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 #include <linux/tboot.h>
51 #include <linux/stackprotector.h>
52 #include <linux/gfp.h>
53 #include <linux/cpuidle.h>
54 
55 #include <asm/acpi.h>
56 #include <asm/desc.h>
57 #include <asm/nmi.h>
58 #include <asm/irq.h>
59 #include <asm/idle.h>
60 #include <asm/trampoline.h>
61 #include <asm/cpu.h>
62 #include <asm/numa.h>
63 #include <asm/pgtable.h>
64 #include <asm/tlbflush.h>
65 #include <asm/mtrr.h>
66 #include <asm/mwait.h>
67 #include <asm/apic.h>
68 #include <asm/io_apic.h>
69 #include <asm/i387.h>
70 #include <asm/fpu-internal.h>
71 #include <asm/setup.h>
72 #include <asm/uv/uv.h>
73 #include <linux/mc146818rtc.h>
74 
75 #include <asm/smpboot_hooks.h>
76 #include <asm/i8259.h>
77 
78 /* State of each CPU */
79 DEFINE_PER_CPU(int, cpu_state) = { 0 };
80 
81 /* Store all idle threads, this can be reused instead of creating
82 * a new thread. Also avoids complicated thread destroy functionality
83 * for idle threads.
84 */
85 #ifdef CONFIG_HOTPLUG_CPU
86 /*
87  * Needed only for CONFIG_HOTPLUG_CPU because __cpuinitdata is
88  * removed after init for !CONFIG_HOTPLUG_CPU.
89  */
90 static DEFINE_PER_CPU(struct task_struct *, idle_thread_array);
91 #define get_idle_for_cpu(x)      (per_cpu(idle_thread_array, x))
92 #define set_idle_for_cpu(x, p)   (per_cpu(idle_thread_array, x) = (p))
93 
94 /*
95  * We need this for trampoline_base protection from concurrent accesses when
96  * off- and onlining cores wildly.
97  */
98 static DEFINE_MUTEX(x86_cpu_hotplug_driver_mutex);
99 
cpu_hotplug_driver_lock(void)100 void cpu_hotplug_driver_lock(void)
101 {
102         mutex_lock(&x86_cpu_hotplug_driver_mutex);
103 }
104 
cpu_hotplug_driver_unlock(void)105 void cpu_hotplug_driver_unlock(void)
106 {
107         mutex_unlock(&x86_cpu_hotplug_driver_mutex);
108 }
109 
arch_cpu_probe(const char * buf,size_t count)110 ssize_t arch_cpu_probe(const char *buf, size_t count) { return -1; }
arch_cpu_release(const char * buf,size_t count)111 ssize_t arch_cpu_release(const char *buf, size_t count) { return -1; }
112 #else
113 static struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
114 #define get_idle_for_cpu(x)      (idle_thread_array[(x)])
115 #define set_idle_for_cpu(x, p)   (idle_thread_array[(x)] = (p))
116 #endif
117 
118 /* Number of siblings per CPU package */
119 int smp_num_siblings = 1;
120 EXPORT_SYMBOL(smp_num_siblings);
121 
122 /* Last level cache ID of each logical CPU */
123 DEFINE_PER_CPU(u16, cpu_llc_id) = BAD_APICID;
124 
125 /* representing HT siblings of each logical CPU */
126 DEFINE_PER_CPU(cpumask_var_t, cpu_sibling_map);
127 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
128 
129 /* representing HT and core siblings of each logical CPU */
130 DEFINE_PER_CPU(cpumask_var_t, cpu_core_map);
131 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
132 
133 DEFINE_PER_CPU(cpumask_var_t, cpu_llc_shared_map);
134 
135 /* Per CPU bogomips and other parameters */
136 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
137 EXPORT_PER_CPU_SYMBOL(cpu_info);
138 
139 atomic_t init_deasserted;
140 
141 /*
142  * Report back to the Boot Processor.
143  * Running on AP.
144  */
smp_callin(void)145 static void __cpuinit smp_callin(void)
146 {
147 	int cpuid, phys_id;
148 	unsigned long timeout;
149 
150 	/*
151 	 * If waken up by an INIT in an 82489DX configuration
152 	 * we may get here before an INIT-deassert IPI reaches
153 	 * our local APIC.  We have to wait for the IPI or we'll
154 	 * lock up on an APIC access.
155 	 */
156 	if (apic->wait_for_init_deassert)
157 		apic->wait_for_init_deassert(&init_deasserted);
158 
159 	/*
160 	 * (This works even if the APIC is not enabled.)
161 	 */
162 	phys_id = read_apic_id();
163 	cpuid = smp_processor_id();
164 	if (cpumask_test_cpu(cpuid, cpu_callin_mask)) {
165 		panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
166 					phys_id, cpuid);
167 	}
168 	pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
169 
170 	/*
171 	 * STARTUP IPIs are fragile beasts as they might sometimes
172 	 * trigger some glue motherboard logic. Complete APIC bus
173 	 * silence for 1 second, this overestimates the time the
174 	 * boot CPU is spending to send the up to 2 STARTUP IPIs
175 	 * by a factor of two. This should be enough.
176 	 */
177 
178 	/*
179 	 * Waiting 2s total for startup (udelay is not yet working)
180 	 */
181 	timeout = jiffies + 2*HZ;
182 	while (time_before(jiffies, timeout)) {
183 		/*
184 		 * Has the boot CPU finished it's STARTUP sequence?
185 		 */
186 		if (cpumask_test_cpu(cpuid, cpu_callout_mask))
187 			break;
188 		cpu_relax();
189 	}
190 
191 	if (!time_before(jiffies, timeout)) {
192 		panic("%s: CPU%d started up but did not get a callout!\n",
193 		      __func__, cpuid);
194 	}
195 
196 	/*
197 	 * the boot CPU has finished the init stage and is spinning
198 	 * on callin_map until we finish. We are free to set up this
199 	 * CPU, first the APIC. (this is probably redundant on most
200 	 * boards)
201 	 */
202 
203 	pr_debug("CALLIN, before setup_local_APIC().\n");
204 	if (apic->smp_callin_clear_local_apic)
205 		apic->smp_callin_clear_local_apic();
206 	setup_local_APIC();
207 	end_local_APIC_setup();
208 
209 	/*
210 	 * Need to setup vector mappings before we enable interrupts.
211 	 */
212 	setup_vector_irq(smp_processor_id());
213 
214 	/*
215 	 * Save our processor parameters. Note: this information
216 	 * is needed for clock calibration.
217 	 */
218 	smp_store_cpu_info(cpuid);
219 
220 	/*
221 	 * Get our bogomips.
222 	 * Update loops_per_jiffy in cpu_data. Previous call to
223 	 * smp_store_cpu_info() stored a value that is close but not as
224 	 * accurate as the value just calculated.
225 	 */
226 	calibrate_delay();
227 	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
228 	pr_debug("Stack at about %p\n", &cpuid);
229 
230 	/*
231 	 * This must be done before setting cpu_online_mask
232 	 * or calling notify_cpu_starting.
233 	 */
234 	set_cpu_sibling_map(raw_smp_processor_id());
235 	wmb();
236 
237 	notify_cpu_starting(cpuid);
238 
239 	/*
240 	 * Allow the master to continue.
241 	 */
242 	cpumask_set_cpu(cpuid, cpu_callin_mask);
243 }
244 
245 /*
246  * Activate a secondary processor.
247  */
start_secondary(void * unused)248 notrace static void __cpuinit start_secondary(void *unused)
249 {
250 	/*
251 	 * Don't put *anything* before cpu_init(), SMP booting is too
252 	 * fragile that we want to limit the things done here to the
253 	 * most necessary things.
254 	 */
255 	cpu_init();
256 	x86_cpuinit.early_percpu_clock_init();
257 	preempt_disable();
258 	smp_callin();
259 
260 #ifdef CONFIG_X86_32
261 	/* switch away from the initial page table */
262 	load_cr3(swapper_pg_dir);
263 	__flush_tlb_all();
264 #endif
265 
266 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
267 	barrier();
268 	/*
269 	 * Check TSC synchronization with the BP:
270 	 */
271 	check_tsc_sync_target();
272 
273 	/*
274 	 * We need to hold call_lock, so there is no inconsistency
275 	 * between the time smp_call_function() determines number of
276 	 * IPI recipients, and the time when the determination is made
277 	 * for which cpus receive the IPI. Holding this
278 	 * lock helps us to not include this cpu in a currently in progress
279 	 * smp_call_function().
280 	 *
281 	 * We need to hold vector_lock so there the set of online cpus
282 	 * does not change while we are assigning vectors to cpus.  Holding
283 	 * this lock ensures we don't half assign or remove an irq from a cpu.
284 	 */
285 	ipi_call_lock();
286 	lock_vector_lock();
287 	set_cpu_online(smp_processor_id(), true);
288 	unlock_vector_lock();
289 	ipi_call_unlock();
290 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
291 	x86_platform.nmi_init();
292 
293 	/* enable local interrupts */
294 	local_irq_enable();
295 
296 	/* to prevent fake stack check failure in clock setup */
297 	boot_init_stack_canary();
298 
299 	x86_cpuinit.setup_percpu_clockev();
300 
301 	wmb();
302 	cpu_idle();
303 }
304 
305 /*
306  * The bootstrap kernel entry code has set these up. Save them for
307  * a given CPU
308  */
309 
smp_store_cpu_info(int id)310 void __cpuinit smp_store_cpu_info(int id)
311 {
312 	struct cpuinfo_x86 *c = &cpu_data(id);
313 
314 	*c = boot_cpu_data;
315 	c->cpu_index = id;
316 	if (id != 0)
317 		identify_secondary_cpu(c);
318 }
319 
link_thread_siblings(int cpu1,int cpu2)320 static void __cpuinit link_thread_siblings(int cpu1, int cpu2)
321 {
322 	cpumask_set_cpu(cpu1, cpu_sibling_mask(cpu2));
323 	cpumask_set_cpu(cpu2, cpu_sibling_mask(cpu1));
324 	cpumask_set_cpu(cpu1, cpu_core_mask(cpu2));
325 	cpumask_set_cpu(cpu2, cpu_core_mask(cpu1));
326 	cpumask_set_cpu(cpu1, cpu_llc_shared_mask(cpu2));
327 	cpumask_set_cpu(cpu2, cpu_llc_shared_mask(cpu1));
328 }
329 
330 
set_cpu_sibling_map(int cpu)331 void __cpuinit set_cpu_sibling_map(int cpu)
332 {
333 	int i;
334 	struct cpuinfo_x86 *c = &cpu_data(cpu);
335 
336 	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
337 
338 	if (smp_num_siblings > 1) {
339 		for_each_cpu(i, cpu_sibling_setup_mask) {
340 			struct cpuinfo_x86 *o = &cpu_data(i);
341 
342 			if (cpu_has(c, X86_FEATURE_TOPOEXT)) {
343 				if (c->phys_proc_id == o->phys_proc_id &&
344 				    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i) &&
345 				    c->compute_unit_id == o->compute_unit_id)
346 					link_thread_siblings(cpu, i);
347 			} else if (c->phys_proc_id == o->phys_proc_id &&
348 				   c->cpu_core_id == o->cpu_core_id) {
349 				link_thread_siblings(cpu, i);
350 			}
351 		}
352 	} else {
353 		cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
354 	}
355 
356 	cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
357 
358 	if (__this_cpu_read(cpu_info.x86_max_cores) == 1) {
359 		cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
360 		c->booted_cores = 1;
361 		return;
362 	}
363 
364 	for_each_cpu(i, cpu_sibling_setup_mask) {
365 		if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
366 		    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
367 			cpumask_set_cpu(i, cpu_llc_shared_mask(cpu));
368 			cpumask_set_cpu(cpu, cpu_llc_shared_mask(i));
369 		}
370 		if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
371 			cpumask_set_cpu(i, cpu_core_mask(cpu));
372 			cpumask_set_cpu(cpu, cpu_core_mask(i));
373 			/*
374 			 *  Does this new cpu bringup a new core?
375 			 */
376 			if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
377 				/*
378 				 * for each core in package, increment
379 				 * the booted_cores for this new cpu
380 				 */
381 				if (cpumask_first(cpu_sibling_mask(i)) == i)
382 					c->booted_cores++;
383 				/*
384 				 * increment the core count for all
385 				 * the other cpus in this package
386 				 */
387 				if (i != cpu)
388 					cpu_data(i).booted_cores++;
389 			} else if (i != cpu && !c->booted_cores)
390 				c->booted_cores = cpu_data(i).booted_cores;
391 		}
392 	}
393 }
394 
395 /* maps the cpu to the sched domain representing multi-core */
cpu_coregroup_mask(int cpu)396 const struct cpumask *cpu_coregroup_mask(int cpu)
397 {
398 	struct cpuinfo_x86 *c = &cpu_data(cpu);
399 	/*
400 	 * For perf, we return last level cache shared map.
401 	 * And for power savings, we return cpu_core_map
402 	 */
403 	if ((sched_mc_power_savings || sched_smt_power_savings) &&
404 	    !(cpu_has(c, X86_FEATURE_AMD_DCM)))
405 		return cpu_core_mask(cpu);
406 	else
407 		return cpu_llc_shared_mask(cpu);
408 }
409 
impress_friends(void)410 static void impress_friends(void)
411 {
412 	int cpu;
413 	unsigned long bogosum = 0;
414 	/*
415 	 * Allow the user to impress friends.
416 	 */
417 	pr_debug("Before bogomips.\n");
418 	for_each_possible_cpu(cpu)
419 		if (cpumask_test_cpu(cpu, cpu_callout_mask))
420 			bogosum += cpu_data(cpu).loops_per_jiffy;
421 	printk(KERN_INFO
422 		"Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
423 		num_online_cpus(),
424 		bogosum/(500000/HZ),
425 		(bogosum/(5000/HZ))%100);
426 
427 	pr_debug("Before bogocount - setting activated=1.\n");
428 }
429 
__inquire_remote_apic(int apicid)430 void __inquire_remote_apic(int apicid)
431 {
432 	unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
433 	const char * const names[] = { "ID", "VERSION", "SPIV" };
434 	int timeout;
435 	u32 status;
436 
437 	printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
438 
439 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
440 		printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
441 
442 		/*
443 		 * Wait for idle.
444 		 */
445 		status = safe_apic_wait_icr_idle();
446 		if (status)
447 			printk(KERN_CONT
448 			       "a previous APIC delivery may have failed\n");
449 
450 		apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
451 
452 		timeout = 0;
453 		do {
454 			udelay(100);
455 			status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
456 		} while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
457 
458 		switch (status) {
459 		case APIC_ICR_RR_VALID:
460 			status = apic_read(APIC_RRR);
461 			printk(KERN_CONT "%08x\n", status);
462 			break;
463 		default:
464 			printk(KERN_CONT "failed\n");
465 		}
466 	}
467 }
468 
469 /*
470  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
471  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
472  * won't ... remember to clear down the APIC, etc later.
473  */
474 int __cpuinit
wakeup_secondary_cpu_via_nmi(int logical_apicid,unsigned long start_eip)475 wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
476 {
477 	unsigned long send_status, accept_status = 0;
478 	int maxlvt;
479 
480 	/* Target chip */
481 	/* Boot on the stack */
482 	/* Kick the second */
483 	apic_icr_write(APIC_DM_NMI | apic->dest_logical, logical_apicid);
484 
485 	pr_debug("Waiting for send to finish...\n");
486 	send_status = safe_apic_wait_icr_idle();
487 
488 	/*
489 	 * Give the other CPU some time to accept the IPI.
490 	 */
491 	udelay(200);
492 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
493 		maxlvt = lapic_get_maxlvt();
494 		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
495 			apic_write(APIC_ESR, 0);
496 		accept_status = (apic_read(APIC_ESR) & 0xEF);
497 	}
498 	pr_debug("NMI sent.\n");
499 
500 	if (send_status)
501 		printk(KERN_ERR "APIC never delivered???\n");
502 	if (accept_status)
503 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
504 
505 	return (send_status | accept_status);
506 }
507 
508 static int __cpuinit
wakeup_secondary_cpu_via_init(int phys_apicid,unsigned long start_eip)509 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
510 {
511 	unsigned long send_status, accept_status = 0;
512 	int maxlvt, num_starts, j;
513 
514 	maxlvt = lapic_get_maxlvt();
515 
516 	/*
517 	 * Be paranoid about clearing APIC errors.
518 	 */
519 	if (APIC_INTEGRATED(apic_version[phys_apicid])) {
520 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
521 			apic_write(APIC_ESR, 0);
522 		apic_read(APIC_ESR);
523 	}
524 
525 	pr_debug("Asserting INIT.\n");
526 
527 	/*
528 	 * Turn INIT on target chip
529 	 */
530 	/*
531 	 * Send IPI
532 	 */
533 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
534 		       phys_apicid);
535 
536 	pr_debug("Waiting for send to finish...\n");
537 	send_status = safe_apic_wait_icr_idle();
538 
539 	mdelay(10);
540 
541 	pr_debug("Deasserting INIT.\n");
542 
543 	/* Target chip */
544 	/* Send IPI */
545 	apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
546 
547 	pr_debug("Waiting for send to finish...\n");
548 	send_status = safe_apic_wait_icr_idle();
549 
550 	mb();
551 	atomic_set(&init_deasserted, 1);
552 
553 	/*
554 	 * Should we send STARTUP IPIs ?
555 	 *
556 	 * Determine this based on the APIC version.
557 	 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
558 	 */
559 	if (APIC_INTEGRATED(apic_version[phys_apicid]))
560 		num_starts = 2;
561 	else
562 		num_starts = 0;
563 
564 	/*
565 	 * Paravirt / VMI wants a startup IPI hook here to set up the
566 	 * target processor state.
567 	 */
568 	startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
569 			 stack_start);
570 
571 	/*
572 	 * Run STARTUP IPI loop.
573 	 */
574 	pr_debug("#startup loops: %d.\n", num_starts);
575 
576 	for (j = 1; j <= num_starts; j++) {
577 		pr_debug("Sending STARTUP #%d.\n", j);
578 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
579 			apic_write(APIC_ESR, 0);
580 		apic_read(APIC_ESR);
581 		pr_debug("After apic_write.\n");
582 
583 		/*
584 		 * STARTUP IPI
585 		 */
586 
587 		/* Target chip */
588 		/* Boot on the stack */
589 		/* Kick the second */
590 		apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
591 			       phys_apicid);
592 
593 		/*
594 		 * Give the other CPU some time to accept the IPI.
595 		 */
596 		udelay(300);
597 
598 		pr_debug("Startup point 1.\n");
599 
600 		pr_debug("Waiting for send to finish...\n");
601 		send_status = safe_apic_wait_icr_idle();
602 
603 		/*
604 		 * Give the other CPU some time to accept the IPI.
605 		 */
606 		udelay(200);
607 		if (maxlvt > 3)		/* Due to the Pentium erratum 3AP.  */
608 			apic_write(APIC_ESR, 0);
609 		accept_status = (apic_read(APIC_ESR) & 0xEF);
610 		if (send_status || accept_status)
611 			break;
612 	}
613 	pr_debug("After Startup.\n");
614 
615 	if (send_status)
616 		printk(KERN_ERR "APIC never delivered???\n");
617 	if (accept_status)
618 		printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
619 
620 	return (send_status | accept_status);
621 }
622 
623 struct create_idle {
624 	struct work_struct work;
625 	struct task_struct *idle;
626 	struct completion done;
627 	int cpu;
628 };
629 
do_fork_idle(struct work_struct * work)630 static void __cpuinit do_fork_idle(struct work_struct *work)
631 {
632 	struct create_idle *c_idle =
633 		container_of(work, struct create_idle, work);
634 
635 	c_idle->idle = fork_idle(c_idle->cpu);
636 	complete(&c_idle->done);
637 }
638 
639 /* reduce the number of lines printed when booting a large cpu count system */
announce_cpu(int cpu,int apicid)640 static void __cpuinit announce_cpu(int cpu, int apicid)
641 {
642 	static int current_node = -1;
643 	int node = early_cpu_to_node(cpu);
644 
645 	if (system_state == SYSTEM_BOOTING) {
646 		if (node != current_node) {
647 			if (current_node > (-1))
648 				pr_cont(" Ok.\n");
649 			current_node = node;
650 			pr_info("Booting Node %3d, Processors ", node);
651 		}
652 		pr_cont(" #%d%s", cpu, cpu == (nr_cpu_ids - 1) ? " Ok.\n" : "");
653 		return;
654 	} else
655 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
656 			node, cpu, apicid);
657 }
658 
659 /*
660  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
661  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
662  * Returns zero if CPU booted OK, else error code from
663  * ->wakeup_secondary_cpu.
664  */
do_boot_cpu(int apicid,int cpu)665 static int __cpuinit do_boot_cpu(int apicid, int cpu)
666 {
667 	unsigned long boot_error = 0;
668 	unsigned long start_ip;
669 	int timeout;
670 	struct create_idle c_idle = {
671 		.cpu	= cpu,
672 		.done	= COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
673 	};
674 
675 	INIT_WORK_ONSTACK(&c_idle.work, do_fork_idle);
676 
677 	alternatives_smp_switch(1);
678 
679 	c_idle.idle = get_idle_for_cpu(cpu);
680 
681 	/*
682 	 * We can't use kernel_thread since we must avoid to
683 	 * reschedule the child.
684 	 */
685 	if (c_idle.idle) {
686 		c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
687 			(THREAD_SIZE +  task_stack_page(c_idle.idle))) - 1);
688 		init_idle(c_idle.idle, cpu);
689 		goto do_rest;
690 	}
691 
692 	schedule_work(&c_idle.work);
693 	wait_for_completion(&c_idle.done);
694 
695 	if (IS_ERR(c_idle.idle)) {
696 		printk("failed fork for CPU %d\n", cpu);
697 		destroy_work_on_stack(&c_idle.work);
698 		return PTR_ERR(c_idle.idle);
699 	}
700 
701 	set_idle_for_cpu(cpu, c_idle.idle);
702 do_rest:
703 	per_cpu(current_task, cpu) = c_idle.idle;
704 #ifdef CONFIG_X86_32
705 	/* Stack for startup_32 can be just as for start_secondary onwards */
706 	irq_ctx_init(cpu);
707 #else
708 	clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
709 	initial_gs = per_cpu_offset(cpu);
710 	per_cpu(kernel_stack, cpu) =
711 		(unsigned long)task_stack_page(c_idle.idle) -
712 		KERNEL_STACK_OFFSET + THREAD_SIZE;
713 #endif
714 	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
715 	initial_code = (unsigned long)start_secondary;
716 	stack_start  = c_idle.idle->thread.sp;
717 
718 	/* start_ip had better be page-aligned! */
719 	start_ip = trampoline_address();
720 
721 	/* So we see what's up */
722 	announce_cpu(cpu, apicid);
723 
724 	/*
725 	 * This grunge runs the startup process for
726 	 * the targeted processor.
727 	 */
728 
729 	atomic_set(&init_deasserted, 0);
730 
731 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
732 
733 		pr_debug("Setting warm reset code and vector.\n");
734 
735 		smpboot_setup_warm_reset_vector(start_ip);
736 		/*
737 		 * Be paranoid about clearing APIC errors.
738 		*/
739 		if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
740 			apic_write(APIC_ESR, 0);
741 			apic_read(APIC_ESR);
742 		}
743 	}
744 
745 	/*
746 	 * Kick the secondary CPU. Use the method in the APIC driver
747 	 * if it's defined - or use an INIT boot APIC message otherwise:
748 	 */
749 	if (apic->wakeup_secondary_cpu)
750 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
751 	else
752 		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
753 
754 	if (!boot_error) {
755 		/*
756 		 * allow APs to start initializing.
757 		 */
758 		pr_debug("Before Callout %d.\n", cpu);
759 		cpumask_set_cpu(cpu, cpu_callout_mask);
760 		pr_debug("After Callout %d.\n", cpu);
761 
762 		/*
763 		 * Wait 5s total for a response
764 		 */
765 		for (timeout = 0; timeout < 50000; timeout++) {
766 			if (cpumask_test_cpu(cpu, cpu_callin_mask))
767 				break;	/* It has booted */
768 			udelay(100);
769 			/*
770 			 * Allow other tasks to run while we wait for the
771 			 * AP to come online. This also gives a chance
772 			 * for the MTRR work(triggered by the AP coming online)
773 			 * to be completed in the stop machine context.
774 			 */
775 			schedule();
776 		}
777 
778 		if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
779 			print_cpu_msr(&cpu_data(cpu));
780 			pr_debug("CPU%d: has booted.\n", cpu);
781 		} else {
782 			boot_error = 1;
783 			if (*(volatile u32 *)TRAMPOLINE_SYM(trampoline_status)
784 			    == 0xA5A5A5A5)
785 				/* trampoline started but...? */
786 				pr_err("CPU%d: Stuck ??\n", cpu);
787 			else
788 				/* trampoline code not run */
789 				pr_err("CPU%d: Not responding.\n", cpu);
790 			if (apic->inquire_remote_apic)
791 				apic->inquire_remote_apic(apicid);
792 		}
793 	}
794 
795 	if (boot_error) {
796 		/* Try to put things back the way they were before ... */
797 		numa_remove_cpu(cpu); /* was set by numa_add_cpu */
798 
799 		/* was set by do_boot_cpu() */
800 		cpumask_clear_cpu(cpu, cpu_callout_mask);
801 
802 		/* was set by cpu_init() */
803 		cpumask_clear_cpu(cpu, cpu_initialized_mask);
804 
805 		set_cpu_present(cpu, false);
806 		per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
807 	}
808 
809 	/* mark "stuck" area as not stuck */
810 	*(volatile u32 *)TRAMPOLINE_SYM(trampoline_status) = 0;
811 
812 	if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
813 		/*
814 		 * Cleanup possible dangling ends...
815 		 */
816 		smpboot_restore_warm_reset_vector();
817 	}
818 
819 	destroy_work_on_stack(&c_idle.work);
820 	return boot_error;
821 }
822 
native_cpu_up(unsigned int cpu)823 int __cpuinit native_cpu_up(unsigned int cpu)
824 {
825 	int apicid = apic->cpu_present_to_apicid(cpu);
826 	unsigned long flags;
827 	int err;
828 
829 	WARN_ON(irqs_disabled());
830 
831 	pr_debug("++++++++++++++++++++=_---CPU UP  %u\n", cpu);
832 
833 	if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
834 	    !physid_isset(apicid, phys_cpu_present_map) ||
835 	    !apic->apic_id_valid(apicid)) {
836 		printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
837 		return -EINVAL;
838 	}
839 
840 	/*
841 	 * Already booted CPU?
842 	 */
843 	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
844 		pr_debug("do_boot_cpu %d Already started\n", cpu);
845 		return -ENOSYS;
846 	}
847 
848 	/*
849 	 * Save current MTRR state in case it was changed since early boot
850 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
851 	 */
852 	mtrr_save_state();
853 
854 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
855 
856 	/* the FPU context is blank, nobody can own it */
857 	__cpu_disable_lazy_restore(cpu);
858 
859 	err = do_boot_cpu(apicid, cpu);
860 	if (err) {
861 		pr_debug("do_boot_cpu failed %d\n", err);
862 		return -EIO;
863 	}
864 
865 	/*
866 	 * Check TSC synchronization with the AP (keep irqs disabled
867 	 * while doing so):
868 	 */
869 	local_irq_save(flags);
870 	check_tsc_sync_source(cpu);
871 	local_irq_restore(flags);
872 
873 	while (!cpu_online(cpu)) {
874 		cpu_relax();
875 		touch_nmi_watchdog();
876 	}
877 
878 	return 0;
879 }
880 
881 /**
882  * arch_disable_smp_support() - disables SMP support for x86 at runtime
883  */
arch_disable_smp_support(void)884 void arch_disable_smp_support(void)
885 {
886 	disable_ioapic_support();
887 }
888 
889 /*
890  * Fall back to non SMP mode after errors.
891  *
892  * RED-PEN audit/test this more. I bet there is more state messed up here.
893  */
disable_smp(void)894 static __init void disable_smp(void)
895 {
896 	init_cpu_present(cpumask_of(0));
897 	init_cpu_possible(cpumask_of(0));
898 	smpboot_clear_io_apic_irqs();
899 
900 	if (smp_found_config)
901 		physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
902 	else
903 		physid_set_mask_of_physid(0, &phys_cpu_present_map);
904 	cpumask_set_cpu(0, cpu_sibling_mask(0));
905 	cpumask_set_cpu(0, cpu_core_mask(0));
906 }
907 
908 /*
909  * Various sanity checks.
910  */
smp_sanity_check(unsigned max_cpus)911 static int __init smp_sanity_check(unsigned max_cpus)
912 {
913 	preempt_disable();
914 
915 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
916 	if (def_to_bigsmp && nr_cpu_ids > 8) {
917 		unsigned int cpu;
918 		unsigned nr;
919 
920 		printk(KERN_WARNING
921 		       "More than 8 CPUs detected - skipping them.\n"
922 		       "Use CONFIG_X86_BIGSMP.\n");
923 
924 		nr = 0;
925 		for_each_present_cpu(cpu) {
926 			if (nr >= 8)
927 				set_cpu_present(cpu, false);
928 			nr++;
929 		}
930 
931 		nr = 0;
932 		for_each_possible_cpu(cpu) {
933 			if (nr >= 8)
934 				set_cpu_possible(cpu, false);
935 			nr++;
936 		}
937 
938 		nr_cpu_ids = 8;
939 	}
940 #endif
941 
942 	if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
943 		printk(KERN_WARNING
944 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
945 			hard_smp_processor_id());
946 
947 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
948 	}
949 
950 	/*
951 	 * If we couldn't find an SMP configuration at boot time,
952 	 * get out of here now!
953 	 */
954 	if (!smp_found_config && !acpi_lapic) {
955 		preempt_enable();
956 		printk(KERN_NOTICE "SMP motherboard not detected.\n");
957 		disable_smp();
958 		if (APIC_init_uniprocessor())
959 			printk(KERN_NOTICE "Local APIC not detected."
960 					   " Using dummy APIC emulation.\n");
961 		return -1;
962 	}
963 
964 	/*
965 	 * Should not be necessary because the MP table should list the boot
966 	 * CPU too, but we do it for the sake of robustness anyway.
967 	 */
968 	if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
969 		printk(KERN_NOTICE
970 			"weird, boot CPU (#%d) not listed by the BIOS.\n",
971 			boot_cpu_physical_apicid);
972 		physid_set(hard_smp_processor_id(), phys_cpu_present_map);
973 	}
974 	preempt_enable();
975 
976 	/*
977 	 * If we couldn't find a local APIC, then get out of here now!
978 	 */
979 	if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
980 	    !cpu_has_apic) {
981 		if (!disable_apic) {
982 			pr_err("BIOS bug, local APIC #%d not detected!...\n",
983 				boot_cpu_physical_apicid);
984 			pr_err("... forcing use of dummy APIC emulation."
985 				"(tell your hw vendor)\n");
986 		}
987 		smpboot_clear_io_apic();
988 		disable_ioapic_support();
989 		return -1;
990 	}
991 
992 	verify_local_APIC();
993 
994 	/*
995 	 * If SMP should be disabled, then really disable it!
996 	 */
997 	if (!max_cpus) {
998 		printk(KERN_INFO "SMP mode deactivated.\n");
999 		smpboot_clear_io_apic();
1000 
1001 		connect_bsp_APIC();
1002 		setup_local_APIC();
1003 		bsp_end_local_APIC_setup();
1004 		return -1;
1005 	}
1006 
1007 	return 0;
1008 }
1009 
smp_cpu_index_default(void)1010 static void __init smp_cpu_index_default(void)
1011 {
1012 	int i;
1013 	struct cpuinfo_x86 *c;
1014 
1015 	for_each_possible_cpu(i) {
1016 		c = &cpu_data(i);
1017 		/* mark all to hotplug */
1018 		c->cpu_index = nr_cpu_ids;
1019 	}
1020 }
1021 
1022 /*
1023  * Prepare for SMP bootup.  The MP table or ACPI has been read
1024  * earlier.  Just do some sanity checking here and enable APIC mode.
1025  */
native_smp_prepare_cpus(unsigned int max_cpus)1026 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1027 {
1028 	unsigned int i;
1029 
1030 	preempt_disable();
1031 	smp_cpu_index_default();
1032 
1033 	/*
1034 	 * Setup boot CPU information
1035 	 */
1036 	smp_store_cpu_info(0); /* Final full version of the data */
1037 	cpumask_copy(cpu_callin_mask, cpumask_of(0));
1038 	mb();
1039 
1040 	current_thread_info()->cpu = 0;  /* needed? */
1041 	for_each_possible_cpu(i) {
1042 		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1043 		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1044 		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1045 	}
1046 	set_cpu_sibling_map(0);
1047 
1048 
1049 	if (smp_sanity_check(max_cpus) < 0) {
1050 		printk(KERN_INFO "SMP disabled\n");
1051 		disable_smp();
1052 		goto out;
1053 	}
1054 
1055 	default_setup_apic_routing();
1056 
1057 	preempt_disable();
1058 	if (read_apic_id() != boot_cpu_physical_apicid) {
1059 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1060 		     read_apic_id(), boot_cpu_physical_apicid);
1061 		/* Or can we switch back to PIC here? */
1062 	}
1063 	preempt_enable();
1064 
1065 	connect_bsp_APIC();
1066 
1067 	/*
1068 	 * Switch from PIC to APIC mode.
1069 	 */
1070 	setup_local_APIC();
1071 
1072 	/*
1073 	 * Enable IO APIC before setting up error vector
1074 	 */
1075 	if (!skip_ioapic_setup && nr_ioapics)
1076 		enable_IO_APIC();
1077 
1078 	bsp_end_local_APIC_setup();
1079 
1080 	if (apic->setup_portio_remap)
1081 		apic->setup_portio_remap();
1082 
1083 	smpboot_setup_io_apic();
1084 	/*
1085 	 * Set up local APIC timer on boot CPU.
1086 	 */
1087 
1088 	printk(KERN_INFO "CPU%d: ", 0);
1089 	print_cpu_info(&cpu_data(0));
1090 	x86_init.timers.setup_percpu_clockev();
1091 
1092 	if (is_uv_system())
1093 		uv_system_init();
1094 
1095 	set_mtrr_aps_delayed_init();
1096 out:
1097 	preempt_enable();
1098 }
1099 
arch_disable_nonboot_cpus_begin(void)1100 void arch_disable_nonboot_cpus_begin(void)
1101 {
1102 	/*
1103 	 * Avoid the smp alternatives switch during the disable_nonboot_cpus().
1104 	 * In the suspend path, we will be back in the SMP mode shortly anyways.
1105 	 */
1106 	skip_smp_alternatives = true;
1107 }
1108 
arch_disable_nonboot_cpus_end(void)1109 void arch_disable_nonboot_cpus_end(void)
1110 {
1111 	skip_smp_alternatives = false;
1112 }
1113 
arch_enable_nonboot_cpus_begin(void)1114 void arch_enable_nonboot_cpus_begin(void)
1115 {
1116 	set_mtrr_aps_delayed_init();
1117 }
1118 
arch_enable_nonboot_cpus_end(void)1119 void arch_enable_nonboot_cpus_end(void)
1120 {
1121 	mtrr_aps_init();
1122 }
1123 
1124 /*
1125  * Early setup to make printk work.
1126  */
native_smp_prepare_boot_cpu(void)1127 void __init native_smp_prepare_boot_cpu(void)
1128 {
1129 	int me = smp_processor_id();
1130 	switch_to_new_gdt(me);
1131 	/* already set me in cpu_online_mask in boot_cpu_init() */
1132 	cpumask_set_cpu(me, cpu_callout_mask);
1133 	per_cpu(cpu_state, me) = CPU_ONLINE;
1134 }
1135 
native_smp_cpus_done(unsigned int max_cpus)1136 void __init native_smp_cpus_done(unsigned int max_cpus)
1137 {
1138 	pr_debug("Boot done.\n");
1139 
1140 	nmi_selftest();
1141 	impress_friends();
1142 #ifdef CONFIG_X86_IO_APIC
1143 	setup_ioapic_dest();
1144 #endif
1145 	mtrr_aps_init();
1146 }
1147 
1148 static int __initdata setup_possible_cpus = -1;
_setup_possible_cpus(char * str)1149 static int __init _setup_possible_cpus(char *str)
1150 {
1151 	get_option(&str, &setup_possible_cpus);
1152 	return 0;
1153 }
1154 early_param("possible_cpus", _setup_possible_cpus);
1155 
1156 
1157 /*
1158  * cpu_possible_mask should be static, it cannot change as cpu's
1159  * are onlined, or offlined. The reason is per-cpu data-structures
1160  * are allocated by some modules at init time, and dont expect to
1161  * do this dynamically on cpu arrival/departure.
1162  * cpu_present_mask on the other hand can change dynamically.
1163  * In case when cpu_hotplug is not compiled, then we resort to current
1164  * behaviour, which is cpu_possible == cpu_present.
1165  * - Ashok Raj
1166  *
1167  * Three ways to find out the number of additional hotplug CPUs:
1168  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1169  * - The user can overwrite it with possible_cpus=NUM
1170  * - Otherwise don't reserve additional CPUs.
1171  * We do this because additional CPUs waste a lot of memory.
1172  * -AK
1173  */
prefill_possible_map(void)1174 __init void prefill_possible_map(void)
1175 {
1176 	int i, possible;
1177 
1178 	/* no processor from mptable or madt */
1179 	if (!num_processors)
1180 		num_processors = 1;
1181 
1182 	i = setup_max_cpus ?: 1;
1183 	if (setup_possible_cpus == -1) {
1184 		possible = num_processors;
1185 #ifdef CONFIG_HOTPLUG_CPU
1186 		if (setup_max_cpus)
1187 			possible += disabled_cpus;
1188 #else
1189 		if (possible > i)
1190 			possible = i;
1191 #endif
1192 	} else
1193 		possible = setup_possible_cpus;
1194 
1195 	total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1196 
1197 	/* nr_cpu_ids could be reduced via nr_cpus= */
1198 	if (possible > nr_cpu_ids) {
1199 		printk(KERN_WARNING
1200 			"%d Processors exceeds NR_CPUS limit of %d\n",
1201 			possible, nr_cpu_ids);
1202 		possible = nr_cpu_ids;
1203 	}
1204 
1205 #ifdef CONFIG_HOTPLUG_CPU
1206 	if (!setup_max_cpus)
1207 #endif
1208 	if (possible > i) {
1209 		printk(KERN_WARNING
1210 			"%d Processors exceeds max_cpus limit of %u\n",
1211 			possible, setup_max_cpus);
1212 		possible = i;
1213 	}
1214 
1215 	printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1216 		possible, max_t(int, possible - num_processors, 0));
1217 
1218 	for (i = 0; i < possible; i++)
1219 		set_cpu_possible(i, true);
1220 	for (; i < NR_CPUS; i++)
1221 		set_cpu_possible(i, false);
1222 
1223 	nr_cpu_ids = possible;
1224 }
1225 
1226 #ifdef CONFIG_HOTPLUG_CPU
1227 
remove_siblinginfo(int cpu)1228 static void remove_siblinginfo(int cpu)
1229 {
1230 	int sibling;
1231 	struct cpuinfo_x86 *c = &cpu_data(cpu);
1232 
1233 	for_each_cpu(sibling, cpu_core_mask(cpu)) {
1234 		cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1235 		/*/
1236 		 * last thread sibling in this cpu core going down
1237 		 */
1238 		if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1239 			cpu_data(sibling).booted_cores--;
1240 	}
1241 
1242 	for_each_cpu(sibling, cpu_sibling_mask(cpu))
1243 		cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1244 	cpumask_clear(cpu_sibling_mask(cpu));
1245 	cpumask_clear(cpu_core_mask(cpu));
1246 	c->phys_proc_id = 0;
1247 	c->cpu_core_id = 0;
1248 	cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1249 }
1250 
remove_cpu_from_maps(int cpu)1251 static void __ref remove_cpu_from_maps(int cpu)
1252 {
1253 	set_cpu_online(cpu, false);
1254 	cpumask_clear_cpu(cpu, cpu_callout_mask);
1255 	cpumask_clear_cpu(cpu, cpu_callin_mask);
1256 	/* was set by cpu_init() */
1257 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
1258 	numa_remove_cpu(cpu);
1259 }
1260 
cpu_disable_common(void)1261 void cpu_disable_common(void)
1262 {
1263 	int cpu = smp_processor_id();
1264 
1265 	remove_siblinginfo(cpu);
1266 
1267 	/* It's now safe to remove this processor from the online map */
1268 	lock_vector_lock();
1269 	remove_cpu_from_maps(cpu);
1270 	unlock_vector_lock();
1271 	fixup_irqs();
1272 }
1273 
native_cpu_disable(void)1274 int native_cpu_disable(void)
1275 {
1276 	int cpu = smp_processor_id();
1277 
1278 	/*
1279 	 * Perhaps use cpufreq to drop frequency, but that could go
1280 	 * into generic code.
1281 	 *
1282 	 * We won't take down the boot processor on i386 due to some
1283 	 * interrupts only being able to be serviced by the BSP.
1284 	 * Especially so if we're not using an IOAPIC	-zwane
1285 	 */
1286 	if (cpu == 0)
1287 		return -EBUSY;
1288 
1289 	clear_local_APIC();
1290 
1291 	cpu_disable_common();
1292 	return 0;
1293 }
1294 
native_cpu_die(unsigned int cpu)1295 void native_cpu_die(unsigned int cpu)
1296 {
1297 	/* We don't do anything here: idle task is faking death itself. */
1298 	unsigned int i;
1299 
1300 	for (i = 0; i < 10; i++) {
1301 		/* They ack this in play_dead by setting CPU_DEAD */
1302 		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1303 			if (system_state == SYSTEM_RUNNING)
1304 				pr_info("CPU %u is now offline\n", cpu);
1305 
1306 			if (1 == num_online_cpus())
1307 				alternatives_smp_switch(0);
1308 			return;
1309 		}
1310 		msleep(100);
1311 	}
1312 	pr_err("CPU %u didn't die...\n", cpu);
1313 }
1314 
play_dead_common(void)1315 void play_dead_common(void)
1316 {
1317 	idle_task_exit();
1318 	reset_lazy_tlbstate();
1319 	amd_e400_remove_cpu(raw_smp_processor_id());
1320 
1321 	mb();
1322 	/* Ack it */
1323 	__this_cpu_write(cpu_state, CPU_DEAD);
1324 
1325 	/*
1326 	 * With physical CPU hotplug, we should halt the cpu
1327 	 */
1328 	local_irq_disable();
1329 }
1330 
1331 /*
1332  * We need to flush the caches before going to sleep, lest we have
1333  * dirty data in our caches when we come back up.
1334  */
mwait_play_dead(void)1335 static inline void mwait_play_dead(void)
1336 {
1337 	unsigned int eax, ebx, ecx, edx;
1338 	unsigned int highest_cstate = 0;
1339 	unsigned int highest_subcstate = 0;
1340 	int i;
1341 	void *mwait_ptr;
1342 	struct cpuinfo_x86 *c = __this_cpu_ptr(&cpu_info);
1343 
1344 	if (!(this_cpu_has(X86_FEATURE_MWAIT) && mwait_usable(c)))
1345 		return;
1346 	if (!this_cpu_has(X86_FEATURE_CLFLSH))
1347 		return;
1348 	if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1349 		return;
1350 
1351 	eax = CPUID_MWAIT_LEAF;
1352 	ecx = 0;
1353 	native_cpuid(&eax, &ebx, &ecx, &edx);
1354 
1355 	/*
1356 	 * eax will be 0 if EDX enumeration is not valid.
1357 	 * Initialized below to cstate, sub_cstate value when EDX is valid.
1358 	 */
1359 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1360 		eax = 0;
1361 	} else {
1362 		edx >>= MWAIT_SUBSTATE_SIZE;
1363 		for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1364 			if (edx & MWAIT_SUBSTATE_MASK) {
1365 				highest_cstate = i;
1366 				highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1367 			}
1368 		}
1369 		eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1370 			(highest_subcstate - 1);
1371 	}
1372 
1373 	/*
1374 	 * This should be a memory location in a cache line which is
1375 	 * unlikely to be touched by other processors.  The actual
1376 	 * content is immaterial as it is not actually modified in any way.
1377 	 */
1378 	mwait_ptr = &current_thread_info()->flags;
1379 
1380 	wbinvd();
1381 
1382 	while (1) {
1383 		/*
1384 		 * The CLFLUSH is a workaround for erratum AAI65 for
1385 		 * the Xeon 7400 series.  It's not clear it is actually
1386 		 * needed, but it should be harmless in either case.
1387 		 * The WBINVD is insufficient due to the spurious-wakeup
1388 		 * case where we return around the loop.
1389 		 */
1390 		clflush(mwait_ptr);
1391 		__monitor(mwait_ptr, 0, 0);
1392 		mb();
1393 		__mwait(eax, 0);
1394 	}
1395 }
1396 
hlt_play_dead(void)1397 static inline void hlt_play_dead(void)
1398 {
1399 	if (__this_cpu_read(cpu_info.x86) >= 4)
1400 		wbinvd();
1401 
1402 	while (1) {
1403 		native_halt();
1404 	}
1405 }
1406 
native_play_dead(void)1407 void native_play_dead(void)
1408 {
1409 	play_dead_common();
1410 	tboot_shutdown(TB_SHUTDOWN_WFS);
1411 
1412 	mwait_play_dead();	/* Only returns on failure */
1413 	if (cpuidle_play_dead())
1414 		hlt_play_dead();
1415 }
1416 
1417 #else /* ... !CONFIG_HOTPLUG_CPU */
native_cpu_disable(void)1418 int native_cpu_disable(void)
1419 {
1420 	return -ENOSYS;
1421 }
1422 
native_cpu_die(unsigned int cpu)1423 void native_cpu_die(unsigned int cpu)
1424 {
1425 	/* We said "no" in __cpu_disable */
1426 	BUG();
1427 }
1428 
native_play_dead(void)1429 void native_play_dead(void)
1430 {
1431 	BUG();
1432 }
1433 
1434 #endif
1435