; j++) {
788 		pr_debug("Sending STARTUP #%d\n", j);
789 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
790 			apic_write(APIC_ESR, 0);
791 		apic_read(APIC_ESR);
792 		pr_debug("After apic_write\n");
793 
794 		/*
795 		 * STARTUP IPI
796 		 */
797 
798 		/* Target chip */
799 		/* Boot on the stack */
800 		/* Kick the second */
801 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
802 			       phys_apicid);
803 
804 		/*
805 		 * Give the other CPU some time to accept the IPI.
806 		 */
807 		if (init_udelay == 0)
808 			udelay(10);
809 		else
810 			udelay(300);
811 
812 		pr_debug("Startup point 1\n");
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 		if (init_udelay == 0)
821 			udelay(10);
822 		else
823 			udelay(200);
824 
825 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
826 			apic_write(APIC_ESR, 0);
827 		accept_status = (apic_read(APIC_ESR) & 0xEF);
828 		if (send_status || accept_status)
829 			break;
830 	}
831 	pr_debug("After Startup\n");
832 
833 	if (send_status)
834 		pr_err("APIC never delivered???\n");
835 	if (accept_status)
836 		pr_err("APIC delivery error (%lx)\n", accept_status);
837 
838 	return (send_status | accept_status);
839 }
840 
841 /* reduce the number of lines printed when booting a large cpu count system */
842 static void announce_cpu(int cpu, int apicid)
843 {
844 	static int current_node = -1;
845 	int node = early_cpu_to_node(cpu);
846 	static int width, node_width;
847 
848 	if (!width)
849 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
850 
851 	if (!node_width)
852 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
853 
854 	if (cpu == 1)
855 		printk(KERN_INFO "x86: Booting SMP configuration:\n");
856 
857 	if (system_state < SYSTEM_RUNNING) {
858 		if (node != current_node) {
859 			if (current_node > (-1))
860 				pr_cont("\n");
861 			current_node = node;
862 
863 			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
864 			       node_width - num_digits(node), " ", node);
865 		}
866 
867 		/* Add padding for the BSP */
868 		if (cpu == 1)
869 			pr_cont("%*s", width + 1, " ");
870 
871 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
872 
873 	} else
874 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
875 			node, cpu, apicid);
876 }
877 
878 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
879 {
880 	int cpu;
881 
882 	cpu = smp_processor_id();
883 	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
884 		return NMI_HANDLED;
885 
886 	return NMI_DONE;
887 }
888 
889 /*
890  * Wake up AP by INIT, INIT, STARTUP sequence.
891  *
892  * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
893  * boot-strap code which is not a desired behavior for waking up BSP. To
894  * void the boot-strap code, wake up CPU0 by NMI instead.
895  *
896  * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
897  * (i.e. physically hot removed and then hot added), NMI won't wake it up.
898  * We'll change this code in the future to wake up hard offlined CPU0 if
899  * real platform and request are available.
900  */
901 static int
902 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
903 	       int *cpu0_nmi_registered)
904 {
905 	int id;
906 	int boot_error;
907 
908 	preempt_disable();
909 
910 	/*
911 	 * Wake up AP by INIT, INIT, STARTUP sequence.
912 	 */
913 	if (cpu) {
914 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
915 		goto out;
916 	}
917 
918 	/*
919 	 * Wake up BSP by nmi.
920 	 *
921 	 * Register a NMI handler to help wake up CPU0.
922 	 */
923 	boot_error = register_nmi_handler(NMI_LOCAL,
924 					  wakeup_cpu0_nmi, 0, "wake_cpu0");
925 
926 	if (!boot_error) {
927 		enable_start_cpu0 = 1;
928 		*cpu0_nmi_registered = 1;
929 		if (apic->dest_logical == APIC_DEST_LOGICAL)
930 			id = cpu0_logical_apicid;
931 		else
932 			id = apicid;
933 		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
934 	}
935 
936 out:
937 	preempt_enable();
938 
939 	return boot_error;
940 }
941 
942 void common_cpu_up(unsigned int cpu, struct task_struct *idle)
943 {
944 	/* Just in case we booted with a single CPU. */
945 	alternatives_enable_smp();
946 
947 	per_cpu(current_task, cpu) = idle;
948 
949 #ifdef CONFIG_X86_32
950 	/* Stack for startup_32 can be just as for start_secondary onwards */
951 	irq_ctx_init(cpu);
952 	per_cpu(cpu_current_top_of_stack, cpu) = task_top_of_stack(idle);
953 #else
954 	initial_gs = per_cpu_offset(cpu);
955 #endif
956 }
957 
958 /*
959  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
960  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
961  * Returns zero if CPU booted OK, else error code from
962  * ->wakeup_secondary_cpu.
963  */
964 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle,
965 		       int *cpu0_nmi_registered)
966 {
967 	volatile u32 *trampoline_status =
968 		(volatile u32 *) __va(real_mode_header->trampoline_status);
969 	/* start_ip had better be page-aligned! */
970 	unsigned long start_ip = real_mode_header->trampoline_start;
971 
972 	unsigned long boot_error = 0;
973 	unsigned long timeout;
974 
975 	idle->thread.sp = (unsigned long)task_pt_regs(idle);
976 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_rw(cpu);
977 	initial_code = (unsigned long)start_secondary;
978 	initial_stack  = idle->thread.sp;
979 
980 	/* Enable the espfix hack for this CPU */
981 	init_espfix_ap(cpu);
982 
983 	/* So we see what's up */
984 	announce_cpu(cpu, apicid);
985 
986 	/*
987 	 * This grunge runs the startup process for
988 	 * the targeted processor.
989 	 */
990 
991 	if (x86_platform.legacy.warm_reset) {
992 
993 		pr_debug("Setting warm reset code and vector.\n");
994 
995 		smpboot_setup_warm_reset_vector(start_ip);
996 		/*
997 		 * Be paranoid about clearing APIC errors.
998 		*/
999 		if (APIC_INTEGRATED(boot_cpu_apic_version)) {
1000 			apic_write(APIC_ESR, 0);
1001 			apic_read(APIC_ESR);
1002 		}
1003 	}
1004 
1005 	/*
1006 	 * AP might wait on cpu_callout_mask in cpu_init() with
1007 	 * cpu_initialized_mask set if previous attempt to online
1008 	 * it timed-out. Clear cpu_initialized_mask so that after
1009 	 * INIT/SIPI it could start with a clean state.
1010 	 */
1011 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1012 	smp_mb();
1013 
1014 	/*
1015 	 * Wake up a CPU in difference cases:
1016 	 * - Use the method in the APIC driver if it's defined
1017 	 * Otherwise,
1018 	 * - Use an INIT boot APIC message for APs or NMI for BSP.
1019 	 */
1020 	if (apic->wakeup_secondary_cpu)
1021 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
1022 	else
1023 		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
1024 						     cpu0_nmi_registered);
1025 
1026 	if (!boot_error) {
1027 		/*
1028 		 * Wait 10s total for first sign of life from AP
1029 		 */
1030 		boot_error = -1;
1031 		timeout = jiffies + 10*HZ;
1032 		while (time_before(jiffies, timeout)) {
1033 			if (cpumask_test_cpu(cpu, cpu_initialized_mask)) {
1034 				/*
1035 				 * Tell AP to proceed with initialization
1036 				 */
1037 				cpumask_set_cpu(cpu, cpu_callout_mask);
1038 				boot_error = 0;
1039 				break;
1040 			}
1041 			schedule();
1042 		}
1043 	}
1044 
1045 	if (!boot_error) {
1046 		/*
1047 		 * Wait till AP completes initial initialization
1048 		 */
1049 		while (!cpumask_test_cpu(cpu, cpu_callin_mask)) {
1050 			/*
1051 			 * Allow other tasks to run while we wait for the
1052 			 * AP to come online. This also gives a chance
1053 			 * for the MTRR work(triggered by the AP coming online)
1054 			 * to be completed in the stop machine context.
1055 			 */
1056 			schedule();
1057 		}
1058 	}
1059 
1060 	/* mark "stuck" area as not stuck */
1061 	*trampoline_status = 0;
1062 
1063 	if (x86_platform.legacy.warm_reset) {
1064 		/*
1065 		 * Cleanup possible dangling ends...
1066 		 */
1067 		smpboot_restore_warm_reset_vector();
1068 	}
1069 
1070 	return boot_error;
1071 }
1072 
1073 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
1074 {
1075 	int apicid = apic->cpu_present_to_apicid(cpu);
1076 	int cpu0_nmi_registered = 0;
1077 	unsigned long flags;
1078 	int err, ret = 0;
1079 
1080 	lockdep_assert_irqs_enabled();
1081 
1082 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
1083 
1084 	if (apicid == BAD_APICID ||
1085 	    !physid_isset(apicid, phys_cpu_present_map) ||
1086 	    !apic->apic_id_valid(apicid)) {
1087 		pr_err("%s: bad cpu %d\n", __func__, cpu);
1088 		return -EINVAL;
1089 	}
1090 
1091 	/*
1092 	 * Already booted CPU?
1093 	 */
1094 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
1095 		pr_debug("do_boot_cpu %d Already started\n", cpu);
1096 		return -ENOSYS;
1097 	}
1098 
1099 	/*
1100 	 * Save current MTRR state in case it was changed since early boot
1101 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
1102 	 */
1103 	mtrr_save_state();
1104 
1105 	/* x86 CPUs take themselves offline, so delayed offline is OK. */
1106 	err = cpu_check_up_prepare(cpu);
1107 	if (err && err != -EBUSY)
1108 		return err;
1109 
1110 	/* the FPU context is blank, nobody can own it */
1111 	per_cpu(fpu_fpregs_owner_ctx, cpu) = NULL;
1112 
1113 	common_cpu_up(cpu, tidle);
1114 
1115 	err = do_boot_cpu(apicid, cpu, tidle, &cpu0_nmi_registered);
1116 	if (err) {
1117 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
1118 		ret = -EIO;
1119 		goto unreg_nmi;
1120 	}
1121 
1122 	/*
1123 	 * Check TSC synchronization with the AP (keep irqs disabled
1124 	 * while doing so):
1125 	 */
1126 	local_irq_save(flags);
1127 	check_tsc_sync_source(cpu);
1128 	local_irq_restore(flags);
1129 
1130 	while (!cpu_online(cpu)) {
1131 		cpu_relax();
1132 		touch_nmi_watchdog();
1133 	}
1134 
1135 unreg_nmi:
1136 	/*
1137 	 * Clean up the nmi handler. Do this after the callin and callout sync
1138 	 * to avoid impact of possible long unregister time.
1139 	 */
1140 	if (cpu0_nmi_registered)
1141 		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
1142 
1143 	return ret;
1144 }
1145 
1146 /**
1147  * arch_disable_smp_support() - disables SMP support for x86 at runtime
1148  */
1149 void arch_disable_smp_support(void)
1150 {
1151 	disable_ioapic_support();
1152 }
1153 
1154 /*
1155  * Fall back to non SMP mode after errors.
1156  *
1157  * RED-PEN audit/test this more. I bet there is more state messed up here.
1158  */
1159 static __init void disable_smp(void)
1160 {
1161 	pr_info("SMP disabled\n");
1162 
1163 	disable_ioapic_support();
1164 
1165 	init_cpu_present(cpumask_of(0));
1166 	init_cpu_possible(cpumask_of(0));
1167 
1168 	if (smp_found_config)
1169 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1170 	else
1171 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1172 	cpumask_set_cpu(0, topology_sibling_cpumask(0));
1173 	cpumask_set_cpu(0, topology_core_cpumask(0));
1174 }
1175 
1176 /*
1177  * Various sanity checks.
1178  */
1179 static void __init smp_sanity_check(void)
1180 {
1181 	preempt_disable();
1182 
1183 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
1184 	if (def_to_bigsmp && nr_cpu_ids > 8) {
1185 		unsigned int cpu;
1186 		unsigned nr;
1187 
1188 		pr_warn("More than 8 CPUs detected - skipping them\n"
1189 			"Use CONFIG_X86_BIGSMP\n");
1190 
1191 		nr = 0;
1192 		for_each_present_cpu(cpu) {
1193 			if (nr >= 8)
1194 				set_cpu_present(cpu, false);
1195 			nr++;
1196 		}
1197 
1198 		nr = 0;
1199 		for_each_possible_cpu(cpu) {
1200 			if (nr >= 8)
1201 				set_cpu_possible(cpu, false);
1202 			nr++;
1203 		}
1204 
1205 		nr_cpu_ids = 8;
1206 	}
1207 #endif
1208 
1209 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1210 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1211 			hard_smp_processor_id());
1212 
1213 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1214 	}
1215 
1216 	/*
1217 	 * Should not be necessary because the MP table should list the boot
1218 	 * CPU too, but we do it for the sake of robustness anyway.
1219 	 */
1220 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1221 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1222 			  boot_cpu_physical_apicid);
1223 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1224 	}
1225 	preempt_enable();
1226 }
1227 
1228 static void __init smp_cpu_index_default(void)
1229 {
1230 	int i;
1231 	struct cpuinfo_x86 *c;
1232 
1233 	for_each_possible_cpu(i) {
1234 		c = &cpu_data(i);
1235 		/* mark all to hotplug */
1236 		c->cpu_index = nr_cpu_ids;
1237 	}
1238 }
1239 
1240 static void __init smp_get_logical_apicid(void)
1241 {
1242 	if (x2apic_mode)
1243 		cpu0_logical_apicid = apic_read(APIC_LDR);
1244 	else
1245 		cpu0_logical_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1246 }
1247 
1248 /*
1249  * Prepare for SMP bootup.
1250  * @max_cpus: configured maximum number of CPUs, It is a legacy parameter
1251  *            for common interface support.
1252  */
1253 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1254 {
1255 	unsigned int i;
1256 
1257 	smp_cpu_index_default();
1258 
1259 	/*
1260 	 * Setup boot CPU information
1261 	 */
1262 	smp_store_boot_cpu_info(); /* Final full version of the data */
1263 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1264 	mb();
1265 
1266 	for_each_possible_cpu(i) {
1267 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1268 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1269 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1270 	}
1271 
1272 	/*
1273 	 * Set 'default' x86 topology, this matches default_topology() in that
1274 	 * it has NUMA nodes as a topology level. See also
1275 	 * native_smp_cpus_done().
1276 	 *
1277 	 * Must be done before set_cpus_sibling_map() is ran.
1278 	 */
1279 	set_sched_topology(x86_topology);
1280 
1281 	set_cpu_sibling_map(0);
1282 
1283 	smp_sanity_check();
1284 
1285 	switch (apic_intr_mode) {
1286 	case APIC_PIC:
1287 	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1288 		disable_smp();
1289 		return;
1290 	case APIC_SYMMETRIC_IO_NO_ROUTING:
1291 		disable_smp();
1292 		/* Setup local timer */
1293 		x86_init.timers.setup_percpu_clockev();
1294 		return;
1295 	case APIC_VIRTUAL_WIRE:
1296 	case APIC_SYMMETRIC_IO:
1297 		break;
1298 	}
1299 
1300 	/* Setup local timer */
1301 	x86_init.timers.setup_percpu_clockev();
1302 
1303 	smp_get_logical_apicid();
1304 
1305 	pr_info("CPU0: ");
1306 	print_cpu_info(&cpu_data(0));
1307 
1308 	native_pv_lock_init();
1309 
1310 	uv_system_init();
1311 
1312 	set_mtrr_aps_delayed_init();
1313 
1314 	smp_quirk_init_udelay();
1315 
1316 	speculative_store_bypass_ht_init();
1317 }
1318 
1319 void arch_enable_nonboot_cpus_begin(void)
1320 {
1321 	set_mtrr_aps_delayed_init();
1322 }
1323 
1324 void arch_enable_nonboot_cpus_end(void)
1325 {
1326 	mtrr_aps_init();
1327 }
1328 
1329 /*
1330  * Early setup to make printk work.
1331  */
1332 void __init native_smp_prepare_boot_cpu(void)
1333 {
1334 	int me = smp_processor_id();
1335 	switch_to_new_gdt(me);
1336 	/* already set me in cpu_online_mask in boot_cpu_init() */
1337 	cpumask_set_cpu(me, cpu_callout_mask);
1338 	cpu_set_state_online(me);
1339 }
1340 
1341 void __init calculate_max_logical_packages(void)
1342 {
1343 	int ncpus;
1344 
1345 	/*
1346 	 * Today neither Intel nor AMD support heterogenous systems so
1347 	 * extrapolate the boot cpu's data to all packages.
1348 	 */
1349 	ncpus = cpu_data(0).booted_cores * topology_max_smt_threads();
1350 	__max_logical_packages = DIV_ROUND_UP(total_cpus, ncpus);
1351 	pr_info("Max logical packages: %u\n", __max_logical_packages);
1352 }
1353 
1354 void __init native_smp_cpus_done(unsigned int max_cpus)
1355 {
1356 	pr_debug("Boot done\n");
1357 
1358 	calculate_max_logical_packages();
1359 
1360 	if (x86_has_numa_in_package)
1361 		set_sched_topology(x86_numa_in_package_topology);
1362 
1363 	nmi_selftest();
1364 	impress_friends();
1365 	mtrr_aps_init();
1366 }
1367 
1368 static int __initdata setup_possible_cpus = -1;
1369 static int __init _setup_possible_cpus(char *str)
1370 {
1371 	get_option(&str, &setup_possible_cpus);
1372 	return 0;
1373 }
1374 early_param("possible_cpus", _setup_possible_cpus);
1375 
1376 
1377 /*
1378  * cpu_possible_mask should be static, it cannot change as cpu's
1379  * are onlined, or offlined. The reason is per-cpu data-structures
1380  * are allocated by some modules at init time, and dont expect to
1381  * do this dynamically on cpu arrival/departure.
1382  * cpu_present_mask on the other hand can change dynamically.
1383  * In case when cpu_hotplug is not compiled, then we resort to current
1384  * behaviour, which is cpu_possible == cpu_present.
1385  * - Ashok Raj
1386  *
1387  * Three ways to find out the number of additional hotplug CPUs:
1388  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1389  * - The user can overwrite it with possible_cpus=NUM
1390  * - Otherwise don't reserve additional CPUs.
1391  * We do this because additional CPUs waste a lot of memory.
1392  * -AK
1393  */
1394 __init void prefill_possible_map(void)
1395 {
1396 	int i, possible;
1397 
1398 	/* No boot processor was found in mptable or ACPI MADT */
1399 	if (!num_processors) {
1400 		if (boot_cpu_has(X86_FEATURE_APIC)) {
1401 			int apicid = boot_cpu_physical_apicid;
1402 			int cpu = hard_smp_processor_id();
1403 
1404 			pr_warn("Boot CPU (id %d) not listed by BIOS\n", cpu);
1405 
1406 			/* Make sure boot cpu is enumerated */
1407 			if (apic->cpu_present_to_apicid(0) == BAD_APICID &&
1408 			    apic->apic_id_valid(apicid))
1409 				generic_processor_info(apicid, boot_cpu_apic_version);
1410 		}
1411 
1412 		if (!num_processors)
1413 			num_processors = 1;
1414 	}
1415 
1416 	i = setup_max_cpus ?: 1;
1417 	if (setup_possible_cpus == -1) {
1418 		possible = num_processors;
1419 #ifdef CONFIG_HOTPLUG_CPU
1420 		if (setup_max_cpus)
1421 			possible += disabled_cpus;
1422 #else
1423 		if (possible > i)
1424 			possible = i;
1425 #endif
1426 	} else
1427 		possible = setup_possible_cpus;
1428 
1429 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1430 
1431 	/* nr_cpu_ids could be reduced via nr_cpus= */
1432 	if (possible > nr_cpu_ids) {
1433 		pr_warn("%d Processors exceeds NR_CPUS limit of %u\n",
1434 			possible, nr_cpu_ids);
1435 		possible = nr_cpu_ids;
1436 	}
1437 
1438 #ifdef CONFIG_HOTPLUG_CPU
1439 	if (!setup_max_cpus)
1440 #endif
1441 	if (possible > i) {
1442 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1443 			possible, setup_max_cpus);
1444 		possible = i;
1445 	}
1446 
1447 	nr_cpu_ids = possible;
1448 
1449 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1450 		possible, max_t(int, possible - num_processors, 0));
1451 
1452 	reset_cpu_possible_mask();
1453 
1454 	for (i = 0; i < possible; i++)
1455 		set_cpu_possible(i, true);
1456 }
1457 
1458 #ifdef CONFIG_HOTPLUG_CPU
1459 
1460 /* Recompute SMT state for all CPUs on offline */
1461 static void recompute_smt_state(void)
1462 {
1463 	int max_threads, cpu;
1464 
1465 	max_threads = 0;
1466 	for_each_online_cpu (cpu) {
1467 		int threads = cpumask_weight(topology_sibling_cpumask(cpu));
1468 
1469 		if (threads > max_threads)
1470 			max_threads = threads;
1471 	}
1472 	__max_smt_threads = max_threads;
1473 }
1474 
1475 static void remove_siblinginfo(int cpu)
1476 {
1477 	int sibling;
1478 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1479 
1480 	for_each_cpu(sibling, topology_core_cpumask(cpu)) {
1481 		cpumask_clear_cpu(cpu, topology_core_cpumask(sibling));
1482 		/*/
1483 		 * last thread sibling in this cpu core going down
1484 		 */
1485 		if (cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
1486 			cpu_data(sibling).booted_cores--;
1487 	}
1488 
1489 	for_each_cpu(sibling, topology_sibling_cpumask(cpu))
1490 		cpumask_clear_cpu(cpu, topology_sibling_cpumask(sibling));
1491 	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1492 		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1493 	cpumask_clear(cpu_llc_shared_mask(cpu));
1494 	cpumask_clear(topology_sibling_cpumask(cpu));
1495 	cpumask_clear(topology_core_cpumask(cpu));
1496 	c->cpu_core_id = 0;
1497 	c->booted_cores = 0;
1498 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1499 	recompute_smt_state();
1500 }
1501 
1502 static void remove_cpu_from_maps(int cpu)
1503 {
1504 	set_cpu_online(cpu, false);
1505 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1506 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1507 	/* was set by cpu_init() */
1508 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1509 	numa_remove_cpu(cpu);
1510 }
1511 
1512 void cpu_disable_common(void)
1513 {
1514 	int cpu = smp_processor_id();
1515 
1516 	remove_siblinginfo(cpu);
1517 
1518 	/* It's now safe to remove this processor from the online map */
1519 	lock_vector_lock();
1520 	remove_cpu_from_maps(cpu);
1521 	unlock_vector_lock();
1522 	fixup_irqs();
1523 	lapic_offline();
1524 }
1525 
1526 int native_cpu_disable(void)
1527 {
1528 	int ret;
1529 
1530 	ret = lapic_can_unplug_cpu();
1531 	if (ret)
1532 		return ret;
1533 
1534 	clear_local_APIC();
1535 	cpu_disable_common();
1536 
1537 	return 0;
1538 }
1539 
1540 int common_cpu_die(unsigned int cpu)
1541 {
1542 	int ret = 0;
1543 
1544 	/* We don't do anything here: idle task is faking death itself. */
1545 
1546 	/* They ack this in play_dead() by setting CPU_DEAD */
1547 	if (cpu_wait_death(cpu, 5)) {
1548 		if (system_state == SYSTEM_RUNNING)
1549 			pr_info("CPU %u is now offline\n", cpu);
1550 	} else {
1551 		pr_err("CPU %u didn't die...\n", cpu);
1552 		ret = -1;
1553 	}
1554 
1555 	return ret;
1556 }
1557 
1558 void native_cpu_die(unsigned int cpu)
1559 {
1560 	common_cpu_die(cpu);
1561 }
1562 
1563 void play_dead_common(void)
1564 {
1565 	idle_task_exit();
1566 
1567 	/* Ack it */
1568 	(void)cpu_report_death();
1569 
1570 	/*
1571 	 * With physical CPU hotplug, we should halt the cpu
1572 	 */
1573 	local_irq_disable();
1574 }
1575 
1576 static bool wakeup_cpu0(void)
1577 {
1578 	if (smp_processor_id() == 0 && enable_start_cpu0)
1579 		return true;
1580 
1581 	return false;
1582 }
1583 
1584 /*
1585  * We need to flush the caches before going to sleep, lest we have
1586  * dirty data in our caches when we come back up.
1587  */
1588 static inline void mwait_play_dead(void)
1589 {
1590 	unsigned int eax, ebx, ecx, edx;
1591 	unsigned int highest_cstate = 0;
1592 	unsigned int highest_subcstate = 0;
1593 	void *mwait_ptr;
1594 	int i;
1595 
1596 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1597 	    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
1598 		return;
1599 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1600 		return;
1601 	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1602 		return;
1603 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1604 		return;
1605 
1606 	eax = CPUID_MWAIT_LEAF;
1607 	ecx = 0;
1608 	native_cpuid(&eax, &ebx, &ecx, &edx);
1609 
1610 	/*
1611 	 * eax will be 0 if EDX enumeration is not valid.
1612 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1613 	 */
1614 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1615 		eax = 0;
1616 	} else {
1617 		edx >>= MWAIT_SUBSTATE_SIZE;
1618 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1619 			if (edx & MWAIT_SUBSTATE_MASK) {
1620 				highest_cstate = i;
1621 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1622 			}
1623 		}
1624 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1625 			(highest_subcstate - 1);
1626 	}
1627 
1628 	/*
1629 	 * This should be a memory location in a cache line which is
1630 	 * unlikely to be touched by other processors.  The actual
1631 	 * content is immaterial as it is not actually modified in any way.
1632 	 */
1633 	mwait_ptr = &current_thread_info()->flags;
1634 
1635 	wbinvd();
1636 
1637 	while (1) {
1638 		/*
1639 		 * The CLFLUSH is a workaround for erratum AAI65 for
1640 		 * the Xeon 7400 series.  It's not clear it is actually
1641 		 * needed, but it should be harmless in either case.
1642 		 * The WBINVD is insufficient due to the spurious-wakeup
1643 		 * case where we return around the loop.
1644 		 */
1645 		mb();
1646 		clflush(mwait_ptr);
1647 		mb();
1648 		__monitor(mwait_ptr, 0, 0);
1649 		mb();
1650 		__mwait(eax, 0);
1651 		/*
1652 		 * If NMI wants to wake up CPU0, start CPU0.
1653 		 */
1654 		if (wakeup_cpu0())
1655 			start_cpu0();
1656 	}
1657 }
1658 
1659 void hlt_play_dead(void)
1660 {
1661 	if (__this_cpu_read(cpu_info.x86) >= 4)
1662 		wbinvd();
1663 
1664 	while (1) {
1665 		native_halt();
1666 		/*
1667 		 * If NMI wants to wake up CPU0, start CPU0.
1668 		 */
1669 		if (wakeup_cpu0())
1670 			start_cpu0();
1671 	}
1672 }
1673 
1674 void native_play_dead(void)
1675 {
1676 	play_dead_common();
1677 	tboot_shutdown(TB_SHUTDOWN_WFS);
1678 
1679 	mwait_play_dead();	/* Only returns on failure */
1680 	if (cpuidle_play_dead())
1681 		hlt_play_dead();
1682 }
1683 
1684 #else /* ... !CONFIG_HOTPLUG_CPU */
1685 int native_cpu_disable(void)
1686 {
1687 	return -ENOSYS;
1688 }
1689 
1690 void native_cpu_die(unsigned int cpu)
1691 {
1692 	/* We said "no" in __cpu_disable */
1693 	BUG();
1694 }
1695 
1696 void native_play_dead(void)
1697 {
1698 	BUG();
1699 }
1700 
1701 #endif
1702