xref: /linux/arch/x86/kernel/smpboot.c (revision 93d546399c2b7d66a54d5fbd5eee17de19246bf6)
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/idle.h>
56 #include <asm/smp.h>
57 #include <asm/trampoline.h>
58 #include <asm/cpu.h>
59 #include <asm/numa.h>
60 #include <asm/pgtable.h>
61 #include <asm/tlbflush.h>
62 #include <asm/mtrr.h>
63 #include <asm/vmi.h>
64 #include <asm/genapic.h>
65 #include <linux/mc146818rtc.h>
66 
67 #include <mach_apic.h>
68 #include <mach_wakecpu.h>
69 #include <smpboot_hooks.h>
70 
71 #ifdef CONFIG_X86_32
72 u8 apicid_2_node[MAX_APICID];
73 static int low_mappings;
74 #endif
75 
76 /* State of each CPU */
77 DEFINE_PER_CPU(int, cpu_state) = { 0 };
78 
79 /* Store all idle threads, this can be reused instead of creating
80 * a new thread. Also avoids complicated thread destroy functionality
81 * for idle threads.
82 */
83 #ifdef CONFIG_HOTPLUG_CPU
84 /*
85  * Needed only for CONFIG_HOTPLUG_CPU because __cpuinitdata is
86  * removed after init for !CONFIG_HOTPLUG_CPU.
87  */
88 static DEFINE_PER_CPU(struct task_struct *, idle_thread_array);
89 #define get_idle_for_cpu(x)      (per_cpu(idle_thread_array, x))
90 #define set_idle_for_cpu(x, p)   (per_cpu(idle_thread_array, x) = (p))
91 #else
92 static struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
93 #define get_idle_for_cpu(x)      (idle_thread_array[(x)])
94 #define set_idle_for_cpu(x, p)   (idle_thread_array[(x)] = (p))
95 #endif
96 
97 /* Number of siblings per CPU package */
98 int smp_num_siblings = 1;
99 EXPORT_SYMBOL(smp_num_siblings);
100 
101 /* Last level cache ID of each logical CPU */
102 DEFINE_PER_CPU(u16, cpu_llc_id) = BAD_APICID;
103 
104 /* bitmap of online cpus */
105 cpumask_t cpu_online_map __read_mostly;
106 EXPORT_SYMBOL(cpu_online_map);
107 
108 cpumask_t cpu_callin_map;
109 cpumask_t cpu_callout_map;
110 cpumask_t cpu_possible_map;
111 EXPORT_SYMBOL(cpu_possible_map);
112 
113 /* representing HT siblings of each logical CPU */
114 DEFINE_PER_CPU(cpumask_t, cpu_sibling_map);
115 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
116 
117 /* representing HT and core siblings of each logical CPU */
118 DEFINE_PER_CPU(cpumask_t, cpu_core_map);
119 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
120 
121 /* Per CPU bogomips and other parameters */
122 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
123 EXPORT_PER_CPU_SYMBOL(cpu_info);
124 
125 static atomic_t init_deasserted;
126 
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 static 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 static int boot_cpu_logical_apicid;
169 
170 u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly =
171 					{ [0 ... NR_CPUS-1] = BAD_APICID };
172 
173 static void map_cpu_to_logical_apicid(void)
174 {
175 	int cpu = smp_processor_id();
176 	int apicid = logical_smp_processor_id();
177 	int node = apicid_to_node(apicid);
178 
179 	if (!node_online(node))
180 		node = first_online_node;
181 
182 	cpu_2_logical_apicid[cpu] = apicid;
183 	map_cpu_to_node(cpu, node);
184 }
185 
186 void numa_remove_cpu(int cpu)
187 {
188 	cpu_2_logical_apicid[cpu] = BAD_APICID;
189 	unmap_cpu_to_node(cpu);
190 }
191 #else
192 #define map_cpu_to_logical_apicid()  do {} while (0)
193 #endif
194 
195 /*
196  * Report back to the Boot Processor.
197  * Running on AP.
198  */
199 static void __cpuinit smp_callin(void)
200 {
201 	int cpuid, phys_id;
202 	unsigned long timeout;
203 
204 	/*
205 	 * If waken up by an INIT in an 82489DX configuration
206 	 * we may get here before an INIT-deassert IPI reaches
207 	 * our local APIC.  We have to wait for the IPI or we'll
208 	 * lock up on an APIC access.
209 	 */
210 	wait_for_init_deassert(&init_deasserted);
211 
212 	/*
213 	 * (This works even if the APIC is not enabled.)
214 	 */
215 	phys_id = read_apic_id();
216 	cpuid = smp_processor_id();
217 	if (cpu_isset(cpuid, cpu_callin_map)) {
218 		panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
219 					phys_id, cpuid);
220 	}
221 	pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
222 
223 	/*
224 	 * STARTUP IPIs are fragile beasts as they might sometimes
225 	 * trigger some glue motherboard logic. Complete APIC bus
226 	 * silence for 1 second, this overestimates the time the
227 	 * boot CPU is spending to send the up to 2 STARTUP IPIs
228 	 * by a factor of two. This should be enough.
229 	 */
230 
231 	/*
232 	 * Waiting 2s total for startup (udelay is not yet working)
233 	 */
234 	timeout = jiffies + 2*HZ;
235 	while (time_before(jiffies, timeout)) {
236 		/*
237 		 * Has the boot CPU finished it's STARTUP sequence?
238 		 */
239 		if (cpu_isset(cpuid, cpu_callout_map))
240 			break;
241 		cpu_relax();
242 	}
243 
244 	if (!time_before(jiffies, timeout)) {
245 		panic("%s: CPU%d started up but did not get a callout!\n",
246 		      __func__, cpuid);
247 	}
248 
249 	/*
250 	 * the boot CPU has finished the init stage and is spinning
251 	 * on callin_map until we finish. We are free to set up this
252 	 * CPU, first the APIC. (this is probably redundant on most
253 	 * boards)
254 	 */
255 
256 	pr_debug("CALLIN, before setup_local_APIC().\n");
257 	smp_callin_clear_local_apic();
258 	setup_local_APIC();
259 	end_local_APIC_setup();
260 	map_cpu_to_logical_apicid();
261 
262 	notify_cpu_starting(cpuid);
263 	/*
264 	 * Get our bogomips.
265 	 *
266 	 * Need to enable IRQs because it can take longer and then
267 	 * the NMI watchdog might kill us.
268 	 */
269 	local_irq_enable();
270 	calibrate_delay();
271 	local_irq_disable();
272 	pr_debug("Stack at about %p\n", &cpuid);
273 
274 	/*
275 	 * Save our processor parameters
276 	 */
277 	smp_store_cpu_info(cpuid);
278 
279 	/*
280 	 * Allow the master to continue.
281 	 */
282 	cpu_set(cpuid, cpu_callin_map);
283 }
284 
285 static int __cpuinitdata unsafe_smp;
286 
287 /*
288  * Activate a secondary processor.
289  */
290 static void __cpuinit start_secondary(void *unused)
291 {
292 	/*
293 	 * Don't put *anything* before cpu_init(), SMP booting is too
294 	 * fragile that we want to limit the things done here to the
295 	 * most necessary things.
296 	 */
297 	vmi_bringup();
298 	cpu_init();
299 	preempt_disable();
300 	smp_callin();
301 
302 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
303 	barrier();
304 	/*
305 	 * Check TSC synchronization with the BP:
306 	 */
307 	check_tsc_sync_target();
308 
309 	if (nmi_watchdog == NMI_IO_APIC) {
310 		disable_8259A_irq(0);
311 		enable_NMI_through_LVT0();
312 		enable_8259A_irq(0);
313 	}
314 
315 #ifdef CONFIG_X86_32
316 	while (low_mappings)
317 		cpu_relax();
318 	__flush_tlb_all();
319 #endif
320 
321 	/* This must be done before setting cpu_online_map */
322 	set_cpu_sibling_map(raw_smp_processor_id());
323 	wmb();
324 
325 	/*
326 	 * We need to hold call_lock, so there is no inconsistency
327 	 * between the time smp_call_function() determines number of
328 	 * IPI recipients, and the time when the determination is made
329 	 * for which cpus receive the IPI. Holding this
330 	 * lock helps us to not include this cpu in a currently in progress
331 	 * smp_call_function().
332 	 *
333 	 * We need to hold vector_lock so there the set of online cpus
334 	 * does not change while we are assigning vectors to cpus.  Holding
335 	 * this lock ensures we don't half assign or remove an irq from a cpu.
336 	 */
337 	ipi_call_lock();
338 	lock_vector_lock();
339 	__setup_vector_irq(smp_processor_id());
340 	cpu_set(smp_processor_id(), cpu_online_map);
341 	unlock_vector_lock();
342 	ipi_call_unlock();
343 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
344 
345 	/* enable local interrupts */
346 	local_irq_enable();
347 
348 	setup_secondary_clock();
349 
350 	wmb();
351 	cpu_idle();
352 }
353 
354 static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c)
355 {
356 	/*
357 	 * Mask B, Pentium, but not Pentium MMX
358 	 */
359 	if (c->x86_vendor == X86_VENDOR_INTEL &&
360 	    c->x86 == 5 &&
361 	    c->x86_mask >= 1 && c->x86_mask <= 4 &&
362 	    c->x86_model <= 3)
363 		/*
364 		 * Remember we have B step Pentia with bugs
365 		 */
366 		smp_b_stepping = 1;
367 
368 	/*
369 	 * Certain Athlons might work (for various values of 'work') in SMP
370 	 * but they are not certified as MP capable.
371 	 */
372 	if ((c->x86_vendor == X86_VENDOR_AMD) && (c->x86 == 6)) {
373 
374 		if (num_possible_cpus() == 1)
375 			goto valid_k7;
376 
377 		/* Athlon 660/661 is valid. */
378 		if ((c->x86_model == 6) && ((c->x86_mask == 0) ||
379 		    (c->x86_mask == 1)))
380 			goto valid_k7;
381 
382 		/* Duron 670 is valid */
383 		if ((c->x86_model == 7) && (c->x86_mask == 0))
384 			goto valid_k7;
385 
386 		/*
387 		 * Athlon 662, Duron 671, and Athlon >model 7 have capability
388 		 * bit. It's worth noting that the A5 stepping (662) of some
389 		 * Athlon XP's have the MP bit set.
390 		 * See http://www.heise.de/newsticker/data/jow-18.10.01-000 for
391 		 * more.
392 		 */
393 		if (((c->x86_model == 6) && (c->x86_mask >= 2)) ||
394 		    ((c->x86_model == 7) && (c->x86_mask >= 1)) ||
395 		     (c->x86_model > 7))
396 			if (cpu_has_mp)
397 				goto valid_k7;
398 
399 		/* If we get here, not a certified SMP capable AMD system. */
400 		unsafe_smp = 1;
401 	}
402 
403 valid_k7:
404 	;
405 }
406 
407 static void __cpuinit smp_checks(void)
408 {
409 	if (smp_b_stepping)
410 		printk(KERN_WARNING "WARNING: SMP operation may be unreliable"
411 				    "with B stepping processors.\n");
412 
413 	/*
414 	 * Don't taint if we are running SMP kernel on a single non-MP
415 	 * approved Athlon
416 	 */
417 	if (unsafe_smp && num_online_cpus() > 1) {
418 		printk(KERN_INFO "WARNING: This combination of AMD"
419 			"processors is not suitable for SMP.\n");
420 		add_taint(TAINT_UNSAFE_SMP);
421 	}
422 }
423 
424 /*
425  * The bootstrap kernel entry code has set these up. Save them for
426  * a given CPU
427  */
428 
429 void __cpuinit smp_store_cpu_info(int id)
430 {
431 	struct cpuinfo_x86 *c = &cpu_data(id);
432 
433 	*c = boot_cpu_data;
434 	c->cpu_index = id;
435 	if (id != 0)
436 		identify_secondary_cpu(c);
437 	smp_apply_quirks(c);
438 }
439 
440 
441 void __cpuinit set_cpu_sibling_map(int cpu)
442 {
443 	int i;
444 	struct cpuinfo_x86 *c = &cpu_data(cpu);
445 
446 	cpu_set(cpu, cpu_sibling_setup_map);
447 
448 	if (smp_num_siblings > 1) {
449 		for_each_cpu_mask_nr(i, cpu_sibling_setup_map) {
450 			if (c->phys_proc_id == cpu_data(i).phys_proc_id &&
451 			    c->cpu_core_id == cpu_data(i).cpu_core_id) {
452 				cpu_set(i, per_cpu(cpu_sibling_map, cpu));
453 				cpu_set(cpu, per_cpu(cpu_sibling_map, i));
454 				cpu_set(i, per_cpu(cpu_core_map, cpu));
455 				cpu_set(cpu, per_cpu(cpu_core_map, i));
456 				cpu_set(i, c->llc_shared_map);
457 				cpu_set(cpu, cpu_data(i).llc_shared_map);
458 			}
459 		}
460 	} else {
461 		cpu_set(cpu, per_cpu(cpu_sibling_map, cpu));
462 	}
463 
464 	cpu_set(cpu, c->llc_shared_map);
465 
466 	if (current_cpu_data.x86_max_cores == 1) {
467 		per_cpu(cpu_core_map, cpu) = per_cpu(cpu_sibling_map, cpu);
468 		c->booted_cores = 1;
469 		return;
470 	}
471 
472 	for_each_cpu_mask_nr(i, cpu_sibling_setup_map) {
473 		if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
474 		    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
475 			cpu_set(i, c->llc_shared_map);
476 			cpu_set(cpu, cpu_data(i).llc_shared_map);
477 		}
478 		if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
479 			cpu_set(i, per_cpu(cpu_core_map, cpu));
480 			cpu_set(cpu, per_cpu(cpu_core_map, i));
481 			/*
482 			 *  Does this new cpu bringup a new core?
483 			 */
484 			if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1) {
485 				/*
486 				 * for each core in package, increment
487 				 * the booted_cores for this new cpu
488 				 */
489 				if (first_cpu(per_cpu(cpu_sibling_map, i)) == i)
490 					c->booted_cores++;
491 				/*
492 				 * increment the core count for all
493 				 * the other cpus in this package
494 				 */
495 				if (i != cpu)
496 					cpu_data(i).booted_cores++;
497 			} else if (i != cpu && !c->booted_cores)
498 				c->booted_cores = cpu_data(i).booted_cores;
499 		}
500 	}
501 }
502 
503 /* maps the cpu to the sched domain representing multi-core */
504 cpumask_t cpu_coregroup_map(int cpu)
505 {
506 	struct cpuinfo_x86 *c = &cpu_data(cpu);
507 	/*
508 	 * For perf, we return last level cache shared map.
509 	 * And for power savings, we return cpu_core_map
510 	 */
511 	if (sched_mc_power_savings || sched_smt_power_savings)
512 		return per_cpu(cpu_core_map, cpu);
513 	else
514 		return c->llc_shared_map;
515 }
516 
517 static void impress_friends(void)
518 {
519 	int cpu;
520 	unsigned long bogosum = 0;
521 	/*
522 	 * Allow the user to impress friends.
523 	 */
524 	pr_debug("Before bogomips.\n");
525 	for_each_possible_cpu(cpu)
526 		if (cpu_isset(cpu, cpu_callout_map))
527 			bogosum += cpu_data(cpu).loops_per_jiffy;
528 	printk(KERN_INFO
529 		"Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
530 		num_online_cpus(),
531 		bogosum/(500000/HZ),
532 		(bogosum/(5000/HZ))%100);
533 
534 	pr_debug("Before bogocount - setting activated=1.\n");
535 }
536 
537 static inline void __inquire_remote_apic(int apicid)
538 {
539 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
540 	char *names[] = { "ID", "VERSION", "SPIV" };
541 	int timeout;
542 	u32 status;
543 
544 	printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
545 
546 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
547 		printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
548 
549 		/*
550 		 * Wait for idle.
551 		 */
552 		status = safe_apic_wait_icr_idle();
553 		if (status)
554 			printk(KERN_CONT
555 			       "a previous APIC delivery may have failed\n");
556 
557 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
558 
559 		timeout = 0;
560 		do {
561 			udelay(100);
562 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
563 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
564 
565 		switch (status) {
566 		case APIC_ICR_RR_VALID:
567 			status = apic_read(APIC_RRR);
568 			printk(KERN_CONT "%08x\n", status);
569 			break;
570 		default:
571 			printk(KERN_CONT "failed\n");
572 		}
573 	}
574 }
575 
576 #ifdef WAKE_SECONDARY_VIA_NMI
577 /*
578  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
579  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
580  * won't ... remember to clear down the APIC, etc later.
581  */
582 static int __devinit
583 wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
584 {
585 	unsigned long send_status, accept_status = 0;
586 	int maxlvt;
587 
588 	/* Target chip */
589 	/* Boot on the stack */
590 	/* Kick the second */
591 	apic_icr_write(APIC_DM_NMI | APIC_DEST_LOGICAL, logical_apicid);
592 
593 	pr_debug("Waiting for send to finish...\n");
594 	send_status = safe_apic_wait_icr_idle();
595 
596 	/*
597 	 * Give the other CPU some time to accept the IPI.
598 	 */
599 	udelay(200);
600 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
601 		maxlvt = lapic_get_maxlvt();
602 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
603 			apic_write(APIC_ESR, 0);
604 		accept_status = (apic_read(APIC_ESR) & 0xEF);
605 	}
606 	pr_debug("NMI sent.\n");
607 
608 	if (send_status)
609 		printk(KERN_ERR "APIC never delivered???\n");
610 	if (accept_status)
611 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
612 
613 	return (send_status | accept_status);
614 }
615 #endif	/* WAKE_SECONDARY_VIA_NMI */
616 
617 #ifdef WAKE_SECONDARY_VIA_INIT
618 static int __devinit
619 wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
620 {
621 	unsigned long send_status, accept_status = 0;
622 	int maxlvt, num_starts, j;
623 
624 	if (get_uv_system_type() == UV_NON_UNIQUE_APIC) {
625 		send_status = uv_wakeup_secondary(phys_apicid, start_eip);
626 		atomic_set(&init_deasserted, 1);
627 		return send_status;
628 	}
629 
630 	maxlvt = lapic_get_maxlvt();
631 
632 	/*
633 	 * Be paranoid about clearing APIC errors.
634 	 */
635 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
636 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
637 			apic_write(APIC_ESR, 0);
638 		apic_read(APIC_ESR);
639 	}
640 
641 	pr_debug("Asserting INIT.\n");
642 
643 	/*
644 	 * Turn INIT on target chip
645 	 */
646 	/*
647 	 * Send IPI
648 	 */
649 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
650 		       phys_apicid);
651 
652 	pr_debug("Waiting for send to finish...\n");
653 	send_status = safe_apic_wait_icr_idle();
654 
655 	mdelay(10);
656 
657 	pr_debug("Deasserting INIT.\n");
658 
659 	/* Target chip */
660 	/* Send IPI */
661 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
662 
663 	pr_debug("Waiting for send to finish...\n");
664 	send_status = safe_apic_wait_icr_idle();
665 
666 	mb();
667 	atomic_set(&init_deasserted, 1);
668 
669 	/*
670 	 * Should we send STARTUP IPIs ?
671 	 *
672 	 * Determine this based on the APIC version.
673 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
674 	 */
675 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
676 		num_starts = 2;
677 	else
678 		num_starts = 0;
679 
680 	/*
681 	 * Paravirt / VMI wants a startup IPI hook here to set up the
682 	 * target processor state.
683 	 */
684 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
685 			 (unsigned long)stack_start.sp);
686 
687 	/*
688 	 * Run STARTUP IPI loop.
689 	 */
690 	pr_debug("#startup loops: %d.\n", num_starts);
691 
692 	for (j = 1; j <= num_starts; j++) {
693 		pr_debug("Sending STARTUP #%d.\n", j);
694 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
695 			apic_write(APIC_ESR, 0);
696 		apic_read(APIC_ESR);
697 		pr_debug("After apic_write.\n");
698 
699 		/*
700 		 * STARTUP IPI
701 		 */
702 
703 		/* Target chip */
704 		/* Boot on the stack */
705 		/* Kick the second */
706 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
707 			       phys_apicid);
708 
709 		/*
710 		 * Give the other CPU some time to accept the IPI.
711 		 */
712 		udelay(300);
713 
714 		pr_debug("Startup point 1.\n");
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 (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
724 			apic_write(APIC_ESR, 0);
725 		accept_status = (apic_read(APIC_ESR) & 0xEF);
726 		if (send_status || accept_status)
727 			break;
728 	}
729 	pr_debug("After Startup.\n");
730 
731 	if (send_status)
732 		printk(KERN_ERR "APIC never delivered???\n");
733 	if (accept_status)
734 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
735 
736 	return (send_status | accept_status);
737 }
738 #endif	/* WAKE_SECONDARY_VIA_INIT */
739 
740 struct create_idle {
741 	struct work_struct work;
742 	struct task_struct *idle;
743 	struct completion done;
744 	int cpu;
745 };
746 
747 static void __cpuinit do_fork_idle(struct work_struct *work)
748 {
749 	struct create_idle *c_idle =
750 		container_of(work, struct create_idle, work);
751 
752 	c_idle->idle = fork_idle(c_idle->cpu);
753 	complete(&c_idle->done);
754 }
755 
756 #ifdef CONFIG_X86_64
757 
758 /* __ref because it's safe to call free_bootmem when after_bootmem == 0. */
759 static void __ref free_bootmem_pda(struct x8664_pda *oldpda)
760 {
761 	if (!after_bootmem)
762 		free_bootmem((unsigned long)oldpda, sizeof(*oldpda));
763 }
764 
765 /*
766  * Allocate node local memory for the AP pda.
767  *
768  * Must be called after the _cpu_pda pointer table is initialized.
769  */
770 int __cpuinit get_local_pda(int cpu)
771 {
772 	struct x8664_pda *oldpda, *newpda;
773 	unsigned long size = sizeof(struct x8664_pda);
774 	int node = cpu_to_node(cpu);
775 
776 	if (cpu_pda(cpu) && !cpu_pda(cpu)->in_bootmem)
777 		return 0;
778 
779 	oldpda = cpu_pda(cpu);
780 	newpda = kmalloc_node(size, GFP_ATOMIC, node);
781 	if (!newpda) {
782 		printk(KERN_ERR "Could not allocate node local PDA "
783 			"for CPU %d on node %d\n", cpu, node);
784 
785 		if (oldpda)
786 			return 0;	/* have a usable pda */
787 		else
788 			return -1;
789 	}
790 
791 	if (oldpda) {
792 		memcpy(newpda, oldpda, size);
793 		free_bootmem_pda(oldpda);
794 	}
795 
796 	newpda->in_bootmem = 0;
797 	cpu_pda(cpu) = newpda;
798 	return 0;
799 }
800 #endif /* CONFIG_X86_64 */
801 
802 static int __cpuinit do_boot_cpu(int apicid, int cpu)
803 /*
804  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
805  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
806  * Returns zero if CPU booted OK, else error code from wakeup_secondary_cpu.
807  */
808 {
809 	unsigned long boot_error = 0;
810 	int timeout;
811 	unsigned long start_ip;
812 	unsigned short nmi_high = 0, nmi_low = 0;
813 	struct create_idle c_idle = {
814 		.cpu = cpu,
815 		.done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
816 	};
817 	INIT_WORK(&c_idle.work, do_fork_idle);
818 
819 #ifdef CONFIG_X86_64
820 	/* Allocate node local memory for AP pdas */
821 	if (cpu > 0) {
822 		boot_error = get_local_pda(cpu);
823 		if (boot_error)
824 			goto restore_state;
825 			/* if can't get pda memory, can't start cpu */
826 	}
827 #endif
828 
829 	alternatives_smp_switch(1);
830 
831 	c_idle.idle = get_idle_for_cpu(cpu);
832 
833 	/*
834 	 * We can't use kernel_thread since we must avoid to
835 	 * reschedule the child.
836 	 */
837 	if (c_idle.idle) {
838 		c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
839 			(THREAD_SIZE +  task_stack_page(c_idle.idle))) - 1);
840 		init_idle(c_idle.idle, cpu);
841 		goto do_rest;
842 	}
843 
844 	if (!keventd_up() || current_is_keventd())
845 		c_idle.work.func(&c_idle.work);
846 	else {
847 		schedule_work(&c_idle.work);
848 		wait_for_completion(&c_idle.done);
849 	}
850 
851 	if (IS_ERR(c_idle.idle)) {
852 		printk("failed fork for CPU %d\n", cpu);
853 		return PTR_ERR(c_idle.idle);
854 	}
855 
856 	set_idle_for_cpu(cpu, c_idle.idle);
857 do_rest:
858 #ifdef CONFIG_X86_32
859 	per_cpu(current_task, cpu) = c_idle.idle;
860 	init_gdt(cpu);
861 	/* Stack for startup_32 can be just as for start_secondary onwards */
862 	irq_ctx_init(cpu);
863 #else
864 	cpu_pda(cpu)->pcurrent = c_idle.idle;
865 	clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
866 #endif
867 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
868 	initial_code = (unsigned long)start_secondary;
869 	stack_start.sp = (void *) c_idle.idle->thread.sp;
870 
871 	/* start_ip had better be page-aligned! */
872 	start_ip = setup_trampoline();
873 
874 	/* So we see what's up   */
875 	printk(KERN_INFO "Booting processor %d APIC 0x%x ip 0x%lx\n",
876 			  cpu, apicid, start_ip);
877 
878 	/*
879 	 * This grunge runs the startup process for
880 	 * the targeted processor.
881 	 */
882 
883 	atomic_set(&init_deasserted, 0);
884 
885 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
886 
887 		pr_debug("Setting warm reset code and vector.\n");
888 
889 		store_NMI_vector(&nmi_high, &nmi_low);
890 
891 		smpboot_setup_warm_reset_vector(start_ip);
892 		/*
893 		 * Be paranoid about clearing APIC errors.
894 		*/
895 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
896 			apic_write(APIC_ESR, 0);
897 			apic_read(APIC_ESR);
898 		}
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 #ifdef CONFIG_X86_32
1179 	boot_cpu_logical_apicid = logical_smp_processor_id();
1180 #endif
1181 	current_thread_info()->cpu = 0;  /* needed? */
1182 	set_cpu_sibling_map(0);
1183 
1184 #ifdef CONFIG_X86_64
1185 	enable_IR_x2apic();
1186 	setup_apic_routing();
1187 #endif
1188 
1189 	if (smp_sanity_check(max_cpus) < 0) {
1190 		printk(KERN_INFO "SMP disabled\n");
1191 		disable_smp();
1192 		goto out;
1193 	}
1194 
1195 	preempt_disable();
1196 	if (read_apic_id() != boot_cpu_physical_apicid) {
1197 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1198 		     read_apic_id(), boot_cpu_physical_apicid);
1199 		/* Or can we switch back to PIC here? */
1200 	}
1201 	preempt_enable();
1202 
1203 	connect_bsp_APIC();
1204 
1205 	/*
1206 	 * Switch from PIC to APIC mode.
1207 	 */
1208 	setup_local_APIC();
1209 
1210 #ifdef CONFIG_X86_64
1211 	/*
1212 	 * Enable IO APIC before setting up error vector
1213 	 */
1214 	if (!skip_ioapic_setup && nr_ioapics)
1215 		enable_IO_APIC();
1216 #endif
1217 	end_local_APIC_setup();
1218 
1219 	map_cpu_to_logical_apicid();
1220 
1221 	setup_portio_remap();
1222 
1223 	smpboot_setup_io_apic();
1224 	/*
1225 	 * Set up local APIC timer on boot CPU.
1226 	 */
1227 
1228 	printk(KERN_INFO "CPU%d: ", 0);
1229 	print_cpu_info(&cpu_data(0));
1230 	setup_boot_clock();
1231 
1232 	if (is_uv_system())
1233 		uv_system_init();
1234 out:
1235 	preempt_enable();
1236 }
1237 /*
1238  * Early setup to make printk work.
1239  */
1240 void __init native_smp_prepare_boot_cpu(void)
1241 {
1242 	int me = smp_processor_id();
1243 #ifdef CONFIG_X86_32
1244 	init_gdt(me);
1245 #endif
1246 	switch_to_new_gdt();
1247 	/* already set me in cpu_online_map in boot_cpu_init() */
1248 	cpu_set(me, cpu_callout_map);
1249 	per_cpu(cpu_state, me) = CPU_ONLINE;
1250 }
1251 
1252 void __init native_smp_cpus_done(unsigned int max_cpus)
1253 {
1254 	pr_debug("Boot done.\n");
1255 
1256 	impress_friends();
1257 	smp_checks();
1258 #ifdef CONFIG_X86_IO_APIC
1259 	setup_ioapic_dest();
1260 #endif
1261 	check_nmi_watchdog();
1262 }
1263 
1264 /*
1265  * cpu_possible_map should be static, it cannot change as cpu's
1266  * are onlined, or offlined. The reason is per-cpu data-structures
1267  * are allocated by some modules at init time, and dont expect to
1268  * do this dynamically on cpu arrival/departure.
1269  * cpu_present_map on the other hand can change dynamically.
1270  * In case when cpu_hotplug is not compiled, then we resort to current
1271  * behaviour, which is cpu_possible == cpu_present.
1272  * - Ashok Raj
1273  *
1274  * Three ways to find out the number of additional hotplug CPUs:
1275  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1276  * - The user can overwrite it with additional_cpus=NUM
1277  * - Otherwise don't reserve additional CPUs.
1278  * We do this because additional CPUs waste a lot of memory.
1279  * -AK
1280  */
1281 __init void prefill_possible_map(void)
1282 {
1283 	int i, possible;
1284 
1285 	/* no processor from mptable or madt */
1286 	if (!num_processors)
1287 		num_processors = 1;
1288 
1289 	possible = num_processors + disabled_cpus;
1290 	if (possible > NR_CPUS)
1291 		possible = NR_CPUS;
1292 
1293 	printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1294 		possible, max_t(int, possible - num_processors, 0));
1295 
1296 	for (i = 0; i < possible; i++)
1297 		cpu_set(i, cpu_possible_map);
1298 
1299 	nr_cpu_ids = possible;
1300 }
1301 
1302 #ifdef CONFIG_HOTPLUG_CPU
1303 
1304 static void remove_siblinginfo(int cpu)
1305 {
1306 	int sibling;
1307 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1308 
1309 	for_each_cpu_mask_nr(sibling, per_cpu(cpu_core_map, cpu)) {
1310 		cpu_clear(cpu, per_cpu(cpu_core_map, sibling));
1311 		/*/
1312 		 * last thread sibling in this cpu core going down
1313 		 */
1314 		if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1)
1315 			cpu_data(sibling).booted_cores--;
1316 	}
1317 
1318 	for_each_cpu_mask_nr(sibling, per_cpu(cpu_sibling_map, cpu))
1319 		cpu_clear(cpu, per_cpu(cpu_sibling_map, sibling));
1320 	cpus_clear(per_cpu(cpu_sibling_map, cpu));
1321 	cpus_clear(per_cpu(cpu_core_map, cpu));
1322 	c->phys_proc_id = 0;
1323 	c->cpu_core_id = 0;
1324 	cpu_clear(cpu, cpu_sibling_setup_map);
1325 }
1326 
1327 static void __ref remove_cpu_from_maps(int cpu)
1328 {
1329 	cpu_clear(cpu, cpu_online_map);
1330 	cpu_clear(cpu, cpu_callout_map);
1331 	cpu_clear(cpu, cpu_callin_map);
1332 	/* was set by cpu_init() */
1333 	cpu_clear(cpu, cpu_initialized);
1334 	numa_remove_cpu(cpu);
1335 }
1336 
1337 void cpu_disable_common(void)
1338 {
1339 	int cpu = smp_processor_id();
1340 	/*
1341 	 * HACK:
1342 	 * Allow any queued timer interrupts to get serviced
1343 	 * This is only a temporary solution until we cleanup
1344 	 * fixup_irqs as we do for IA64.
1345 	 */
1346 	local_irq_enable();
1347 	mdelay(1);
1348 
1349 	local_irq_disable();
1350 	remove_siblinginfo(cpu);
1351 
1352 	/* It's now safe to remove this processor from the online map */
1353 	lock_vector_lock();
1354 	remove_cpu_from_maps(cpu);
1355 	unlock_vector_lock();
1356 	fixup_irqs(cpu_online_map);
1357 }
1358 
1359 int native_cpu_disable(void)
1360 {
1361 	int cpu = smp_processor_id();
1362 
1363 	/*
1364 	 * Perhaps use cpufreq to drop frequency, but that could go
1365 	 * into generic code.
1366 	 *
1367 	 * We won't take down the boot processor on i386 due to some
1368 	 * interrupts only being able to be serviced by the BSP.
1369 	 * Especially so if we're not using an IOAPIC	-zwane
1370 	 */
1371 	if (cpu == 0)
1372 		return -EBUSY;
1373 
1374 	if (nmi_watchdog == NMI_LOCAL_APIC)
1375 		stop_apic_nmi_watchdog(NULL);
1376 	clear_local_APIC();
1377 
1378 	cpu_disable_common();
1379 	return 0;
1380 }
1381 
1382 void native_cpu_die(unsigned int cpu)
1383 {
1384 	/* We don't do anything here: idle task is faking death itself. */
1385 	unsigned int i;
1386 
1387 	for (i = 0; i < 10; i++) {
1388 		/* They ack this in play_dead by setting CPU_DEAD */
1389 		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1390 			printk(KERN_INFO "CPU %d is now offline\n", cpu);
1391 			if (1 == num_online_cpus())
1392 				alternatives_smp_switch(0);
1393 			return;
1394 		}
1395 		msleep(100);
1396 	}
1397 	printk(KERN_ERR "CPU %u didn't die...\n", cpu);
1398 }
1399 
1400 void play_dead_common(void)
1401 {
1402 	idle_task_exit();
1403 	reset_lazy_tlbstate();
1404 	irq_ctx_exit(raw_smp_processor_id());
1405 	c1e_remove_cpu(raw_smp_processor_id());
1406 
1407 	mb();
1408 	/* Ack it */
1409 	__get_cpu_var(cpu_state) = CPU_DEAD;
1410 
1411 	/*
1412 	 * With physical CPU hotplug, we should halt the cpu
1413 	 */
1414 	local_irq_disable();
1415 }
1416 
1417 void native_play_dead(void)
1418 {
1419 	play_dead_common();
1420 	wbinvd_halt();
1421 }
1422 
1423 #else /* ... !CONFIG_HOTPLUG_CPU */
1424 int native_cpu_disable(void)
1425 {
1426 	return -ENOSYS;
1427 }
1428 
1429 void native_cpu_die(unsigned int cpu)
1430 {
1431 	/* We said "no" in __cpu_disable */
1432 	BUG();
1433 }
1434 
1435 void native_play_dead(void)
1436 {
1437 	BUG();
1438 }
1439 
1440 #endif
1441