xref: /openbmc/linux/arch/x86/kernel/smpboot.c (revision 95777591)
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 #include <linux/numa.h>
60 
61 #include <asm/acpi.h>
62 #include <asm/desc.h>
63 #include <asm/nmi.h>
64 #include <asm/irq.h>
65 #include <asm/realmode.h>
66 #include <asm/cpu.h>
67 #include <asm/numa.h>
68 #include <asm/pgtable.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/internal.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 
86 /* representing HT siblings of each logical CPU */
87 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
88 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
89 
90 /* representing HT and core siblings of each logical CPU */
91 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_map);
92 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
93 
94 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_llc_shared_map);
95 
96 /* Per CPU bogomips and other parameters */
97 DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
98 EXPORT_PER_CPU_SYMBOL(cpu_info);
99 
100 /* Logical package management. We might want to allocate that dynamically */
101 unsigned int __max_logical_packages __read_mostly;
102 EXPORT_SYMBOL(__max_logical_packages);
103 static unsigned int logical_packages __read_mostly;
104 
105 /* Maximum number of SMT threads on any online core */
106 int __read_mostly __max_smt_threads = 1;
107 
108 /* Flag to indicate if a complete sched domain rebuild is required */
109 bool x86_topology_update;
110 
111 int arch_update_cpu_topology(void)
112 {
113 	int retval = x86_topology_update;
114 
115 	x86_topology_update = false;
116 	return retval;
117 }
118 
119 static inline void smpboot_setup_warm_reset_vector(unsigned long start_eip)
120 {
121 	unsigned long flags;
122 
123 	spin_lock_irqsave(&rtc_lock, flags);
124 	CMOS_WRITE(0xa, 0xf);
125 	spin_unlock_irqrestore(&rtc_lock, flags);
126 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_HIGH)) =
127 							start_eip >> 4;
128 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) =
129 							start_eip & 0xf;
130 }
131 
132 static inline void smpboot_restore_warm_reset_vector(void)
133 {
134 	unsigned long flags;
135 
136 	/*
137 	 * Paranoid:  Set warm reset code and vector here back
138 	 * to default values.
139 	 */
140 	spin_lock_irqsave(&rtc_lock, flags);
141 	CMOS_WRITE(0, 0xf);
142 	spin_unlock_irqrestore(&rtc_lock, flags);
143 
144 	*((volatile u32 *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = 0;
145 }
146 
147 /*
148  * Report back to the Boot Processor during boot time or to the caller processor
149  * during CPU online.
150  */
151 static void smp_callin(void)
152 {
153 	int cpuid, phys_id;
154 
155 	/*
156 	 * If waken up by an INIT in an 82489DX configuration
157 	 * cpu_callout_mask guarantees we don't get here before
158 	 * an INIT_deassert IPI reaches our local APIC, so it is
159 	 * now safe to touch our local APIC.
160 	 */
161 	cpuid = smp_processor_id();
162 
163 	/*
164 	 * (This works even if the APIC is not enabled.)
165 	 */
166 	phys_id = read_apic_id();
167 
168 	/*
169 	 * the boot CPU has finished the init stage and is spinning
170 	 * on callin_map until we finish. We are free to set up this
171 	 * CPU, first the APIC. (this is probably redundant on most
172 	 * boards)
173 	 */
174 	apic_ap_setup();
175 
176 	/*
177 	 * Save our processor parameters. Note: this information
178 	 * is needed for clock calibration.
179 	 */
180 	smp_store_cpu_info(cpuid);
181 
182 	/*
183 	 * The topology information must be up to date before
184 	 * calibrate_delay() and notify_cpu_starting().
185 	 */
186 	set_cpu_sibling_map(raw_smp_processor_id());
187 
188 	/*
189 	 * Get our bogomips.
190 	 * Update loops_per_jiffy in cpu_data. Previous call to
191 	 * smp_store_cpu_info() stored a value that is close but not as
192 	 * accurate as the value just calculated.
193 	 */
194 	calibrate_delay();
195 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
196 	pr_debug("Stack at about %p\n", &cpuid);
197 
198 	wmb();
199 
200 	notify_cpu_starting(cpuid);
201 
202 	/*
203 	 * Allow the master to continue.
204 	 */
205 	cpumask_set_cpu(cpuid, cpu_callin_mask);
206 }
207 
208 static int cpu0_logical_apicid;
209 static int enable_start_cpu0;
210 /*
211  * Activate a secondary processor.
212  */
213 static void notrace start_secondary(void *unused)
214 {
215 	/*
216 	 * Don't put *anything* except direct CPU state initialization
217 	 * before cpu_init(), SMP booting is too fragile that we want to
218 	 * limit the things done here to the most necessary things.
219 	 */
220 	if (boot_cpu_has(X86_FEATURE_PCID))
221 		__write_cr4(__read_cr4() | X86_CR4_PCIDE);
222 
223 #ifdef CONFIG_X86_32
224 	/* switch away from the initial page table */
225 	load_cr3(swapper_pg_dir);
226 	/*
227 	 * Initialize the CR4 shadow before doing anything that could
228 	 * try to read it.
229 	 */
230 	cr4_init_shadow();
231 	__flush_tlb_all();
232 #endif
233 	load_current_idt();
234 	cpu_init();
235 	x86_cpuinit.early_percpu_clock_init();
236 	preempt_disable();
237 	smp_callin();
238 
239 	enable_start_cpu0 = 0;
240 
241 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
242 	barrier();
243 	/*
244 	 * Check TSC synchronization with the boot CPU:
245 	 */
246 	check_tsc_sync_target();
247 
248 	speculative_store_bypass_ht_init();
249 
250 	/*
251 	 * Lock vector_lock, set CPU online and bring the vector
252 	 * allocator online. Online must be set with vector_lock held
253 	 * to prevent a concurrent irq setup/teardown from seeing a
254 	 * half valid vector space.
255 	 */
256 	lock_vector_lock();
257 	set_cpu_online(smp_processor_id(), true);
258 	lapic_online();
259 	unlock_vector_lock();
260 	cpu_set_state_online(smp_processor_id());
261 	x86_platform.nmi_init();
262 
263 	/* enable local interrupts */
264 	local_irq_enable();
265 
266 	/* to prevent fake stack check failure in clock setup */
267 	boot_init_stack_canary();
268 
269 	x86_cpuinit.setup_percpu_clockev();
270 
271 	wmb();
272 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
273 }
274 
275 /**
276  * topology_is_primary_thread - Check whether CPU is the primary SMT thread
277  * @cpu:	CPU to check
278  */
279 bool topology_is_primary_thread(unsigned int cpu)
280 {
281 	return apic_id_is_primary_thread(per_cpu(x86_cpu_to_apicid, cpu));
282 }
283 
284 /**
285  * topology_smt_supported - Check whether SMT is supported by the CPUs
286  */
287 bool topology_smt_supported(void)
288 {
289 	return smp_num_siblings > 1;
290 }
291 
292 /**
293  * topology_phys_to_logical_pkg - Map a physical package id to a logical
294  *
295  * Returns logical package id or -1 if not found
296  */
297 int topology_phys_to_logical_pkg(unsigned int phys_pkg)
298 {
299 	int cpu;
300 
301 	for_each_possible_cpu(cpu) {
302 		struct cpuinfo_x86 *c = &cpu_data(cpu);
303 
304 		if (c->initialized && c->phys_proc_id == phys_pkg)
305 			return c->logical_proc_id;
306 	}
307 	return -1;
308 }
309 EXPORT_SYMBOL(topology_phys_to_logical_pkg);
310 
311 /**
312  * topology_update_package_map - Update the physical to logical package map
313  * @pkg:	The physical package id as retrieved via CPUID
314  * @cpu:	The cpu for which this is updated
315  */
316 int topology_update_package_map(unsigned int pkg, unsigned int cpu)
317 {
318 	int new;
319 
320 	/* Already available somewhere? */
321 	new = topology_phys_to_logical_pkg(pkg);
322 	if (new >= 0)
323 		goto found;
324 
325 	new = logical_packages++;
326 	if (new != pkg) {
327 		pr_info("CPU %u Converting physical %u to logical package %u\n",
328 			cpu, pkg, new);
329 	}
330 found:
331 	cpu_data(cpu).logical_proc_id = new;
332 	return 0;
333 }
334 
335 void __init smp_store_boot_cpu_info(void)
336 {
337 	int id = 0; /* CPU 0 */
338 	struct cpuinfo_x86 *c = &cpu_data(id);
339 
340 	*c = boot_cpu_data;
341 	c->cpu_index = id;
342 	topology_update_package_map(c->phys_proc_id, id);
343 	c->initialized = true;
344 }
345 
346 /*
347  * The bootstrap kernel entry code has set these up. Save them for
348  * a given CPU
349  */
350 void smp_store_cpu_info(int id)
351 {
352 	struct cpuinfo_x86 *c = &cpu_data(id);
353 
354 	/* Copy boot_cpu_data only on the first bringup */
355 	if (!c->initialized)
356 		*c = boot_cpu_data;
357 	c->cpu_index = id;
358 	/*
359 	 * During boot time, CPU0 has this setup already. Save the info when
360 	 * bringing up AP or offlined CPU0.
361 	 */
362 	identify_secondary_cpu(c);
363 	c->initialized = true;
364 }
365 
366 static bool
367 topology_same_node(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
368 {
369 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
370 
371 	return (cpu_to_node(cpu1) == cpu_to_node(cpu2));
372 }
373 
374 static bool
375 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
376 {
377 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
378 
379 	return !WARN_ONCE(!topology_same_node(c, o),
380 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
381 		"[node: %d != %d]. Ignoring dependency.\n",
382 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
383 }
384 
385 #define link_mask(mfunc, c1, c2)					\
386 do {									\
387 	cpumask_set_cpu((c1), mfunc(c2));				\
388 	cpumask_set_cpu((c2), mfunc(c1));				\
389 } while (0)
390 
391 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
392 {
393 	if (boot_cpu_has(X86_FEATURE_TOPOEXT)) {
394 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
395 
396 		if (c->phys_proc_id == o->phys_proc_id &&
397 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2)) {
398 			if (c->cpu_core_id == o->cpu_core_id)
399 				return topology_sane(c, o, "smt");
400 
401 			if ((c->cu_id != 0xff) &&
402 			    (o->cu_id != 0xff) &&
403 			    (c->cu_id == o->cu_id))
404 				return topology_sane(c, o, "smt");
405 		}
406 
407 	} else if (c->phys_proc_id == o->phys_proc_id &&
408 		   c->cpu_core_id == o->cpu_core_id) {
409 		return topology_sane(c, o, "smt");
410 	}
411 
412 	return false;
413 }
414 
415 /*
416  * Define snc_cpu[] for SNC (Sub-NUMA Cluster) CPUs.
417  *
418  * These are Intel CPUs that enumerate an LLC that is shared by
419  * multiple NUMA nodes. The LLC on these systems is shared for
420  * off-package data access but private to the NUMA node (half
421  * of the package) for on-package access.
422  *
423  * CPUID (the source of the information about the LLC) can only
424  * enumerate the cache as being shared *or* unshared, but not
425  * this particular configuration. The CPU in this case enumerates
426  * the cache to be shared across the entire package (spanning both
427  * NUMA nodes).
428  */
429 
430 static const struct x86_cpu_id snc_cpu[] = {
431 	{ X86_VENDOR_INTEL, 6, INTEL_FAM6_SKYLAKE_X },
432 	{}
433 };
434 
435 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
436 {
437 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
438 
439 	/* Do not match if we do not have a valid APICID for cpu: */
440 	if (per_cpu(cpu_llc_id, cpu1) == BAD_APICID)
441 		return false;
442 
443 	/* Do not match if LLC id does not match: */
444 	if (per_cpu(cpu_llc_id, cpu1) != per_cpu(cpu_llc_id, cpu2))
445 		return false;
446 
447 	/*
448 	 * Allow the SNC topology without warning. Return of false
449 	 * means 'c' does not share the LLC of 'o'. This will be
450 	 * reflected to userspace.
451 	 */
452 	if (!topology_same_node(c, o) && x86_match_cpu(snc_cpu))
453 		return false;
454 
455 	return topology_sane(c, o, "llc");
456 }
457 
458 /*
459  * Unlike the other levels, we do not enforce keeping a
460  * multicore group inside a NUMA node.  If this happens, we will
461  * discard the MC level of the topology later.
462  */
463 static bool match_die(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
464 {
465 	if (c->phys_proc_id == o->phys_proc_id)
466 		return true;
467 	return false;
468 }
469 
470 #if defined(CONFIG_SCHED_SMT) || defined(CONFIG_SCHED_MC)
471 static inline int x86_sched_itmt_flags(void)
472 {
473 	return sysctl_sched_itmt_enabled ? SD_ASYM_PACKING : 0;
474 }
475 
476 #ifdef CONFIG_SCHED_MC
477 static int x86_core_flags(void)
478 {
479 	return cpu_core_flags() | x86_sched_itmt_flags();
480 }
481 #endif
482 #ifdef CONFIG_SCHED_SMT
483 static int x86_smt_flags(void)
484 {
485 	return cpu_smt_flags() | x86_sched_itmt_flags();
486 }
487 #endif
488 #endif
489 
490 static struct sched_domain_topology_level x86_numa_in_package_topology[] = {
491 #ifdef CONFIG_SCHED_SMT
492 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
493 #endif
494 #ifdef CONFIG_SCHED_MC
495 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
496 #endif
497 	{ NULL, },
498 };
499 
500 static struct sched_domain_topology_level x86_topology[] = {
501 #ifdef CONFIG_SCHED_SMT
502 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
503 #endif
504 #ifdef CONFIG_SCHED_MC
505 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
506 #endif
507 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
508 	{ NULL, },
509 };
510 
511 /*
512  * Set if a package/die has multiple NUMA nodes inside.
513  * AMD Magny-Cours, Intel Cluster-on-Die, and Intel
514  * Sub-NUMA Clustering have this.
515  */
516 static bool x86_has_numa_in_package;
517 
518 void set_cpu_sibling_map(int cpu)
519 {
520 	bool has_smt = smp_num_siblings > 1;
521 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
522 	struct cpuinfo_x86 *c = &cpu_data(cpu);
523 	struct cpuinfo_x86 *o;
524 	int i, threads;
525 
526 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
527 
528 	if (!has_mp) {
529 		cpumask_set_cpu(cpu, topology_sibling_cpumask(cpu));
530 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
531 		cpumask_set_cpu(cpu, topology_core_cpumask(cpu));
532 		c->booted_cores = 1;
533 		return;
534 	}
535 
536 	for_each_cpu(i, cpu_sibling_setup_mask) {
537 		o = &cpu_data(i);
538 
539 		if ((i == cpu) || (has_smt && match_smt(c, o)))
540 			link_mask(topology_sibling_cpumask, cpu, i);
541 
542 		if ((i == cpu) || (has_mp && match_llc(c, o)))
543 			link_mask(cpu_llc_shared_mask, cpu, i);
544 
545 	}
546 
547 	/*
548 	 * This needs a separate iteration over the cpus because we rely on all
549 	 * topology_sibling_cpumask links to be set-up.
550 	 */
551 	for_each_cpu(i, cpu_sibling_setup_mask) {
552 		o = &cpu_data(i);
553 
554 		if ((i == cpu) || (has_mp && match_die(c, o))) {
555 			link_mask(topology_core_cpumask, cpu, i);
556 
557 			/*
558 			 *  Does this new cpu bringup a new core?
559 			 */
560 			if (cpumask_weight(
561 			    topology_sibling_cpumask(cpu)) == 1) {
562 				/*
563 				 * for each core in package, increment
564 				 * the booted_cores for this new cpu
565 				 */
566 				if (cpumask_first(
567 				    topology_sibling_cpumask(i)) == i)
568 					c->booted_cores++;
569 				/*
570 				 * increment the core count for all
571 				 * the other cpus in this package
572 				 */
573 				if (i != cpu)
574 					cpu_data(i).booted_cores++;
575 			} else if (i != cpu && !c->booted_cores)
576 				c->booted_cores = cpu_data(i).booted_cores;
577 		}
578 		if (match_die(c, o) && !topology_same_node(c, o))
579 			x86_has_numa_in_package = true;
580 	}
581 
582 	threads = cpumask_weight(topology_sibling_cpumask(cpu));
583 	if (threads > __max_smt_threads)
584 		__max_smt_threads = threads;
585 }
586 
587 /* maps the cpu to the sched domain representing multi-core */
588 const struct cpumask *cpu_coregroup_mask(int cpu)
589 {
590 	return cpu_llc_shared_mask(cpu);
591 }
592 
593 static void impress_friends(void)
594 {
595 	int cpu;
596 	unsigned long bogosum = 0;
597 	/*
598 	 * Allow the user to impress friends.
599 	 */
600 	pr_debug("Before bogomips\n");
601 	for_each_possible_cpu(cpu)
602 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
603 			bogosum += cpu_data(cpu).loops_per_jiffy;
604 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
605 		num_online_cpus(),
606 		bogosum/(500000/HZ),
607 		(bogosum/(5000/HZ))%100);
608 
609 	pr_debug("Before bogocount - setting activated=1\n");
610 }
611 
612 void __inquire_remote_apic(int apicid)
613 {
614 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
615 	const char * const names[] = { "ID", "VERSION", "SPIV" };
616 	int timeout;
617 	u32 status;
618 
619 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
620 
621 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
622 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
623 
624 		/*
625 		 * Wait for idle.
626 		 */
627 		status = safe_apic_wait_icr_idle();
628 		if (status)
629 			pr_cont("a previous APIC delivery may have failed\n");
630 
631 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
632 
633 		timeout = 0;
634 		do {
635 			udelay(100);
636 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
637 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
638 
639 		switch (status) {
640 		case APIC_ICR_RR_VALID:
641 			status = apic_read(APIC_RRR);
642 			pr_cont("%08x\n", status);
643 			break;
644 		default:
645 			pr_cont("failed\n");
646 		}
647 	}
648 }
649 
650 /*
651  * The Multiprocessor Specification 1.4 (1997) example code suggests
652  * that there should be a 10ms delay between the BSP asserting INIT
653  * and de-asserting INIT, when starting a remote processor.
654  * But that slows boot and resume on modern processors, which include
655  * many cores and don't require that delay.
656  *
657  * Cmdline "init_cpu_udelay=" is available to over-ride this delay.
658  * Modern processor families are quirked to remove the delay entirely.
659  */
660 #define UDELAY_10MS_DEFAULT 10000
661 
662 static unsigned int init_udelay = UINT_MAX;
663 
664 static int __init cpu_init_udelay(char *str)
665 {
666 	get_option(&str, &init_udelay);
667 
668 	return 0;
669 }
670 early_param("cpu_init_udelay", cpu_init_udelay);
671 
672 static void __init smp_quirk_init_udelay(void)
673 {
674 	/* if cmdline changed it from default, leave it alone */
675 	if (init_udelay != UINT_MAX)
676 		return;
677 
678 	/* if modern processor, use no delay */
679 	if (((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && (boot_cpu_data.x86 == 6)) ||
680 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) && (boot_cpu_data.x86 >= 0x18)) ||
681 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && (boot_cpu_data.x86 >= 0xF))) {
682 		init_udelay = 0;
683 		return;
684 	}
685 	/* else, use legacy delay */
686 	init_udelay = UDELAY_10MS_DEFAULT;
687 }
688 
689 /*
690  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
691  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
692  * won't ... remember to clear down the APIC, etc later.
693  */
694 int
695 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
696 {
697 	unsigned long send_status, accept_status = 0;
698 	int maxlvt;
699 
700 	/* Target chip */
701 	/* Boot on the stack */
702 	/* Kick the second */
703 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, apicid);
704 
705 	pr_debug("Waiting for send to finish...\n");
706 	send_status = safe_apic_wait_icr_idle();
707 
708 	/*
709 	 * Give the other CPU some time to accept the IPI.
710 	 */
711 	udelay(200);
712 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
713 		maxlvt = lapic_get_maxlvt();
714 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
715 			apic_write(APIC_ESR, 0);
716 		accept_status = (apic_read(APIC_ESR) & 0xEF);
717 	}
718 	pr_debug("NMI sent\n");
719 
720 	if (send_status)
721 		pr_err("APIC never delivered???\n");
722 	if (accept_status)
723 		pr_err("APIC delivery error (%lx)\n", accept_status);
724 
725 	return (send_status | accept_status);
726 }
727 
728 static int
729 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
730 {
731 	unsigned long send_status = 0, accept_status = 0;
732 	int maxlvt, num_starts, j;
733 
734 	maxlvt = lapic_get_maxlvt();
735 
736 	/*
737 	 * Be paranoid about clearing APIC errors.
738 	 */
739 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
740 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
741 			apic_write(APIC_ESR, 0);
742 		apic_read(APIC_ESR);
743 	}
744 
745 	pr_debug("Asserting INIT\n");
746 
747 	/*
748 	 * Turn INIT on target chip
749 	 */
750 	/*
751 	 * Send IPI
752 	 */
753 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
754 		       phys_apicid);
755 
756 	pr_debug("Waiting for send to finish...\n");
757 	send_status = safe_apic_wait_icr_idle();
758 
759 	udelay(init_udelay);
760 
761 	pr_debug("Deasserting INIT\n");
762 
763 	/* Target chip */
764 	/* Send IPI */
765 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
766 
767 	pr_debug("Waiting for send to finish...\n");
768 	send_status = safe_apic_wait_icr_idle();
769 
770 	mb();
771 
772 	/*
773 	 * Should we send STARTUP IPIs ?
774 	 *
775 	 * Determine this based on the APIC version.
776 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
777 	 */
778 	if (APIC_INTEGRATED(boot_cpu_apic_version))
779 		num_starts = 2;
780 	else
781 		num_starts = 0;
782 
783 	/*
784 	 * Run STARTUP IPI loop.
785 	 */
786 	pr_debug("#startup loops: %d\n", num_starts);
787 
788 	for (j = 1; j <= num_starts; j++) {
789 		pr_debug("Sending STARTUP #%d\n", j);
790 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
791 			apic_write(APIC_ESR, 0);
792 		apic_read(APIC_ESR);
793 		pr_debug("After apic_write\n");
794 
795 		/*
796 		 * STARTUP IPI
797 		 */
798 
799 		/* Target chip */
800 		/* Boot on the stack */
801 		/* Kick the second */
802 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
803 			       phys_apicid);
804 
805 		/*
806 		 * Give the other CPU some time to accept the IPI.
807 		 */
808 		if (init_udelay == 0)
809 			udelay(10);
810 		else
811 			udelay(300);
812 
813 		pr_debug("Startup point 1\n");
814 
815 		pr_debug("Waiting for send to finish...\n");
816 		send_status = safe_apic_wait_icr_idle();
817 
818 		/*
819 		 * Give the other CPU some time to accept the IPI.
820 		 */
821 		if (init_udelay == 0)
822 			udelay(10);
823 		else
824 			udelay(200);
825 
826 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
827 			apic_write(APIC_ESR, 0);
828 		accept_status = (apic_read(APIC_ESR) & 0xEF);
829 		if (send_status || accept_status)
830 			break;
831 	}
832 	pr_debug("After Startup\n");
833 
834 	if (send_status)
835 		pr_err("APIC never delivered???\n");
836 	if (accept_status)
837 		pr_err("APIC delivery error (%lx)\n", accept_status);
838 
839 	return (send_status | accept_status);
840 }
841 
842 /* reduce the number of lines printed when booting a large cpu count system */
843 static void announce_cpu(int cpu, int apicid)
844 {
845 	static int current_node = NUMA_NO_NODE;
846 	int node = early_cpu_to_node(cpu);
847 	static int width, node_width;
848 
849 	if (!width)
850 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
851 
852 	if (!node_width)
853 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
854 
855 	if (cpu == 1)
856 		printk(KERN_INFO "x86: Booting SMP configuration:\n");
857 
858 	if (system_state < SYSTEM_RUNNING) {
859 		if (node != current_node) {
860 			if (current_node > (-1))
861 				pr_cont("\n");
862 			current_node = node;
863 
864 			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
865 			       node_width - num_digits(node), " ", node);
866 		}
867 
868 		/* Add padding for the BSP */
869 		if (cpu == 1)
870 			pr_cont("%*s", width + 1, " ");
871 
872 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
873 
874 	} else
875 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
876 			node, cpu, apicid);
877 }
878 
879 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
880 {
881 	int cpu;
882 
883 	cpu = smp_processor_id();
884 	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
885 		return NMI_HANDLED;
886 
887 	return NMI_DONE;
888 }
889 
890 /*
891  * Wake up AP by INIT, INIT, STARTUP sequence.
892  *
893  * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
894  * boot-strap code which is not a desired behavior for waking up BSP. To
895  * void the boot-strap code, wake up CPU0 by NMI instead.
896  *
897  * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
898  * (i.e. physically hot removed and then hot added), NMI won't wake it up.
899  * We'll change this code in the future to wake up hard offlined CPU0 if
900  * real platform and request are available.
901  */
902 static int
903 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
904 	       int *cpu0_nmi_registered)
905 {
906 	int id;
907 	int boot_error;
908 
909 	preempt_disable();
910 
911 	/*
912 	 * Wake up AP by INIT, INIT, STARTUP sequence.
913 	 */
914 	if (cpu) {
915 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
916 		goto out;
917 	}
918 
919 	/*
920 	 * Wake up BSP by nmi.
921 	 *
922 	 * Register a NMI handler to help wake up CPU0.
923 	 */
924 	boot_error = register_nmi_handler(NMI_LOCAL,
925 					  wakeup_cpu0_nmi, 0, "wake_cpu0");
926 
927 	if (!boot_error) {
928 		enable_start_cpu0 = 1;
929 		*cpu0_nmi_registered = 1;
930 		if (apic->dest_logical == APIC_DEST_LOGICAL)
931 			id = cpu0_logical_apicid;
932 		else
933 			id = apicid;
934 		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
935 	}
936 
937 out:
938 	preempt_enable();
939 
940 	return boot_error;
941 }
942 
943 void common_cpu_up(unsigned int cpu, struct task_struct *idle)
944 {
945 	/* Just in case we booted with a single CPU. */
946 	alternatives_enable_smp();
947 
948 	per_cpu(current_task, cpu) = idle;
949 
950 #ifdef CONFIG_X86_32
951 	/* Stack for startup_32 can be just as for start_secondary onwards */
952 	irq_ctx_init(cpu);
953 	per_cpu(cpu_current_top_of_stack, cpu) = task_top_of_stack(idle);
954 #else
955 	initial_gs = per_cpu_offset(cpu);
956 #endif
957 }
958 
959 /*
960  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
961  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
962  * Returns zero if CPU booted OK, else error code from
963  * ->wakeup_secondary_cpu.
964  */
965 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle,
966 		       int *cpu0_nmi_registered)
967 {
968 	volatile u32 *trampoline_status =
969 		(volatile u32 *) __va(real_mode_header->trampoline_status);
970 	/* start_ip had better be page-aligned! */
971 	unsigned long start_ip = real_mode_header->trampoline_start;
972 
973 	unsigned long boot_error = 0;
974 	unsigned long timeout;
975 
976 	idle->thread.sp = (unsigned long)task_pt_regs(idle);
977 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_rw(cpu);
978 	initial_code = (unsigned long)start_secondary;
979 	initial_stack  = idle->thread.sp;
980 
981 	/* Enable the espfix hack for this CPU */
982 	init_espfix_ap(cpu);
983 
984 	/* So we see what's up */
985 	announce_cpu(cpu, apicid);
986 
987 	/*
988 	 * This grunge runs the startup process for
989 	 * the targeted processor.
990 	 */
991 
992 	if (x86_platform.legacy.warm_reset) {
993 
994 		pr_debug("Setting warm reset code and vector.\n");
995 
996 		smpboot_setup_warm_reset_vector(start_ip);
997 		/*
998 		 * Be paranoid about clearing APIC errors.
999 		*/
1000 		if (APIC_INTEGRATED(boot_cpu_apic_version)) {
1001 			apic_write(APIC_ESR, 0);
1002 			apic_read(APIC_ESR);
1003 		}
1004 	}
1005 
1006 	/*
1007 	 * AP might wait on cpu_callout_mask in cpu_init() with
1008 	 * cpu_initialized_mask set if previous attempt to online
1009 	 * it timed-out. Clear cpu_initialized_mask so that after
1010 	 * INIT/SIPI it could start with a clean state.
1011 	 */
1012 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1013 	smp_mb();
1014 
1015 	/*
1016 	 * Wake up a CPU in difference cases:
1017 	 * - Use the method in the APIC driver if it's defined
1018 	 * Otherwise,
1019 	 * - Use an INIT boot APIC message for APs or NMI for BSP.
1020 	 */
1021 	if (apic->wakeup_secondary_cpu)
1022 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
1023 	else
1024 		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
1025 						     cpu0_nmi_registered);
1026 
1027 	if (!boot_error) {
1028 		/*
1029 		 * Wait 10s total for first sign of life from AP
1030 		 */
1031 		boot_error = -1;
1032 		timeout = jiffies + 10*HZ;
1033 		while (time_before(jiffies, timeout)) {
1034 			if (cpumask_test_cpu(cpu, cpu_initialized_mask)) {
1035 				/*
1036 				 * Tell AP to proceed with initialization
1037 				 */
1038 				cpumask_set_cpu(cpu, cpu_callout_mask);
1039 				boot_error = 0;
1040 				break;
1041 			}
1042 			schedule();
1043 		}
1044 	}
1045 
1046 	if (!boot_error) {
1047 		/*
1048 		 * Wait till AP completes initial initialization
1049 		 */
1050 		while (!cpumask_test_cpu(cpu, cpu_callin_mask)) {
1051 			/*
1052 			 * Allow other tasks to run while we wait for the
1053 			 * AP to come online. This also gives a chance
1054 			 * for the MTRR work(triggered by the AP coming online)
1055 			 * to be completed in the stop machine context.
1056 			 */
1057 			schedule();
1058 		}
1059 	}
1060 
1061 	/* mark "stuck" area as not stuck */
1062 	*trampoline_status = 0;
1063 
1064 	if (x86_platform.legacy.warm_reset) {
1065 		/*
1066 		 * Cleanup possible dangling ends...
1067 		 */
1068 		smpboot_restore_warm_reset_vector();
1069 	}
1070 
1071 	return boot_error;
1072 }
1073 
1074 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
1075 {
1076 	int apicid = apic->cpu_present_to_apicid(cpu);
1077 	int cpu0_nmi_registered = 0;
1078 	unsigned long flags;
1079 	int err, ret = 0;
1080 
1081 	lockdep_assert_irqs_enabled();
1082 
1083 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
1084 
1085 	if (apicid == BAD_APICID ||
1086 	    !physid_isset(apicid, phys_cpu_present_map) ||
1087 	    !apic->apic_id_valid(apicid)) {
1088 		pr_err("%s: bad cpu %d\n", __func__, cpu);
1089 		return -EINVAL;
1090 	}
1091 
1092 	/*
1093 	 * Already booted CPU?
1094 	 */
1095 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
1096 		pr_debug("do_boot_cpu %d Already started\n", cpu);
1097 		return -ENOSYS;
1098 	}
1099 
1100 	/*
1101 	 * Save current MTRR state in case it was changed since early boot
1102 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
1103 	 */
1104 	mtrr_save_state();
1105 
1106 	/* x86 CPUs take themselves offline, so delayed offline is OK. */
1107 	err = cpu_check_up_prepare(cpu);
1108 	if (err && err != -EBUSY)
1109 		return err;
1110 
1111 	/* the FPU context is blank, nobody can own it */
1112 	per_cpu(fpu_fpregs_owner_ctx, cpu) = NULL;
1113 
1114 	common_cpu_up(cpu, tidle);
1115 
1116 	err = do_boot_cpu(apicid, cpu, tidle, &cpu0_nmi_registered);
1117 	if (err) {
1118 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
1119 		ret = -EIO;
1120 		goto unreg_nmi;
1121 	}
1122 
1123 	/*
1124 	 * Check TSC synchronization with the AP (keep irqs disabled
1125 	 * while doing so):
1126 	 */
1127 	local_irq_save(flags);
1128 	check_tsc_sync_source(cpu);
1129 	local_irq_restore(flags);
1130 
1131 	while (!cpu_online(cpu)) {
1132 		cpu_relax();
1133 		touch_nmi_watchdog();
1134 	}
1135 
1136 unreg_nmi:
1137 	/*
1138 	 * Clean up the nmi handler. Do this after the callin and callout sync
1139 	 * to avoid impact of possible long unregister time.
1140 	 */
1141 	if (cpu0_nmi_registered)
1142 		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
1143 
1144 	return ret;
1145 }
1146 
1147 /**
1148  * arch_disable_smp_support() - disables SMP support for x86 at runtime
1149  */
1150 void arch_disable_smp_support(void)
1151 {
1152 	disable_ioapic_support();
1153 }
1154 
1155 /*
1156  * Fall back to non SMP mode after errors.
1157  *
1158  * RED-PEN audit/test this more. I bet there is more state messed up here.
1159  */
1160 static __init void disable_smp(void)
1161 {
1162 	pr_info("SMP disabled\n");
1163 
1164 	disable_ioapic_support();
1165 
1166 	init_cpu_present(cpumask_of(0));
1167 	init_cpu_possible(cpumask_of(0));
1168 
1169 	if (smp_found_config)
1170 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1171 	else
1172 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1173 	cpumask_set_cpu(0, topology_sibling_cpumask(0));
1174 	cpumask_set_cpu(0, topology_core_cpumask(0));
1175 }
1176 
1177 /*
1178  * Various sanity checks.
1179  */
1180 static void __init smp_sanity_check(void)
1181 {
1182 	preempt_disable();
1183 
1184 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
1185 	if (def_to_bigsmp && nr_cpu_ids > 8) {
1186 		unsigned int cpu;
1187 		unsigned nr;
1188 
1189 		pr_warn("More than 8 CPUs detected - skipping them\n"
1190 			"Use CONFIG_X86_BIGSMP\n");
1191 
1192 		nr = 0;
1193 		for_each_present_cpu(cpu) {
1194 			if (nr >= 8)
1195 				set_cpu_present(cpu, false);
1196 			nr++;
1197 		}
1198 
1199 		nr = 0;
1200 		for_each_possible_cpu(cpu) {
1201 			if (nr >= 8)
1202 				set_cpu_possible(cpu, false);
1203 			nr++;
1204 		}
1205 
1206 		nr_cpu_ids = 8;
1207 	}
1208 #endif
1209 
1210 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1211 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1212 			hard_smp_processor_id());
1213 
1214 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1215 	}
1216 
1217 	/*
1218 	 * Should not be necessary because the MP table should list the boot
1219 	 * CPU too, but we do it for the sake of robustness anyway.
1220 	 */
1221 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1222 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1223 			  boot_cpu_physical_apicid);
1224 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1225 	}
1226 	preempt_enable();
1227 }
1228 
1229 static void __init smp_cpu_index_default(void)
1230 {
1231 	int i;
1232 	struct cpuinfo_x86 *c;
1233 
1234 	for_each_possible_cpu(i) {
1235 		c = &cpu_data(i);
1236 		/* mark all to hotplug */
1237 		c->cpu_index = nr_cpu_ids;
1238 	}
1239 }
1240 
1241 static void __init smp_get_logical_apicid(void)
1242 {
1243 	if (x2apic_mode)
1244 		cpu0_logical_apicid = apic_read(APIC_LDR);
1245 	else
1246 		cpu0_logical_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1247 }
1248 
1249 /*
1250  * Prepare for SMP bootup.
1251  * @max_cpus: configured maximum number of CPUs, It is a legacy parameter
1252  *            for common interface support.
1253  */
1254 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1255 {
1256 	unsigned int i;
1257 
1258 	smp_cpu_index_default();
1259 
1260 	/*
1261 	 * Setup boot CPU information
1262 	 */
1263 	smp_store_boot_cpu_info(); /* Final full version of the data */
1264 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1265 	mb();
1266 
1267 	for_each_possible_cpu(i) {
1268 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1269 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1270 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1271 	}
1272 
1273 	/*
1274 	 * Set 'default' x86 topology, this matches default_topology() in that
1275 	 * it has NUMA nodes as a topology level. See also
1276 	 * native_smp_cpus_done().
1277 	 *
1278 	 * Must be done before set_cpus_sibling_map() is ran.
1279 	 */
1280 	set_sched_topology(x86_topology);
1281 
1282 	set_cpu_sibling_map(0);
1283 
1284 	smp_sanity_check();
1285 
1286 	switch (apic_intr_mode) {
1287 	case APIC_PIC:
1288 	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1289 		disable_smp();
1290 		return;
1291 	case APIC_SYMMETRIC_IO_NO_ROUTING:
1292 		disable_smp();
1293 		/* Setup local timer */
1294 		x86_init.timers.setup_percpu_clockev();
1295 		return;
1296 	case APIC_VIRTUAL_WIRE:
1297 	case APIC_SYMMETRIC_IO:
1298 		break;
1299 	}
1300 
1301 	/* Setup local timer */
1302 	x86_init.timers.setup_percpu_clockev();
1303 
1304 	smp_get_logical_apicid();
1305 
1306 	pr_info("CPU0: ");
1307 	print_cpu_info(&cpu_data(0));
1308 
1309 	native_pv_lock_init();
1310 
1311 	uv_system_init();
1312 
1313 	set_mtrr_aps_delayed_init();
1314 
1315 	smp_quirk_init_udelay();
1316 
1317 	speculative_store_bypass_ht_init();
1318 }
1319 
1320 void arch_enable_nonboot_cpus_begin(void)
1321 {
1322 	set_mtrr_aps_delayed_init();
1323 }
1324 
1325 void arch_enable_nonboot_cpus_end(void)
1326 {
1327 	mtrr_aps_init();
1328 }
1329 
1330 /*
1331  * Early setup to make printk work.
1332  */
1333 void __init native_smp_prepare_boot_cpu(void)
1334 {
1335 	int me = smp_processor_id();
1336 	switch_to_new_gdt(me);
1337 	/* already set me in cpu_online_mask in boot_cpu_init() */
1338 	cpumask_set_cpu(me, cpu_callout_mask);
1339 	cpu_set_state_online(me);
1340 }
1341 
1342 void __init calculate_max_logical_packages(void)
1343 {
1344 	int ncpus;
1345 
1346 	/*
1347 	 * Today neither Intel nor AMD support heterogenous systems so
1348 	 * extrapolate the boot cpu's data to all packages.
1349 	 */
1350 	ncpus = cpu_data(0).booted_cores * topology_max_smt_threads();
1351 	__max_logical_packages = DIV_ROUND_UP(total_cpus, ncpus);
1352 	pr_info("Max logical packages: %u\n", __max_logical_packages);
1353 }
1354 
1355 void __init native_smp_cpus_done(unsigned int max_cpus)
1356 {
1357 	pr_debug("Boot done\n");
1358 
1359 	calculate_max_logical_packages();
1360 
1361 	if (x86_has_numa_in_package)
1362 		set_sched_topology(x86_numa_in_package_topology);
1363 
1364 	nmi_selftest();
1365 	impress_friends();
1366 	mtrr_aps_init();
1367 }
1368 
1369 static int __initdata setup_possible_cpus = -1;
1370 static int __init _setup_possible_cpus(char *str)
1371 {
1372 	get_option(&str, &setup_possible_cpus);
1373 	return 0;
1374 }
1375 early_param("possible_cpus", _setup_possible_cpus);
1376 
1377 
1378 /*
1379  * cpu_possible_mask should be static, it cannot change as cpu's
1380  * are onlined, or offlined. The reason is per-cpu data-structures
1381  * are allocated by some modules at init time, and dont expect to
1382  * do this dynamically on cpu arrival/departure.
1383  * cpu_present_mask on the other hand can change dynamically.
1384  * In case when cpu_hotplug is not compiled, then we resort to current
1385  * behaviour, which is cpu_possible == cpu_present.
1386  * - Ashok Raj
1387  *
1388  * Three ways to find out the number of additional hotplug CPUs:
1389  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1390  * - The user can overwrite it with possible_cpus=NUM
1391  * - Otherwise don't reserve additional CPUs.
1392  * We do this because additional CPUs waste a lot of memory.
1393  * -AK
1394  */
1395 __init void prefill_possible_map(void)
1396 {
1397 	int i, possible;
1398 
1399 	/* No boot processor was found in mptable or ACPI MADT */
1400 	if (!num_processors) {
1401 		if (boot_cpu_has(X86_FEATURE_APIC)) {
1402 			int apicid = boot_cpu_physical_apicid;
1403 			int cpu = hard_smp_processor_id();
1404 
1405 			pr_warn("Boot CPU (id %d) not listed by BIOS\n", cpu);
1406 
1407 			/* Make sure boot cpu is enumerated */
1408 			if (apic->cpu_present_to_apicid(0) == BAD_APICID &&
1409 			    apic->apic_id_valid(apicid))
1410 				generic_processor_info(apicid, boot_cpu_apic_version);
1411 		}
1412 
1413 		if (!num_processors)
1414 			num_processors = 1;
1415 	}
1416 
1417 	i = setup_max_cpus ?: 1;
1418 	if (setup_possible_cpus == -1) {
1419 		possible = num_processors;
1420 #ifdef CONFIG_HOTPLUG_CPU
1421 		if (setup_max_cpus)
1422 			possible += disabled_cpus;
1423 #else
1424 		if (possible > i)
1425 			possible = i;
1426 #endif
1427 	} else
1428 		possible = setup_possible_cpus;
1429 
1430 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1431 
1432 	/* nr_cpu_ids could be reduced via nr_cpus= */
1433 	if (possible > nr_cpu_ids) {
1434 		pr_warn("%d Processors exceeds NR_CPUS limit of %u\n",
1435 			possible, nr_cpu_ids);
1436 		possible = nr_cpu_ids;
1437 	}
1438 
1439 #ifdef CONFIG_HOTPLUG_CPU
1440 	if (!setup_max_cpus)
1441 #endif
1442 	if (possible > i) {
1443 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1444 			possible, setup_max_cpus);
1445 		possible = i;
1446 	}
1447 
1448 	nr_cpu_ids = possible;
1449 
1450 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1451 		possible, max_t(int, possible - num_processors, 0));
1452 
1453 	reset_cpu_possible_mask();
1454 
1455 	for (i = 0; i < possible; i++)
1456 		set_cpu_possible(i, true);
1457 }
1458 
1459 #ifdef CONFIG_HOTPLUG_CPU
1460 
1461 /* Recompute SMT state for all CPUs on offline */
1462 static void recompute_smt_state(void)
1463 {
1464 	int max_threads, cpu;
1465 
1466 	max_threads = 0;
1467 	for_each_online_cpu (cpu) {
1468 		int threads = cpumask_weight(topology_sibling_cpumask(cpu));
1469 
1470 		if (threads > max_threads)
1471 			max_threads = threads;
1472 	}
1473 	__max_smt_threads = max_threads;
1474 }
1475 
1476 static void remove_siblinginfo(int cpu)
1477 {
1478 	int sibling;
1479 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1480 
1481 	for_each_cpu(sibling, topology_core_cpumask(cpu)) {
1482 		cpumask_clear_cpu(cpu, topology_core_cpumask(sibling));
1483 		/*/
1484 		 * last thread sibling in this cpu core going down
1485 		 */
1486 		if (cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
1487 			cpu_data(sibling).booted_cores--;
1488 	}
1489 
1490 	for_each_cpu(sibling, topology_sibling_cpumask(cpu))
1491 		cpumask_clear_cpu(cpu, topology_sibling_cpumask(sibling));
1492 	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1493 		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1494 	cpumask_clear(cpu_llc_shared_mask(cpu));
1495 	cpumask_clear(topology_sibling_cpumask(cpu));
1496 	cpumask_clear(topology_core_cpumask(cpu));
1497 	c->cpu_core_id = 0;
1498 	c->booted_cores = 0;
1499 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1500 	recompute_smt_state();
1501 }
1502 
1503 static void remove_cpu_from_maps(int cpu)
1504 {
1505 	set_cpu_online(cpu, false);
1506 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1507 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1508 	/* was set by cpu_init() */
1509 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1510 	numa_remove_cpu(cpu);
1511 }
1512 
1513 void cpu_disable_common(void)
1514 {
1515 	int cpu = smp_processor_id();
1516 
1517 	remove_siblinginfo(cpu);
1518 
1519 	/* It's now safe to remove this processor from the online map */
1520 	lock_vector_lock();
1521 	remove_cpu_from_maps(cpu);
1522 	unlock_vector_lock();
1523 	fixup_irqs();
1524 	lapic_offline();
1525 }
1526 
1527 int native_cpu_disable(void)
1528 {
1529 	int ret;
1530 
1531 	ret = lapic_can_unplug_cpu();
1532 	if (ret)
1533 		return ret;
1534 
1535 	clear_local_APIC();
1536 	cpu_disable_common();
1537 
1538 	return 0;
1539 }
1540 
1541 int common_cpu_die(unsigned int cpu)
1542 {
1543 	int ret = 0;
1544 
1545 	/* We don't do anything here: idle task is faking death itself. */
1546 
1547 	/* They ack this in play_dead() by setting CPU_DEAD */
1548 	if (cpu_wait_death(cpu, 5)) {
1549 		if (system_state == SYSTEM_RUNNING)
1550 			pr_info("CPU %u is now offline\n", cpu);
1551 	} else {
1552 		pr_err("CPU %u didn't die...\n", cpu);
1553 		ret = -1;
1554 	}
1555 
1556 	return ret;
1557 }
1558 
1559 void native_cpu_die(unsigned int cpu)
1560 {
1561 	common_cpu_die(cpu);
1562 }
1563 
1564 void play_dead_common(void)
1565 {
1566 	idle_task_exit();
1567 
1568 	/* Ack it */
1569 	(void)cpu_report_death();
1570 
1571 	/*
1572 	 * With physical CPU hotplug, we should halt the cpu
1573 	 */
1574 	local_irq_disable();
1575 }
1576 
1577 static bool wakeup_cpu0(void)
1578 {
1579 	if (smp_processor_id() == 0 && enable_start_cpu0)
1580 		return true;
1581 
1582 	return false;
1583 }
1584 
1585 /*
1586  * We need to flush the caches before going to sleep, lest we have
1587  * dirty data in our caches when we come back up.
1588  */
1589 static inline void mwait_play_dead(void)
1590 {
1591 	unsigned int eax, ebx, ecx, edx;
1592 	unsigned int highest_cstate = 0;
1593 	unsigned int highest_subcstate = 0;
1594 	void *mwait_ptr;
1595 	int i;
1596 
1597 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1598 	    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
1599 		return;
1600 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1601 		return;
1602 	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1603 		return;
1604 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1605 		return;
1606 
1607 	eax = CPUID_MWAIT_LEAF;
1608 	ecx = 0;
1609 	native_cpuid(&eax, &ebx, &ecx, &edx);
1610 
1611 	/*
1612 	 * eax will be 0 if EDX enumeration is not valid.
1613 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1614 	 */
1615 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1616 		eax = 0;
1617 	} else {
1618 		edx >>= MWAIT_SUBSTATE_SIZE;
1619 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1620 			if (edx & MWAIT_SUBSTATE_MASK) {
1621 				highest_cstate = i;
1622 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1623 			}
1624 		}
1625 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1626 			(highest_subcstate - 1);
1627 	}
1628 
1629 	/*
1630 	 * This should be a memory location in a cache line which is
1631 	 * unlikely to be touched by other processors.  The actual
1632 	 * content is immaterial as it is not actually modified in any way.
1633 	 */
1634 	mwait_ptr = &current_thread_info()->flags;
1635 
1636 	wbinvd();
1637 
1638 	while (1) {
1639 		/*
1640 		 * The CLFLUSH is a workaround for erratum AAI65 for
1641 		 * the Xeon 7400 series.  It's not clear it is actually
1642 		 * needed, but it should be harmless in either case.
1643 		 * The WBINVD is insufficient due to the spurious-wakeup
1644 		 * case where we return around the loop.
1645 		 */
1646 		mb();
1647 		clflush(mwait_ptr);
1648 		mb();
1649 		__monitor(mwait_ptr, 0, 0);
1650 		mb();
1651 		__mwait(eax, 0);
1652 		/*
1653 		 * If NMI wants to wake up CPU0, start CPU0.
1654 		 */
1655 		if (wakeup_cpu0())
1656 			start_cpu0();
1657 	}
1658 }
1659 
1660 void hlt_play_dead(void)
1661 {
1662 	if (__this_cpu_read(cpu_info.x86) >= 4)
1663 		wbinvd();
1664 
1665 	while (1) {
1666 		native_halt();
1667 		/*
1668 		 * If NMI wants to wake up CPU0, start CPU0.
1669 		 */
1670 		if (wakeup_cpu0())
1671 			start_cpu0();
1672 	}
1673 }
1674 
1675 void native_play_dead(void)
1676 {
1677 	play_dead_common();
1678 	tboot_shutdown(TB_SHUTDOWN_WFS);
1679 
1680 	mwait_play_dead();	/* Only returns on failure */
1681 	if (cpuidle_play_dead())
1682 		hlt_play_dead();
1683 }
1684 
1685 #else /* ... !CONFIG_HOTPLUG_CPU */
1686 int native_cpu_disable(void)
1687 {
1688 	return -ENOSYS;
1689 }
1690 
1691 void native_cpu_die(unsigned int cpu)
1692 {
1693 	/* We said "no" in __cpu_disable */
1694 	BUG();
1695 }
1696 
1697 void native_play_dead(void)
1698 {
1699 	BUG();
1700 }
1701 
1702 #endif
1703