xref: /openbmc/linux/arch/x86/kernel/smpboot.c (revision b8bb76713ec50df2f11efee386e16f93d51e1076)
1 /*
2  *	x86 SMP booting functions
3  *
4  *	(c) 1995 Alan Cox, Building #3 <alan@lxorguk.ukuu.org.uk>
5  *	(c) 1998, 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
6  *	Copyright 2001 Andi Kleen, SuSE Labs.
7  *
8  *	Much of the core SMP work is based on previous work by Thomas Radke, to
9  *	whom a great many thanks are extended.
10  *
11  *	Thanks to Intel for making available several different Pentium,
12  *	Pentium Pro and Pentium-II/Xeon MP machines.
13  *	Original development of Linux SMP code supported by Caldera.
14  *
15  *	This code is released under the GNU General Public License version 2 or
16  *	later.
17  *
18  *	Fixes
19  *		Felix Koop	:	NR_CPUS used properly
20  *		Jose Renau	:	Handle single CPU case.
21  *		Alan Cox	:	By repeated request 8) - Total BogoMIPS report.
22  *		Greg Wright	:	Fix for kernel stacks panic.
23  *		Erich Boleyn	:	MP v1.4 and additional changes.
24  *	Matthias Sattler	:	Changes for 2.1 kernel map.
25  *	Michel Lespinasse	:	Changes for 2.1 kernel map.
26  *	Michael Chastain	:	Change trampoline.S to gnu as.
27  *		Alan Cox	:	Dumb bug: 'B' step PPro's are fine
28  *		Ingo Molnar	:	Added APIC timers, based on code
29  *					from Jose Renau
30  *		Ingo Molnar	:	various cleanups and rewrites
31  *		Tigran Aivazian	:	fixed "0.00 in /proc/uptime on SMP" bug.
32  *	Maciej W. Rozycki	:	Bits for genuine 82489DX APICs
33  *	Andi Kleen		:	Changed for SMP boot into long mode.
34  *		Martin J. Bligh	: 	Added support for multi-quad systems
35  *		Dave Jones	:	Report invalid combinations of Athlon CPUs.
36  *		Rusty Russell	:	Hacked into shape for new "hotplug" boot process.
37  *      Andi Kleen              :       Converted to new state machine.
38  *	Ashok Raj		: 	CPU hotplug support
39  *	Glauber Costa		:	i386 and x86_64 integration
40  */
41 
42 #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/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/apic.h>
64 #include <asm/setup.h>
65 #include <asm/uv/uv.h>
66 #include <linux/mc146818rtc.h>
67 
68 #include <asm/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 static 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 /* representing HT siblings of each logical CPU */
104 DEFINE_PER_CPU(cpumask_t, cpu_sibling_map);
105 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
106 
107 /* representing HT and core siblings of each logical CPU */
108 DEFINE_PER_CPU(cpumask_t, cpu_core_map);
109 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
110 
111 /* Per CPU bogomips and other parameters */
112 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
113 EXPORT_PER_CPU_SYMBOL(cpu_info);
114 
115 atomic_t init_deasserted;
116 
117 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_32)
118 
119 /* which logical CPUs are on which nodes */
120 cpumask_t node_to_cpumask_map[MAX_NUMNODES] __read_mostly =
121 				{ [0 ... MAX_NUMNODES-1] = CPU_MASK_NONE };
122 EXPORT_SYMBOL(node_to_cpumask_map);
123 /* which node each logical CPU is on */
124 int cpu_to_node_map[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 };
125 EXPORT_SYMBOL(cpu_to_node_map);
126 
127 /* set up a mapping between cpu and node. */
128 static void map_cpu_to_node(int cpu, int node)
129 {
130 	printk(KERN_INFO "Mapping cpu %d to node %d\n", cpu, node);
131 	cpumask_set_cpu(cpu, &node_to_cpumask_map[node]);
132 	cpu_to_node_map[cpu] = node;
133 }
134 
135 /* undo a mapping between cpu and node. */
136 static void unmap_cpu_to_node(int cpu)
137 {
138 	int node;
139 
140 	printk(KERN_INFO "Unmapping cpu %d from all nodes\n", cpu);
141 	for (node = 0; node < MAX_NUMNODES; node++)
142 		cpumask_clear_cpu(cpu, &node_to_cpumask_map[node]);
143 	cpu_to_node_map[cpu] = 0;
144 }
145 #else /* !(CONFIG_NUMA && CONFIG_X86_32) */
146 #define map_cpu_to_node(cpu, node)	({})
147 #define unmap_cpu_to_node(cpu)	({})
148 #endif
149 
150 #ifdef CONFIG_X86_32
151 static int boot_cpu_logical_apicid;
152 
153 u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly =
154 					{ [0 ... NR_CPUS-1] = BAD_APICID };
155 
156 static void map_cpu_to_logical_apicid(void)
157 {
158 	int cpu = smp_processor_id();
159 	int apicid = logical_smp_processor_id();
160 	int node = apic->apicid_to_node(apicid);
161 
162 	if (!node_online(node))
163 		node = first_online_node;
164 
165 	cpu_2_logical_apicid[cpu] = apicid;
166 	map_cpu_to_node(cpu, node);
167 }
168 
169 void numa_remove_cpu(int cpu)
170 {
171 	cpu_2_logical_apicid[cpu] = BAD_APICID;
172 	unmap_cpu_to_node(cpu);
173 }
174 #else
175 #define map_cpu_to_logical_apicid()  do {} while (0)
176 #endif
177 
178 /*
179  * Report back to the Boot Processor.
180  * Running on AP.
181  */
182 static void __cpuinit smp_callin(void)
183 {
184 	int cpuid, phys_id;
185 	unsigned long timeout;
186 
187 	/*
188 	 * If waken up by an INIT in an 82489DX configuration
189 	 * we may get here before an INIT-deassert IPI reaches
190 	 * our local APIC.  We have to wait for the IPI or we'll
191 	 * lock up on an APIC access.
192 	 */
193 	if (apic->wait_for_init_deassert)
194 		apic->wait_for_init_deassert(&init_deasserted);
195 
196 	/*
197 	 * (This works even if the APIC is not enabled.)
198 	 */
199 	phys_id = read_apic_id();
200 	cpuid = smp_processor_id();
201 	if (cpumask_test_cpu(cpuid, cpu_callin_mask)) {
202 		panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
203 					phys_id, cpuid);
204 	}
205 	pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
206 
207 	/*
208 	 * STARTUP IPIs are fragile beasts as they might sometimes
209 	 * trigger some glue motherboard logic. Complete APIC bus
210 	 * silence for 1 second, this overestimates the time the
211 	 * boot CPU is spending to send the up to 2 STARTUP IPIs
212 	 * by a factor of two. This should be enough.
213 	 */
214 
215 	/*
216 	 * Waiting 2s total for startup (udelay is not yet working)
217 	 */
218 	timeout = jiffies + 2*HZ;
219 	while (time_before(jiffies, timeout)) {
220 		/*
221 		 * Has the boot CPU finished it's STARTUP sequence?
222 		 */
223 		if (cpumask_test_cpu(cpuid, cpu_callout_mask))
224 			break;
225 		cpu_relax();
226 	}
227 
228 	if (!time_before(jiffies, timeout)) {
229 		panic("%s: CPU%d started up but did not get a callout!\n",
230 		      __func__, cpuid);
231 	}
232 
233 	/*
234 	 * the boot CPU has finished the init stage and is spinning
235 	 * on callin_map until we finish. We are free to set up this
236 	 * CPU, first the APIC. (this is probably redundant on most
237 	 * boards)
238 	 */
239 
240 	pr_debug("CALLIN, before setup_local_APIC().\n");
241 	if (apic->smp_callin_clear_local_apic)
242 		apic->smp_callin_clear_local_apic();
243 	setup_local_APIC();
244 	end_local_APIC_setup();
245 	map_cpu_to_logical_apicid();
246 
247 	notify_cpu_starting(cpuid);
248 	/*
249 	 * Get our bogomips.
250 	 *
251 	 * Need to enable IRQs because it can take longer and then
252 	 * the NMI watchdog might kill us.
253 	 */
254 	local_irq_enable();
255 	calibrate_delay();
256 	local_irq_disable();
257 	pr_debug("Stack at about %p\n", &cpuid);
258 
259 	/*
260 	 * Save our processor parameters
261 	 */
262 	smp_store_cpu_info(cpuid);
263 
264 	/*
265 	 * Allow the master to continue.
266 	 */
267 	cpumask_set_cpu(cpuid, cpu_callin_mask);
268 }
269 
270 /*
271  * Activate a secondary processor.
272  */
273 notrace static void __cpuinit start_secondary(void *unused)
274 {
275 	/*
276 	 * Don't put *anything* before cpu_init(), SMP booting is too
277 	 * fragile that we want to limit the things done here to the
278 	 * most necessary things.
279 	 */
280 	vmi_bringup();
281 	cpu_init();
282 	preempt_disable();
283 	smp_callin();
284 
285 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
286 	barrier();
287 	/*
288 	 * Check TSC synchronization with the BP:
289 	 */
290 	check_tsc_sync_target();
291 
292 	if (nmi_watchdog == NMI_IO_APIC) {
293 		disable_8259A_irq(0);
294 		enable_NMI_through_LVT0();
295 		enable_8259A_irq(0);
296 	}
297 
298 #ifdef CONFIG_X86_32
299 	while (low_mappings)
300 		cpu_relax();
301 	__flush_tlb_all();
302 #endif
303 
304 	/* This must be done before setting cpu_online_map */
305 	set_cpu_sibling_map(raw_smp_processor_id());
306 	wmb();
307 
308 	/*
309 	 * We need to hold call_lock, so there is no inconsistency
310 	 * between the time smp_call_function() determines number of
311 	 * IPI recipients, and the time when the determination is made
312 	 * for which cpus receive the IPI. Holding this
313 	 * lock helps us to not include this cpu in a currently in progress
314 	 * smp_call_function().
315 	 *
316 	 * We need to hold vector_lock so there the set of online cpus
317 	 * does not change while we are assigning vectors to cpus.  Holding
318 	 * this lock ensures we don't half assign or remove an irq from a cpu.
319 	 */
320 	ipi_call_lock();
321 	lock_vector_lock();
322 	__setup_vector_irq(smp_processor_id());
323 	set_cpu_online(smp_processor_id(), true);
324 	unlock_vector_lock();
325 	ipi_call_unlock();
326 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
327 
328 	/* enable local interrupts */
329 	local_irq_enable();
330 
331 	setup_secondary_clock();
332 
333 	wmb();
334 	cpu_idle();
335 }
336 
337 /*
338  * The bootstrap kernel entry code has set these up. Save them for
339  * a given CPU
340  */
341 
342 void __cpuinit smp_store_cpu_info(int id)
343 {
344 	struct cpuinfo_x86 *c = &cpu_data(id);
345 
346 	*c = boot_cpu_data;
347 	c->cpu_index = id;
348 	if (id != 0)
349 		identify_secondary_cpu(c);
350 }
351 
352 
353 void __cpuinit set_cpu_sibling_map(int cpu)
354 {
355 	int i;
356 	struct cpuinfo_x86 *c = &cpu_data(cpu);
357 
358 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
359 
360 	if (smp_num_siblings > 1) {
361 		for_each_cpu(i, cpu_sibling_setup_mask) {
362 			struct cpuinfo_x86 *o = &cpu_data(i);
363 
364 			if (c->phys_proc_id == o->phys_proc_id &&
365 			    c->cpu_core_id == o->cpu_core_id) {
366 				cpumask_set_cpu(i, cpu_sibling_mask(cpu));
367 				cpumask_set_cpu(cpu, cpu_sibling_mask(i));
368 				cpumask_set_cpu(i, cpu_core_mask(cpu));
369 				cpumask_set_cpu(cpu, cpu_core_mask(i));
370 				cpumask_set_cpu(i, &c->llc_shared_map);
371 				cpumask_set_cpu(cpu, &o->llc_shared_map);
372 			}
373 		}
374 	} else {
375 		cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
376 	}
377 
378 	cpumask_set_cpu(cpu, &c->llc_shared_map);
379 
380 	if (current_cpu_data.x86_max_cores == 1) {
381 		cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
382 		c->booted_cores = 1;
383 		return;
384 	}
385 
386 	for_each_cpu(i, cpu_sibling_setup_mask) {
387 		if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
388 		    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
389 			cpumask_set_cpu(i, &c->llc_shared_map);
390 			cpumask_set_cpu(cpu, &cpu_data(i).llc_shared_map);
391 		}
392 		if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
393 			cpumask_set_cpu(i, cpu_core_mask(cpu));
394 			cpumask_set_cpu(cpu, cpu_core_mask(i));
395 			/*
396 			 *  Does this new cpu bringup a new core?
397 			 */
398 			if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
399 				/*
400 				 * for each core in package, increment
401 				 * the booted_cores for this new cpu
402 				 */
403 				if (cpumask_first(cpu_sibling_mask(i)) == i)
404 					c->booted_cores++;
405 				/*
406 				 * increment the core count for all
407 				 * the other cpus in this package
408 				 */
409 				if (i != cpu)
410 					cpu_data(i).booted_cores++;
411 			} else if (i != cpu && !c->booted_cores)
412 				c->booted_cores = cpu_data(i).booted_cores;
413 		}
414 	}
415 }
416 
417 /* maps the cpu to the sched domain representing multi-core */
418 const struct cpumask *cpu_coregroup_mask(int cpu)
419 {
420 	struct cpuinfo_x86 *c = &cpu_data(cpu);
421 	/*
422 	 * For perf, we return last level cache shared map.
423 	 * And for power savings, we return cpu_core_map
424 	 */
425 	if (sched_mc_power_savings || sched_smt_power_savings)
426 		return cpu_core_mask(cpu);
427 	else
428 		return &c->llc_shared_map;
429 }
430 
431 cpumask_t cpu_coregroup_map(int cpu)
432 {
433 	return *cpu_coregroup_mask(cpu);
434 }
435 
436 static void impress_friends(void)
437 {
438 	int cpu;
439 	unsigned long bogosum = 0;
440 	/*
441 	 * Allow the user to impress friends.
442 	 */
443 	pr_debug("Before bogomips.\n");
444 	for_each_possible_cpu(cpu)
445 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
446 			bogosum += cpu_data(cpu).loops_per_jiffy;
447 	printk(KERN_INFO
448 		"Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
449 		num_online_cpus(),
450 		bogosum/(500000/HZ),
451 		(bogosum/(5000/HZ))%100);
452 
453 	pr_debug("Before bogocount - setting activated=1.\n");
454 }
455 
456 void __inquire_remote_apic(int apicid)
457 {
458 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
459 	char *names[] = { "ID", "VERSION", "SPIV" };
460 	int timeout;
461 	u32 status;
462 
463 	printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
464 
465 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
466 		printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
467 
468 		/*
469 		 * Wait for idle.
470 		 */
471 		status = safe_apic_wait_icr_idle();
472 		if (status)
473 			printk(KERN_CONT
474 			       "a previous APIC delivery may have failed\n");
475 
476 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
477 
478 		timeout = 0;
479 		do {
480 			udelay(100);
481 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
482 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
483 
484 		switch (status) {
485 		case APIC_ICR_RR_VALID:
486 			status = apic_read(APIC_RRR);
487 			printk(KERN_CONT "%08x\n", status);
488 			break;
489 		default:
490 			printk(KERN_CONT "failed\n");
491 		}
492 	}
493 }
494 
495 /*
496  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
497  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
498  * won't ... remember to clear down the APIC, etc later.
499  */
500 int __devinit
501 wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
502 {
503 	unsigned long send_status, accept_status = 0;
504 	int maxlvt;
505 
506 	/* Target chip */
507 	/* Boot on the stack */
508 	/* Kick the second */
509 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, logical_apicid);
510 
511 	pr_debug("Waiting for send to finish...\n");
512 	send_status = safe_apic_wait_icr_idle();
513 
514 	/*
515 	 * Give the other CPU some time to accept the IPI.
516 	 */
517 	udelay(200);
518 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
519 		maxlvt = lapic_get_maxlvt();
520 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
521 			apic_write(APIC_ESR, 0);
522 		accept_status = (apic_read(APIC_ESR) & 0xEF);
523 	}
524 	pr_debug("NMI sent.\n");
525 
526 	if (send_status)
527 		printk(KERN_ERR "APIC never delivered???\n");
528 	if (accept_status)
529 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
530 
531 	return (send_status | accept_status);
532 }
533 
534 int __devinit
535 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
536 {
537 	unsigned long send_status, accept_status = 0;
538 	int maxlvt, num_starts, j;
539 
540 	maxlvt = lapic_get_maxlvt();
541 
542 	/*
543 	 * Be paranoid about clearing APIC errors.
544 	 */
545 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
546 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
547 			apic_write(APIC_ESR, 0);
548 		apic_read(APIC_ESR);
549 	}
550 
551 	pr_debug("Asserting INIT.\n");
552 
553 	/*
554 	 * Turn INIT on target chip
555 	 */
556 	/*
557 	 * Send IPI
558 	 */
559 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
560 		       phys_apicid);
561 
562 	pr_debug("Waiting for send to finish...\n");
563 	send_status = safe_apic_wait_icr_idle();
564 
565 	mdelay(10);
566 
567 	pr_debug("Deasserting INIT.\n");
568 
569 	/* Target chip */
570 	/* Send IPI */
571 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
572 
573 	pr_debug("Waiting for send to finish...\n");
574 	send_status = safe_apic_wait_icr_idle();
575 
576 	mb();
577 	atomic_set(&init_deasserted, 1);
578 
579 	/*
580 	 * Should we send STARTUP IPIs ?
581 	 *
582 	 * Determine this based on the APIC version.
583 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
584 	 */
585 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
586 		num_starts = 2;
587 	else
588 		num_starts = 0;
589 
590 	/*
591 	 * Paravirt / VMI wants a startup IPI hook here to set up the
592 	 * target processor state.
593 	 */
594 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
595 			 (unsigned long)stack_start.sp);
596 
597 	/*
598 	 * Run STARTUP IPI loop.
599 	 */
600 	pr_debug("#startup loops: %d.\n", num_starts);
601 
602 	for (j = 1; j <= num_starts; j++) {
603 		pr_debug("Sending STARTUP #%d.\n", j);
604 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
605 			apic_write(APIC_ESR, 0);
606 		apic_read(APIC_ESR);
607 		pr_debug("After apic_write.\n");
608 
609 		/*
610 		 * STARTUP IPI
611 		 */
612 
613 		/* Target chip */
614 		/* Boot on the stack */
615 		/* Kick the second */
616 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
617 			       phys_apicid);
618 
619 		/*
620 		 * Give the other CPU some time to accept the IPI.
621 		 */
622 		udelay(300);
623 
624 		pr_debug("Startup point 1.\n");
625 
626 		pr_debug("Waiting for send to finish...\n");
627 		send_status = safe_apic_wait_icr_idle();
628 
629 		/*
630 		 * Give the other CPU some time to accept the IPI.
631 		 */
632 		udelay(200);
633 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
634 			apic_write(APIC_ESR, 0);
635 		accept_status = (apic_read(APIC_ESR) & 0xEF);
636 		if (send_status || accept_status)
637 			break;
638 	}
639 	pr_debug("After Startup.\n");
640 
641 	if (send_status)
642 		printk(KERN_ERR "APIC never delivered???\n");
643 	if (accept_status)
644 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
645 
646 	return (send_status | accept_status);
647 }
648 
649 struct create_idle {
650 	struct work_struct work;
651 	struct task_struct *idle;
652 	struct completion done;
653 	int cpu;
654 };
655 
656 static void __cpuinit do_fork_idle(struct work_struct *work)
657 {
658 	struct create_idle *c_idle =
659 		container_of(work, struct create_idle, work);
660 
661 	c_idle->idle = fork_idle(c_idle->cpu);
662 	complete(&c_idle->done);
663 }
664 
665 /*
666  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
667  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
668  * Returns zero if CPU booted OK, else error code from
669  * ->wakeup_secondary_cpu.
670  */
671 static int __cpuinit do_boot_cpu(int apicid, int cpu)
672 {
673 	unsigned long boot_error = 0;
674 	unsigned long start_ip;
675 	int timeout;
676 	struct create_idle c_idle = {
677 		.cpu	= cpu,
678 		.done	= COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
679 	};
680 
681 	INIT_WORK(&c_idle.work, do_fork_idle);
682 
683 	alternatives_smp_switch(1);
684 
685 	c_idle.idle = get_idle_for_cpu(cpu);
686 
687 	/*
688 	 * We can't use kernel_thread since we must avoid to
689 	 * reschedule the child.
690 	 */
691 	if (c_idle.idle) {
692 		c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
693 			(THREAD_SIZE +  task_stack_page(c_idle.idle))) - 1);
694 		init_idle(c_idle.idle, cpu);
695 		goto do_rest;
696 	}
697 
698 	if (!keventd_up() || current_is_keventd())
699 		c_idle.work.func(&c_idle.work);
700 	else {
701 		schedule_work(&c_idle.work);
702 		wait_for_completion(&c_idle.done);
703 	}
704 
705 	if (IS_ERR(c_idle.idle)) {
706 		printk("failed fork for CPU %d\n", cpu);
707 		return PTR_ERR(c_idle.idle);
708 	}
709 
710 	set_idle_for_cpu(cpu, c_idle.idle);
711 do_rest:
712 	per_cpu(current_task, cpu) = c_idle.idle;
713 #ifdef CONFIG_X86_32
714 	/* Stack for startup_32 can be just as for start_secondary onwards */
715 	irq_ctx_init(cpu);
716 #else
717 	clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
718 	initial_gs = per_cpu_offset(cpu);
719 	per_cpu(kernel_stack, cpu) =
720 		(unsigned long)task_stack_page(c_idle.idle) -
721 		KERNEL_STACK_OFFSET + THREAD_SIZE;
722 #endif
723 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
724 	initial_code = (unsigned long)start_secondary;
725 	stack_start.sp = (void *) c_idle.idle->thread.sp;
726 
727 	/* start_ip had better be page-aligned! */
728 	start_ip = setup_trampoline();
729 
730 	/* So we see what's up   */
731 	printk(KERN_INFO "Booting processor %d APIC 0x%x ip 0x%lx\n",
732 			  cpu, apicid, start_ip);
733 
734 	/*
735 	 * This grunge runs the startup process for
736 	 * the targeted processor.
737 	 */
738 
739 	atomic_set(&init_deasserted, 0);
740 
741 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
742 
743 		pr_debug("Setting warm reset code and vector.\n");
744 
745 		smpboot_setup_warm_reset_vector(start_ip);
746 		/*
747 		 * Be paranoid about clearing APIC errors.
748 		*/
749 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
750 			apic_write(APIC_ESR, 0);
751 			apic_read(APIC_ESR);
752 		}
753 	}
754 
755 	/*
756 	 * Kick the secondary CPU. Use the method in the APIC driver
757 	 * if it's defined - or use an INIT boot APIC message otherwise:
758 	 */
759 	if (apic->wakeup_secondary_cpu)
760 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
761 	else
762 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
763 
764 	if (!boot_error) {
765 		/*
766 		 * allow APs to start initializing.
767 		 */
768 		pr_debug("Before Callout %d.\n", cpu);
769 		cpumask_set_cpu(cpu, cpu_callout_mask);
770 		pr_debug("After Callout %d.\n", cpu);
771 
772 		/*
773 		 * Wait 5s total for a response
774 		 */
775 		for (timeout = 0; timeout < 50000; timeout++) {
776 			if (cpumask_test_cpu(cpu, cpu_callin_mask))
777 				break;	/* It has booted */
778 			udelay(100);
779 		}
780 
781 		if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
782 			/* number CPUs logically, starting from 1 (BSP is 0) */
783 			pr_debug("OK.\n");
784 			printk(KERN_INFO "CPU%d: ", cpu);
785 			print_cpu_info(&cpu_data(cpu));
786 			pr_debug("CPU has booted.\n");
787 		} else {
788 			boot_error = 1;
789 			if (*((volatile unsigned char *)trampoline_base)
790 					== 0xA5)
791 				/* trampoline started but...? */
792 				printk(KERN_ERR "Stuck ??\n");
793 			else
794 				/* trampoline code not run */
795 				printk(KERN_ERR "Not responding.\n");
796 			if (apic->inquire_remote_apic)
797 				apic->inquire_remote_apic(apicid);
798 		}
799 	}
800 
801 	if (boot_error) {
802 		/* Try to put things back the way they were before ... */
803 		numa_remove_cpu(cpu); /* was set by numa_add_cpu */
804 
805 		/* was set by do_boot_cpu() */
806 		cpumask_clear_cpu(cpu, cpu_callout_mask);
807 
808 		/* was set by cpu_init() */
809 		cpumask_clear_cpu(cpu, cpu_initialized_mask);
810 
811 		set_cpu_present(cpu, false);
812 		per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
813 	}
814 
815 	/* mark "stuck" area as not stuck */
816 	*((volatile unsigned long *)trampoline_base) = 0;
817 
818 	/*
819 	 * Cleanup possible dangling ends...
820 	 */
821 	smpboot_restore_warm_reset_vector();
822 
823 	return boot_error;
824 }
825 
826 int __cpuinit native_cpu_up(unsigned int cpu)
827 {
828 	int apicid = apic->cpu_present_to_apicid(cpu);
829 	unsigned long flags;
830 	int err;
831 
832 	WARN_ON(irqs_disabled());
833 
834 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
835 
836 	if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
837 	    !physid_isset(apicid, phys_cpu_present_map)) {
838 		printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
839 		return -EINVAL;
840 	}
841 
842 	/*
843 	 * Already booted CPU?
844 	 */
845 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
846 		pr_debug("do_boot_cpu %d Already started\n", cpu);
847 		return -ENOSYS;
848 	}
849 
850 	/*
851 	 * Save current MTRR state in case it was changed since early boot
852 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
853 	 */
854 	mtrr_save_state();
855 
856 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
857 
858 #ifdef CONFIG_X86_32
859 	/* init low mem mapping */
860 	clone_pgd_range(swapper_pg_dir, swapper_pg_dir + KERNEL_PGD_BOUNDARY,
861 		min_t(unsigned long, KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
862 	flush_tlb_all();
863 	low_mappings = 1;
864 
865 	err = do_boot_cpu(apicid, cpu);
866 
867 	zap_low_mappings();
868 	low_mappings = 0;
869 #else
870 	err = do_boot_cpu(apicid, cpu);
871 #endif
872 	if (err) {
873 		pr_debug("do_boot_cpu failed %d\n", err);
874 		return -EIO;
875 	}
876 
877 	/*
878 	 * Check TSC synchronization with the AP (keep irqs disabled
879 	 * while doing so):
880 	 */
881 	local_irq_save(flags);
882 	check_tsc_sync_source(cpu);
883 	local_irq_restore(flags);
884 
885 	while (!cpu_online(cpu)) {
886 		cpu_relax();
887 		touch_nmi_watchdog();
888 	}
889 
890 	return 0;
891 }
892 
893 /*
894  * Fall back to non SMP mode after errors.
895  *
896  * RED-PEN audit/test this more. I bet there is more state messed up here.
897  */
898 static __init void disable_smp(void)
899 {
900 	/* use the read/write pointers to the present and possible maps */
901 	cpumask_copy(&cpu_present_map, cpumask_of(0));
902 	cpumask_copy(&cpu_possible_map, cpumask_of(0));
903 	smpboot_clear_io_apic_irqs();
904 
905 	if (smp_found_config)
906 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
907 	else
908 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
909 	map_cpu_to_logical_apicid();
910 	cpumask_set_cpu(0, cpu_sibling_mask(0));
911 	cpumask_set_cpu(0, cpu_core_mask(0));
912 }
913 
914 /*
915  * Various sanity checks.
916  */
917 static int __init smp_sanity_check(unsigned max_cpus)
918 {
919 	preempt_disable();
920 
921 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
922 	if (def_to_bigsmp && nr_cpu_ids > 8) {
923 		unsigned int cpu;
924 		unsigned nr;
925 
926 		printk(KERN_WARNING
927 		       "More than 8 CPUs detected - skipping them.\n"
928 		       "Use CONFIG_X86_BIGSMP.\n");
929 
930 		nr = 0;
931 		for_each_present_cpu(cpu) {
932 			if (nr >= 8)
933 				set_cpu_present(cpu, false);
934 			nr++;
935 		}
936 
937 		nr = 0;
938 		for_each_possible_cpu(cpu) {
939 			if (nr >= 8)
940 				set_cpu_possible(cpu, false);
941 			nr++;
942 		}
943 
944 		nr_cpu_ids = 8;
945 	}
946 #endif
947 
948 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
949 		printk(KERN_WARNING
950 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
951 			hard_smp_processor_id());
952 
953 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
954 	}
955 
956 	/*
957 	 * If we couldn't find an SMP configuration at boot time,
958 	 * get out of here now!
959 	 */
960 	if (!smp_found_config && !acpi_lapic) {
961 		preempt_enable();
962 		printk(KERN_NOTICE "SMP motherboard not detected.\n");
963 		disable_smp();
964 		if (APIC_init_uniprocessor())
965 			printk(KERN_NOTICE "Local APIC not detected."
966 					   " Using dummy APIC emulation.\n");
967 		return -1;
968 	}
969 
970 	/*
971 	 * Should not be necessary because the MP table should list the boot
972 	 * CPU too, but we do it for the sake of robustness anyway.
973 	 */
974 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
975 		printk(KERN_NOTICE
976 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
977 			boot_cpu_physical_apicid);
978 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
979 	}
980 	preempt_enable();
981 
982 	/*
983 	 * If we couldn't find a local APIC, then get out of here now!
984 	 */
985 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
986 	    !cpu_has_apic) {
987 		printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
988 			boot_cpu_physical_apicid);
989 		printk(KERN_ERR "... forcing use of dummy APIC emulation."
990 				"(tell your hw vendor)\n");
991 		smpboot_clear_io_apic();
992 		arch_disable_smp_support();
993 		return -1;
994 	}
995 
996 	verify_local_APIC();
997 
998 	/*
999 	 * If SMP should be disabled, then really disable it!
1000 	 */
1001 	if (!max_cpus) {
1002 		printk(KERN_INFO "SMP mode deactivated.\n");
1003 		smpboot_clear_io_apic();
1004 
1005 		localise_nmi_watchdog();
1006 
1007 		connect_bsp_APIC();
1008 		setup_local_APIC();
1009 		end_local_APIC_setup();
1010 		return -1;
1011 	}
1012 
1013 	return 0;
1014 }
1015 
1016 static void __init smp_cpu_index_default(void)
1017 {
1018 	int i;
1019 	struct cpuinfo_x86 *c;
1020 
1021 	for_each_possible_cpu(i) {
1022 		c = &cpu_data(i);
1023 		/* mark all to hotplug */
1024 		c->cpu_index = nr_cpu_ids;
1025 	}
1026 }
1027 
1028 /*
1029  * Prepare for SMP bootup.  The MP table or ACPI has been read
1030  * earlier.  Just do some sanity checking here and enable APIC mode.
1031  */
1032 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1033 {
1034 	preempt_disable();
1035 	smp_cpu_index_default();
1036 	current_cpu_data = boot_cpu_data;
1037 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1038 	mb();
1039 	/*
1040 	 * Setup boot CPU information
1041 	 */
1042 	smp_store_cpu_info(0); /* Final full version of the data */
1043 #ifdef CONFIG_X86_32
1044 	boot_cpu_logical_apicid = logical_smp_processor_id();
1045 #endif
1046 	current_thread_info()->cpu = 0;  /* needed? */
1047 	set_cpu_sibling_map(0);
1048 
1049 	enable_IR_x2apic();
1050 #ifdef CONFIG_X86_64
1051 	default_setup_apic_routing();
1052 #endif
1053 
1054 	if (smp_sanity_check(max_cpus) < 0) {
1055 		printk(KERN_INFO "SMP disabled\n");
1056 		disable_smp();
1057 		goto out;
1058 	}
1059 
1060 	preempt_disable();
1061 	if (read_apic_id() != boot_cpu_physical_apicid) {
1062 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1063 		     read_apic_id(), boot_cpu_physical_apicid);
1064 		/* Or can we switch back to PIC here? */
1065 	}
1066 	preempt_enable();
1067 
1068 	connect_bsp_APIC();
1069 
1070 	/*
1071 	 * Switch from PIC to APIC mode.
1072 	 */
1073 	setup_local_APIC();
1074 
1075 	/*
1076 	 * Enable IO APIC before setting up error vector
1077 	 */
1078 	if (!skip_ioapic_setup && nr_ioapics)
1079 		enable_IO_APIC();
1080 
1081 	end_local_APIC_setup();
1082 
1083 	map_cpu_to_logical_apicid();
1084 
1085 	if (apic->setup_portio_remap)
1086 		apic->setup_portio_remap();
1087 
1088 	smpboot_setup_io_apic();
1089 	/*
1090 	 * Set up local APIC timer on boot CPU.
1091 	 */
1092 
1093 	printk(KERN_INFO "CPU%d: ", 0);
1094 	print_cpu_info(&cpu_data(0));
1095 	setup_boot_clock();
1096 
1097 	if (is_uv_system())
1098 		uv_system_init();
1099 out:
1100 	preempt_enable();
1101 }
1102 /*
1103  * Early setup to make printk work.
1104  */
1105 void __init native_smp_prepare_boot_cpu(void)
1106 {
1107 	int me = smp_processor_id();
1108 	switch_to_new_gdt(me);
1109 	/* already set me in cpu_online_mask in boot_cpu_init() */
1110 	cpumask_set_cpu(me, cpu_callout_mask);
1111 	per_cpu(cpu_state, me) = CPU_ONLINE;
1112 }
1113 
1114 void __init native_smp_cpus_done(unsigned int max_cpus)
1115 {
1116 	pr_debug("Boot done.\n");
1117 
1118 	impress_friends();
1119 #ifdef CONFIG_X86_IO_APIC
1120 	setup_ioapic_dest();
1121 #endif
1122 	check_nmi_watchdog();
1123 }
1124 
1125 static int __initdata setup_possible_cpus = -1;
1126 static int __init _setup_possible_cpus(char *str)
1127 {
1128 	get_option(&str, &setup_possible_cpus);
1129 	return 0;
1130 }
1131 early_param("possible_cpus", _setup_possible_cpus);
1132 
1133 
1134 /*
1135  * cpu_possible_map should be static, it cannot change as cpu's
1136  * are onlined, or offlined. The reason is per-cpu data-structures
1137  * are allocated by some modules at init time, and dont expect to
1138  * do this dynamically on cpu arrival/departure.
1139  * cpu_present_map on the other hand can change dynamically.
1140  * In case when cpu_hotplug is not compiled, then we resort to current
1141  * behaviour, which is cpu_possible == cpu_present.
1142  * - Ashok Raj
1143  *
1144  * Three ways to find out the number of additional hotplug CPUs:
1145  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1146  * - The user can overwrite it with possible_cpus=NUM
1147  * - Otherwise don't reserve additional CPUs.
1148  * We do this because additional CPUs waste a lot of memory.
1149  * -AK
1150  */
1151 __init void prefill_possible_map(void)
1152 {
1153 	int i, possible;
1154 
1155 	/* no processor from mptable or madt */
1156 	if (!num_processors)
1157 		num_processors = 1;
1158 
1159 	if (setup_possible_cpus == -1)
1160 		possible = num_processors + disabled_cpus;
1161 	else
1162 		possible = setup_possible_cpus;
1163 
1164 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1165 
1166 	if (possible > CONFIG_NR_CPUS) {
1167 		printk(KERN_WARNING
1168 			"%d Processors exceeds NR_CPUS limit of %d\n",
1169 			possible, CONFIG_NR_CPUS);
1170 		possible = CONFIG_NR_CPUS;
1171 	}
1172 
1173 	printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1174 		possible, max_t(int, possible - num_processors, 0));
1175 
1176 	for (i = 0; i < possible; i++)
1177 		set_cpu_possible(i, true);
1178 
1179 	nr_cpu_ids = possible;
1180 }
1181 
1182 #ifdef CONFIG_HOTPLUG_CPU
1183 
1184 static void remove_siblinginfo(int cpu)
1185 {
1186 	int sibling;
1187 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1188 
1189 	for_each_cpu(sibling, cpu_core_mask(cpu)) {
1190 		cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1191 		/*/
1192 		 * last thread sibling in this cpu core going down
1193 		 */
1194 		if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1195 			cpu_data(sibling).booted_cores--;
1196 	}
1197 
1198 	for_each_cpu(sibling, cpu_sibling_mask(cpu))
1199 		cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1200 	cpumask_clear(cpu_sibling_mask(cpu));
1201 	cpumask_clear(cpu_core_mask(cpu));
1202 	c->phys_proc_id = 0;
1203 	c->cpu_core_id = 0;
1204 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1205 }
1206 
1207 static void __ref remove_cpu_from_maps(int cpu)
1208 {
1209 	set_cpu_online(cpu, false);
1210 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1211 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1212 	/* was set by cpu_init() */
1213 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1214 	numa_remove_cpu(cpu);
1215 }
1216 
1217 void cpu_disable_common(void)
1218 {
1219 	int cpu = smp_processor_id();
1220 	/*
1221 	 * HACK:
1222 	 * Allow any queued timer interrupts to get serviced
1223 	 * This is only a temporary solution until we cleanup
1224 	 * fixup_irqs as we do for IA64.
1225 	 */
1226 	local_irq_enable();
1227 	mdelay(1);
1228 
1229 	local_irq_disable();
1230 	remove_siblinginfo(cpu);
1231 
1232 	/* It's now safe to remove this processor from the online map */
1233 	lock_vector_lock();
1234 	remove_cpu_from_maps(cpu);
1235 	unlock_vector_lock();
1236 	fixup_irqs();
1237 }
1238 
1239 int native_cpu_disable(void)
1240 {
1241 	int cpu = smp_processor_id();
1242 
1243 	/*
1244 	 * Perhaps use cpufreq to drop frequency, but that could go
1245 	 * into generic code.
1246 	 *
1247 	 * We won't take down the boot processor on i386 due to some
1248 	 * interrupts only being able to be serviced by the BSP.
1249 	 * Especially so if we're not using an IOAPIC	-zwane
1250 	 */
1251 	if (cpu == 0)
1252 		return -EBUSY;
1253 
1254 	if (nmi_watchdog == NMI_LOCAL_APIC)
1255 		stop_apic_nmi_watchdog(NULL);
1256 	clear_local_APIC();
1257 
1258 	cpu_disable_common();
1259 	return 0;
1260 }
1261 
1262 void native_cpu_die(unsigned int cpu)
1263 {
1264 	/* We don't do anything here: idle task is faking death itself. */
1265 	unsigned int i;
1266 
1267 	for (i = 0; i < 10; i++) {
1268 		/* They ack this in play_dead by setting CPU_DEAD */
1269 		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1270 			printk(KERN_INFO "CPU %d is now offline\n", cpu);
1271 			if (1 == num_online_cpus())
1272 				alternatives_smp_switch(0);
1273 			return;
1274 		}
1275 		msleep(100);
1276 	}
1277 	printk(KERN_ERR "CPU %u didn't die...\n", cpu);
1278 }
1279 
1280 void play_dead_common(void)
1281 {
1282 	idle_task_exit();
1283 	reset_lazy_tlbstate();
1284 	irq_ctx_exit(raw_smp_processor_id());
1285 	c1e_remove_cpu(raw_smp_processor_id());
1286 
1287 	mb();
1288 	/* Ack it */
1289 	__get_cpu_var(cpu_state) = CPU_DEAD;
1290 
1291 	/*
1292 	 * With physical CPU hotplug, we should halt the cpu
1293 	 */
1294 	local_irq_disable();
1295 }
1296 
1297 void native_play_dead(void)
1298 {
1299 	play_dead_common();
1300 	wbinvd_halt();
1301 }
1302 
1303 #else /* ... !CONFIG_HOTPLUG_CPU */
1304 int native_cpu_disable(void)
1305 {
1306 	return -ENOSYS;
1307 }
1308 
1309 void native_cpu_die(unsigned int cpu)
1310 {
1311 	/* We said "no" in __cpu_disable */
1312 	BUG();
1313 }
1314 
1315 void native_play_dead(void)
1316 {
1317 	BUG();
1318 }
1319 
1320 #endif
1321