xref: /openbmc/linux/arch/x86/kernel/smpboot.c (revision 8fa5723aa7e053d498336b48448b292fc2e0458b)
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 #ifdef CONFIG_VMI
298 	vmi_bringup();
299 #endif
300 	cpu_init();
301 	preempt_disable();
302 	smp_callin();
303 
304 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
305 	barrier();
306 	/*
307 	 * Check TSC synchronization with the BP:
308 	 */
309 	check_tsc_sync_target();
310 
311 	if (nmi_watchdog == NMI_IO_APIC) {
312 		disable_8259A_irq(0);
313 		enable_NMI_through_LVT0();
314 		enable_8259A_irq(0);
315 	}
316 
317 #ifdef CONFIG_X86_32
318 	while (low_mappings)
319 		cpu_relax();
320 	__flush_tlb_all();
321 #endif
322 
323 	/* This must be done before setting cpu_online_map */
324 	set_cpu_sibling_map(raw_smp_processor_id());
325 	wmb();
326 
327 	/*
328 	 * We need to hold call_lock, so there is no inconsistency
329 	 * between the time smp_call_function() determines number of
330 	 * IPI recipients, and the time when the determination is made
331 	 * for which cpus receive the IPI. Holding this
332 	 * lock helps us to not include this cpu in a currently in progress
333 	 * smp_call_function().
334 	 *
335 	 * We need to hold vector_lock so there the set of online cpus
336 	 * does not change while we are assigning vectors to cpus.  Holding
337 	 * this lock ensures we don't half assign or remove an irq from a cpu.
338 	 */
339 	ipi_call_lock();
340 	lock_vector_lock();
341 	__setup_vector_irq(smp_processor_id());
342 	cpu_set(smp_processor_id(), cpu_online_map);
343 	unlock_vector_lock();
344 	ipi_call_unlock();
345 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
346 
347 	/* enable local interrupts */
348 	local_irq_enable();
349 
350 	setup_secondary_clock();
351 
352 	wmb();
353 	cpu_idle();
354 }
355 
356 static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c)
357 {
358 	/*
359 	 * Mask B, Pentium, but not Pentium MMX
360 	 */
361 	if (c->x86_vendor == X86_VENDOR_INTEL &&
362 	    c->x86 == 5 &&
363 	    c->x86_mask >= 1 && c->x86_mask <= 4 &&
364 	    c->x86_model <= 3)
365 		/*
366 		 * Remember we have B step Pentia with bugs
367 		 */
368 		smp_b_stepping = 1;
369 
370 	/*
371 	 * Certain Athlons might work (for various values of 'work') in SMP
372 	 * but they are not certified as MP capable.
373 	 */
374 	if ((c->x86_vendor == X86_VENDOR_AMD) && (c->x86 == 6)) {
375 
376 		if (num_possible_cpus() == 1)
377 			goto valid_k7;
378 
379 		/* Athlon 660/661 is valid. */
380 		if ((c->x86_model == 6) && ((c->x86_mask == 0) ||
381 		    (c->x86_mask == 1)))
382 			goto valid_k7;
383 
384 		/* Duron 670 is valid */
385 		if ((c->x86_model == 7) && (c->x86_mask == 0))
386 			goto valid_k7;
387 
388 		/*
389 		 * Athlon 662, Duron 671, and Athlon >model 7 have capability
390 		 * bit. It's worth noting that the A5 stepping (662) of some
391 		 * Athlon XP's have the MP bit set.
392 		 * See http://www.heise.de/newsticker/data/jow-18.10.01-000 for
393 		 * more.
394 		 */
395 		if (((c->x86_model == 6) && (c->x86_mask >= 2)) ||
396 		    ((c->x86_model == 7) && (c->x86_mask >= 1)) ||
397 		     (c->x86_model > 7))
398 			if (cpu_has_mp)
399 				goto valid_k7;
400 
401 		/* If we get here, not a certified SMP capable AMD system. */
402 		unsafe_smp = 1;
403 	}
404 
405 valid_k7:
406 	;
407 }
408 
409 static void __cpuinit smp_checks(void)
410 {
411 	if (smp_b_stepping)
412 		printk(KERN_WARNING "WARNING: SMP operation may be unreliable"
413 				    "with B stepping processors.\n");
414 
415 	/*
416 	 * Don't taint if we are running SMP kernel on a single non-MP
417 	 * approved Athlon
418 	 */
419 	if (unsafe_smp && num_online_cpus() > 1) {
420 		printk(KERN_INFO "WARNING: This combination of AMD"
421 			"processors is not suitable for SMP.\n");
422 		add_taint(TAINT_UNSAFE_SMP);
423 	}
424 }
425 
426 /*
427  * The bootstrap kernel entry code has set these up. Save them for
428  * a given CPU
429  */
430 
431 void __cpuinit smp_store_cpu_info(int id)
432 {
433 	struct cpuinfo_x86 *c = &cpu_data(id);
434 
435 	*c = boot_cpu_data;
436 	c->cpu_index = id;
437 	if (id != 0)
438 		identify_secondary_cpu(c);
439 	smp_apply_quirks(c);
440 }
441 
442 
443 void __cpuinit set_cpu_sibling_map(int cpu)
444 {
445 	int i;
446 	struct cpuinfo_x86 *c = &cpu_data(cpu);
447 
448 	cpu_set(cpu, cpu_sibling_setup_map);
449 
450 	if (smp_num_siblings > 1) {
451 		for_each_cpu_mask_nr(i, cpu_sibling_setup_map) {
452 			if (c->phys_proc_id == cpu_data(i).phys_proc_id &&
453 			    c->cpu_core_id == cpu_data(i).cpu_core_id) {
454 				cpu_set(i, per_cpu(cpu_sibling_map, cpu));
455 				cpu_set(cpu, per_cpu(cpu_sibling_map, i));
456 				cpu_set(i, per_cpu(cpu_core_map, cpu));
457 				cpu_set(cpu, per_cpu(cpu_core_map, i));
458 				cpu_set(i, c->llc_shared_map);
459 				cpu_set(cpu, cpu_data(i).llc_shared_map);
460 			}
461 		}
462 	} else {
463 		cpu_set(cpu, per_cpu(cpu_sibling_map, cpu));
464 	}
465 
466 	cpu_set(cpu, c->llc_shared_map);
467 
468 	if (current_cpu_data.x86_max_cores == 1) {
469 		per_cpu(cpu_core_map, cpu) = per_cpu(cpu_sibling_map, cpu);
470 		c->booted_cores = 1;
471 		return;
472 	}
473 
474 	for_each_cpu_mask_nr(i, cpu_sibling_setup_map) {
475 		if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
476 		    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
477 			cpu_set(i, c->llc_shared_map);
478 			cpu_set(cpu, cpu_data(i).llc_shared_map);
479 		}
480 		if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
481 			cpu_set(i, per_cpu(cpu_core_map, cpu));
482 			cpu_set(cpu, per_cpu(cpu_core_map, i));
483 			/*
484 			 *  Does this new cpu bringup a new core?
485 			 */
486 			if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1) {
487 				/*
488 				 * for each core in package, increment
489 				 * the booted_cores for this new cpu
490 				 */
491 				if (first_cpu(per_cpu(cpu_sibling_map, i)) == i)
492 					c->booted_cores++;
493 				/*
494 				 * increment the core count for all
495 				 * the other cpus in this package
496 				 */
497 				if (i != cpu)
498 					cpu_data(i).booted_cores++;
499 			} else if (i != cpu && !c->booted_cores)
500 				c->booted_cores = cpu_data(i).booted_cores;
501 		}
502 	}
503 }
504 
505 /* maps the cpu to the sched domain representing multi-core */
506 cpumask_t cpu_coregroup_map(int cpu)
507 {
508 	struct cpuinfo_x86 *c = &cpu_data(cpu);
509 	/*
510 	 * For perf, we return last level cache shared map.
511 	 * And for power savings, we return cpu_core_map
512 	 */
513 	if (sched_mc_power_savings || sched_smt_power_savings)
514 		return per_cpu(cpu_core_map, cpu);
515 	else
516 		return c->llc_shared_map;
517 }
518 
519 static void impress_friends(void)
520 {
521 	int cpu;
522 	unsigned long bogosum = 0;
523 	/*
524 	 * Allow the user to impress friends.
525 	 */
526 	pr_debug("Before bogomips.\n");
527 	for_each_possible_cpu(cpu)
528 		if (cpu_isset(cpu, cpu_callout_map))
529 			bogosum += cpu_data(cpu).loops_per_jiffy;
530 	printk(KERN_INFO
531 		"Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
532 		num_online_cpus(),
533 		bogosum/(500000/HZ),
534 		(bogosum/(5000/HZ))%100);
535 
536 	pr_debug("Before bogocount - setting activated=1.\n");
537 }
538 
539 static inline void __inquire_remote_apic(int apicid)
540 {
541 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
542 	char *names[] = { "ID", "VERSION", "SPIV" };
543 	int timeout;
544 	u32 status;
545 
546 	printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
547 
548 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
549 		printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
550 
551 		/*
552 		 * Wait for idle.
553 		 */
554 		status = safe_apic_wait_icr_idle();
555 		if (status)
556 			printk(KERN_CONT
557 			       "a previous APIC delivery may have failed\n");
558 
559 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
560 
561 		timeout = 0;
562 		do {
563 			udelay(100);
564 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
565 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
566 
567 		switch (status) {
568 		case APIC_ICR_RR_VALID:
569 			status = apic_read(APIC_RRR);
570 			printk(KERN_CONT "%08x\n", status);
571 			break;
572 		default:
573 			printk(KERN_CONT "failed\n");
574 		}
575 	}
576 }
577 
578 #ifdef WAKE_SECONDARY_VIA_NMI
579 /*
580  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
581  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
582  * won't ... remember to clear down the APIC, etc later.
583  */
584 static int __devinit
585 wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
586 {
587 	unsigned long send_status, accept_status = 0;
588 	int maxlvt;
589 
590 	/* Target chip */
591 	/* Boot on the stack */
592 	/* Kick the second */
593 	apic_icr_write(APIC_DM_NMI | APIC_DEST_LOGICAL, logical_apicid);
594 
595 	pr_debug("Waiting for send to finish...\n");
596 	send_status = safe_apic_wait_icr_idle();
597 
598 	/*
599 	 * Give the other CPU some time to accept the IPI.
600 	 */
601 	udelay(200);
602 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
603 		maxlvt = lapic_get_maxlvt();
604 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
605 			apic_write(APIC_ESR, 0);
606 		accept_status = (apic_read(APIC_ESR) & 0xEF);
607 	}
608 	pr_debug("NMI sent.\n");
609 
610 	if (send_status)
611 		printk(KERN_ERR "APIC never delivered???\n");
612 	if (accept_status)
613 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
614 
615 	return (send_status | accept_status);
616 }
617 #endif	/* WAKE_SECONDARY_VIA_NMI */
618 
619 #ifdef WAKE_SECONDARY_VIA_INIT
620 static int __devinit
621 wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
622 {
623 	unsigned long send_status, accept_status = 0;
624 	int maxlvt, num_starts, j;
625 
626 	if (get_uv_system_type() == UV_NON_UNIQUE_APIC) {
627 		send_status = uv_wakeup_secondary(phys_apicid, start_eip);
628 		atomic_set(&init_deasserted, 1);
629 		return send_status;
630 	}
631 
632 	maxlvt = lapic_get_maxlvt();
633 
634 	/*
635 	 * Be paranoid about clearing APIC errors.
636 	 */
637 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
638 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
639 			apic_write(APIC_ESR, 0);
640 		apic_read(APIC_ESR);
641 	}
642 
643 	pr_debug("Asserting INIT.\n");
644 
645 	/*
646 	 * Turn INIT on target chip
647 	 */
648 	/*
649 	 * Send IPI
650 	 */
651 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
652 		       phys_apicid);
653 
654 	pr_debug("Waiting for send to finish...\n");
655 	send_status = safe_apic_wait_icr_idle();
656 
657 	mdelay(10);
658 
659 	pr_debug("Deasserting INIT.\n");
660 
661 	/* Target chip */
662 	/* Send IPI */
663 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
664 
665 	pr_debug("Waiting for send to finish...\n");
666 	send_status = safe_apic_wait_icr_idle();
667 
668 	mb();
669 	atomic_set(&init_deasserted, 1);
670 
671 	/*
672 	 * Should we send STARTUP IPIs ?
673 	 *
674 	 * Determine this based on the APIC version.
675 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
676 	 */
677 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
678 		num_starts = 2;
679 	else
680 		num_starts = 0;
681 
682 	/*
683 	 * Paravirt / VMI wants a startup IPI hook here to set up the
684 	 * target processor state.
685 	 */
686 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
687 			 (unsigned long)stack_start.sp);
688 
689 	/*
690 	 * Run STARTUP IPI loop.
691 	 */
692 	pr_debug("#startup loops: %d.\n", num_starts);
693 
694 	for (j = 1; j <= num_starts; j++) {
695 		pr_debug("Sending STARTUP #%d.\n", j);
696 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
697 			apic_write(APIC_ESR, 0);
698 		apic_read(APIC_ESR);
699 		pr_debug("After apic_write.\n");
700 
701 		/*
702 		 * STARTUP IPI
703 		 */
704 
705 		/* Target chip */
706 		/* Boot on the stack */
707 		/* Kick the second */
708 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
709 			       phys_apicid);
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 APIC 0x%x ip 0x%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 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
898 			apic_write(APIC_ESR, 0);
899 			apic_read(APIC_ESR);
900 		}
901 	}
902 
903 	/*
904 	 * Starting actual IPI sequence...
905 	 */
906 	boot_error = wakeup_secondary_cpu(apicid, start_ip);
907 
908 	if (!boot_error) {
909 		/*
910 		 * allow APs to start initializing.
911 		 */
912 		pr_debug("Before Callout %d.\n", cpu);
913 		cpu_set(cpu, cpu_callout_map);
914 		pr_debug("After Callout %d.\n", cpu);
915 
916 		/*
917 		 * Wait 5s total for a response
918 		 */
919 		for (timeout = 0; timeout < 50000; timeout++) {
920 			if (cpu_isset(cpu, cpu_callin_map))
921 				break;	/* It has booted */
922 			udelay(100);
923 		}
924 
925 		if (cpu_isset(cpu, cpu_callin_map)) {
926 			/* number CPUs logically, starting from 1 (BSP is 0) */
927 			pr_debug("OK.\n");
928 			printk(KERN_INFO "CPU%d: ", cpu);
929 			print_cpu_info(&cpu_data(cpu));
930 			pr_debug("CPU has booted.\n");
931 		} else {
932 			boot_error = 1;
933 			if (*((volatile unsigned char *)trampoline_base)
934 					== 0xA5)
935 				/* trampoline started but...? */
936 				printk(KERN_ERR "Stuck ??\n");
937 			else
938 				/* trampoline code not run */
939 				printk(KERN_ERR "Not responding.\n");
940 			if (get_uv_system_type() != UV_NON_UNIQUE_APIC)
941 				inquire_remote_apic(apicid);
942 		}
943 	}
944 #ifdef CONFIG_X86_64
945 restore_state:
946 #endif
947 	if (boot_error) {
948 		/* Try to put things back the way they were before ... */
949 		numa_remove_cpu(cpu); /* was set by numa_add_cpu */
950 		cpu_clear(cpu, cpu_callout_map); /* was set by do_boot_cpu() */
951 		cpu_clear(cpu, cpu_initialized); /* was set by cpu_init() */
952 		cpu_clear(cpu, cpu_present_map);
953 		per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
954 	}
955 
956 	/* mark "stuck" area as not stuck */
957 	*((volatile unsigned long *)trampoline_base) = 0;
958 
959 	/*
960 	 * Cleanup possible dangling ends...
961 	 */
962 	smpboot_restore_warm_reset_vector();
963 
964 	return boot_error;
965 }
966 
967 int __cpuinit native_cpu_up(unsigned int cpu)
968 {
969 	int apicid = cpu_present_to_apicid(cpu);
970 	unsigned long flags;
971 	int err;
972 
973 	WARN_ON(irqs_disabled());
974 
975 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
976 
977 	if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
978 	    !physid_isset(apicid, phys_cpu_present_map)) {
979 		printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
980 		return -EINVAL;
981 	}
982 
983 	/*
984 	 * Already booted CPU?
985 	 */
986 	if (cpu_isset(cpu, cpu_callin_map)) {
987 		pr_debug("do_boot_cpu %d Already started\n", cpu);
988 		return -ENOSYS;
989 	}
990 
991 	/*
992 	 * Save current MTRR state in case it was changed since early boot
993 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
994 	 */
995 	mtrr_save_state();
996 
997 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
998 
999 #ifdef CONFIG_X86_32
1000 	/* init low mem mapping */
1001 	clone_pgd_range(swapper_pg_dir, swapper_pg_dir + KERNEL_PGD_BOUNDARY,
1002 		min_t(unsigned long, KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
1003 	flush_tlb_all();
1004 	low_mappings = 1;
1005 
1006 	err = do_boot_cpu(apicid, cpu);
1007 
1008 	zap_low_mappings();
1009 	low_mappings = 0;
1010 #else
1011 	err = do_boot_cpu(apicid, cpu);
1012 #endif
1013 	if (err) {
1014 		pr_debug("do_boot_cpu failed %d\n", err);
1015 		return -EIO;
1016 	}
1017 
1018 	/*
1019 	 * Check TSC synchronization with the AP (keep irqs disabled
1020 	 * while doing so):
1021 	 */
1022 	local_irq_save(flags);
1023 	check_tsc_sync_source(cpu);
1024 	local_irq_restore(flags);
1025 
1026 	while (!cpu_online(cpu)) {
1027 		cpu_relax();
1028 		touch_nmi_watchdog();
1029 	}
1030 
1031 	return 0;
1032 }
1033 
1034 /*
1035  * Fall back to non SMP mode after errors.
1036  *
1037  * RED-PEN audit/test this more. I bet there is more state messed up here.
1038  */
1039 static __init void disable_smp(void)
1040 {
1041 	cpu_present_map = cpumask_of_cpu(0);
1042 	cpu_possible_map = cpumask_of_cpu(0);
1043 	smpboot_clear_io_apic_irqs();
1044 
1045 	if (smp_found_config)
1046 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1047 	else
1048 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
1049 	map_cpu_to_logical_apicid();
1050 	cpu_set(0, per_cpu(cpu_sibling_map, 0));
1051 	cpu_set(0, per_cpu(cpu_core_map, 0));
1052 }
1053 
1054 /*
1055  * Various sanity checks.
1056  */
1057 static int __init smp_sanity_check(unsigned max_cpus)
1058 {
1059 	preempt_disable();
1060 
1061 #if defined(CONFIG_X86_PC) && defined(CONFIG_X86_32)
1062 	if (def_to_bigsmp && nr_cpu_ids > 8) {
1063 		unsigned int cpu;
1064 		unsigned nr;
1065 
1066 		printk(KERN_WARNING
1067 		       "More than 8 CPUs detected - skipping them.\n"
1068 		       "Use CONFIG_X86_GENERICARCH and CONFIG_X86_BIGSMP.\n");
1069 
1070 		nr = 0;
1071 		for_each_present_cpu(cpu) {
1072 			if (nr >= 8)
1073 				cpu_clear(cpu, cpu_present_map);
1074 			nr++;
1075 		}
1076 
1077 		nr = 0;
1078 		for_each_possible_cpu(cpu) {
1079 			if (nr >= 8)
1080 				cpu_clear(cpu, cpu_possible_map);
1081 			nr++;
1082 		}
1083 
1084 		nr_cpu_ids = 8;
1085 	}
1086 #endif
1087 
1088 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1089 		printk(KERN_WARNING "weird, boot CPU (#%d) not listed"
1090 				    "by the BIOS.\n", hard_smp_processor_id());
1091 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1092 	}
1093 
1094 	/*
1095 	 * If we couldn't find an SMP configuration at boot time,
1096 	 * get out of here now!
1097 	 */
1098 	if (!smp_found_config && !acpi_lapic) {
1099 		preempt_enable();
1100 		printk(KERN_NOTICE "SMP motherboard not detected.\n");
1101 		disable_smp();
1102 		if (APIC_init_uniprocessor())
1103 			printk(KERN_NOTICE "Local APIC not detected."
1104 					   " Using dummy APIC emulation.\n");
1105 		return -1;
1106 	}
1107 
1108 	/*
1109 	 * Should not be necessary because the MP table should list the boot
1110 	 * CPU too, but we do it for the sake of robustness anyway.
1111 	 */
1112 	if (!check_phys_apicid_present(boot_cpu_physical_apicid)) {
1113 		printk(KERN_NOTICE
1114 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
1115 			boot_cpu_physical_apicid);
1116 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1117 	}
1118 	preempt_enable();
1119 
1120 	/*
1121 	 * If we couldn't find a local APIC, then get out of here now!
1122 	 */
1123 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1124 	    !cpu_has_apic) {
1125 		printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
1126 			boot_cpu_physical_apicid);
1127 		printk(KERN_ERR "... forcing use of dummy APIC emulation."
1128 				"(tell your hw vendor)\n");
1129 		smpboot_clear_io_apic();
1130 		return -1;
1131 	}
1132 
1133 	verify_local_APIC();
1134 
1135 	/*
1136 	 * If SMP should be disabled, then really disable it!
1137 	 */
1138 	if (!max_cpus) {
1139 		printk(KERN_INFO "SMP mode deactivated.\n");
1140 		smpboot_clear_io_apic();
1141 
1142 		localise_nmi_watchdog();
1143 
1144 		connect_bsp_APIC();
1145 		setup_local_APIC();
1146 		end_local_APIC_setup();
1147 		return -1;
1148 	}
1149 
1150 	return 0;
1151 }
1152 
1153 static void __init smp_cpu_index_default(void)
1154 {
1155 	int i;
1156 	struct cpuinfo_x86 *c;
1157 
1158 	for_each_possible_cpu(i) {
1159 		c = &cpu_data(i);
1160 		/* mark all to hotplug */
1161 		c->cpu_index = NR_CPUS;
1162 	}
1163 }
1164 
1165 /*
1166  * Prepare for SMP bootup.  The MP table or ACPI has been read
1167  * earlier.  Just do some sanity checking here and enable APIC mode.
1168  */
1169 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1170 {
1171 	preempt_disable();
1172 	smp_cpu_index_default();
1173 	current_cpu_data = boot_cpu_data;
1174 	cpu_callin_map = cpumask_of_cpu(0);
1175 	mb();
1176 	/*
1177 	 * Setup boot CPU information
1178 	 */
1179 	smp_store_cpu_info(0); /* Final full version of the data */
1180 #ifdef CONFIG_X86_32
1181 	boot_cpu_logical_apicid = logical_smp_processor_id();
1182 #endif
1183 	current_thread_info()->cpu = 0;  /* needed? */
1184 	set_cpu_sibling_map(0);
1185 
1186 #ifdef CONFIG_X86_64
1187 	enable_IR_x2apic();
1188 	setup_apic_routing();
1189 #endif
1190 
1191 	if (smp_sanity_check(max_cpus) < 0) {
1192 		printk(KERN_INFO "SMP disabled\n");
1193 		disable_smp();
1194 		goto out;
1195 	}
1196 
1197 	preempt_disable();
1198 	if (read_apic_id() != boot_cpu_physical_apicid) {
1199 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1200 		     read_apic_id(), boot_cpu_physical_apicid);
1201 		/* Or can we switch back to PIC here? */
1202 	}
1203 	preempt_enable();
1204 
1205 	connect_bsp_APIC();
1206 
1207 	/*
1208 	 * Switch from PIC to APIC mode.
1209 	 */
1210 	setup_local_APIC();
1211 
1212 #ifdef CONFIG_X86_64
1213 	/*
1214 	 * Enable IO APIC before setting up error vector
1215 	 */
1216 	if (!skip_ioapic_setup && nr_ioapics)
1217 		enable_IO_APIC();
1218 #endif
1219 	end_local_APIC_setup();
1220 
1221 	map_cpu_to_logical_apicid();
1222 
1223 	setup_portio_remap();
1224 
1225 	smpboot_setup_io_apic();
1226 	/*
1227 	 * Set up local APIC timer on boot CPU.
1228 	 */
1229 
1230 	printk(KERN_INFO "CPU%d: ", 0);
1231 	print_cpu_info(&cpu_data(0));
1232 	setup_boot_clock();
1233 
1234 	if (is_uv_system())
1235 		uv_system_init();
1236 out:
1237 	preempt_enable();
1238 }
1239 /*
1240  * Early setup to make printk work.
1241  */
1242 void __init native_smp_prepare_boot_cpu(void)
1243 {
1244 	int me = smp_processor_id();
1245 #ifdef CONFIG_X86_32
1246 	init_gdt(me);
1247 #endif
1248 	switch_to_new_gdt();
1249 	/* already set me in cpu_online_map in boot_cpu_init() */
1250 	cpu_set(me, cpu_callout_map);
1251 	per_cpu(cpu_state, me) = CPU_ONLINE;
1252 }
1253 
1254 void __init native_smp_cpus_done(unsigned int max_cpus)
1255 {
1256 	pr_debug("Boot done.\n");
1257 
1258 	impress_friends();
1259 	smp_checks();
1260 #ifdef CONFIG_X86_IO_APIC
1261 	setup_ioapic_dest();
1262 #endif
1263 	check_nmi_watchdog();
1264 }
1265 
1266 /*
1267  * cpu_possible_map should be static, it cannot change as cpu's
1268  * are onlined, or offlined. The reason is per-cpu data-structures
1269  * are allocated by some modules at init time, and dont expect to
1270  * do this dynamically on cpu arrival/departure.
1271  * cpu_present_map on the other hand can change dynamically.
1272  * In case when cpu_hotplug is not compiled, then we resort to current
1273  * behaviour, which is cpu_possible == cpu_present.
1274  * - Ashok Raj
1275  *
1276  * Three ways to find out the number of additional hotplug CPUs:
1277  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1278  * - The user can overwrite it with additional_cpus=NUM
1279  * - Otherwise don't reserve additional CPUs.
1280  * We do this because additional CPUs waste a lot of memory.
1281  * -AK
1282  */
1283 __init void prefill_possible_map(void)
1284 {
1285 	int i, possible;
1286 
1287 	/* no processor from mptable or madt */
1288 	if (!num_processors)
1289 		num_processors = 1;
1290 
1291 	possible = num_processors + disabled_cpus;
1292 	if (possible > NR_CPUS)
1293 		possible = NR_CPUS;
1294 
1295 	printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1296 		possible, max_t(int, possible - num_processors, 0));
1297 
1298 	for (i = 0; i < possible; i++)
1299 		cpu_set(i, cpu_possible_map);
1300 
1301 	nr_cpu_ids = possible;
1302 }
1303 
1304 #ifdef CONFIG_HOTPLUG_CPU
1305 
1306 static void remove_siblinginfo(int cpu)
1307 {
1308 	int sibling;
1309 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1310 
1311 	for_each_cpu_mask_nr(sibling, per_cpu(cpu_core_map, cpu)) {
1312 		cpu_clear(cpu, per_cpu(cpu_core_map, sibling));
1313 		/*/
1314 		 * last thread sibling in this cpu core going down
1315 		 */
1316 		if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1)
1317 			cpu_data(sibling).booted_cores--;
1318 	}
1319 
1320 	for_each_cpu_mask_nr(sibling, per_cpu(cpu_sibling_map, cpu))
1321 		cpu_clear(cpu, per_cpu(cpu_sibling_map, sibling));
1322 	cpus_clear(per_cpu(cpu_sibling_map, cpu));
1323 	cpus_clear(per_cpu(cpu_core_map, cpu));
1324 	c->phys_proc_id = 0;
1325 	c->cpu_core_id = 0;
1326 	cpu_clear(cpu, cpu_sibling_setup_map);
1327 }
1328 
1329 static void __ref remove_cpu_from_maps(int cpu)
1330 {
1331 	cpu_clear(cpu, cpu_online_map);
1332 	cpu_clear(cpu, cpu_callout_map);
1333 	cpu_clear(cpu, cpu_callin_map);
1334 	/* was set by cpu_init() */
1335 	cpu_clear(cpu, cpu_initialized);
1336 	numa_remove_cpu(cpu);
1337 }
1338 
1339 void cpu_disable_common(void)
1340 {
1341 	int cpu = smp_processor_id();
1342 	/*
1343 	 * HACK:
1344 	 * Allow any queued timer interrupts to get serviced
1345 	 * This is only a temporary solution until we cleanup
1346 	 * fixup_irqs as we do for IA64.
1347 	 */
1348 	local_irq_enable();
1349 	mdelay(1);
1350 
1351 	local_irq_disable();
1352 	remove_siblinginfo(cpu);
1353 
1354 	/* It's now safe to remove this processor from the online map */
1355 	lock_vector_lock();
1356 	remove_cpu_from_maps(cpu);
1357 	unlock_vector_lock();
1358 	fixup_irqs(cpu_online_map);
1359 }
1360 
1361 int native_cpu_disable(void)
1362 {
1363 	int cpu = smp_processor_id();
1364 
1365 	/*
1366 	 * Perhaps use cpufreq to drop frequency, but that could go
1367 	 * into generic code.
1368 	 *
1369 	 * We won't take down the boot processor on i386 due to some
1370 	 * interrupts only being able to be serviced by the BSP.
1371 	 * Especially so if we're not using an IOAPIC	-zwane
1372 	 */
1373 	if (cpu == 0)
1374 		return -EBUSY;
1375 
1376 	if (nmi_watchdog == NMI_LOCAL_APIC)
1377 		stop_apic_nmi_watchdog(NULL);
1378 	clear_local_APIC();
1379 
1380 	cpu_disable_common();
1381 	return 0;
1382 }
1383 
1384 void native_cpu_die(unsigned int cpu)
1385 {
1386 	/* We don't do anything here: idle task is faking death itself. */
1387 	unsigned int i;
1388 
1389 	for (i = 0; i < 10; i++) {
1390 		/* They ack this in play_dead by setting CPU_DEAD */
1391 		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1392 			printk(KERN_INFO "CPU %d is now offline\n", cpu);
1393 			if (1 == num_online_cpus())
1394 				alternatives_smp_switch(0);
1395 			return;
1396 		}
1397 		msleep(100);
1398 	}
1399 	printk(KERN_ERR "CPU %u didn't die...\n", cpu);
1400 }
1401 
1402 void play_dead_common(void)
1403 {
1404 	idle_task_exit();
1405 	reset_lazy_tlbstate();
1406 	irq_ctx_exit(raw_smp_processor_id());
1407 	c1e_remove_cpu(raw_smp_processor_id());
1408 
1409 	mb();
1410 	/* Ack it */
1411 	__get_cpu_var(cpu_state) = CPU_DEAD;
1412 
1413 	/*
1414 	 * With physical CPU hotplug, we should halt the cpu
1415 	 */
1416 	local_irq_disable();
1417 }
1418 
1419 void native_play_dead(void)
1420 {
1421 	play_dead_common();
1422 	wbinvd_halt();
1423 }
1424 
1425 #else /* ... !CONFIG_HOTPLUG_CPU */
1426 int native_cpu_disable(void)
1427 {
1428 	return -ENOSYS;
1429 }
1430 
1431 void native_cpu_die(unsigned int cpu)
1432 {
1433 	/* We said "no" in __cpu_disable */
1434 	BUG();
1435 }
1436 
1437 void native_play_dead(void)
1438 {
1439 	BUG();
1440 }
1441 
1442 #endif
1443