2 * x86 SMP booting functions
6 * Copyright 2001 Andi Kleen, SuSE Labs.
8 * Much of the core SMP work is based on previous work by Thomas Radke, to
9 * whom a great many thanks are extended.
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.
15 * This code is released under the GNU General Public License version 2 or
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
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
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>
59 #include <asm/trampoline.h>
62 #include <asm/pgtable.h>
63 #include <asm/tlbflush.h>
67 #include <asm/setup.h>
68 #include <asm/uv/uv.h>
69 #include <linux/mc146818rtc.h>
71 #include <asm/smpboot_hooks.h>
72 #include <asm/i8259.h>
75 u8 apicid_2_node[MAX_APICID];
76 static int low_mappings;
79 /* State of each CPU */
80 DEFINE_PER_CPU(int, cpu_state) = { 0 };
82 /* Store all idle threads, this can be reused instead of creating
83 * a new thread. Also avoids complicated thread destroy functionality
86 #ifdef CONFIG_HOTPLUG_CPU
88 * Needed only for CONFIG_HOTPLUG_CPU because __cpuinitdata is
89 * removed after init for !CONFIG_HOTPLUG_CPU.
91 static DEFINE_PER_CPU(struct task_struct *, idle_thread_array);
92 #define get_idle_for_cpu(x) (per_cpu(idle_thread_array, x))
93 #define set_idle_for_cpu(x, p) (per_cpu(idle_thread_array, x) = (p))
95 static struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
96 #define get_idle_for_cpu(x) (idle_thread_array[(x)])
97 #define set_idle_for_cpu(x, p) (idle_thread_array[(x)] = (p))
100 /* Number of siblings per CPU package */
101 int smp_num_siblings = 1;
102 EXPORT_SYMBOL(smp_num_siblings);
104 /* Last level cache ID of each logical CPU */
105 DEFINE_PER_CPU(u16, cpu_llc_id) = BAD_APICID;
107 /* representing HT siblings of each logical CPU */
108 DEFINE_PER_CPU(cpumask_var_t, cpu_sibling_map);
109 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
111 /* representing HT and core siblings of each logical CPU */
112 DEFINE_PER_CPU(cpumask_var_t, cpu_core_map);
113 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
115 /* Per CPU bogomips and other parameters */
116 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
117 EXPORT_PER_CPU_SYMBOL(cpu_info);
119 atomic_t init_deasserted;
121 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_32)
122 /* which node each logical CPU is on */
123 int cpu_to_node_map[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 };
124 EXPORT_SYMBOL(cpu_to_node_map);
126 /* set up a mapping between cpu and node. */
127 static void map_cpu_to_node(int cpu, int node)
129 printk(KERN_INFO "Mapping cpu %d to node %d\n", cpu, node);
130 cpumask_set_cpu(cpu, node_to_cpumask_map[node]);
131 cpu_to_node_map[cpu] = node;
134 /* undo a mapping between cpu and node. */
135 static void unmap_cpu_to_node(int cpu)
139 printk(KERN_INFO "Unmapping cpu %d from all nodes\n", cpu);
140 for (node = 0; node < MAX_NUMNODES; node++)
141 cpumask_clear_cpu(cpu, node_to_cpumask_map[node]);
142 cpu_to_node_map[cpu] = 0;
144 #else /* !(CONFIG_NUMA && CONFIG_X86_32) */
145 #define map_cpu_to_node(cpu, node) ({})
146 #define unmap_cpu_to_node(cpu) ({})
150 static int boot_cpu_logical_apicid;
152 u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly =
153 { [0 ... NR_CPUS-1] = BAD_APICID };
155 static void map_cpu_to_logical_apicid(void)
157 int cpu = smp_processor_id();
158 int apicid = logical_smp_processor_id();
159 int node = apic->apicid_to_node(apicid);
161 if (!node_online(node))
162 node = first_online_node;
164 cpu_2_logical_apicid[cpu] = apicid;
165 map_cpu_to_node(cpu, node);
168 void numa_remove_cpu(int cpu)
170 cpu_2_logical_apicid[cpu] = BAD_APICID;
171 unmap_cpu_to_node(cpu);
174 #define map_cpu_to_logical_apicid() do {} while (0)
178 * Report back to the Boot Processor.
181 static void __cpuinit smp_callin(void)
184 unsigned long timeout;
187 * If waken up by an INIT in an 82489DX configuration
188 * we may get here before an INIT-deassert IPI reaches
189 * our local APIC. We have to wait for the IPI or we'll
190 * lock up on an APIC access.
192 if (apic->wait_for_init_deassert)
193 apic->wait_for_init_deassert(&init_deasserted);
196 * (This works even if the APIC is not enabled.)
198 phys_id = read_apic_id();
199 cpuid = smp_processor_id();
200 if (cpumask_test_cpu(cpuid, cpu_callin_mask)) {
201 panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
204 pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
207 * STARTUP IPIs are fragile beasts as they might sometimes
208 * trigger some glue motherboard logic. Complete APIC bus
209 * silence for 1 second, this overestimates the time the
210 * boot CPU is spending to send the up to 2 STARTUP IPIs
211 * by a factor of two. This should be enough.
215 * Waiting 2s total for startup (udelay is not yet working)
217 timeout = jiffies + 2*HZ;
218 while (time_before(jiffies, timeout)) {
220 * Has the boot CPU finished it's STARTUP sequence?
222 if (cpumask_test_cpu(cpuid, cpu_callout_mask))
227 if (!time_before(jiffies, timeout)) {
228 panic("%s: CPU%d started up but did not get a callout!\n",
233 * the boot CPU has finished the init stage and is spinning
234 * on callin_map until we finish. We are free to set up this
235 * CPU, first the APIC. (this is probably redundant on most
239 pr_debug("CALLIN, before setup_local_APIC().\n");
240 if (apic->smp_callin_clear_local_apic)
241 apic->smp_callin_clear_local_apic();
243 end_local_APIC_setup();
244 map_cpu_to_logical_apicid();
247 * Need to setup vector mappings before we enable interrupts.
249 setup_vector_irq(smp_processor_id());
253 * Need to enable IRQs because it can take longer and then
254 * the NMI watchdog might kill us.
259 pr_debug("Stack at about %p\n", &cpuid);
262 * Save our processor parameters
264 smp_store_cpu_info(cpuid);
266 notify_cpu_starting(cpuid);
269 * Allow the master to continue.
271 cpumask_set_cpu(cpuid, cpu_callin_mask);
275 * Activate a secondary processor.
277 notrace static void __cpuinit start_secondary(void *unused)
280 * Don't put *anything* before cpu_init(), SMP booting is too
281 * fragile that we want to limit the things done here to the
282 * most necessary things.
289 /* otherwise gcc will move up smp_processor_id before the cpu_init */
292 * Check TSC synchronization with the BP:
294 check_tsc_sync_target();
296 if (nmi_watchdog == NMI_IO_APIC) {
297 legacy_pic->chip->mask(0);
298 enable_NMI_through_LVT0();
299 legacy_pic->chip->unmask(0);
308 /* This must be done before setting cpu_online_mask */
309 set_cpu_sibling_map(raw_smp_processor_id());
313 * We need to hold call_lock, so there is no inconsistency
314 * between the time smp_call_function() determines number of
315 * IPI recipients, and the time when the determination is made
316 * for which cpus receive the IPI. Holding this
317 * lock helps us to not include this cpu in a currently in progress
318 * smp_call_function().
320 * We need to hold vector_lock so there the set of online cpus
321 * does not change while we are assigning vectors to cpus. Holding
322 * this lock ensures we don't half assign or remove an irq from a cpu.
326 set_cpu_online(smp_processor_id(), true);
327 unlock_vector_lock();
329 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
330 x86_platform.nmi_init();
332 /* enable local interrupts */
335 /* to prevent fake stack check failure in clock setup */
336 boot_init_stack_canary();
338 x86_cpuinit.setup_percpu_clockev();
344 #ifdef CONFIG_CPUMASK_OFFSTACK
345 /* In this case, llc_shared_map is a pointer to a cpumask. */
346 static inline void copy_cpuinfo_x86(struct cpuinfo_x86 *dst,
347 const struct cpuinfo_x86 *src)
349 struct cpumask *llc = dst->llc_shared_map;
351 dst->llc_shared_map = llc;
354 static inline void copy_cpuinfo_x86(struct cpuinfo_x86 *dst,
355 const struct cpuinfo_x86 *src)
359 #endif /* CONFIG_CPUMASK_OFFSTACK */
362 * The bootstrap kernel entry code has set these up. Save them for
366 void __cpuinit smp_store_cpu_info(int id)
368 struct cpuinfo_x86 *c = &cpu_data(id);
370 copy_cpuinfo_x86(c, &boot_cpu_data);
373 identify_secondary_cpu(c);
377 void __cpuinit set_cpu_sibling_map(int cpu)
380 struct cpuinfo_x86 *c = &cpu_data(cpu);
382 cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
384 if (smp_num_siblings > 1) {
385 for_each_cpu(i, cpu_sibling_setup_mask) {
386 struct cpuinfo_x86 *o = &cpu_data(i);
388 if (c->phys_proc_id == o->phys_proc_id &&
389 c->cpu_core_id == o->cpu_core_id) {
390 cpumask_set_cpu(i, cpu_sibling_mask(cpu));
391 cpumask_set_cpu(cpu, cpu_sibling_mask(i));
392 cpumask_set_cpu(i, cpu_core_mask(cpu));
393 cpumask_set_cpu(cpu, cpu_core_mask(i));
394 cpumask_set_cpu(i, c->llc_shared_map);
395 cpumask_set_cpu(cpu, o->llc_shared_map);
399 cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
402 cpumask_set_cpu(cpu, c->llc_shared_map);
404 if (current_cpu_data.x86_max_cores == 1) {
405 cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
410 for_each_cpu(i, cpu_sibling_setup_mask) {
411 if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
412 per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
413 cpumask_set_cpu(i, c->llc_shared_map);
414 cpumask_set_cpu(cpu, cpu_data(i).llc_shared_map);
416 if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
417 cpumask_set_cpu(i, cpu_core_mask(cpu));
418 cpumask_set_cpu(cpu, cpu_core_mask(i));
420 * Does this new cpu bringup a new core?
422 if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
424 * for each core in package, increment
425 * the booted_cores for this new cpu
427 if (cpumask_first(cpu_sibling_mask(i)) == i)
430 * increment the core count for all
431 * the other cpus in this package
434 cpu_data(i).booted_cores++;
435 } else if (i != cpu && !c->booted_cores)
436 c->booted_cores = cpu_data(i).booted_cores;
441 /* maps the cpu to the sched domain representing multi-core */
442 const struct cpumask *cpu_coregroup_mask(int cpu)
444 struct cpuinfo_x86 *c = &cpu_data(cpu);
446 * For perf, we return last level cache shared map.
447 * And for power savings, we return cpu_core_map
449 if ((sched_mc_power_savings || sched_smt_power_savings) &&
450 !(cpu_has(c, X86_FEATURE_AMD_DCM)))
451 return cpu_core_mask(cpu);
453 return c->llc_shared_map;
456 static void impress_friends(void)
459 unsigned long bogosum = 0;
461 * Allow the user to impress friends.
463 pr_debug("Before bogomips.\n");
464 for_each_possible_cpu(cpu)
465 if (cpumask_test_cpu(cpu, cpu_callout_mask))
466 bogosum += cpu_data(cpu).loops_per_jiffy;
468 "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
471 (bogosum/(5000/HZ))%100);
473 pr_debug("Before bogocount - setting activated=1.\n");
476 void __inquire_remote_apic(int apicid)
478 unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
479 char *names[] = { "ID", "VERSION", "SPIV" };
483 printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
485 for (i = 0; i < ARRAY_SIZE(regs); i++) {
486 printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
491 status = safe_apic_wait_icr_idle();
494 "a previous APIC delivery may have failed\n");
496 apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
501 status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
502 } while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
505 case APIC_ICR_RR_VALID:
506 status = apic_read(APIC_RRR);
507 printk(KERN_CONT "%08x\n", status);
510 printk(KERN_CONT "failed\n");
516 * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
517 * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
518 * won't ... remember to clear down the APIC, etc later.
521 wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
523 unsigned long send_status, accept_status = 0;
527 /* Boot on the stack */
528 /* Kick the second */
529 apic_icr_write(APIC_DM_NMI | apic->dest_logical, logical_apicid);
531 pr_debug("Waiting for send to finish...\n");
532 send_status = safe_apic_wait_icr_idle();
535 * Give the other CPU some time to accept the IPI.
538 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
539 maxlvt = lapic_get_maxlvt();
540 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
541 apic_write(APIC_ESR, 0);
542 accept_status = (apic_read(APIC_ESR) & 0xEF);
544 pr_debug("NMI sent.\n");
547 printk(KERN_ERR "APIC never delivered???\n");
549 printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
551 return (send_status | accept_status);
555 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
557 unsigned long send_status, accept_status = 0;
558 int maxlvt, num_starts, j;
560 maxlvt = lapic_get_maxlvt();
563 * Be paranoid about clearing APIC errors.
565 if (APIC_INTEGRATED(apic_version[phys_apicid])) {
566 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
567 apic_write(APIC_ESR, 0);
571 pr_debug("Asserting INIT.\n");
574 * Turn INIT on target chip
579 apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
582 pr_debug("Waiting for send to finish...\n");
583 send_status = safe_apic_wait_icr_idle();
587 pr_debug("Deasserting INIT.\n");
591 apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
593 pr_debug("Waiting for send to finish...\n");
594 send_status = safe_apic_wait_icr_idle();
597 atomic_set(&init_deasserted, 1);
600 * Should we send STARTUP IPIs ?
602 * Determine this based on the APIC version.
603 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
605 if (APIC_INTEGRATED(apic_version[phys_apicid]))
611 * Paravirt / VMI wants a startup IPI hook here to set up the
612 * target processor state.
614 startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
615 (unsigned long)stack_start.sp);
618 * Run STARTUP IPI loop.
620 pr_debug("#startup loops: %d.\n", num_starts);
622 for (j = 1; j <= num_starts; j++) {
623 pr_debug("Sending STARTUP #%d.\n", j);
624 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
625 apic_write(APIC_ESR, 0);
627 pr_debug("After apic_write.\n");
634 /* Boot on the stack */
635 /* Kick the second */
636 apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
640 * Give the other CPU some time to accept the IPI.
644 pr_debug("Startup point 1.\n");
646 pr_debug("Waiting for send to finish...\n");
647 send_status = safe_apic_wait_icr_idle();
650 * Give the other CPU some time to accept the IPI.
653 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
654 apic_write(APIC_ESR, 0);
655 accept_status = (apic_read(APIC_ESR) & 0xEF);
656 if (send_status || accept_status)
659 pr_debug("After Startup.\n");
662 printk(KERN_ERR "APIC never delivered???\n");
664 printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
666 return (send_status | accept_status);
670 struct work_struct work;
671 struct task_struct *idle;
672 struct completion done;
676 static void __cpuinit do_fork_idle(struct work_struct *work)
678 struct create_idle *c_idle =
679 container_of(work, struct create_idle, work);
681 c_idle->idle = fork_idle(c_idle->cpu);
682 complete(&c_idle->done);
685 /* reduce the number of lines printed when booting a large cpu count system */
686 static void __cpuinit announce_cpu(int cpu, int apicid)
688 static int current_node = -1;
689 int node = early_cpu_to_node(cpu);
691 if (system_state == SYSTEM_BOOTING) {
692 if (node != current_node) {
693 if (current_node > (-1))
696 pr_info("Booting Node %3d, Processors ", node);
698 pr_cont(" #%d%s", cpu, cpu == (nr_cpu_ids - 1) ? " Ok.\n" : "");
701 pr_info("Booting Node %d Processor %d APIC 0x%x\n",
706 * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
707 * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
708 * Returns zero if CPU booted OK, else error code from
709 * ->wakeup_secondary_cpu.
711 static int __cpuinit do_boot_cpu(int apicid, int cpu)
713 unsigned long boot_error = 0;
714 unsigned long start_ip;
716 struct create_idle c_idle = {
718 .done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
721 INIT_WORK_ON_STACK(&c_idle.work, do_fork_idle);
723 alternatives_smp_switch(1);
725 c_idle.idle = get_idle_for_cpu(cpu);
728 * We can't use kernel_thread since we must avoid to
729 * reschedule the child.
732 c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
733 (THREAD_SIZE + task_stack_page(c_idle.idle))) - 1);
734 init_idle(c_idle.idle, cpu);
738 if (!keventd_up() || current_is_keventd())
739 c_idle.work.func(&c_idle.work);
741 schedule_work(&c_idle.work);
742 wait_for_completion(&c_idle.done);
745 if (IS_ERR(c_idle.idle)) {
746 printk("failed fork for CPU %d\n", cpu);
747 destroy_work_on_stack(&c_idle.work);
748 return PTR_ERR(c_idle.idle);
751 set_idle_for_cpu(cpu, c_idle.idle);
753 per_cpu(current_task, cpu) = c_idle.idle;
755 /* Stack for startup_32 can be just as for start_secondary onwards */
758 clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
759 initial_gs = per_cpu_offset(cpu);
760 per_cpu(kernel_stack, cpu) =
761 (unsigned long)task_stack_page(c_idle.idle) -
762 KERNEL_STACK_OFFSET + THREAD_SIZE;
764 early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
765 initial_code = (unsigned long)start_secondary;
766 stack_start.sp = (void *) c_idle.idle->thread.sp;
768 /* start_ip had better be page-aligned! */
769 start_ip = setup_trampoline();
771 /* So we see what's up */
772 announce_cpu(cpu, apicid);
775 * This grunge runs the startup process for
776 * the targeted processor.
779 atomic_set(&init_deasserted, 0);
781 if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
783 pr_debug("Setting warm reset code and vector.\n");
785 smpboot_setup_warm_reset_vector(start_ip);
787 * Be paranoid about clearing APIC errors.
789 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
790 apic_write(APIC_ESR, 0);
796 * Kick the secondary CPU. Use the method in the APIC driver
797 * if it's defined - or use an INIT boot APIC message otherwise:
799 if (apic->wakeup_secondary_cpu)
800 boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
802 boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
806 * allow APs to start initializing.
808 pr_debug("Before Callout %d.\n", cpu);
809 cpumask_set_cpu(cpu, cpu_callout_mask);
810 pr_debug("After Callout %d.\n", cpu);
813 * Wait 5s total for a response
815 for (timeout = 0; timeout < 50000; timeout++) {
816 if (cpumask_test_cpu(cpu, cpu_callin_mask))
817 break; /* It has booted */
820 * Allow other tasks to run while we wait for the
821 * AP to come online. This also gives a chance
822 * for the MTRR work(triggered by the AP coming online)
823 * to be completed in the stop machine context.
828 if (cpumask_test_cpu(cpu, cpu_callin_mask))
829 pr_debug("CPU%d: has booted.\n", cpu);
832 if (*((volatile unsigned char *)trampoline_base)
834 /* trampoline started but...? */
835 pr_err("CPU%d: Stuck ??\n", cpu);
837 /* trampoline code not run */
838 pr_err("CPU%d: Not responding.\n", cpu);
839 if (apic->inquire_remote_apic)
840 apic->inquire_remote_apic(apicid);
845 /* Try to put things back the way they were before ... */
846 numa_remove_cpu(cpu); /* was set by numa_add_cpu */
848 /* was set by do_boot_cpu() */
849 cpumask_clear_cpu(cpu, cpu_callout_mask);
851 /* was set by cpu_init() */
852 cpumask_clear_cpu(cpu, cpu_initialized_mask);
854 set_cpu_present(cpu, false);
855 per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
858 /* mark "stuck" area as not stuck */
859 *((volatile unsigned long *)trampoline_base) = 0;
861 if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
863 * Cleanup possible dangling ends...
865 smpboot_restore_warm_reset_vector();
868 destroy_work_on_stack(&c_idle.work);
872 int __cpuinit native_cpu_up(unsigned int cpu)
874 int apicid = apic->cpu_present_to_apicid(cpu);
878 WARN_ON(irqs_disabled());
880 pr_debug("++++++++++++++++++++=_---CPU UP %u\n", cpu);
882 if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
883 !physid_isset(apicid, phys_cpu_present_map)) {
884 printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
889 * Already booted CPU?
891 if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
892 pr_debug("do_boot_cpu %d Already started\n", cpu);
897 * Save current MTRR state in case it was changed since early boot
898 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
902 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
905 /* init low mem mapping */
906 clone_pgd_range(swapper_pg_dir, swapper_pg_dir + KERNEL_PGD_BOUNDARY,
907 min_t(unsigned long, KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
911 err = do_boot_cpu(apicid, cpu);
913 zap_low_mappings(false);
916 err = do_boot_cpu(apicid, cpu);
919 pr_debug("do_boot_cpu failed %d\n", err);
924 * Check TSC synchronization with the AP (keep irqs disabled
927 local_irq_save(flags);
928 check_tsc_sync_source(cpu);
929 local_irq_restore(flags);
931 while (!cpu_online(cpu)) {
933 touch_nmi_watchdog();
940 * Fall back to non SMP mode after errors.
942 * RED-PEN audit/test this more. I bet there is more state messed up here.
944 static __init void disable_smp(void)
946 init_cpu_present(cpumask_of(0));
947 init_cpu_possible(cpumask_of(0));
948 smpboot_clear_io_apic_irqs();
950 if (smp_found_config)
951 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
953 physid_set_mask_of_physid(0, &phys_cpu_present_map);
954 map_cpu_to_logical_apicid();
955 cpumask_set_cpu(0, cpu_sibling_mask(0));
956 cpumask_set_cpu(0, cpu_core_mask(0));
960 * Various sanity checks.
962 static int __init smp_sanity_check(unsigned max_cpus)
966 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
967 if (def_to_bigsmp && nr_cpu_ids > 8) {
972 "More than 8 CPUs detected - skipping them.\n"
973 "Use CONFIG_X86_BIGSMP.\n");
976 for_each_present_cpu(cpu) {
978 set_cpu_present(cpu, false);
983 for_each_possible_cpu(cpu) {
985 set_cpu_possible(cpu, false);
993 if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
995 "weird, boot CPU (#%d) not listed by the BIOS.\n",
996 hard_smp_processor_id());
998 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1002 * If we couldn't find an SMP configuration at boot time,
1003 * get out of here now!
1005 if (!smp_found_config && !acpi_lapic) {
1007 printk(KERN_NOTICE "SMP motherboard not detected.\n");
1009 if (APIC_init_uniprocessor())
1010 printk(KERN_NOTICE "Local APIC not detected."
1011 " Using dummy APIC emulation.\n");
1016 * Should not be necessary because the MP table should list the boot
1017 * CPU too, but we do it for the sake of robustness anyway.
1019 if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1021 "weird, boot CPU (#%d) not listed by the BIOS.\n",
1022 boot_cpu_physical_apicid);
1023 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1028 * If we couldn't find a local APIC, then get out of here now!
1030 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1032 if (!disable_apic) {
1033 pr_err("BIOS bug, local APIC #%d not detected!...\n",
1034 boot_cpu_physical_apicid);
1035 pr_err("... forcing use of dummy APIC emulation."
1036 "(tell your hw vendor)\n");
1038 smpboot_clear_io_apic();
1039 arch_disable_smp_support();
1043 verify_local_APIC();
1046 * If SMP should be disabled, then really disable it!
1049 printk(KERN_INFO "SMP mode deactivated.\n");
1050 smpboot_clear_io_apic();
1052 localise_nmi_watchdog();
1056 end_local_APIC_setup();
1063 static void __init smp_cpu_index_default(void)
1066 struct cpuinfo_x86 *c;
1068 for_each_possible_cpu(i) {
1070 /* mark all to hotplug */
1071 c->cpu_index = nr_cpu_ids;
1076 * Prepare for SMP bootup. The MP table or ACPI has been read
1077 * earlier. Just do some sanity checking here and enable APIC mode.
1079 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1084 smp_cpu_index_default();
1085 current_cpu_data = boot_cpu_data;
1086 cpumask_copy(cpu_callin_mask, cpumask_of(0));
1089 * Setup boot CPU information
1091 smp_store_cpu_info(0); /* Final full version of the data */
1092 #ifdef CONFIG_X86_32
1093 boot_cpu_logical_apicid = logical_smp_processor_id();
1095 current_thread_info()->cpu = 0; /* needed? */
1096 for_each_possible_cpu(i) {
1097 zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1098 zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1099 zalloc_cpumask_var(&cpu_data(i).llc_shared_map, GFP_KERNEL);
1101 set_cpu_sibling_map(0);
1104 default_setup_apic_routing();
1106 if (smp_sanity_check(max_cpus) < 0) {
1107 printk(KERN_INFO "SMP disabled\n");
1113 if (read_apic_id() != boot_cpu_physical_apicid) {
1114 panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1115 read_apic_id(), boot_cpu_physical_apicid);
1116 /* Or can we switch back to PIC here? */
1123 * Switch from PIC to APIC mode.
1128 * Enable IO APIC before setting up error vector
1130 if (!skip_ioapic_setup && nr_ioapics)
1133 end_local_APIC_setup();
1135 map_cpu_to_logical_apicid();
1137 if (apic->setup_portio_remap)
1138 apic->setup_portio_remap();
1140 smpboot_setup_io_apic();
1142 * Set up local APIC timer on boot CPU.
1145 printk(KERN_INFO "CPU%d: ", 0);
1146 print_cpu_info(&cpu_data(0));
1147 x86_init.timers.setup_percpu_clockev();
1152 set_mtrr_aps_delayed_init();
1157 void arch_enable_nonboot_cpus_begin(void)
1159 set_mtrr_aps_delayed_init();
1162 void arch_enable_nonboot_cpus_end(void)
1168 * Early setup to make printk work.
1170 void __init native_smp_prepare_boot_cpu(void)
1172 int me = smp_processor_id();
1173 switch_to_new_gdt(me);
1174 /* already set me in cpu_online_mask in boot_cpu_init() */
1175 cpumask_set_cpu(me, cpu_callout_mask);
1176 per_cpu(cpu_state, me) = CPU_ONLINE;
1179 void __init native_smp_cpus_done(unsigned int max_cpus)
1181 pr_debug("Boot done.\n");
1184 #ifdef CONFIG_X86_IO_APIC
1185 setup_ioapic_dest();
1187 check_nmi_watchdog();
1191 static int __initdata setup_possible_cpus = -1;
1192 static int __init _setup_possible_cpus(char *str)
1194 get_option(&str, &setup_possible_cpus);
1197 early_param("possible_cpus", _setup_possible_cpus);
1201 * cpu_possible_mask should be static, it cannot change as cpu's
1202 * are onlined, or offlined. The reason is per-cpu data-structures
1203 * are allocated by some modules at init time, and dont expect to
1204 * do this dynamically on cpu arrival/departure.
1205 * cpu_present_mask on the other hand can change dynamically.
1206 * In case when cpu_hotplug is not compiled, then we resort to current
1207 * behaviour, which is cpu_possible == cpu_present.
1210 * Three ways to find out the number of additional hotplug CPUs:
1211 * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1212 * - The user can overwrite it with possible_cpus=NUM
1213 * - Otherwise don't reserve additional CPUs.
1214 * We do this because additional CPUs waste a lot of memory.
1217 __init void prefill_possible_map(void)
1221 /* no processor from mptable or madt */
1222 if (!num_processors)
1225 i = setup_max_cpus ?: 1;
1226 if (setup_possible_cpus == -1) {
1227 possible = num_processors;
1228 #ifdef CONFIG_HOTPLUG_CPU
1230 possible += disabled_cpus;
1236 possible = setup_possible_cpus;
1238 total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1240 /* nr_cpu_ids could be reduced via nr_cpus= */
1241 if (possible > nr_cpu_ids) {
1243 "%d Processors exceeds NR_CPUS limit of %d\n",
1244 possible, nr_cpu_ids);
1245 possible = nr_cpu_ids;
1248 #ifdef CONFIG_HOTPLUG_CPU
1249 if (!setup_max_cpus)
1253 "%d Processors exceeds max_cpus limit of %u\n",
1254 possible, setup_max_cpus);
1258 printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1259 possible, max_t(int, possible - num_processors, 0));
1261 for (i = 0; i < possible; i++)
1262 set_cpu_possible(i, true);
1263 for (; i < NR_CPUS; i++)
1264 set_cpu_possible(i, false);
1266 nr_cpu_ids = possible;
1269 #ifdef CONFIG_HOTPLUG_CPU
1271 static void remove_siblinginfo(int cpu)
1274 struct cpuinfo_x86 *c = &cpu_data(cpu);
1276 for_each_cpu(sibling, cpu_core_mask(cpu)) {
1277 cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1279 * last thread sibling in this cpu core going down
1281 if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1282 cpu_data(sibling).booted_cores--;
1285 for_each_cpu(sibling, cpu_sibling_mask(cpu))
1286 cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1287 cpumask_clear(cpu_sibling_mask(cpu));
1288 cpumask_clear(cpu_core_mask(cpu));
1289 c->phys_proc_id = 0;
1291 cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1294 static void __ref remove_cpu_from_maps(int cpu)
1296 set_cpu_online(cpu, false);
1297 cpumask_clear_cpu(cpu, cpu_callout_mask);
1298 cpumask_clear_cpu(cpu, cpu_callin_mask);
1299 /* was set by cpu_init() */
1300 cpumask_clear_cpu(cpu, cpu_initialized_mask);
1301 numa_remove_cpu(cpu);
1304 void cpu_disable_common(void)
1306 int cpu = smp_processor_id();
1308 remove_siblinginfo(cpu);
1310 /* It's now safe to remove this processor from the online map */
1312 remove_cpu_from_maps(cpu);
1313 unlock_vector_lock();
1317 int native_cpu_disable(void)
1319 int cpu = smp_processor_id();
1322 * Perhaps use cpufreq to drop frequency, but that could go
1323 * into generic code.
1325 * We won't take down the boot processor on i386 due to some
1326 * interrupts only being able to be serviced by the BSP.
1327 * Especially so if we're not using an IOAPIC -zwane
1332 if (nmi_watchdog == NMI_LOCAL_APIC)
1333 stop_apic_nmi_watchdog(NULL);
1336 cpu_disable_common();
1340 void native_cpu_die(unsigned int cpu)
1342 /* We don't do anything here: idle task is faking death itself. */
1345 for (i = 0; i < 10; i++) {
1346 /* They ack this in play_dead by setting CPU_DEAD */
1347 if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1348 if (system_state == SYSTEM_RUNNING)
1349 pr_info("CPU %u is now offline\n", cpu);
1351 if (1 == num_online_cpus())
1352 alternatives_smp_switch(0);
1357 pr_err("CPU %u didn't die...\n", cpu);
1360 void play_dead_common(void)
1363 reset_lazy_tlbstate();
1364 irq_ctx_exit(raw_smp_processor_id());
1365 c1e_remove_cpu(raw_smp_processor_id());
1369 __get_cpu_var(cpu_state) = CPU_DEAD;
1372 * With physical CPU hotplug, we should halt the cpu
1374 local_irq_disable();
1377 void native_play_dead(void)
1380 tboot_shutdown(TB_SHUTDOWN_WFS);
1384 #else /* ... !CONFIG_HOTPLUG_CPU */
1385 int native_cpu_disable(void)
1390 void native_cpu_die(unsigned int cpu)
1392 /* We said "no" in __cpu_disable */
1396 void native_play_dead(void)