xref: /linux/arch/x86/kernel/smpboot.c (revision c0c9209ddd96bc4f1d70a8b9958710671e076080)
1 /*
2  *	x86 SMP booting functions
3  *
4  *	(c) 1995 Alan Cox, Building #3 <alan@redhat.com>
5  *	(c) 1998, 1999, 2000 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 #include <linux/init.h>
43 #include <linux/smp.h>
44 #include <linux/module.h>
45 #include <linux/sched.h>
46 #include <linux/percpu.h>
47 #include <linux/bootmem.h>
48 #include <linux/err.h>
49 #include <linux/nmi.h>
50 
51 #include <asm/acpi.h>
52 #include <asm/desc.h>
53 #include <asm/nmi.h>
54 #include <asm/irq.h>
55 #include <asm/smp.h>
56 #include <asm/trampoline.h>
57 #include <asm/cpu.h>
58 #include <asm/numa.h>
59 #include <asm/pgtable.h>
60 #include <asm/tlbflush.h>
61 #include <asm/mtrr.h>
62 #include <asm/vmi.h>
63 #include <asm/genapic.h>
64 #include <linux/mc146818rtc.h>
65 
66 #include <mach_apic.h>
67 #include <mach_wakecpu.h>
68 #include <smpboot_hooks.h>
69 
70 #ifdef CONFIG_X86_32
71 u8 apicid_2_node[MAX_APICID];
72 static int low_mappings;
73 #endif
74 
75 /* State of each CPU */
76 DEFINE_PER_CPU(int, cpu_state) = { 0 };
77 
78 /* Store all idle threads, this can be reused instead of creating
79 * a new thread. Also avoids complicated thread destroy functionality
80 * for idle threads.
81 */
82 #ifdef CONFIG_HOTPLUG_CPU
83 /*
84  * Needed only for CONFIG_HOTPLUG_CPU because __cpuinitdata is
85  * removed after init for !CONFIG_HOTPLUG_CPU.
86  */
87 static DEFINE_PER_CPU(struct task_struct *, idle_thread_array);
88 #define get_idle_for_cpu(x)      (per_cpu(idle_thread_array, x))
89 #define set_idle_for_cpu(x, p)   (per_cpu(idle_thread_array, x) = (p))
90 #else
91 struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
92 #define get_idle_for_cpu(x)      (idle_thread_array[(x)])
93 #define set_idle_for_cpu(x, p)   (idle_thread_array[(x)] = (p))
94 #endif
95 
96 /* Number of siblings per CPU package */
97 int smp_num_siblings = 1;
98 EXPORT_SYMBOL(smp_num_siblings);
99 
100 /* Last level cache ID of each logical CPU */
101 DEFINE_PER_CPU(u16, cpu_llc_id) = BAD_APICID;
102 
103 /* bitmap of online cpus */
104 cpumask_t cpu_online_map __read_mostly;
105 EXPORT_SYMBOL(cpu_online_map);
106 
107 cpumask_t cpu_callin_map;
108 cpumask_t cpu_callout_map;
109 cpumask_t cpu_possible_map;
110 EXPORT_SYMBOL(cpu_possible_map);
111 
112 /* representing HT siblings of each logical CPU */
113 DEFINE_PER_CPU(cpumask_t, cpu_sibling_map);
114 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
115 
116 /* representing HT and core siblings of each logical CPU */
117 DEFINE_PER_CPU(cpumask_t, cpu_core_map);
118 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
119 
120 /* Per CPU bogomips and other parameters */
121 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
122 EXPORT_PER_CPU_SYMBOL(cpu_info);
123 
124 static atomic_t init_deasserted;
125 
126 static int boot_cpu_logical_apicid;
127 
128 /* representing cpus for which sibling maps can be computed */
129 static cpumask_t cpu_sibling_setup_map;
130 
131 /* Set if we find a B stepping CPU */
132 int __cpuinitdata smp_b_stepping;
133 
134 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_32)
135 
136 /* which logical CPUs are on which nodes */
137 cpumask_t node_to_cpumask_map[MAX_NUMNODES] __read_mostly =
138 				{ [0 ... MAX_NUMNODES-1] = CPU_MASK_NONE };
139 EXPORT_SYMBOL(node_to_cpumask_map);
140 /* which node each logical CPU is on */
141 int cpu_to_node_map[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 };
142 EXPORT_SYMBOL(cpu_to_node_map);
143 
144 /* set up a mapping between cpu and node. */
145 static void map_cpu_to_node(int cpu, int node)
146 {
147 	printk(KERN_INFO "Mapping cpu %d to node %d\n", cpu, node);
148 	cpu_set(cpu, node_to_cpumask_map[node]);
149 	cpu_to_node_map[cpu] = node;
150 }
151 
152 /* undo a mapping between cpu and node. */
153 static void unmap_cpu_to_node(int cpu)
154 {
155 	int node;
156 
157 	printk(KERN_INFO "Unmapping cpu %d from all nodes\n", cpu);
158 	for (node = 0; node < MAX_NUMNODES; node++)
159 		cpu_clear(cpu, node_to_cpumask_map[node]);
160 	cpu_to_node_map[cpu] = 0;
161 }
162 #else /* !(CONFIG_NUMA && CONFIG_X86_32) */
163 #define map_cpu_to_node(cpu, node)	({})
164 #define unmap_cpu_to_node(cpu)	({})
165 #endif
166 
167 #ifdef CONFIG_X86_32
168 u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly =
169 					{ [0 ... NR_CPUS-1] = BAD_APICID };
170 
171 static void map_cpu_to_logical_apicid(void)
172 {
173 	int cpu = smp_processor_id();
174 	int apicid = logical_smp_processor_id();
175 	int node = apicid_to_node(apicid);
176 
177 	if (!node_online(node))
178 		node = first_online_node;
179 
180 	cpu_2_logical_apicid[cpu] = apicid;
181 	map_cpu_to_node(cpu, node);
182 }
183 
184 void numa_remove_cpu(int cpu)
185 {
186 	cpu_2_logical_apicid[cpu] = BAD_APICID;
187 	unmap_cpu_to_node(cpu);
188 }
189 #else
190 #define map_cpu_to_logical_apicid()  do {} while (0)
191 #endif
192 
193 /*
194  * Report back to the Boot Processor.
195  * Running on AP.
196  */
197 static void __cpuinit smp_callin(void)
198 {
199 	int cpuid, phys_id;
200 	unsigned long timeout;
201 
202 	/*
203 	 * If waken up by an INIT in an 82489DX configuration
204 	 * we may get here before an INIT-deassert IPI reaches
205 	 * our local APIC.  We have to wait for the IPI or we'll
206 	 * lock up on an APIC access.
207 	 */
208 	wait_for_init_deassert(&init_deasserted);
209 
210 	/*
211 	 * (This works even if the APIC is not enabled.)
212 	 */
213 	phys_id = GET_APIC_ID(read_apic_id());
214 	cpuid = smp_processor_id();
215 	if (cpu_isset(cpuid, cpu_callin_map)) {
216 		panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
217 					phys_id, cpuid);
218 	}
219 	pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
220 
221 	/*
222 	 * STARTUP IPIs are fragile beasts as they might sometimes
223 	 * trigger some glue motherboard logic. Complete APIC bus
224 	 * silence for 1 second, this overestimates the time the
225 	 * boot CPU is spending to send the up to 2 STARTUP IPIs
226 	 * by a factor of two. This should be enough.
227 	 */
228 
229 	/*
230 	 * Waiting 2s total for startup (udelay is not yet working)
231 	 */
232 	timeout = jiffies + 2*HZ;
233 	while (time_before(jiffies, timeout)) {
234 		/*
235 		 * Has the boot CPU finished it's STARTUP sequence?
236 		 */
237 		if (cpu_isset(cpuid, cpu_callout_map))
238 			break;
239 		cpu_relax();
240 	}
241 
242 	if (!time_before(jiffies, timeout)) {
243 		panic("%s: CPU%d started up but did not get a callout!\n",
244 		      __func__, cpuid);
245 	}
246 
247 	/*
248 	 * the boot CPU has finished the init stage and is spinning
249 	 * on callin_map until we finish. We are free to set up this
250 	 * CPU, first the APIC. (this is probably redundant on most
251 	 * boards)
252 	 */
253 
254 	pr_debug("CALLIN, before setup_local_APIC().\n");
255 	smp_callin_clear_local_apic();
256 	setup_local_APIC();
257 	end_local_APIC_setup();
258 	map_cpu_to_logical_apicid();
259 
260 	/*
261 	 * Get our bogomips.
262 	 *
263 	 * Need to enable IRQs because it can take longer and then
264 	 * the NMI watchdog might kill us.
265 	 */
266 	local_irq_enable();
267 	calibrate_delay();
268 	local_irq_disable();
269 	pr_debug("Stack at about %p\n", &cpuid);
270 
271 	/*
272 	 * Save our processor parameters
273 	 */
274 	smp_store_cpu_info(cpuid);
275 
276 	/*
277 	 * Allow the master to continue.
278 	 */
279 	cpu_set(cpuid, cpu_callin_map);
280 }
281 
282 /*
283  * Activate a secondary processor.
284  */
285 static void __cpuinit start_secondary(void *unused)
286 {
287 	/*
288 	 * Don't put *anything* before cpu_init(), SMP booting is too
289 	 * fragile that we want to limit the things done here to the
290 	 * most necessary things.
291 	 */
292 #ifdef CONFIG_VMI
293 	vmi_bringup();
294 #endif
295 	cpu_init();
296 	preempt_disable();
297 	smp_callin();
298 
299 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
300 	barrier();
301 	/*
302 	 * Check TSC synchronization with the BP:
303 	 */
304 	check_tsc_sync_target();
305 
306 	if (nmi_watchdog == NMI_IO_APIC) {
307 		disable_8259A_irq(0);
308 		enable_NMI_through_LVT0();
309 		enable_8259A_irq(0);
310 	}
311 
312 #ifdef CONFIG_X86_32
313 	while (low_mappings)
314 		cpu_relax();
315 	__flush_tlb_all();
316 #endif
317 
318 	/* This must be done before setting cpu_online_map */
319 	set_cpu_sibling_map(raw_smp_processor_id());
320 	wmb();
321 
322 	/*
323 	 * We need to hold call_lock, so there is no inconsistency
324 	 * between the time smp_call_function() determines number of
325 	 * IPI recipients, and the time when the determination is made
326 	 * for which cpus receive the IPI. Holding this
327 	 * lock helps us to not include this cpu in a currently in progress
328 	 * smp_call_function().
329 	 *
330 	 * We need to hold vector_lock so there the set of online cpus
331 	 * does not change while we are assigning vectors to cpus.  Holding
332 	 * this lock ensures we don't half assign or remove an irq from a cpu.
333 	 */
334 	ipi_call_lock_irq();
335 	lock_vector_lock();
336 	__setup_vector_irq(smp_processor_id());
337 	cpu_set(smp_processor_id(), cpu_online_map);
338 	unlock_vector_lock();
339 	ipi_call_unlock_irq();
340 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
341 
342 	setup_secondary_clock();
343 
344 	wmb();
345 	cpu_idle();
346 }
347 
348 static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c)
349 {
350 	/*
351 	 * Mask B, Pentium, but not Pentium MMX
352 	 */
353 	if (c->x86_vendor == X86_VENDOR_INTEL &&
354 	    c->x86 == 5 &&
355 	    c->x86_mask >= 1 && c->x86_mask <= 4 &&
356 	    c->x86_model <= 3)
357 		/*
358 		 * Remember we have B step Pentia with bugs
359 		 */
360 		smp_b_stepping = 1;
361 
362 	/*
363 	 * Certain Athlons might work (for various values of 'work') in SMP
364 	 * but they are not certified as MP capable.
365 	 */
366 	if ((c->x86_vendor == X86_VENDOR_AMD) && (c->x86 == 6)) {
367 
368 		if (num_possible_cpus() == 1)
369 			goto valid_k7;
370 
371 		/* Athlon 660/661 is valid. */
372 		if ((c->x86_model == 6) && ((c->x86_mask == 0) ||
373 		    (c->x86_mask == 1)))
374 			goto valid_k7;
375 
376 		/* Duron 670 is valid */
377 		if ((c->x86_model == 7) && (c->x86_mask == 0))
378 			goto valid_k7;
379 
380 		/*
381 		 * Athlon 662, Duron 671, and Athlon >model 7 have capability
382 		 * bit. It's worth noting that the A5 stepping (662) of some
383 		 * Athlon XP's have the MP bit set.
384 		 * See http://www.heise.de/newsticker/data/jow-18.10.01-000 for
385 		 * more.
386 		 */
387 		if (((c->x86_model == 6) && (c->x86_mask >= 2)) ||
388 		    ((c->x86_model == 7) && (c->x86_mask >= 1)) ||
389 		     (c->x86_model > 7))
390 			if (cpu_has_mp)
391 				goto valid_k7;
392 
393 		/* If we get here, not a certified SMP capable AMD system. */
394 		add_taint(TAINT_UNSAFE_SMP);
395 	}
396 
397 valid_k7:
398 	;
399 }
400 
401 static void __cpuinit smp_checks(void)
402 {
403 	if (smp_b_stepping)
404 		printk(KERN_WARNING "WARNING: SMP operation may be unreliable"
405 				    "with B stepping processors.\n");
406 
407 	/*
408 	 * Don't taint if we are running SMP kernel on a single non-MP
409 	 * approved Athlon
410 	 */
411 	if (tainted & TAINT_UNSAFE_SMP) {
412 		if (num_online_cpus())
413 			printk(KERN_INFO "WARNING: This combination of AMD"
414 				"processors is not suitable for SMP.\n");
415 		else
416 			tainted &= ~TAINT_UNSAFE_SMP;
417 	}
418 }
419 
420 /*
421  * The bootstrap kernel entry code has set these up. Save them for
422  * a given CPU
423  */
424 
425 void __cpuinit smp_store_cpu_info(int id)
426 {
427 	struct cpuinfo_x86 *c = &cpu_data(id);
428 
429 	*c = boot_cpu_data;
430 	c->cpu_index = id;
431 	if (id != 0)
432 		identify_secondary_cpu(c);
433 	smp_apply_quirks(c);
434 }
435 
436 
437 void __cpuinit set_cpu_sibling_map(int cpu)
438 {
439 	int i;
440 	struct cpuinfo_x86 *c = &cpu_data(cpu);
441 
442 	cpu_set(cpu, cpu_sibling_setup_map);
443 
444 	if (smp_num_siblings > 1) {
445 		for_each_cpu_mask_nr(i, cpu_sibling_setup_map) {
446 			if (c->phys_proc_id == cpu_data(i).phys_proc_id &&
447 			    c->cpu_core_id == cpu_data(i).cpu_core_id) {
448 				cpu_set(i, per_cpu(cpu_sibling_map, cpu));
449 				cpu_set(cpu, per_cpu(cpu_sibling_map, i));
450 				cpu_set(i, per_cpu(cpu_core_map, cpu));
451 				cpu_set(cpu, per_cpu(cpu_core_map, i));
452 				cpu_set(i, c->llc_shared_map);
453 				cpu_set(cpu, cpu_data(i).llc_shared_map);
454 			}
455 		}
456 	} else {
457 		cpu_set(cpu, per_cpu(cpu_sibling_map, cpu));
458 	}
459 
460 	cpu_set(cpu, c->llc_shared_map);
461 
462 	if (current_cpu_data.x86_max_cores == 1) {
463 		per_cpu(cpu_core_map, cpu) = per_cpu(cpu_sibling_map, cpu);
464 		c->booted_cores = 1;
465 		return;
466 	}
467 
468 	for_each_cpu_mask_nr(i, cpu_sibling_setup_map) {
469 		if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
470 		    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
471 			cpu_set(i, c->llc_shared_map);
472 			cpu_set(cpu, cpu_data(i).llc_shared_map);
473 		}
474 		if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
475 			cpu_set(i, per_cpu(cpu_core_map, cpu));
476 			cpu_set(cpu, per_cpu(cpu_core_map, i));
477 			/*
478 			 *  Does this new cpu bringup a new core?
479 			 */
480 			if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1) {
481 				/*
482 				 * for each core in package, increment
483 				 * the booted_cores for this new cpu
484 				 */
485 				if (first_cpu(per_cpu(cpu_sibling_map, i)) == i)
486 					c->booted_cores++;
487 				/*
488 				 * increment the core count for all
489 				 * the other cpus in this package
490 				 */
491 				if (i != cpu)
492 					cpu_data(i).booted_cores++;
493 			} else if (i != cpu && !c->booted_cores)
494 				c->booted_cores = cpu_data(i).booted_cores;
495 		}
496 	}
497 }
498 
499 /* maps the cpu to the sched domain representing multi-core */
500 cpumask_t cpu_coregroup_map(int cpu)
501 {
502 	struct cpuinfo_x86 *c = &cpu_data(cpu);
503 	/*
504 	 * For perf, we return last level cache shared map.
505 	 * And for power savings, we return cpu_core_map
506 	 */
507 	if (sched_mc_power_savings || sched_smt_power_savings)
508 		return per_cpu(cpu_core_map, cpu);
509 	else
510 		return c->llc_shared_map;
511 }
512 
513 static void impress_friends(void)
514 {
515 	int cpu;
516 	unsigned long bogosum = 0;
517 	/*
518 	 * Allow the user to impress friends.
519 	 */
520 	pr_debug("Before bogomips.\n");
521 	for_each_possible_cpu(cpu)
522 		if (cpu_isset(cpu, cpu_callout_map))
523 			bogosum += cpu_data(cpu).loops_per_jiffy;
524 	printk(KERN_INFO
525 		"Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
526 		num_online_cpus(),
527 		bogosum/(500000/HZ),
528 		(bogosum/(5000/HZ))%100);
529 
530 	pr_debug("Before bogocount - setting activated=1.\n");
531 }
532 
533 static inline void __inquire_remote_apic(int apicid)
534 {
535 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
536 	char *names[] = { "ID", "VERSION", "SPIV" };
537 	int timeout;
538 	u32 status;
539 
540 	printk(KERN_INFO "Inquiring remote APIC #%d...\n", apicid);
541 
542 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
543 		printk(KERN_INFO "... APIC #%d %s: ", apicid, names[i]);
544 
545 		/*
546 		 * Wait for idle.
547 		 */
548 		status = safe_apic_wait_icr_idle();
549 		if (status)
550 			printk(KERN_CONT
551 			       "a previous APIC delivery may have failed\n");
552 
553 		apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(apicid));
554 		apic_write(APIC_ICR, APIC_DM_REMRD | regs[i]);
555 
556 		timeout = 0;
557 		do {
558 			udelay(100);
559 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
560 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
561 
562 		switch (status) {
563 		case APIC_ICR_RR_VALID:
564 			status = apic_read(APIC_RRR);
565 			printk(KERN_CONT "%08x\n", status);
566 			break;
567 		default:
568 			printk(KERN_CONT "failed\n");
569 		}
570 	}
571 }
572 
573 #ifdef WAKE_SECONDARY_VIA_NMI
574 /*
575  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
576  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
577  * won't ... remember to clear down the APIC, etc later.
578  */
579 static int __devinit
580 wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
581 {
582 	unsigned long send_status, accept_status = 0;
583 	int maxlvt;
584 
585 	/* Target chip */
586 	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(logical_apicid));
587 
588 	/* Boot on the stack */
589 	/* Kick the second */
590 	apic_write(APIC_ICR, APIC_DM_NMI | APIC_DEST_LOGICAL);
591 
592 	pr_debug("Waiting for send to finish...\n");
593 	send_status = safe_apic_wait_icr_idle();
594 
595 	/*
596 	 * Give the other CPU some time to accept the IPI.
597 	 */
598 	udelay(200);
599 	maxlvt = lapic_get_maxlvt();
600 	if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
601 		apic_write(APIC_ESR, 0);
602 	accept_status = (apic_read(APIC_ESR) & 0xEF);
603 	pr_debug("NMI sent.\n");
604 
605 	if (send_status)
606 		printk(KERN_ERR "APIC never delivered???\n");
607 	if (accept_status)
608 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
609 
610 	return (send_status | accept_status);
611 }
612 #endif	/* WAKE_SECONDARY_VIA_NMI */
613 
614 #ifdef WAKE_SECONDARY_VIA_INIT
615 static int __devinit
616 wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
617 {
618 	unsigned long send_status, accept_status = 0;
619 	int maxlvt, num_starts, j;
620 
621 	if (get_uv_system_type() == UV_NON_UNIQUE_APIC) {
622 		send_status = uv_wakeup_secondary(phys_apicid, start_eip);
623 		atomic_set(&init_deasserted, 1);
624 		return send_status;
625 	}
626 
627 	maxlvt = lapic_get_maxlvt();
628 
629 	/*
630 	 * Be paranoid about clearing APIC errors.
631 	 */
632 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
633 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
634 			apic_write(APIC_ESR, 0);
635 		apic_read(APIC_ESR);
636 	}
637 
638 	pr_debug("Asserting INIT.\n");
639 
640 	/*
641 	 * Turn INIT on target chip
642 	 */
643 	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
644 
645 	/*
646 	 * Send IPI
647 	 */
648 	apic_write(APIC_ICR,
649 		   APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT);
650 
651 	pr_debug("Waiting for send to finish...\n");
652 	send_status = safe_apic_wait_icr_idle();
653 
654 	mdelay(10);
655 
656 	pr_debug("Deasserting INIT.\n");
657 
658 	/* Target chip */
659 	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
660 
661 	/* Send IPI */
662 	apic_write(APIC_ICR, APIC_INT_LEVELTRIG | APIC_DM_INIT);
663 
664 	pr_debug("Waiting for send to finish...\n");
665 	send_status = safe_apic_wait_icr_idle();
666 
667 	mb();
668 	atomic_set(&init_deasserted, 1);
669 
670 	/*
671 	 * Should we send STARTUP IPIs ?
672 	 *
673 	 * Determine this based on the APIC version.
674 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
675 	 */
676 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
677 		num_starts = 2;
678 	else
679 		num_starts = 0;
680 
681 	/*
682 	 * Paravirt / VMI wants a startup IPI hook here to set up the
683 	 * target processor state.
684 	 */
685 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
686 			 (unsigned long)stack_start.sp);
687 
688 	/*
689 	 * Run STARTUP IPI loop.
690 	 */
691 	pr_debug("#startup loops: %d.\n", num_starts);
692 
693 	for (j = 1; j <= num_starts; j++) {
694 		pr_debug("Sending STARTUP #%d.\n", j);
695 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
696 			apic_write(APIC_ESR, 0);
697 		apic_read(APIC_ESR);
698 		pr_debug("After apic_write.\n");
699 
700 		/*
701 		 * STARTUP IPI
702 		 */
703 
704 		/* Target chip */
705 		apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
706 
707 		/* Boot on the stack */
708 		/* Kick the second */
709 		apic_write(APIC_ICR, APIC_DM_STARTUP | (start_eip >> 12));
710 
711 		/*
712 		 * Give the other CPU some time to accept the IPI.
713 		 */
714 		udelay(300);
715 
716 		pr_debug("Startup point 1.\n");
717 
718 		pr_debug("Waiting for send to finish...\n");
719 		send_status = safe_apic_wait_icr_idle();
720 
721 		/*
722 		 * Give the other CPU some time to accept the IPI.
723 		 */
724 		udelay(200);
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 		if (send_status || accept_status)
729 			break;
730 	}
731 	pr_debug("After Startup.\n");
732 
733 	if (send_status)
734 		printk(KERN_ERR "APIC never delivered???\n");
735 	if (accept_status)
736 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
737 
738 	return (send_status | accept_status);
739 }
740 #endif	/* WAKE_SECONDARY_VIA_INIT */
741 
742 struct create_idle {
743 	struct work_struct work;
744 	struct task_struct *idle;
745 	struct completion done;
746 	int cpu;
747 };
748 
749 static void __cpuinit do_fork_idle(struct work_struct *work)
750 {
751 	struct create_idle *c_idle =
752 		container_of(work, struct create_idle, work);
753 
754 	c_idle->idle = fork_idle(c_idle->cpu);
755 	complete(&c_idle->done);
756 }
757 
758 #ifdef CONFIG_X86_64
759 
760 /* __ref because it's safe to call free_bootmem when after_bootmem == 0. */
761 static void __ref free_bootmem_pda(struct x8664_pda *oldpda)
762 {
763 	if (!after_bootmem)
764 		free_bootmem((unsigned long)oldpda, sizeof(*oldpda));
765 }
766 
767 /*
768  * Allocate node local memory for the AP pda.
769  *
770  * Must be called after the _cpu_pda pointer table is initialized.
771  */
772 int __cpuinit get_local_pda(int cpu)
773 {
774 	struct x8664_pda *oldpda, *newpda;
775 	unsigned long size = sizeof(struct x8664_pda);
776 	int node = cpu_to_node(cpu);
777 
778 	if (cpu_pda(cpu) && !cpu_pda(cpu)->in_bootmem)
779 		return 0;
780 
781 	oldpda = cpu_pda(cpu);
782 	newpda = kmalloc_node(size, GFP_ATOMIC, node);
783 	if (!newpda) {
784 		printk(KERN_ERR "Could not allocate node local PDA "
785 			"for CPU %d on node %d\n", cpu, node);
786 
787 		if (oldpda)
788 			return 0;	/* have a usable pda */
789 		else
790 			return -1;
791 	}
792 
793 	if (oldpda) {
794 		memcpy(newpda, oldpda, size);
795 		free_bootmem_pda(oldpda);
796 	}
797 
798 	newpda->in_bootmem = 0;
799 	cpu_pda(cpu) = newpda;
800 	return 0;
801 }
802 #endif /* CONFIG_X86_64 */
803 
804 static int __cpuinit do_boot_cpu(int apicid, int cpu)
805 /*
806  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
807  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
808  * Returns zero if CPU booted OK, else error code from wakeup_secondary_cpu.
809  */
810 {
811 	unsigned long boot_error = 0;
812 	int timeout;
813 	unsigned long start_ip;
814 	unsigned short nmi_high = 0, nmi_low = 0;
815 	struct create_idle c_idle = {
816 		.cpu = cpu,
817 		.done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
818 	};
819 	INIT_WORK(&c_idle.work, do_fork_idle);
820 
821 #ifdef CONFIG_X86_64
822 	/* Allocate node local memory for AP pdas */
823 	if (cpu > 0) {
824 		boot_error = get_local_pda(cpu);
825 		if (boot_error)
826 			goto restore_state;
827 			/* if can't get pda memory, can't start cpu */
828 	}
829 #endif
830 
831 	alternatives_smp_switch(1);
832 
833 	c_idle.idle = get_idle_for_cpu(cpu);
834 
835 	/*
836 	 * We can't use kernel_thread since we must avoid to
837 	 * reschedule the child.
838 	 */
839 	if (c_idle.idle) {
840 		c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
841 			(THREAD_SIZE +  task_stack_page(c_idle.idle))) - 1);
842 		init_idle(c_idle.idle, cpu);
843 		goto do_rest;
844 	}
845 
846 	if (!keventd_up() || current_is_keventd())
847 		c_idle.work.func(&c_idle.work);
848 	else {
849 		schedule_work(&c_idle.work);
850 		wait_for_completion(&c_idle.done);
851 	}
852 
853 	if (IS_ERR(c_idle.idle)) {
854 		printk("failed fork for CPU %d\n", cpu);
855 		return PTR_ERR(c_idle.idle);
856 	}
857 
858 	set_idle_for_cpu(cpu, c_idle.idle);
859 do_rest:
860 #ifdef CONFIG_X86_32
861 	per_cpu(current_task, cpu) = c_idle.idle;
862 	init_gdt(cpu);
863 	/* Stack for startup_32 can be just as for start_secondary onwards */
864 	irq_ctx_init(cpu);
865 #else
866 	cpu_pda(cpu)->pcurrent = c_idle.idle;
867 	clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
868 #endif
869 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
870 	initial_code = (unsigned long)start_secondary;
871 	stack_start.sp = (void *) c_idle.idle->thread.sp;
872 
873 	/* start_ip had better be page-aligned! */
874 	start_ip = setup_trampoline();
875 
876 	/* So we see what's up   */
877 	printk(KERN_INFO "Booting processor %d/%d ip %lx\n",
878 			  cpu, apicid, start_ip);
879 
880 	/*
881 	 * This grunge runs the startup process for
882 	 * the targeted processor.
883 	 */
884 
885 	atomic_set(&init_deasserted, 0);
886 
887 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
888 
889 		pr_debug("Setting warm reset code and vector.\n");
890 
891 		store_NMI_vector(&nmi_high, &nmi_low);
892 
893 		smpboot_setup_warm_reset_vector(start_ip);
894 		/*
895 		 * Be paranoid about clearing APIC errors.
896 	 	*/
897 		apic_write(APIC_ESR, 0);
898 		apic_read(APIC_ESR);
899 	}
900 
901 	/*
902 	 * Starting actual IPI sequence...
903 	 */
904 	boot_error = wakeup_secondary_cpu(apicid, start_ip);
905 
906 	if (!boot_error) {
907 		/*
908 		 * allow APs to start initializing.
909 		 */
910 		pr_debug("Before Callout %d.\n", cpu);
911 		cpu_set(cpu, cpu_callout_map);
912 		pr_debug("After Callout %d.\n", cpu);
913 
914 		/*
915 		 * Wait 5s total for a response
916 		 */
917 		for (timeout = 0; timeout < 50000; timeout++) {
918 			if (cpu_isset(cpu, cpu_callin_map))
919 				break;	/* It has booted */
920 			udelay(100);
921 		}
922 
923 		if (cpu_isset(cpu, cpu_callin_map)) {
924 			/* number CPUs logically, starting from 1 (BSP is 0) */
925 			pr_debug("OK.\n");
926 			printk(KERN_INFO "CPU%d: ", cpu);
927 			print_cpu_info(&cpu_data(cpu));
928 			pr_debug("CPU has booted.\n");
929 		} else {
930 			boot_error = 1;
931 			if (*((volatile unsigned char *)trampoline_base)
932 					== 0xA5)
933 				/* trampoline started but...? */
934 				printk(KERN_ERR "Stuck ??\n");
935 			else
936 				/* trampoline code not run */
937 				printk(KERN_ERR "Not responding.\n");
938 			if (get_uv_system_type() != UV_NON_UNIQUE_APIC)
939 				inquire_remote_apic(apicid);
940 		}
941 	}
942 #ifdef CONFIG_X86_64
943 restore_state:
944 #endif
945 	if (boot_error) {
946 		/* Try to put things back the way they were before ... */
947 		numa_remove_cpu(cpu); /* was set by numa_add_cpu */
948 		cpu_clear(cpu, cpu_callout_map); /* was set by do_boot_cpu() */
949 		cpu_clear(cpu, cpu_initialized); /* was set by cpu_init() */
950 		cpu_clear(cpu, cpu_present_map);
951 		per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
952 	}
953 
954 	/* mark "stuck" area as not stuck */
955 	*((volatile unsigned long *)trampoline_base) = 0;
956 
957 	/*
958 	 * Cleanup possible dangling ends...
959 	 */
960 	smpboot_restore_warm_reset_vector();
961 
962 	return boot_error;
963 }
964 
965 int __cpuinit native_cpu_up(unsigned int cpu)
966 {
967 	int apicid = cpu_present_to_apicid(cpu);
968 	unsigned long flags;
969 	int err;
970 
971 	WARN_ON(irqs_disabled());
972 
973 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
974 
975 	if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
976 	    !physid_isset(apicid, phys_cpu_present_map)) {
977 		printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
978 		return -EINVAL;
979 	}
980 
981 	/*
982 	 * Already booted CPU?
983 	 */
984 	if (cpu_isset(cpu, cpu_callin_map)) {
985 		pr_debug("do_boot_cpu %d Already started\n", cpu);
986 		return -ENOSYS;
987 	}
988 
989 	/*
990 	 * Save current MTRR state in case it was changed since early boot
991 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
992 	 */
993 	mtrr_save_state();
994 
995 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
996 
997 #ifdef CONFIG_X86_32
998 	/* init low mem mapping */
999 	clone_pgd_range(swapper_pg_dir, swapper_pg_dir + KERNEL_PGD_BOUNDARY,
1000 		min_t(unsigned long, KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
1001 	flush_tlb_all();
1002 	low_mappings = 1;
1003 
1004 	err = do_boot_cpu(apicid, cpu);
1005 
1006 	zap_low_mappings();
1007 	low_mappings = 0;
1008 #else
1009 	err = do_boot_cpu(apicid, cpu);
1010 #endif
1011 	if (err) {
1012 		pr_debug("do_boot_cpu failed %d\n", err);
1013 		return -EIO;
1014 	}
1015 
1016 	/*
1017 	 * Check TSC synchronization with the AP (keep irqs disabled
1018 	 * while doing so):
1019 	 */
1020 	local_irq_save(flags);
1021 	check_tsc_sync_source(cpu);
1022 	local_irq_restore(flags);
1023 
1024 	while (!cpu_online(cpu)) {
1025 		cpu_relax();
1026 		touch_nmi_watchdog();
1027 	}
1028 
1029 	return 0;
1030 }
1031 
1032 /*
1033  * Fall back to non SMP mode after errors.
1034  *
1035  * RED-PEN audit/test this more. I bet there is more state messed up here.
1036  */
1037 static __init void disable_smp(void)
1038 {
1039 	cpu_present_map = cpumask_of_cpu(0);
1040 	cpu_possible_map = cpumask_of_cpu(0);
1041 	smpboot_clear_io_apic_irqs();
1042 
1043 	if (smp_found_config)
1044 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1045 	else
1046 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1047 	map_cpu_to_logical_apicid();
1048 	cpu_set(0, per_cpu(cpu_sibling_map, 0));
1049 	cpu_set(0, per_cpu(cpu_core_map, 0));
1050 }
1051 
1052 /*
1053  * Various sanity checks.
1054  */
1055 static int __init smp_sanity_check(unsigned max_cpus)
1056 {
1057 	preempt_disable();
1058 
1059 #if defined(CONFIG_X86_PC) && defined(CONFIG_X86_32)
1060 	if (def_to_bigsmp && nr_cpu_ids > 8) {
1061 		unsigned int cpu;
1062 		unsigned nr;
1063 
1064 		printk(KERN_WARNING
1065 		       "More than 8 CPUs detected - skipping them.\n"
1066 		       "Use CONFIG_X86_GENERICARCH and CONFIG_X86_BIGSMP.\n");
1067 
1068 		nr = 0;
1069 		for_each_present_cpu(cpu) {
1070 			if (nr >= 8)
1071 				cpu_clear(cpu, cpu_present_map);
1072 			nr++;
1073 		}
1074 
1075 		nr = 0;
1076 		for_each_possible_cpu(cpu) {
1077 			if (nr >= 8)
1078 				cpu_clear(cpu, cpu_possible_map);
1079 			nr++;
1080 		}
1081 
1082 		nr_cpu_ids = 8;
1083 	}
1084 #endif
1085 
1086 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1087 		printk(KERN_WARNING "weird, boot CPU (#%d) not listed"
1088 				    "by the BIOS.\n", hard_smp_processor_id());
1089 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1090 	}
1091 
1092 	/*
1093 	 * If we couldn't find an SMP configuration at boot time,
1094 	 * get out of here now!
1095 	 */
1096 	if (!smp_found_config && !acpi_lapic) {
1097 		preempt_enable();
1098 		printk(KERN_NOTICE "SMP motherboard not detected.\n");
1099 		disable_smp();
1100 		if (APIC_init_uniprocessor())
1101 			printk(KERN_NOTICE "Local APIC not detected."
1102 					   " Using dummy APIC emulation.\n");
1103 		return -1;
1104 	}
1105 
1106 	/*
1107 	 * Should not be necessary because the MP table should list the boot
1108 	 * CPU too, but we do it for the sake of robustness anyway.
1109 	 */
1110 	if (!check_phys_apicid_present(boot_cpu_physical_apicid)) {
1111 		printk(KERN_NOTICE
1112 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
1113 			boot_cpu_physical_apicid);
1114 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1115 	}
1116 	preempt_enable();
1117 
1118 	/*
1119 	 * If we couldn't find a local APIC, then get out of here now!
1120 	 */
1121 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1122 	    !cpu_has_apic) {
1123 		printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
1124 			boot_cpu_physical_apicid);
1125 		printk(KERN_ERR "... forcing use of dummy APIC emulation."
1126 				"(tell your hw vendor)\n");
1127 		smpboot_clear_io_apic();
1128 		return -1;
1129 	}
1130 
1131 	verify_local_APIC();
1132 
1133 	/*
1134 	 * If SMP should be disabled, then really disable it!
1135 	 */
1136 	if (!max_cpus) {
1137 		printk(KERN_INFO "SMP mode deactivated.\n");
1138 		smpboot_clear_io_apic();
1139 
1140 		localise_nmi_watchdog();
1141 
1142 		connect_bsp_APIC();
1143 		setup_local_APIC();
1144 		end_local_APIC_setup();
1145 		return -1;
1146 	}
1147 
1148 	return 0;
1149 }
1150 
1151 static void __init smp_cpu_index_default(void)
1152 {
1153 	int i;
1154 	struct cpuinfo_x86 *c;
1155 
1156 	for_each_possible_cpu(i) {
1157 		c = &cpu_data(i);
1158 		/* mark all to hotplug */
1159 		c->cpu_index = NR_CPUS;
1160 	}
1161 }
1162 
1163 /*
1164  * Prepare for SMP bootup.  The MP table or ACPI has been read
1165  * earlier.  Just do some sanity checking here and enable APIC mode.
1166  */
1167 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1168 {
1169 	preempt_disable();
1170 	smp_cpu_index_default();
1171 	current_cpu_data = boot_cpu_data;
1172 	cpu_callin_map = cpumask_of_cpu(0);
1173 	mb();
1174 	/*
1175 	 * Setup boot CPU information
1176 	 */
1177 	smp_store_cpu_info(0); /* Final full version of the data */
1178 	boot_cpu_logical_apicid = logical_smp_processor_id();
1179 	current_thread_info()->cpu = 0;  /* needed? */
1180 	set_cpu_sibling_map(0);
1181 
1182 	if (smp_sanity_check(max_cpus) < 0) {
1183 		printk(KERN_INFO "SMP disabled\n");
1184 		disable_smp();
1185 		goto out;
1186 	}
1187 
1188 	preempt_disable();
1189 	if (GET_APIC_ID(read_apic_id()) != boot_cpu_physical_apicid) {
1190 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1191 		     GET_APIC_ID(read_apic_id()), boot_cpu_physical_apicid);
1192 		/* Or can we switch back to PIC here? */
1193 	}
1194 	preempt_enable();
1195 
1196 	connect_bsp_APIC();
1197 
1198 	/*
1199 	 * Switch from PIC to APIC mode.
1200 	 */
1201 	setup_local_APIC();
1202 
1203 #ifdef CONFIG_X86_64
1204 	/*
1205 	 * Enable IO APIC before setting up error vector
1206 	 */
1207 	if (!skip_ioapic_setup && nr_ioapics)
1208 		enable_IO_APIC();
1209 #endif
1210 	end_local_APIC_setup();
1211 
1212 	map_cpu_to_logical_apicid();
1213 
1214 	setup_portio_remap();
1215 
1216 	smpboot_setup_io_apic();
1217 	/*
1218 	 * Set up local APIC timer on boot CPU.
1219 	 */
1220 
1221 	printk(KERN_INFO "CPU%d: ", 0);
1222 	print_cpu_info(&cpu_data(0));
1223 	setup_boot_clock();
1224 
1225 	if (is_uv_system())
1226 		uv_system_init();
1227 out:
1228 	preempt_enable();
1229 }
1230 /*
1231  * Early setup to make printk work.
1232  */
1233 void __init native_smp_prepare_boot_cpu(void)
1234 {
1235 	int me = smp_processor_id();
1236 #ifdef CONFIG_X86_32
1237 	init_gdt(me);
1238 #endif
1239 	switch_to_new_gdt();
1240 	/* already set me in cpu_online_map in boot_cpu_init() */
1241 	cpu_set(me, cpu_callout_map);
1242 	per_cpu(cpu_state, me) = CPU_ONLINE;
1243 }
1244 
1245 void __init native_smp_cpus_done(unsigned int max_cpus)
1246 {
1247 	pr_debug("Boot done.\n");
1248 
1249 	impress_friends();
1250 	smp_checks();
1251 #ifdef CONFIG_X86_IO_APIC
1252 	setup_ioapic_dest();
1253 #endif
1254 	check_nmi_watchdog();
1255 }
1256 
1257 #ifdef CONFIG_HOTPLUG_CPU
1258 
1259 static void remove_siblinginfo(int cpu)
1260 {
1261 	int sibling;
1262 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1263 
1264 	for_each_cpu_mask_nr(sibling, per_cpu(cpu_core_map, cpu)) {
1265 		cpu_clear(cpu, per_cpu(cpu_core_map, sibling));
1266 		/*/
1267 		 * last thread sibling in this cpu core going down
1268 		 */
1269 		if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1)
1270 			cpu_data(sibling).booted_cores--;
1271 	}
1272 
1273 	for_each_cpu_mask_nr(sibling, per_cpu(cpu_sibling_map, cpu))
1274 		cpu_clear(cpu, per_cpu(cpu_sibling_map, sibling));
1275 	cpus_clear(per_cpu(cpu_sibling_map, cpu));
1276 	cpus_clear(per_cpu(cpu_core_map, cpu));
1277 	c->phys_proc_id = 0;
1278 	c->cpu_core_id = 0;
1279 	cpu_clear(cpu, cpu_sibling_setup_map);
1280 }
1281 
1282 static int additional_cpus __initdata = -1;
1283 
1284 static __init int setup_additional_cpus(char *s)
1285 {
1286 	return s && get_option(&s, &additional_cpus) ? 0 : -EINVAL;
1287 }
1288 early_param("additional_cpus", setup_additional_cpus);
1289 
1290 /*
1291  * cpu_possible_map should be static, it cannot change as cpu's
1292  * are onlined, or offlined. The reason is per-cpu data-structures
1293  * are allocated by some modules at init time, and dont expect to
1294  * do this dynamically on cpu arrival/departure.
1295  * cpu_present_map on the other hand can change dynamically.
1296  * In case when cpu_hotplug is not compiled, then we resort to current
1297  * behaviour, which is cpu_possible == cpu_present.
1298  * - Ashok Raj
1299  *
1300  * Three ways to find out the number of additional hotplug CPUs:
1301  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1302  * - The user can overwrite it with additional_cpus=NUM
1303  * - Otherwise don't reserve additional CPUs.
1304  * We do this because additional CPUs waste a lot of memory.
1305  * -AK
1306  */
1307 __init void prefill_possible_map(void)
1308 {
1309 	int i;
1310 	int possible;
1311 
1312 	/* no processor from mptable or madt */
1313 	if (!num_processors)
1314 		num_processors = 1;
1315 
1316 #ifdef CONFIG_HOTPLUG_CPU
1317 	if (additional_cpus == -1) {
1318 		if (disabled_cpus > 0)
1319 			additional_cpus = disabled_cpus;
1320 		else
1321 			additional_cpus = 0;
1322 	}
1323 #else
1324 	additional_cpus = 0;
1325 #endif
1326 	possible = num_processors + additional_cpus;
1327 	if (possible > NR_CPUS)
1328 		possible = NR_CPUS;
1329 
1330 	printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1331 		possible, max_t(int, possible - num_processors, 0));
1332 
1333 	for (i = 0; i < possible; i++)
1334 		cpu_set(i, cpu_possible_map);
1335 
1336 	nr_cpu_ids = possible;
1337 }
1338 
1339 static void __ref remove_cpu_from_maps(int cpu)
1340 {
1341 	cpu_clear(cpu, cpu_online_map);
1342 	cpu_clear(cpu, cpu_callout_map);
1343 	cpu_clear(cpu, cpu_callin_map);
1344 	/* was set by cpu_init() */
1345 	cpu_clear(cpu, cpu_initialized);
1346 	numa_remove_cpu(cpu);
1347 }
1348 
1349 int __cpu_disable(void)
1350 {
1351 	int cpu = smp_processor_id();
1352 
1353 	/*
1354 	 * Perhaps use cpufreq to drop frequency, but that could go
1355 	 * into generic code.
1356 	 *
1357 	 * We won't take down the boot processor on i386 due to some
1358 	 * interrupts only being able to be serviced by the BSP.
1359 	 * Especially so if we're not using an IOAPIC	-zwane
1360 	 */
1361 	if (cpu == 0)
1362 		return -EBUSY;
1363 
1364 	if (nmi_watchdog == NMI_LOCAL_APIC)
1365 		stop_apic_nmi_watchdog(NULL);
1366 	clear_local_APIC();
1367 
1368 	/*
1369 	 * HACK:
1370 	 * Allow any queued timer interrupts to get serviced
1371 	 * This is only a temporary solution until we cleanup
1372 	 * fixup_irqs as we do for IA64.
1373 	 */
1374 	local_irq_enable();
1375 	mdelay(1);
1376 
1377 	local_irq_disable();
1378 	remove_siblinginfo(cpu);
1379 
1380 	/* It's now safe to remove this processor from the online map */
1381 	lock_vector_lock();
1382 	remove_cpu_from_maps(cpu);
1383 	unlock_vector_lock();
1384 	fixup_irqs(cpu_online_map);
1385 	return 0;
1386 }
1387 
1388 void __cpu_die(unsigned int cpu)
1389 {
1390 	/* We don't do anything here: idle task is faking death itself. */
1391 	unsigned int i;
1392 
1393 	for (i = 0; i < 10; i++) {
1394 		/* They ack this in play_dead by setting CPU_DEAD */
1395 		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1396 			printk(KERN_INFO "CPU %d is now offline\n", cpu);
1397 			if (1 == num_online_cpus())
1398 				alternatives_smp_switch(0);
1399 			return;
1400 		}
1401 		msleep(100);
1402 	}
1403 	printk(KERN_ERR "CPU %u didn't die...\n", cpu);
1404 }
1405 #else /* ... !CONFIG_HOTPLUG_CPU */
1406 int __cpu_disable(void)
1407 {
1408 	return -ENOSYS;
1409 }
1410 
1411 void __cpu_die(unsigned int cpu)
1412 {
1413 	/* We said "no" in __cpu_disable */
1414 	BUG();
1415 }
1416 #endif
1417