xref: /openbmc/linux/arch/x86/kernel/smpboot.c (revision 727dede0)
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/bootmem.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/realmode.h>
79 #include <asm/misc.h>
80 
81 /* Number of siblings per CPU package */
82 int smp_num_siblings = 1;
83 EXPORT_SYMBOL(smp_num_siblings);
84 
85 /* Last level cache ID of each logical CPU */
86 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
87 
88 /* representing HT siblings of each logical CPU */
89 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
90 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
91 
92 /* representing HT and core siblings of each logical CPU */
93 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_map);
94 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
95 
96 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_llc_shared_map);
97 
98 /* Per CPU bogomips and other parameters */
99 DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
100 EXPORT_PER_CPU_SYMBOL(cpu_info);
101 
102 /* Logical package management. We might want to allocate that dynamically */
103 static int *physical_to_logical_pkg __read_mostly;
104 static unsigned long *physical_package_map __read_mostly;;
105 static unsigned int max_physical_pkg_id __read_mostly;
106 unsigned int __max_logical_packages __read_mostly;
107 EXPORT_SYMBOL(__max_logical_packages);
108 static unsigned int logical_packages __read_mostly;
109 
110 /* Maximum number of SMT threads on any online core */
111 int __max_smt_threads __read_mostly;
112 
113 /* Flag to indicate if a complete sched domain rebuild is required */
114 bool x86_topology_update;
115 
116 int arch_update_cpu_topology(void)
117 {
118 	int retval = x86_topology_update;
119 
120 	x86_topology_update = false;
121 	return retval;
122 }
123 
124 static inline void smpboot_setup_warm_reset_vector(unsigned long start_eip)
125 {
126 	unsigned long flags;
127 
128 	spin_lock_irqsave(&rtc_lock, flags);
129 	CMOS_WRITE(0xa, 0xf);
130 	spin_unlock_irqrestore(&rtc_lock, flags);
131 	local_flush_tlb();
132 	pr_debug("1.\n");
133 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_HIGH)) =
134 							start_eip >> 4;
135 	pr_debug("2.\n");
136 	*((volatile unsigned short *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) =
137 							start_eip & 0xf;
138 	pr_debug("3.\n");
139 }
140 
141 static inline void smpboot_restore_warm_reset_vector(void)
142 {
143 	unsigned long flags;
144 
145 	/*
146 	 * Install writable page 0 entry to set BIOS data area.
147 	 */
148 	local_flush_tlb();
149 
150 	/*
151 	 * Paranoid:  Set warm reset code and vector here back
152 	 * to default values.
153 	 */
154 	spin_lock_irqsave(&rtc_lock, flags);
155 	CMOS_WRITE(0, 0xf);
156 	spin_unlock_irqrestore(&rtc_lock, flags);
157 
158 	*((volatile u32 *)phys_to_virt(TRAMPOLINE_PHYS_LOW)) = 0;
159 }
160 
161 /*
162  * Report back to the Boot Processor during boot time or to the caller processor
163  * during CPU online.
164  */
165 static void smp_callin(void)
166 {
167 	int cpuid, phys_id;
168 
169 	/*
170 	 * If waken up by an INIT in an 82489DX configuration
171 	 * cpu_callout_mask guarantees we don't get here before
172 	 * an INIT_deassert IPI reaches our local APIC, so it is
173 	 * now safe to touch our local APIC.
174 	 */
175 	cpuid = smp_processor_id();
176 
177 	/*
178 	 * (This works even if the APIC is not enabled.)
179 	 */
180 	phys_id = read_apic_id();
181 
182 	/*
183 	 * the boot CPU has finished the init stage and is spinning
184 	 * on callin_map until we finish. We are free to set up this
185 	 * CPU, first the APIC. (this is probably redundant on most
186 	 * boards)
187 	 */
188 	apic_ap_setup();
189 
190 	/*
191 	 * Save our processor parameters. Note: this information
192 	 * is needed for clock calibration.
193 	 */
194 	smp_store_cpu_info(cpuid);
195 
196 	/*
197 	 * Get our bogomips.
198 	 * Update loops_per_jiffy in cpu_data. Previous call to
199 	 * smp_store_cpu_info() stored a value that is close but not as
200 	 * accurate as the value just calculated.
201 	 */
202 	calibrate_delay();
203 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
204 	pr_debug("Stack at about %p\n", &cpuid);
205 
206 	/*
207 	 * This must be done before setting cpu_online_mask
208 	 * or calling notify_cpu_starting.
209 	 */
210 	set_cpu_sibling_map(raw_smp_processor_id());
211 	wmb();
212 
213 	notify_cpu_starting(cpuid);
214 
215 	/*
216 	 * Allow the master to continue.
217 	 */
218 	cpumask_set_cpu(cpuid, cpu_callin_mask);
219 }
220 
221 static int cpu0_logical_apicid;
222 static int enable_start_cpu0;
223 /*
224  * Activate a secondary processor.
225  */
226 static void notrace start_secondary(void *unused)
227 {
228 	/*
229 	 * Don't put *anything* except direct CPU state initialization
230 	 * before cpu_init(), SMP booting is too fragile that we want to
231 	 * limit the things done here to the most necessary things.
232 	 */
233 	if (boot_cpu_has(X86_FEATURE_PCID))
234 		__write_cr4(__read_cr4() | X86_CR4_PCIDE);
235 
236 #ifdef CONFIG_X86_32
237 	/* switch away from the initial page table */
238 	load_cr3(swapper_pg_dir);
239 	__flush_tlb_all();
240 #endif
241 
242 	cpu_init();
243 	x86_cpuinit.early_percpu_clock_init();
244 	preempt_disable();
245 	smp_callin();
246 
247 	enable_start_cpu0 = 0;
248 
249 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
250 	barrier();
251 	/*
252 	 * Check TSC synchronization with the BP:
253 	 */
254 	check_tsc_sync_target();
255 
256 	/*
257 	 * Lock vector_lock and initialize the vectors on this cpu
258 	 * before setting the cpu online. We must set it online with
259 	 * vector_lock held to prevent a concurrent setup/teardown
260 	 * from seeing a half valid vector space.
261 	 */
262 	lock_vector_lock();
263 	setup_vector_irq(smp_processor_id());
264 	set_cpu_online(smp_processor_id(), true);
265 	unlock_vector_lock();
266 	cpu_set_state_online(smp_processor_id());
267 	x86_platform.nmi_init();
268 
269 	/* enable local interrupts */
270 	local_irq_enable();
271 
272 	/* to prevent fake stack check failure in clock setup */
273 	boot_init_stack_canary();
274 
275 	x86_cpuinit.setup_percpu_clockev();
276 
277 	wmb();
278 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
279 }
280 
281 /**
282  * topology_update_package_map - Update the physical to logical package map
283  * @pkg:	The physical package id as retrieved via CPUID
284  * @cpu:	The cpu for which this is updated
285  */
286 int topology_update_package_map(unsigned int pkg, unsigned int cpu)
287 {
288 	unsigned int new;
289 
290 	/* Called from early boot ? */
291 	if (!physical_package_map)
292 		return 0;
293 
294 	if (pkg >= max_physical_pkg_id)
295 		return -EINVAL;
296 
297 	/* Set the logical package id */
298 	if (test_and_set_bit(pkg, physical_package_map))
299 		goto found;
300 
301 	if (logical_packages >= __max_logical_packages) {
302 		pr_warn("Package %u of CPU %u exceeds BIOS package data %u.\n",
303 			logical_packages, cpu, __max_logical_packages);
304 		return -ENOSPC;
305 	}
306 
307 	new = logical_packages++;
308 	if (new != pkg) {
309 		pr_info("CPU %u Converting physical %u to logical package %u\n",
310 			cpu, pkg, new);
311 	}
312 	physical_to_logical_pkg[pkg] = new;
313 
314 found:
315 	cpu_data(cpu).logical_proc_id = physical_to_logical_pkg[pkg];
316 	return 0;
317 }
318 
319 /**
320  * topology_phys_to_logical_pkg - Map a physical package id to a logical
321  *
322  * Returns logical package id or -1 if not found
323  */
324 int topology_phys_to_logical_pkg(unsigned int phys_pkg)
325 {
326 	if (phys_pkg >= max_physical_pkg_id)
327 		return -1;
328 	return physical_to_logical_pkg[phys_pkg];
329 }
330 EXPORT_SYMBOL(topology_phys_to_logical_pkg);
331 
332 static void __init smp_init_package_map(struct cpuinfo_x86 *c, unsigned int cpu)
333 {
334 	unsigned int ncpus;
335 	size_t size;
336 
337 	/*
338 	 * Today neither Intel nor AMD support heterogenous systems. That
339 	 * might change in the future....
340 	 *
341 	 * While ideally we'd want '* smp_num_siblings' in the below @ncpus
342 	 * computation, this won't actually work since some Intel BIOSes
343 	 * report inconsistent HT data when they disable HT.
344 	 *
345 	 * In particular, they reduce the APIC-IDs to only include the cores,
346 	 * but leave the CPUID topology to say there are (2) siblings.
347 	 * This means we don't know how many threads there will be until
348 	 * after the APIC enumeration.
349 	 *
350 	 * By not including this we'll sometimes over-estimate the number of
351 	 * logical packages by the amount of !present siblings, but this is
352 	 * still better than MAX_LOCAL_APIC.
353 	 *
354 	 * We use total_cpus not nr_cpu_ids because nr_cpu_ids can be limited
355 	 * on the command line leading to a similar issue as the HT disable
356 	 * problem because the hyperthreads are usually enumerated after the
357 	 * primary cores.
358 	 */
359 	ncpus = boot_cpu_data.x86_max_cores;
360 	if (!ncpus) {
361 		pr_warn("x86_max_cores == zero !?!?");
362 		ncpus = 1;
363 	}
364 
365 	__max_logical_packages = DIV_ROUND_UP(total_cpus, ncpus);
366 	logical_packages = 0;
367 
368 	/*
369 	 * Possibly larger than what we need as the number of apic ids per
370 	 * package can be smaller than the actual used apic ids.
371 	 */
372 	max_physical_pkg_id = DIV_ROUND_UP(MAX_LOCAL_APIC, ncpus);
373 	size = max_physical_pkg_id * sizeof(unsigned int);
374 	physical_to_logical_pkg = kmalloc(size, GFP_KERNEL);
375 	memset(physical_to_logical_pkg, 0xff, size);
376 	size = BITS_TO_LONGS(max_physical_pkg_id) * sizeof(unsigned long);
377 	physical_package_map = kzalloc(size, GFP_KERNEL);
378 
379 	pr_info("Max logical packages: %u\n", __max_logical_packages);
380 
381 	topology_update_package_map(c->phys_proc_id, cpu);
382 }
383 
384 void __init smp_store_boot_cpu_info(void)
385 {
386 	int id = 0; /* CPU 0 */
387 	struct cpuinfo_x86 *c = &cpu_data(id);
388 
389 	*c = boot_cpu_data;
390 	c->cpu_index = id;
391 	smp_init_package_map(c, id);
392 }
393 
394 /*
395  * The bootstrap kernel entry code has set these up. Save them for
396  * a given CPU
397  */
398 void smp_store_cpu_info(int id)
399 {
400 	struct cpuinfo_x86 *c = &cpu_data(id);
401 
402 	*c = boot_cpu_data;
403 	c->cpu_index = id;
404 	/*
405 	 * During boot time, CPU0 has this setup already. Save the info when
406 	 * bringing up AP or offlined CPU0.
407 	 */
408 	identify_secondary_cpu(c);
409 }
410 
411 static bool
412 topology_same_node(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
413 {
414 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
415 
416 	return (cpu_to_node(cpu1) == cpu_to_node(cpu2));
417 }
418 
419 static bool
420 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
421 {
422 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
423 
424 	return !WARN_ONCE(!topology_same_node(c, o),
425 		"sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
426 		"[node: %d != %d]. Ignoring dependency.\n",
427 		cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
428 }
429 
430 #define link_mask(mfunc, c1, c2)					\
431 do {									\
432 	cpumask_set_cpu((c1), mfunc(c2));				\
433 	cpumask_set_cpu((c2), mfunc(c1));				\
434 } while (0)
435 
436 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
437 {
438 	if (boot_cpu_has(X86_FEATURE_TOPOEXT)) {
439 		int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
440 
441 		if (c->phys_proc_id == o->phys_proc_id &&
442 		    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2)) {
443 			if (c->cpu_core_id == o->cpu_core_id)
444 				return topology_sane(c, o, "smt");
445 
446 			if ((c->cu_id != 0xff) &&
447 			    (o->cu_id != 0xff) &&
448 			    (c->cu_id == o->cu_id))
449 				return topology_sane(c, o, "smt");
450 		}
451 
452 	} else if (c->phys_proc_id == o->phys_proc_id &&
453 		   c->cpu_core_id == o->cpu_core_id) {
454 		return topology_sane(c, o, "smt");
455 	}
456 
457 	return false;
458 }
459 
460 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
461 {
462 	int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
463 
464 	if (per_cpu(cpu_llc_id, cpu1) != BAD_APICID &&
465 	    per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2))
466 		return topology_sane(c, o, "llc");
467 
468 	return false;
469 }
470 
471 /*
472  * Unlike the other levels, we do not enforce keeping a
473  * multicore group inside a NUMA node.  If this happens, we will
474  * discard the MC level of the topology later.
475  */
476 static bool match_die(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
477 {
478 	if (c->phys_proc_id == o->phys_proc_id)
479 		return true;
480 	return false;
481 }
482 
483 #if defined(CONFIG_SCHED_SMT) || defined(CONFIG_SCHED_MC)
484 static inline int x86_sched_itmt_flags(void)
485 {
486 	return sysctl_sched_itmt_enabled ? SD_ASYM_PACKING : 0;
487 }
488 
489 #ifdef CONFIG_SCHED_MC
490 static int x86_core_flags(void)
491 {
492 	return cpu_core_flags() | x86_sched_itmt_flags();
493 }
494 #endif
495 #ifdef CONFIG_SCHED_SMT
496 static int x86_smt_flags(void)
497 {
498 	return cpu_smt_flags() | x86_sched_itmt_flags();
499 }
500 #endif
501 #endif
502 
503 static struct sched_domain_topology_level x86_numa_in_package_topology[] = {
504 #ifdef CONFIG_SCHED_SMT
505 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
506 #endif
507 #ifdef CONFIG_SCHED_MC
508 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
509 #endif
510 	{ NULL, },
511 };
512 
513 static struct sched_domain_topology_level x86_topology[] = {
514 #ifdef CONFIG_SCHED_SMT
515 	{ cpu_smt_mask, x86_smt_flags, SD_INIT_NAME(SMT) },
516 #endif
517 #ifdef CONFIG_SCHED_MC
518 	{ cpu_coregroup_mask, x86_core_flags, SD_INIT_NAME(MC) },
519 #endif
520 	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
521 	{ NULL, },
522 };
523 
524 /*
525  * Set if a package/die has multiple NUMA nodes inside.
526  * AMD Magny-Cours and Intel Cluster-on-Die have this.
527  */
528 static bool x86_has_numa_in_package;
529 
530 void set_cpu_sibling_map(int cpu)
531 {
532 	bool has_smt = smp_num_siblings > 1;
533 	bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
534 	struct cpuinfo_x86 *c = &cpu_data(cpu);
535 	struct cpuinfo_x86 *o;
536 	int i, threads;
537 
538 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
539 
540 	if (!has_mp) {
541 		cpumask_set_cpu(cpu, topology_sibling_cpumask(cpu));
542 		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
543 		cpumask_set_cpu(cpu, topology_core_cpumask(cpu));
544 		c->booted_cores = 1;
545 		return;
546 	}
547 
548 	for_each_cpu(i, cpu_sibling_setup_mask) {
549 		o = &cpu_data(i);
550 
551 		if ((i == cpu) || (has_smt && match_smt(c, o)))
552 			link_mask(topology_sibling_cpumask, cpu, i);
553 
554 		if ((i == cpu) || (has_mp && match_llc(c, o)))
555 			link_mask(cpu_llc_shared_mask, cpu, i);
556 
557 	}
558 
559 	/*
560 	 * This needs a separate iteration over the cpus because we rely on all
561 	 * topology_sibling_cpumask links to be set-up.
562 	 */
563 	for_each_cpu(i, cpu_sibling_setup_mask) {
564 		o = &cpu_data(i);
565 
566 		if ((i == cpu) || (has_mp && match_die(c, o))) {
567 			link_mask(topology_core_cpumask, cpu, i);
568 
569 			/*
570 			 *  Does this new cpu bringup a new core?
571 			 */
572 			if (cpumask_weight(
573 			    topology_sibling_cpumask(cpu)) == 1) {
574 				/*
575 				 * for each core in package, increment
576 				 * the booted_cores for this new cpu
577 				 */
578 				if (cpumask_first(
579 				    topology_sibling_cpumask(i)) == i)
580 					c->booted_cores++;
581 				/*
582 				 * increment the core count for all
583 				 * the other cpus in this package
584 				 */
585 				if (i != cpu)
586 					cpu_data(i).booted_cores++;
587 			} else if (i != cpu && !c->booted_cores)
588 				c->booted_cores = cpu_data(i).booted_cores;
589 		}
590 		if (match_die(c, o) && !topology_same_node(c, o))
591 			x86_has_numa_in_package = true;
592 	}
593 
594 	threads = cpumask_weight(topology_sibling_cpumask(cpu));
595 	if (threads > __max_smt_threads)
596 		__max_smt_threads = threads;
597 }
598 
599 /* maps the cpu to the sched domain representing multi-core */
600 const struct cpumask *cpu_coregroup_mask(int cpu)
601 {
602 	return cpu_llc_shared_mask(cpu);
603 }
604 
605 static void impress_friends(void)
606 {
607 	int cpu;
608 	unsigned long bogosum = 0;
609 	/*
610 	 * Allow the user to impress friends.
611 	 */
612 	pr_debug("Before bogomips\n");
613 	for_each_possible_cpu(cpu)
614 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
615 			bogosum += cpu_data(cpu).loops_per_jiffy;
616 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
617 		num_online_cpus(),
618 		bogosum/(500000/HZ),
619 		(bogosum/(5000/HZ))%100);
620 
621 	pr_debug("Before bogocount - setting activated=1\n");
622 }
623 
624 void __inquire_remote_apic(int apicid)
625 {
626 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
627 	const char * const names[] = { "ID", "VERSION", "SPIV" };
628 	int timeout;
629 	u32 status;
630 
631 	pr_info("Inquiring remote APIC 0x%x...\n", apicid);
632 
633 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
634 		pr_info("... APIC 0x%x %s: ", apicid, names[i]);
635 
636 		/*
637 		 * Wait for idle.
638 		 */
639 		status = safe_apic_wait_icr_idle();
640 		if (status)
641 			pr_cont("a previous APIC delivery may have failed\n");
642 
643 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
644 
645 		timeout = 0;
646 		do {
647 			udelay(100);
648 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
649 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
650 
651 		switch (status) {
652 		case APIC_ICR_RR_VALID:
653 			status = apic_read(APIC_RRR);
654 			pr_cont("%08x\n", status);
655 			break;
656 		default:
657 			pr_cont("failed\n");
658 		}
659 	}
660 }
661 
662 /*
663  * The Multiprocessor Specification 1.4 (1997) example code suggests
664  * that there should be a 10ms delay between the BSP asserting INIT
665  * and de-asserting INIT, when starting a remote processor.
666  * But that slows boot and resume on modern processors, which include
667  * many cores and don't require that delay.
668  *
669  * Cmdline "init_cpu_udelay=" is available to over-ride this delay.
670  * Modern processor families are quirked to remove the delay entirely.
671  */
672 #define UDELAY_10MS_DEFAULT 10000
673 
674 static unsigned int init_udelay = UINT_MAX;
675 
676 static int __init cpu_init_udelay(char *str)
677 {
678 	get_option(&str, &init_udelay);
679 
680 	return 0;
681 }
682 early_param("cpu_init_udelay", cpu_init_udelay);
683 
684 static void __init smp_quirk_init_udelay(void)
685 {
686 	/* if cmdline changed it from default, leave it alone */
687 	if (init_udelay != UINT_MAX)
688 		return;
689 
690 	/* if modern processor, use no delay */
691 	if (((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && (boot_cpu_data.x86 == 6)) ||
692 	    ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && (boot_cpu_data.x86 >= 0xF))) {
693 		init_udelay = 0;
694 		return;
695 	}
696 	/* else, use legacy delay */
697 	init_udelay = UDELAY_10MS_DEFAULT;
698 }
699 
700 /*
701  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
702  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
703  * won't ... remember to clear down the APIC, etc later.
704  */
705 int
706 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
707 {
708 	unsigned long send_status, accept_status = 0;
709 	int maxlvt;
710 
711 	/* Target chip */
712 	/* Boot on the stack */
713 	/* Kick the second */
714 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, apicid);
715 
716 	pr_debug("Waiting for send to finish...\n");
717 	send_status = safe_apic_wait_icr_idle();
718 
719 	/*
720 	 * Give the other CPU some time to accept the IPI.
721 	 */
722 	udelay(200);
723 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
724 		maxlvt = lapic_get_maxlvt();
725 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
726 			apic_write(APIC_ESR, 0);
727 		accept_status = (apic_read(APIC_ESR) & 0xEF);
728 	}
729 	pr_debug("NMI sent\n");
730 
731 	if (send_status)
732 		pr_err("APIC never delivered???\n");
733 	if (accept_status)
734 		pr_err("APIC delivery error (%lx)\n", accept_status);
735 
736 	return (send_status | accept_status);
737 }
738 
739 static int
740 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
741 {
742 	unsigned long send_status = 0, accept_status = 0;
743 	int maxlvt, num_starts, j;
744 
745 	maxlvt = lapic_get_maxlvt();
746 
747 	/*
748 	 * Be paranoid about clearing APIC errors.
749 	 */
750 	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
751 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
752 			apic_write(APIC_ESR, 0);
753 		apic_read(APIC_ESR);
754 	}
755 
756 	pr_debug("Asserting INIT\n");
757 
758 	/*
759 	 * Turn INIT on target chip
760 	 */
761 	/*
762 	 * Send IPI
763 	 */
764 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
765 		       phys_apicid);
766 
767 	pr_debug("Waiting for send to finish...\n");
768 	send_status = safe_apic_wait_icr_idle();
769 
770 	udelay(init_udelay);
771 
772 	pr_debug("Deasserting INIT\n");
773 
774 	/* Target chip */
775 	/* Send IPI */
776 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
777 
778 	pr_debug("Waiting for send to finish...\n");
779 	send_status = safe_apic_wait_icr_idle();
780 
781 	mb();
782 
783 	/*
784 	 * Should we send STARTUP IPIs ?
785 	 *
786 	 * Determine this based on the APIC version.
787 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
788 	 */
789 	if (APIC_INTEGRATED(boot_cpu_apic_version))
790 		num_starts = 2;
791 	else
792 		num_starts = 0;
793 
794 	/*
795 	 * Run STARTUP IPI loop.
796 	 */
797 	pr_debug("#startup loops: %d\n", num_starts);
798 
799 	for (j = 1; j <= num_starts; j++) {
800 		pr_debug("Sending STARTUP #%d\n", j);
801 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
802 			apic_write(APIC_ESR, 0);
803 		apic_read(APIC_ESR);
804 		pr_debug("After apic_write\n");
805 
806 		/*
807 		 * STARTUP IPI
808 		 */
809 
810 		/* Target chip */
811 		/* Boot on the stack */
812 		/* Kick the second */
813 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
814 			       phys_apicid);
815 
816 		/*
817 		 * Give the other CPU some time to accept the IPI.
818 		 */
819 		if (init_udelay == 0)
820 			udelay(10);
821 		else
822 			udelay(300);
823 
824 		pr_debug("Startup point 1\n");
825 
826 		pr_debug("Waiting for send to finish...\n");
827 		send_status = safe_apic_wait_icr_idle();
828 
829 		/*
830 		 * Give the other CPU some time to accept the IPI.
831 		 */
832 		if (init_udelay == 0)
833 			udelay(10);
834 		else
835 			udelay(200);
836 
837 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
838 			apic_write(APIC_ESR, 0);
839 		accept_status = (apic_read(APIC_ESR) & 0xEF);
840 		if (send_status || accept_status)
841 			break;
842 	}
843 	pr_debug("After Startup\n");
844 
845 	if (send_status)
846 		pr_err("APIC never delivered???\n");
847 	if (accept_status)
848 		pr_err("APIC delivery error (%lx)\n", accept_status);
849 
850 	return (send_status | accept_status);
851 }
852 
853 /* reduce the number of lines printed when booting a large cpu count system */
854 static void announce_cpu(int cpu, int apicid)
855 {
856 	static int current_node = -1;
857 	int node = early_cpu_to_node(cpu);
858 	static int width, node_width;
859 
860 	if (!width)
861 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
862 
863 	if (!node_width)
864 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
865 
866 	if (cpu == 1)
867 		printk(KERN_INFO "x86: Booting SMP configuration:\n");
868 
869 	if (system_state < SYSTEM_RUNNING) {
870 		if (node != current_node) {
871 			if (current_node > (-1))
872 				pr_cont("\n");
873 			current_node = node;
874 
875 			printk(KERN_INFO ".... node %*s#%d, CPUs:  ",
876 			       node_width - num_digits(node), " ", node);
877 		}
878 
879 		/* Add padding for the BSP */
880 		if (cpu == 1)
881 			pr_cont("%*s", width + 1, " ");
882 
883 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
884 
885 	} else
886 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
887 			node, cpu, apicid);
888 }
889 
890 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
891 {
892 	int cpu;
893 
894 	cpu = smp_processor_id();
895 	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
896 		return NMI_HANDLED;
897 
898 	return NMI_DONE;
899 }
900 
901 /*
902  * Wake up AP by INIT, INIT, STARTUP sequence.
903  *
904  * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
905  * boot-strap code which is not a desired behavior for waking up BSP. To
906  * void the boot-strap code, wake up CPU0 by NMI instead.
907  *
908  * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
909  * (i.e. physically hot removed and then hot added), NMI won't wake it up.
910  * We'll change this code in the future to wake up hard offlined CPU0 if
911  * real platform and request are available.
912  */
913 static int
914 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
915 	       int *cpu0_nmi_registered)
916 {
917 	int id;
918 	int boot_error;
919 
920 	preempt_disable();
921 
922 	/*
923 	 * Wake up AP by INIT, INIT, STARTUP sequence.
924 	 */
925 	if (cpu) {
926 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
927 		goto out;
928 	}
929 
930 	/*
931 	 * Wake up BSP by nmi.
932 	 *
933 	 * Register a NMI handler to help wake up CPU0.
934 	 */
935 	boot_error = register_nmi_handler(NMI_LOCAL,
936 					  wakeup_cpu0_nmi, 0, "wake_cpu0");
937 
938 	if (!boot_error) {
939 		enable_start_cpu0 = 1;
940 		*cpu0_nmi_registered = 1;
941 		if (apic->dest_logical == APIC_DEST_LOGICAL)
942 			id = cpu0_logical_apicid;
943 		else
944 			id = apicid;
945 		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
946 	}
947 
948 out:
949 	preempt_enable();
950 
951 	return boot_error;
952 }
953 
954 void common_cpu_up(unsigned int cpu, struct task_struct *idle)
955 {
956 	/* Just in case we booted with a single CPU. */
957 	alternatives_enable_smp();
958 
959 	per_cpu(current_task, cpu) = idle;
960 
961 #ifdef CONFIG_X86_32
962 	/* Stack for startup_32 can be just as for start_secondary onwards */
963 	irq_ctx_init(cpu);
964 	per_cpu(cpu_current_top_of_stack, cpu) =
965 		(unsigned long)task_stack_page(idle) + THREAD_SIZE;
966 #else
967 	initial_gs = per_cpu_offset(cpu);
968 #endif
969 }
970 
971 /*
972  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
973  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
974  * Returns zero if CPU booted OK, else error code from
975  * ->wakeup_secondary_cpu.
976  */
977 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle,
978 		       int *cpu0_nmi_registered)
979 {
980 	volatile u32 *trampoline_status =
981 		(volatile u32 *) __va(real_mode_header->trampoline_status);
982 	/* start_ip had better be page-aligned! */
983 	unsigned long start_ip = real_mode_header->trampoline_start;
984 
985 	unsigned long boot_error = 0;
986 	unsigned long timeout;
987 
988 	idle->thread.sp = (unsigned long)task_pt_regs(idle);
989 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_rw(cpu);
990 	initial_code = (unsigned long)start_secondary;
991 	initial_stack  = idle->thread.sp;
992 
993 	/*
994 	 * Enable the espfix hack for this CPU
995 	*/
996 #ifdef CONFIG_X86_ESPFIX64
997 	init_espfix_ap(cpu);
998 #endif
999 
1000 	/* So we see what's up */
1001 	announce_cpu(cpu, apicid);
1002 
1003 	/*
1004 	 * This grunge runs the startup process for
1005 	 * the targeted processor.
1006 	 */
1007 
1008 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
1009 
1010 		pr_debug("Setting warm reset code and vector.\n");
1011 
1012 		smpboot_setup_warm_reset_vector(start_ip);
1013 		/*
1014 		 * Be paranoid about clearing APIC errors.
1015 		*/
1016 		if (APIC_INTEGRATED(boot_cpu_apic_version)) {
1017 			apic_write(APIC_ESR, 0);
1018 			apic_read(APIC_ESR);
1019 		}
1020 	}
1021 
1022 	/*
1023 	 * AP might wait on cpu_callout_mask in cpu_init() with
1024 	 * cpu_initialized_mask set if previous attempt to online
1025 	 * it timed-out. Clear cpu_initialized_mask so that after
1026 	 * INIT/SIPI it could start with a clean state.
1027 	 */
1028 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1029 	smp_mb();
1030 
1031 	/*
1032 	 * Wake up a CPU in difference cases:
1033 	 * - Use the method in the APIC driver if it's defined
1034 	 * Otherwise,
1035 	 * - Use an INIT boot APIC message for APs or NMI for BSP.
1036 	 */
1037 	if (apic->wakeup_secondary_cpu)
1038 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
1039 	else
1040 		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
1041 						     cpu0_nmi_registered);
1042 
1043 	if (!boot_error) {
1044 		/*
1045 		 * Wait 10s total for first sign of life from AP
1046 		 */
1047 		boot_error = -1;
1048 		timeout = jiffies + 10*HZ;
1049 		while (time_before(jiffies, timeout)) {
1050 			if (cpumask_test_cpu(cpu, cpu_initialized_mask)) {
1051 				/*
1052 				 * Tell AP to proceed with initialization
1053 				 */
1054 				cpumask_set_cpu(cpu, cpu_callout_mask);
1055 				boot_error = 0;
1056 				break;
1057 			}
1058 			schedule();
1059 		}
1060 	}
1061 
1062 	if (!boot_error) {
1063 		/*
1064 		 * Wait till AP completes initial initialization
1065 		 */
1066 		while (!cpumask_test_cpu(cpu, cpu_callin_mask)) {
1067 			/*
1068 			 * Allow other tasks to run while we wait for the
1069 			 * AP to come online. This also gives a chance
1070 			 * for the MTRR work(triggered by the AP coming online)
1071 			 * to be completed in the stop machine context.
1072 			 */
1073 			schedule();
1074 		}
1075 	}
1076 
1077 	/* mark "stuck" area as not stuck */
1078 	*trampoline_status = 0;
1079 
1080 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
1081 		/*
1082 		 * Cleanup possible dangling ends...
1083 		 */
1084 		smpboot_restore_warm_reset_vector();
1085 	}
1086 
1087 	return boot_error;
1088 }
1089 
1090 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
1091 {
1092 	int apicid = apic->cpu_present_to_apicid(cpu);
1093 	int cpu0_nmi_registered = 0;
1094 	unsigned long flags;
1095 	int err, ret = 0;
1096 
1097 	WARN_ON(irqs_disabled());
1098 
1099 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
1100 
1101 	if (apicid == BAD_APICID ||
1102 	    !physid_isset(apicid, phys_cpu_present_map) ||
1103 	    !apic->apic_id_valid(apicid)) {
1104 		pr_err("%s: bad cpu %d\n", __func__, cpu);
1105 		return -EINVAL;
1106 	}
1107 
1108 	/*
1109 	 * Already booted CPU?
1110 	 */
1111 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
1112 		pr_debug("do_boot_cpu %d Already started\n", cpu);
1113 		return -ENOSYS;
1114 	}
1115 
1116 	/*
1117 	 * Save current MTRR state in case it was changed since early boot
1118 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
1119 	 */
1120 	mtrr_save_state();
1121 
1122 	/* x86 CPUs take themselves offline, so delayed offline is OK. */
1123 	err = cpu_check_up_prepare(cpu);
1124 	if (err && err != -EBUSY)
1125 		return err;
1126 
1127 	/* the FPU context is blank, nobody can own it */
1128 	per_cpu(fpu_fpregs_owner_ctx, cpu) = NULL;
1129 
1130 	common_cpu_up(cpu, tidle);
1131 
1132 	err = do_boot_cpu(apicid, cpu, tidle, &cpu0_nmi_registered);
1133 	if (err) {
1134 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
1135 		ret = -EIO;
1136 		goto unreg_nmi;
1137 	}
1138 
1139 	/*
1140 	 * Check TSC synchronization with the AP (keep irqs disabled
1141 	 * while doing so):
1142 	 */
1143 	local_irq_save(flags);
1144 	check_tsc_sync_source(cpu);
1145 	local_irq_restore(flags);
1146 
1147 	while (!cpu_online(cpu)) {
1148 		cpu_relax();
1149 		touch_nmi_watchdog();
1150 	}
1151 
1152 unreg_nmi:
1153 	/*
1154 	 * Clean up the nmi handler. Do this after the callin and callout sync
1155 	 * to avoid impact of possible long unregister time.
1156 	 */
1157 	if (cpu0_nmi_registered)
1158 		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
1159 
1160 	return ret;
1161 }
1162 
1163 /**
1164  * arch_disable_smp_support() - disables SMP support for x86 at runtime
1165  */
1166 void arch_disable_smp_support(void)
1167 {
1168 	disable_ioapic_support();
1169 }
1170 
1171 /*
1172  * Fall back to non SMP mode after errors.
1173  *
1174  * RED-PEN audit/test this more. I bet there is more state messed up here.
1175  */
1176 static __init void disable_smp(void)
1177 {
1178 	pr_info("SMP disabled\n");
1179 
1180 	disable_ioapic_support();
1181 
1182 	init_cpu_present(cpumask_of(0));
1183 	init_cpu_possible(cpumask_of(0));
1184 
1185 	if (smp_found_config)
1186 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1187 	else
1188 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1189 	cpumask_set_cpu(0, topology_sibling_cpumask(0));
1190 	cpumask_set_cpu(0, topology_core_cpumask(0));
1191 }
1192 
1193 enum {
1194 	SMP_OK,
1195 	SMP_NO_CONFIG,
1196 	SMP_NO_APIC,
1197 	SMP_FORCE_UP,
1198 };
1199 
1200 /*
1201  * Various sanity checks.
1202  */
1203 static int __init smp_sanity_check(unsigned max_cpus)
1204 {
1205 	preempt_disable();
1206 
1207 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
1208 	if (def_to_bigsmp && nr_cpu_ids > 8) {
1209 		unsigned int cpu;
1210 		unsigned nr;
1211 
1212 		pr_warn("More than 8 CPUs detected - skipping them\n"
1213 			"Use CONFIG_X86_BIGSMP\n");
1214 
1215 		nr = 0;
1216 		for_each_present_cpu(cpu) {
1217 			if (nr >= 8)
1218 				set_cpu_present(cpu, false);
1219 			nr++;
1220 		}
1221 
1222 		nr = 0;
1223 		for_each_possible_cpu(cpu) {
1224 			if (nr >= 8)
1225 				set_cpu_possible(cpu, false);
1226 			nr++;
1227 		}
1228 
1229 		nr_cpu_ids = 8;
1230 	}
1231 #endif
1232 
1233 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1234 		pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1235 			hard_smp_processor_id());
1236 
1237 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1238 	}
1239 
1240 	/*
1241 	 * If we couldn't find an SMP configuration at boot time,
1242 	 * get out of here now!
1243 	 */
1244 	if (!smp_found_config && !acpi_lapic) {
1245 		preempt_enable();
1246 		pr_notice("SMP motherboard not detected\n");
1247 		return SMP_NO_CONFIG;
1248 	}
1249 
1250 	/*
1251 	 * Should not be necessary because the MP table should list the boot
1252 	 * CPU too, but we do it for the sake of robustness anyway.
1253 	 */
1254 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1255 		pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1256 			  boot_cpu_physical_apicid);
1257 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1258 	}
1259 	preempt_enable();
1260 
1261 	/*
1262 	 * If we couldn't find a local APIC, then get out of here now!
1263 	 */
1264 	if (APIC_INTEGRATED(boot_cpu_apic_version) &&
1265 	    !boot_cpu_has(X86_FEATURE_APIC)) {
1266 		if (!disable_apic) {
1267 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
1268 				boot_cpu_physical_apicid);
1269 			pr_err("... forcing use of dummy APIC emulation (tell your hw vendor)\n");
1270 		}
1271 		return SMP_NO_APIC;
1272 	}
1273 
1274 	/*
1275 	 * If SMP should be disabled, then really disable it!
1276 	 */
1277 	if (!max_cpus) {
1278 		pr_info("SMP mode deactivated\n");
1279 		return SMP_FORCE_UP;
1280 	}
1281 
1282 	return SMP_OK;
1283 }
1284 
1285 static void __init smp_cpu_index_default(void)
1286 {
1287 	int i;
1288 	struct cpuinfo_x86 *c;
1289 
1290 	for_each_possible_cpu(i) {
1291 		c = &cpu_data(i);
1292 		/* mark all to hotplug */
1293 		c->cpu_index = nr_cpu_ids;
1294 	}
1295 }
1296 
1297 /*
1298  * Prepare for SMP bootup.  The MP table or ACPI has been read
1299  * earlier.  Just do some sanity checking here and enable APIC mode.
1300  */
1301 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1302 {
1303 	unsigned int i;
1304 
1305 	smp_cpu_index_default();
1306 
1307 	/*
1308 	 * Setup boot CPU information
1309 	 */
1310 	smp_store_boot_cpu_info(); /* Final full version of the data */
1311 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1312 	mb();
1313 
1314 	for_each_possible_cpu(i) {
1315 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1316 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1317 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1318 	}
1319 
1320 	/*
1321 	 * Set 'default' x86 topology, this matches default_topology() in that
1322 	 * it has NUMA nodes as a topology level. See also
1323 	 * native_smp_cpus_done().
1324 	 *
1325 	 * Must be done before set_cpus_sibling_map() is ran.
1326 	 */
1327 	set_sched_topology(x86_topology);
1328 
1329 	set_cpu_sibling_map(0);
1330 
1331 	switch (smp_sanity_check(max_cpus)) {
1332 	case SMP_NO_CONFIG:
1333 		disable_smp();
1334 		if (APIC_init_uniprocessor())
1335 			pr_notice("Local APIC not detected. Using dummy APIC emulation.\n");
1336 		return;
1337 	case SMP_NO_APIC:
1338 		disable_smp();
1339 		return;
1340 	case SMP_FORCE_UP:
1341 		disable_smp();
1342 		apic_bsp_setup(false);
1343 		return;
1344 	case SMP_OK:
1345 		break;
1346 	}
1347 
1348 	if (read_apic_id() != boot_cpu_physical_apicid) {
1349 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1350 		     read_apic_id(), boot_cpu_physical_apicid);
1351 		/* Or can we switch back to PIC here? */
1352 	}
1353 
1354 	default_setup_apic_routing();
1355 	cpu0_logical_apicid = apic_bsp_setup(false);
1356 
1357 	pr_info("CPU0: ");
1358 	print_cpu_info(&cpu_data(0));
1359 
1360 	uv_system_init();
1361 
1362 	set_mtrr_aps_delayed_init();
1363 
1364 	smp_quirk_init_udelay();
1365 }
1366 
1367 void arch_enable_nonboot_cpus_begin(void)
1368 {
1369 	set_mtrr_aps_delayed_init();
1370 }
1371 
1372 void arch_enable_nonboot_cpus_end(void)
1373 {
1374 	mtrr_aps_init();
1375 }
1376 
1377 /*
1378  * Early setup to make printk work.
1379  */
1380 void __init native_smp_prepare_boot_cpu(void)
1381 {
1382 	int me = smp_processor_id();
1383 	switch_to_new_gdt(me);
1384 	/* already set me in cpu_online_mask in boot_cpu_init() */
1385 	cpumask_set_cpu(me, cpu_callout_mask);
1386 	cpu_set_state_online(me);
1387 }
1388 
1389 void __init native_smp_cpus_done(unsigned int max_cpus)
1390 {
1391 	pr_debug("Boot done\n");
1392 
1393 	if (x86_has_numa_in_package)
1394 		set_sched_topology(x86_numa_in_package_topology);
1395 
1396 	nmi_selftest();
1397 	impress_friends();
1398 	setup_ioapic_dest();
1399 	mtrr_aps_init();
1400 }
1401 
1402 static int __initdata setup_possible_cpus = -1;
1403 static int __init _setup_possible_cpus(char *str)
1404 {
1405 	get_option(&str, &setup_possible_cpus);
1406 	return 0;
1407 }
1408 early_param("possible_cpus", _setup_possible_cpus);
1409 
1410 
1411 /*
1412  * cpu_possible_mask should be static, it cannot change as cpu's
1413  * are onlined, or offlined. The reason is per-cpu data-structures
1414  * are allocated by some modules at init time, and dont expect to
1415  * do this dynamically on cpu arrival/departure.
1416  * cpu_present_mask on the other hand can change dynamically.
1417  * In case when cpu_hotplug is not compiled, then we resort to current
1418  * behaviour, which is cpu_possible == cpu_present.
1419  * - Ashok Raj
1420  *
1421  * Three ways to find out the number of additional hotplug CPUs:
1422  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1423  * - The user can overwrite it with possible_cpus=NUM
1424  * - Otherwise don't reserve additional CPUs.
1425  * We do this because additional CPUs waste a lot of memory.
1426  * -AK
1427  */
1428 __init void prefill_possible_map(void)
1429 {
1430 	int i, possible;
1431 
1432 	/* No boot processor was found in mptable or ACPI MADT */
1433 	if (!num_processors) {
1434 		if (boot_cpu_has(X86_FEATURE_APIC)) {
1435 			int apicid = boot_cpu_physical_apicid;
1436 			int cpu = hard_smp_processor_id();
1437 
1438 			pr_warn("Boot CPU (id %d) not listed by BIOS\n", cpu);
1439 
1440 			/* Make sure boot cpu is enumerated */
1441 			if (apic->cpu_present_to_apicid(0) == BAD_APICID &&
1442 			    apic->apic_id_valid(apicid))
1443 				generic_processor_info(apicid, boot_cpu_apic_version);
1444 		}
1445 
1446 		if (!num_processors)
1447 			num_processors = 1;
1448 	}
1449 
1450 	i = setup_max_cpus ?: 1;
1451 	if (setup_possible_cpus == -1) {
1452 		possible = num_processors;
1453 #ifdef CONFIG_HOTPLUG_CPU
1454 		if (setup_max_cpus)
1455 			possible += disabled_cpus;
1456 #else
1457 		if (possible > i)
1458 			possible = i;
1459 #endif
1460 	} else
1461 		possible = setup_possible_cpus;
1462 
1463 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1464 
1465 	/* nr_cpu_ids could be reduced via nr_cpus= */
1466 	if (possible > nr_cpu_ids) {
1467 		pr_warn("%d Processors exceeds NR_CPUS limit of %u\n",
1468 			possible, nr_cpu_ids);
1469 		possible = nr_cpu_ids;
1470 	}
1471 
1472 #ifdef CONFIG_HOTPLUG_CPU
1473 	if (!setup_max_cpus)
1474 #endif
1475 	if (possible > i) {
1476 		pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1477 			possible, setup_max_cpus);
1478 		possible = i;
1479 	}
1480 
1481 	nr_cpu_ids = possible;
1482 
1483 	pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1484 		possible, max_t(int, possible - num_processors, 0));
1485 
1486 	reset_cpu_possible_mask();
1487 
1488 	for (i = 0; i < possible; i++)
1489 		set_cpu_possible(i, true);
1490 }
1491 
1492 #ifdef CONFIG_HOTPLUG_CPU
1493 
1494 /* Recompute SMT state for all CPUs on offline */
1495 static void recompute_smt_state(void)
1496 {
1497 	int max_threads, cpu;
1498 
1499 	max_threads = 0;
1500 	for_each_online_cpu (cpu) {
1501 		int threads = cpumask_weight(topology_sibling_cpumask(cpu));
1502 
1503 		if (threads > max_threads)
1504 			max_threads = threads;
1505 	}
1506 	__max_smt_threads = max_threads;
1507 }
1508 
1509 static void remove_siblinginfo(int cpu)
1510 {
1511 	int sibling;
1512 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1513 
1514 	for_each_cpu(sibling, topology_core_cpumask(cpu)) {
1515 		cpumask_clear_cpu(cpu, topology_core_cpumask(sibling));
1516 		/*/
1517 		 * last thread sibling in this cpu core going down
1518 		 */
1519 		if (cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
1520 			cpu_data(sibling).booted_cores--;
1521 	}
1522 
1523 	for_each_cpu(sibling, topology_sibling_cpumask(cpu))
1524 		cpumask_clear_cpu(cpu, topology_sibling_cpumask(sibling));
1525 	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1526 		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1527 	cpumask_clear(cpu_llc_shared_mask(cpu));
1528 	cpumask_clear(topology_sibling_cpumask(cpu));
1529 	cpumask_clear(topology_core_cpumask(cpu));
1530 	c->phys_proc_id = 0;
1531 	c->cpu_core_id = 0;
1532 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1533 	recompute_smt_state();
1534 }
1535 
1536 static void remove_cpu_from_maps(int cpu)
1537 {
1538 	set_cpu_online(cpu, false);
1539 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1540 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1541 	/* was set by cpu_init() */
1542 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1543 	numa_remove_cpu(cpu);
1544 }
1545 
1546 void cpu_disable_common(void)
1547 {
1548 	int cpu = smp_processor_id();
1549 
1550 	remove_siblinginfo(cpu);
1551 
1552 	/* It's now safe to remove this processor from the online map */
1553 	lock_vector_lock();
1554 	remove_cpu_from_maps(cpu);
1555 	unlock_vector_lock();
1556 	fixup_irqs();
1557 }
1558 
1559 int native_cpu_disable(void)
1560 {
1561 	int ret;
1562 
1563 	ret = check_irq_vectors_for_cpu_disable();
1564 	if (ret)
1565 		return ret;
1566 
1567 	clear_local_APIC();
1568 	cpu_disable_common();
1569 
1570 	return 0;
1571 }
1572 
1573 int common_cpu_die(unsigned int cpu)
1574 {
1575 	int ret = 0;
1576 
1577 	/* We don't do anything here: idle task is faking death itself. */
1578 
1579 	/* They ack this in play_dead() by setting CPU_DEAD */
1580 	if (cpu_wait_death(cpu, 5)) {
1581 		if (system_state == SYSTEM_RUNNING)
1582 			pr_info("CPU %u is now offline\n", cpu);
1583 	} else {
1584 		pr_err("CPU %u didn't die...\n", cpu);
1585 		ret = -1;
1586 	}
1587 
1588 	return ret;
1589 }
1590 
1591 void native_cpu_die(unsigned int cpu)
1592 {
1593 	common_cpu_die(cpu);
1594 }
1595 
1596 void play_dead_common(void)
1597 {
1598 	idle_task_exit();
1599 
1600 	/* Ack it */
1601 	(void)cpu_report_death();
1602 
1603 	/*
1604 	 * With physical CPU hotplug, we should halt the cpu
1605 	 */
1606 	local_irq_disable();
1607 }
1608 
1609 static bool wakeup_cpu0(void)
1610 {
1611 	if (smp_processor_id() == 0 && enable_start_cpu0)
1612 		return true;
1613 
1614 	return false;
1615 }
1616 
1617 /*
1618  * We need to flush the caches before going to sleep, lest we have
1619  * dirty data in our caches when we come back up.
1620  */
1621 static inline void mwait_play_dead(void)
1622 {
1623 	unsigned int eax, ebx, ecx, edx;
1624 	unsigned int highest_cstate = 0;
1625 	unsigned int highest_subcstate = 0;
1626 	void *mwait_ptr;
1627 	int i;
1628 
1629 	if (!this_cpu_has(X86_FEATURE_MWAIT))
1630 		return;
1631 	if (!this_cpu_has(X86_FEATURE_CLFLUSH))
1632 		return;
1633 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1634 		return;
1635 
1636 	eax = CPUID_MWAIT_LEAF;
1637 	ecx = 0;
1638 	native_cpuid(&eax, &ebx, &ecx, &edx);
1639 
1640 	/*
1641 	 * eax will be 0 if EDX enumeration is not valid.
1642 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1643 	 */
1644 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1645 		eax = 0;
1646 	} else {
1647 		edx >>= MWAIT_SUBSTATE_SIZE;
1648 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1649 			if (edx & MWAIT_SUBSTATE_MASK) {
1650 				highest_cstate = i;
1651 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1652 			}
1653 		}
1654 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1655 			(highest_subcstate - 1);
1656 	}
1657 
1658 	/*
1659 	 * This should be a memory location in a cache line which is
1660 	 * unlikely to be touched by other processors.  The actual
1661 	 * content is immaterial as it is not actually modified in any way.
1662 	 */
1663 	mwait_ptr = &current_thread_info()->flags;
1664 
1665 	wbinvd();
1666 
1667 	while (1) {
1668 		/*
1669 		 * The CLFLUSH is a workaround for erratum AAI65 for
1670 		 * the Xeon 7400 series.  It's not clear it is actually
1671 		 * needed, but it should be harmless in either case.
1672 		 * The WBINVD is insufficient due to the spurious-wakeup
1673 		 * case where we return around the loop.
1674 		 */
1675 		mb();
1676 		clflush(mwait_ptr);
1677 		mb();
1678 		__monitor(mwait_ptr, 0, 0);
1679 		mb();
1680 		__mwait(eax, 0);
1681 		/*
1682 		 * If NMI wants to wake up CPU0, start CPU0.
1683 		 */
1684 		if (wakeup_cpu0())
1685 			start_cpu0();
1686 	}
1687 }
1688 
1689 void hlt_play_dead(void)
1690 {
1691 	if (__this_cpu_read(cpu_info.x86) >= 4)
1692 		wbinvd();
1693 
1694 	while (1) {
1695 		native_halt();
1696 		/*
1697 		 * If NMI wants to wake up CPU0, start CPU0.
1698 		 */
1699 		if (wakeup_cpu0())
1700 			start_cpu0();
1701 	}
1702 }
1703 
1704 void native_play_dead(void)
1705 {
1706 	play_dead_common();
1707 	tboot_shutdown(TB_SHUTDOWN_WFS);
1708 
1709 	mwait_play_dead();	/* Only returns on failure */
1710 	if (cpuidle_play_dead())
1711 		hlt_play_dead();
1712 }
1713 
1714 #else /* ... !CONFIG_HOTPLUG_CPU */
1715 int native_cpu_disable(void)
1716 {
1717 	return -ENOSYS;
1718 }
1719 
1720 void native_cpu_die(unsigned int cpu)
1721 {
1722 	/* We said "no" in __cpu_disable */
1723 	BUG();
1724 }
1725 
1726 void native_play_dead(void)
1727 {
1728 	BUG();
1729 }
1730 
1731 #endif
1732