2 * x86 SMP booting functions
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.
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>
65 #include <asm/mwait.h>
67 #include <asm/io_apic.h>
68 #include <asm/setup.h>
69 #include <asm/uv/uv.h>
70 #include <linux/mc146818rtc.h>
72 #include <asm/smpboot_hooks.h>
73 #include <asm/i8259.h>
75 /* State of each CPU */
76 DEFINE_PER_CPU(int, cpu_state) = { 0 };
78 /* Store all idle threads, this can be reused instead of creating
79 * a new thread. Also avoids complicated thread destroy functionality
82 #ifdef CONFIG_HOTPLUG_CPU
84 * Needed only for CONFIG_HOTPLUG_CPU because __cpuinitdata is
85 * removed after init for !CONFIG_HOTPLUG_CPU.
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))
92 * We need this for trampoline_base protection from concurrent accesses when
93 * off- and onlining cores wildly.
95 static DEFINE_MUTEX(x86_cpu_hotplug_driver_mutex);
97 void cpu_hotplug_driver_lock(void)
99 mutex_lock(&x86_cpu_hotplug_driver_mutex);
102 void cpu_hotplug_driver_unlock(void)
104 mutex_unlock(&x86_cpu_hotplug_driver_mutex);
107 ssize_t arch_cpu_probe(const char *buf, size_t count) { return -1; }
108 ssize_t arch_cpu_release(const char *buf, size_t count) { return -1; }
110 static struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
111 #define get_idle_for_cpu(x) (idle_thread_array[(x)])
112 #define set_idle_for_cpu(x, p) (idle_thread_array[(x)] = (p))
115 /* Number of siblings per CPU package */
116 int smp_num_siblings = 1;
117 EXPORT_SYMBOL(smp_num_siblings);
119 /* Last level cache ID of each logical CPU */
120 DEFINE_PER_CPU(u16, cpu_llc_id) = BAD_APICID;
122 /* representing HT siblings of each logical CPU */
123 DEFINE_PER_CPU(cpumask_var_t, cpu_sibling_map);
124 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
126 /* representing HT and core siblings of each logical CPU */
127 DEFINE_PER_CPU(cpumask_var_t, cpu_core_map);
128 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
130 DEFINE_PER_CPU(cpumask_var_t, cpu_llc_shared_map);
132 /* Per CPU bogomips and other parameters */
133 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
134 EXPORT_PER_CPU_SYMBOL(cpu_info);
136 atomic_t init_deasserted;
139 * Report back to the Boot Processor.
142 static void __cpuinit smp_callin(void)
145 unsigned long timeout;
148 * If waken up by an INIT in an 82489DX configuration
149 * we may get here before an INIT-deassert IPI reaches
150 * our local APIC. We have to wait for the IPI or we'll
151 * lock up on an APIC access.
153 if (apic->wait_for_init_deassert)
154 apic->wait_for_init_deassert(&init_deasserted);
157 * (This works even if the APIC is not enabled.)
159 phys_id = read_apic_id();
160 cpuid = smp_processor_id();
161 if (cpumask_test_cpu(cpuid, cpu_callin_mask)) {
162 panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
165 pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
168 * STARTUP IPIs are fragile beasts as they might sometimes
169 * trigger some glue motherboard logic. Complete APIC bus
170 * silence for 1 second, this overestimates the time the
171 * boot CPU is spending to send the up to 2 STARTUP IPIs
172 * by a factor of two. This should be enough.
176 * Waiting 2s total for startup (udelay is not yet working)
178 timeout = jiffies + 2*HZ;
179 while (time_before(jiffies, timeout)) {
181 * Has the boot CPU finished it's STARTUP sequence?
183 if (cpumask_test_cpu(cpuid, cpu_callout_mask))
188 if (!time_before(jiffies, timeout)) {
189 panic("%s: CPU%d started up but did not get a callout!\n",
194 * the boot CPU has finished the init stage and is spinning
195 * on callin_map until we finish. We are free to set up this
196 * CPU, first the APIC. (this is probably redundant on most
200 pr_debug("CALLIN, before setup_local_APIC().\n");
201 if (apic->smp_callin_clear_local_apic)
202 apic->smp_callin_clear_local_apic();
204 end_local_APIC_setup();
207 * Need to setup vector mappings before we enable interrupts.
209 setup_vector_irq(smp_processor_id());
213 * Need to enable IRQs because it can take longer and then
214 * the NMI watchdog might kill us.
219 pr_debug("Stack at about %p\n", &cpuid);
222 * Save our processor parameters
224 smp_store_cpu_info(cpuid);
227 * This must be done before setting cpu_online_mask
228 * or calling notify_cpu_starting.
230 set_cpu_sibling_map(raw_smp_processor_id());
233 notify_cpu_starting(cpuid);
236 * Allow the master to continue.
238 cpumask_set_cpu(cpuid, cpu_callin_mask);
242 * Activate a secondary processor.
244 notrace static void __cpuinit start_secondary(void *unused)
247 * Don't put *anything* before cpu_init(), SMP booting is too
248 * fragile that we want to limit the things done here to the
249 * most necessary things.
256 /* switch away from the initial page table */
257 load_cr3(swapper_pg_dir);
261 /* otherwise gcc will move up smp_processor_id before the cpu_init */
264 * Check TSC synchronization with the BP:
266 check_tsc_sync_target();
269 * We need to hold call_lock, so there is no inconsistency
270 * between the time smp_call_function() determines number of
271 * IPI recipients, and the time when the determination is made
272 * for which cpus receive the IPI. Holding this
273 * lock helps us to not include this cpu in a currently in progress
274 * smp_call_function().
276 * We need to hold vector_lock so there the set of online cpus
277 * does not change while we are assigning vectors to cpus. Holding
278 * this lock ensures we don't half assign or remove an irq from a cpu.
282 set_cpu_online(smp_processor_id(), true);
283 unlock_vector_lock();
285 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
286 x86_platform.nmi_init();
288 /* enable local interrupts */
291 /* to prevent fake stack check failure in clock setup */
292 boot_init_stack_canary();
294 x86_cpuinit.setup_percpu_clockev();
301 * The bootstrap kernel entry code has set these up. Save them for
305 void __cpuinit smp_store_cpu_info(int id)
307 struct cpuinfo_x86 *c = &cpu_data(id);
312 identify_secondary_cpu(c);
315 static void __cpuinit check_cpu_siblings_on_same_node(int cpu1, int cpu2)
317 int node1 = early_cpu_to_node(cpu1);
318 int node2 = early_cpu_to_node(cpu2);
321 * Our CPU scheduler assumes all logical cpus in the same physical cpu
322 * share the same node. But, buggy ACPI or NUMA emulation might assign
323 * them to different node. Fix it.
325 if (node1 != node2) {
326 pr_warning("CPU %d in node %d and CPU %d in node %d are in the same physical CPU. forcing same node %d\n",
327 cpu1, node1, cpu2, node2, node2);
329 numa_remove_cpu(cpu1);
330 numa_set_node(cpu1, node2);
335 static void __cpuinit link_thread_siblings(int cpu1, int cpu2)
337 cpumask_set_cpu(cpu1, cpu_sibling_mask(cpu2));
338 cpumask_set_cpu(cpu2, cpu_sibling_mask(cpu1));
339 cpumask_set_cpu(cpu1, cpu_core_mask(cpu2));
340 cpumask_set_cpu(cpu2, cpu_core_mask(cpu1));
341 cpumask_set_cpu(cpu1, cpu_llc_shared_mask(cpu2));
342 cpumask_set_cpu(cpu2, cpu_llc_shared_mask(cpu1));
343 check_cpu_siblings_on_same_node(cpu1, cpu2);
347 void __cpuinit set_cpu_sibling_map(int cpu)
350 struct cpuinfo_x86 *c = &cpu_data(cpu);
352 cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
354 if (smp_num_siblings > 1) {
355 for_each_cpu(i, cpu_sibling_setup_mask) {
356 struct cpuinfo_x86 *o = &cpu_data(i);
358 if (cpu_has(c, X86_FEATURE_TOPOEXT)) {
359 if (c->phys_proc_id == o->phys_proc_id &&
360 per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i) &&
361 c->compute_unit_id == o->compute_unit_id)
362 link_thread_siblings(cpu, i);
363 } else if (c->phys_proc_id == o->phys_proc_id &&
364 c->cpu_core_id == o->cpu_core_id) {
365 link_thread_siblings(cpu, i);
369 cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
372 cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
374 if (__this_cpu_read(cpu_info.x86_max_cores) == 1) {
375 cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
380 for_each_cpu(i, cpu_sibling_setup_mask) {
381 if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
382 per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
383 cpumask_set_cpu(i, cpu_llc_shared_mask(cpu));
384 cpumask_set_cpu(cpu, cpu_llc_shared_mask(i));
385 check_cpu_siblings_on_same_node(cpu, i);
387 if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
388 cpumask_set_cpu(i, cpu_core_mask(cpu));
389 cpumask_set_cpu(cpu, cpu_core_mask(i));
390 check_cpu_siblings_on_same_node(cpu, i);
392 * Does this new cpu bringup a new core?
394 if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
396 * for each core in package, increment
397 * the booted_cores for this new cpu
399 if (cpumask_first(cpu_sibling_mask(i)) == i)
402 * increment the core count for all
403 * the other cpus in this package
406 cpu_data(i).booted_cores++;
407 } else if (i != cpu && !c->booted_cores)
408 c->booted_cores = cpu_data(i).booted_cores;
413 /* maps the cpu to the sched domain representing multi-core */
414 const struct cpumask *cpu_coregroup_mask(int cpu)
416 struct cpuinfo_x86 *c = &cpu_data(cpu);
418 * For perf, we return last level cache shared map.
419 * And for power savings, we return cpu_core_map
421 if ((sched_mc_power_savings || sched_smt_power_savings) &&
422 !(cpu_has(c, X86_FEATURE_AMD_DCM)))
423 return cpu_core_mask(cpu);
425 return cpu_llc_shared_mask(cpu);
428 static void impress_friends(void)
431 unsigned long bogosum = 0;
433 * Allow the user to impress friends.
435 pr_debug("Before bogomips.\n");
436 for_each_possible_cpu(cpu)
437 if (cpumask_test_cpu(cpu, cpu_callout_mask))
438 bogosum += cpu_data(cpu).loops_per_jiffy;
440 "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
443 (bogosum/(5000/HZ))%100);
445 pr_debug("Before bogocount - setting activated=1.\n");
448 void __inquire_remote_apic(int apicid)
450 unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
451 char *names[] = { "ID", "VERSION", "SPIV" };
455 printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
457 for (i = 0; i < ARRAY_SIZE(regs); i++) {
458 printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
463 status = safe_apic_wait_icr_idle();
466 "a previous APIC delivery may have failed\n");
468 apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
473 status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
474 } while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
477 case APIC_ICR_RR_VALID:
478 status = apic_read(APIC_RRR);
479 printk(KERN_CONT "%08x\n", status);
482 printk(KERN_CONT "failed\n");
488 * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
489 * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
490 * won't ... remember to clear down the APIC, etc later.
493 wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
495 unsigned long send_status, accept_status = 0;
499 /* Boot on the stack */
500 /* Kick the second */
501 apic_icr_write(APIC_DM_NMI | apic->dest_logical, logical_apicid);
503 pr_debug("Waiting for send to finish...\n");
504 send_status = safe_apic_wait_icr_idle();
507 * Give the other CPU some time to accept the IPI.
510 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
511 maxlvt = lapic_get_maxlvt();
512 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
513 apic_write(APIC_ESR, 0);
514 accept_status = (apic_read(APIC_ESR) & 0xEF);
516 pr_debug("NMI sent.\n");
519 printk(KERN_ERR "APIC never delivered???\n");
521 printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
523 return (send_status | accept_status);
527 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
529 unsigned long send_status, accept_status = 0;
530 int maxlvt, num_starts, j;
532 maxlvt = lapic_get_maxlvt();
535 * Be paranoid about clearing APIC errors.
537 if (APIC_INTEGRATED(apic_version[phys_apicid])) {
538 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
539 apic_write(APIC_ESR, 0);
543 pr_debug("Asserting INIT.\n");
546 * Turn INIT on target chip
551 apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
554 pr_debug("Waiting for send to finish...\n");
555 send_status = safe_apic_wait_icr_idle();
559 pr_debug("Deasserting INIT.\n");
563 apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
565 pr_debug("Waiting for send to finish...\n");
566 send_status = safe_apic_wait_icr_idle();
569 atomic_set(&init_deasserted, 1);
572 * Should we send STARTUP IPIs ?
574 * Determine this based on the APIC version.
575 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
577 if (APIC_INTEGRATED(apic_version[phys_apicid]))
583 * Paravirt / VMI wants a startup IPI hook here to set up the
584 * target processor state.
586 startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
590 * Run STARTUP IPI loop.
592 pr_debug("#startup loops: %d.\n", num_starts);
594 for (j = 1; j <= num_starts; j++) {
595 pr_debug("Sending STARTUP #%d.\n", j);
596 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
597 apic_write(APIC_ESR, 0);
599 pr_debug("After apic_write.\n");
606 /* Boot on the stack */
607 /* Kick the second */
608 apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
612 * Give the other CPU some time to accept the IPI.
616 pr_debug("Startup point 1.\n");
618 pr_debug("Waiting for send to finish...\n");
619 send_status = safe_apic_wait_icr_idle();
622 * Give the other CPU some time to accept the IPI.
625 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
626 apic_write(APIC_ESR, 0);
627 accept_status = (apic_read(APIC_ESR) & 0xEF);
628 if (send_status || accept_status)
631 pr_debug("After Startup.\n");
634 printk(KERN_ERR "APIC never delivered???\n");
636 printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
638 return (send_status | accept_status);
642 struct work_struct work;
643 struct task_struct *idle;
644 struct completion done;
648 static void __cpuinit do_fork_idle(struct work_struct *work)
650 struct create_idle *c_idle =
651 container_of(work, struct create_idle, work);
653 c_idle->idle = fork_idle(c_idle->cpu);
654 complete(&c_idle->done);
657 /* reduce the number of lines printed when booting a large cpu count system */
658 static void __cpuinit announce_cpu(int cpu, int apicid)
660 static int current_node = -1;
661 int node = early_cpu_to_node(cpu);
663 if (system_state == SYSTEM_BOOTING) {
664 if (node != current_node) {
665 if (current_node > (-1))
668 pr_info("Booting Node %3d, Processors ", node);
670 pr_cont(" #%d%s", cpu, cpu == (nr_cpu_ids - 1) ? " Ok.\n" : "");
673 pr_info("Booting Node %d Processor %d APIC 0x%x\n",
678 * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
679 * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
680 * Returns zero if CPU booted OK, else error code from
681 * ->wakeup_secondary_cpu.
683 static int __cpuinit do_boot_cpu(int apicid, int cpu)
685 unsigned long boot_error = 0;
686 unsigned long start_ip;
688 struct create_idle c_idle = {
690 .done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
693 INIT_WORK_ONSTACK(&c_idle.work, do_fork_idle);
695 alternatives_smp_switch(1);
697 c_idle.idle = get_idle_for_cpu(cpu);
700 * We can't use kernel_thread since we must avoid to
701 * reschedule the child.
704 c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
705 (THREAD_SIZE + task_stack_page(c_idle.idle))) - 1);
706 init_idle(c_idle.idle, cpu);
710 schedule_work(&c_idle.work);
711 wait_for_completion(&c_idle.done);
713 if (IS_ERR(c_idle.idle)) {
714 printk("failed fork for CPU %d\n", cpu);
715 destroy_work_on_stack(&c_idle.work);
716 return PTR_ERR(c_idle.idle);
719 set_idle_for_cpu(cpu, c_idle.idle);
721 per_cpu(current_task, cpu) = c_idle.idle;
723 /* Stack for startup_32 can be just as for start_secondary onwards */
726 clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
727 initial_gs = per_cpu_offset(cpu);
728 per_cpu(kernel_stack, cpu) =
729 (unsigned long)task_stack_page(c_idle.idle) -
730 KERNEL_STACK_OFFSET + THREAD_SIZE;
732 early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
733 initial_code = (unsigned long)start_secondary;
734 stack_start = c_idle.idle->thread.sp;
736 /* start_ip had better be page-aligned! */
737 start_ip = trampoline_address();
739 /* So we see what's up */
740 announce_cpu(cpu, apicid);
743 * This grunge runs the startup process for
744 * the targeted processor.
747 printk(KERN_DEBUG "smpboot cpu %d: start_ip = %lx\n", cpu, start_ip);
749 atomic_set(&init_deasserted, 0);
751 if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
753 pr_debug("Setting warm reset code and vector.\n");
755 smpboot_setup_warm_reset_vector(start_ip);
757 * Be paranoid about clearing APIC errors.
759 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
760 apic_write(APIC_ESR, 0);
766 * Kick the secondary CPU. Use the method in the APIC driver
767 * if it's defined - or use an INIT boot APIC message otherwise:
769 if (apic->wakeup_secondary_cpu)
770 boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
772 boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
776 * allow APs to start initializing.
778 pr_debug("Before Callout %d.\n", cpu);
779 cpumask_set_cpu(cpu, cpu_callout_mask);
780 pr_debug("After Callout %d.\n", cpu);
783 * Wait 5s total for a response
785 for (timeout = 0; timeout < 50000; timeout++) {
786 if (cpumask_test_cpu(cpu, cpu_callin_mask))
787 break; /* It has booted */
790 * Allow other tasks to run while we wait for the
791 * AP to come online. This also gives a chance
792 * for the MTRR work(triggered by the AP coming online)
793 * to be completed in the stop machine context.
798 if (cpumask_test_cpu(cpu, cpu_callin_mask))
799 pr_debug("CPU%d: has booted.\n", cpu);
802 if (*(volatile u32 *)TRAMPOLINE_SYM(trampoline_status)
804 /* trampoline started but...? */
805 pr_err("CPU%d: Stuck ??\n", cpu);
807 /* trampoline code not run */
808 pr_err("CPU%d: Not responding.\n", cpu);
809 if (apic->inquire_remote_apic)
810 apic->inquire_remote_apic(apicid);
815 /* Try to put things back the way they were before ... */
816 numa_remove_cpu(cpu); /* was set by numa_add_cpu */
818 /* was set by do_boot_cpu() */
819 cpumask_clear_cpu(cpu, cpu_callout_mask);
821 /* was set by cpu_init() */
822 cpumask_clear_cpu(cpu, cpu_initialized_mask);
824 set_cpu_present(cpu, false);
825 per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
828 /* mark "stuck" area as not stuck */
829 *(volatile u32 *)TRAMPOLINE_SYM(trampoline_status) = 0;
831 if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
833 * Cleanup possible dangling ends...
835 smpboot_restore_warm_reset_vector();
838 destroy_work_on_stack(&c_idle.work);
842 int __cpuinit native_cpu_up(unsigned int cpu)
844 int apicid = apic->cpu_present_to_apicid(cpu);
848 WARN_ON(irqs_disabled());
850 pr_debug("++++++++++++++++++++=_---CPU UP %u\n", cpu);
852 if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
853 !physid_isset(apicid, phys_cpu_present_map)) {
854 printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
859 * Already booted CPU?
861 if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
862 pr_debug("do_boot_cpu %d Already started\n", cpu);
867 * Save current MTRR state in case it was changed since early boot
868 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
872 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
874 err = do_boot_cpu(apicid, cpu);
876 pr_debug("do_boot_cpu failed %d\n", err);
881 * Check TSC synchronization with the AP (keep irqs disabled
884 local_irq_save(flags);
885 check_tsc_sync_source(cpu);
886 local_irq_restore(flags);
888 while (!cpu_online(cpu)) {
890 touch_nmi_watchdog();
897 * arch_disable_smp_support() - disables SMP support for x86 at runtime
899 void arch_disable_smp_support(void)
901 disable_ioapic_support();
905 * Fall back to non SMP mode after errors.
907 * RED-PEN audit/test this more. I bet there is more state messed up here.
909 static __init void disable_smp(void)
911 init_cpu_present(cpumask_of(0));
912 init_cpu_possible(cpumask_of(0));
913 smpboot_clear_io_apic_irqs();
915 if (smp_found_config)
916 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
918 physid_set_mask_of_physid(0, &phys_cpu_present_map);
919 cpumask_set_cpu(0, cpu_sibling_mask(0));
920 cpumask_set_cpu(0, cpu_core_mask(0));
924 * Various sanity checks.
926 static int __init smp_sanity_check(unsigned max_cpus)
930 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
931 if (def_to_bigsmp && nr_cpu_ids > 8) {
936 "More than 8 CPUs detected - skipping them.\n"
937 "Use CONFIG_X86_BIGSMP.\n");
940 for_each_present_cpu(cpu) {
942 set_cpu_present(cpu, false);
947 for_each_possible_cpu(cpu) {
949 set_cpu_possible(cpu, false);
957 if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
959 "weird, boot CPU (#%d) not listed by the BIOS.\n",
960 hard_smp_processor_id());
962 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
966 * If we couldn't find an SMP configuration at boot time,
967 * get out of here now!
969 if (!smp_found_config && !acpi_lapic) {
971 printk(KERN_NOTICE "SMP motherboard not detected.\n");
973 if (APIC_init_uniprocessor())
974 printk(KERN_NOTICE "Local APIC not detected."
975 " Using dummy APIC emulation.\n");
980 * Should not be necessary because the MP table should list the boot
981 * CPU too, but we do it for the sake of robustness anyway.
983 if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
985 "weird, boot CPU (#%d) not listed by the BIOS.\n",
986 boot_cpu_physical_apicid);
987 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
992 * If we couldn't find a local APIC, then get out of here now!
994 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
997 pr_err("BIOS bug, local APIC #%d not detected!...\n",
998 boot_cpu_physical_apicid);
999 pr_err("... forcing use of dummy APIC emulation."
1000 "(tell your hw vendor)\n");
1002 smpboot_clear_io_apic();
1003 disable_ioapic_support();
1007 verify_local_APIC();
1010 * If SMP should be disabled, then really disable it!
1013 printk(KERN_INFO "SMP mode deactivated.\n");
1014 smpboot_clear_io_apic();
1018 bsp_end_local_APIC_setup();
1025 static void __init smp_cpu_index_default(void)
1028 struct cpuinfo_x86 *c;
1030 for_each_possible_cpu(i) {
1032 /* mark all to hotplug */
1033 c->cpu_index = nr_cpu_ids;
1038 * Prepare for SMP bootup. The MP table or ACPI has been read
1039 * earlier. Just do some sanity checking here and enable APIC mode.
1041 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1046 smp_cpu_index_default();
1049 * Setup boot CPU information
1051 smp_store_cpu_info(0); /* Final full version of the data */
1052 cpumask_copy(cpu_callin_mask, cpumask_of(0));
1055 current_thread_info()->cpu = 0; /* needed? */
1056 for_each_possible_cpu(i) {
1057 zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1058 zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1059 zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1061 set_cpu_sibling_map(0);
1064 if (smp_sanity_check(max_cpus) < 0) {
1065 printk(KERN_INFO "SMP disabled\n");
1070 default_setup_apic_routing();
1073 if (read_apic_id() != boot_cpu_physical_apicid) {
1074 panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1075 read_apic_id(), boot_cpu_physical_apicid);
1076 /* Or can we switch back to PIC here? */
1083 * Switch from PIC to APIC mode.
1088 * Enable IO APIC before setting up error vector
1090 if (!skip_ioapic_setup && nr_ioapics)
1093 bsp_end_local_APIC_setup();
1095 if (apic->setup_portio_remap)
1096 apic->setup_portio_remap();
1098 smpboot_setup_io_apic();
1100 * Set up local APIC timer on boot CPU.
1103 printk(KERN_INFO "CPU%d: ", 0);
1104 print_cpu_info(&cpu_data(0));
1105 x86_init.timers.setup_percpu_clockev();
1110 set_mtrr_aps_delayed_init();
1115 void arch_disable_nonboot_cpus_begin(void)
1118 * Avoid the smp alternatives switch during the disable_nonboot_cpus().
1119 * In the suspend path, we will be back in the SMP mode shortly anyways.
1121 skip_smp_alternatives = true;
1124 void arch_disable_nonboot_cpus_end(void)
1126 skip_smp_alternatives = false;
1129 void arch_enable_nonboot_cpus_begin(void)
1131 set_mtrr_aps_delayed_init();
1134 void arch_enable_nonboot_cpus_end(void)
1140 * Early setup to make printk work.
1142 void __init native_smp_prepare_boot_cpu(void)
1144 int me = smp_processor_id();
1145 switch_to_new_gdt(me);
1146 /* already set me in cpu_online_mask in boot_cpu_init() */
1147 cpumask_set_cpu(me, cpu_callout_mask);
1148 per_cpu(cpu_state, me) = CPU_ONLINE;
1151 void __init native_smp_cpus_done(unsigned int max_cpus)
1153 pr_debug("Boot done.\n");
1156 #ifdef CONFIG_X86_IO_APIC
1157 setup_ioapic_dest();
1162 static int __initdata setup_possible_cpus = -1;
1163 static int __init _setup_possible_cpus(char *str)
1165 get_option(&str, &setup_possible_cpus);
1168 early_param("possible_cpus", _setup_possible_cpus);
1172 * cpu_possible_mask should be static, it cannot change as cpu's
1173 * are onlined, or offlined. The reason is per-cpu data-structures
1174 * are allocated by some modules at init time, and dont expect to
1175 * do this dynamically on cpu arrival/departure.
1176 * cpu_present_mask on the other hand can change dynamically.
1177 * In case when cpu_hotplug is not compiled, then we resort to current
1178 * behaviour, which is cpu_possible == cpu_present.
1181 * Three ways to find out the number of additional hotplug CPUs:
1182 * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1183 * - The user can overwrite it with possible_cpus=NUM
1184 * - Otherwise don't reserve additional CPUs.
1185 * We do this because additional CPUs waste a lot of memory.
1188 __init void prefill_possible_map(void)
1192 /* no processor from mptable or madt */
1193 if (!num_processors)
1196 i = setup_max_cpus ?: 1;
1197 if (setup_possible_cpus == -1) {
1198 possible = num_processors;
1199 #ifdef CONFIG_HOTPLUG_CPU
1201 possible += disabled_cpus;
1207 possible = setup_possible_cpus;
1209 total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1211 /* nr_cpu_ids could be reduced via nr_cpus= */
1212 if (possible > nr_cpu_ids) {
1214 "%d Processors exceeds NR_CPUS limit of %d\n",
1215 possible, nr_cpu_ids);
1216 possible = nr_cpu_ids;
1219 #ifdef CONFIG_HOTPLUG_CPU
1220 if (!setup_max_cpus)
1224 "%d Processors exceeds max_cpus limit of %u\n",
1225 possible, setup_max_cpus);
1229 printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1230 possible, max_t(int, possible - num_processors, 0));
1232 for (i = 0; i < possible; i++)
1233 set_cpu_possible(i, true);
1234 for (; i < NR_CPUS; i++)
1235 set_cpu_possible(i, false);
1237 nr_cpu_ids = possible;
1240 #ifdef CONFIG_HOTPLUG_CPU
1242 static void remove_siblinginfo(int cpu)
1245 struct cpuinfo_x86 *c = &cpu_data(cpu);
1247 for_each_cpu(sibling, cpu_core_mask(cpu)) {
1248 cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1250 * last thread sibling in this cpu core going down
1252 if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1253 cpu_data(sibling).booted_cores--;
1256 for_each_cpu(sibling, cpu_sibling_mask(cpu))
1257 cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1258 cpumask_clear(cpu_sibling_mask(cpu));
1259 cpumask_clear(cpu_core_mask(cpu));
1260 c->phys_proc_id = 0;
1262 cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1265 static void __ref remove_cpu_from_maps(int cpu)
1267 set_cpu_online(cpu, false);
1268 cpumask_clear_cpu(cpu, cpu_callout_mask);
1269 cpumask_clear_cpu(cpu, cpu_callin_mask);
1270 /* was set by cpu_init() */
1271 cpumask_clear_cpu(cpu, cpu_initialized_mask);
1272 numa_remove_cpu(cpu);
1275 void cpu_disable_common(void)
1277 int cpu = smp_processor_id();
1279 remove_siblinginfo(cpu);
1281 /* It's now safe to remove this processor from the online map */
1283 remove_cpu_from_maps(cpu);
1284 unlock_vector_lock();
1288 int native_cpu_disable(void)
1290 int cpu = smp_processor_id();
1293 * Perhaps use cpufreq to drop frequency, but that could go
1294 * into generic code.
1296 * We won't take down the boot processor on i386 due to some
1297 * interrupts only being able to be serviced by the BSP.
1298 * Especially so if we're not using an IOAPIC -zwane
1305 cpu_disable_common();
1309 void native_cpu_die(unsigned int cpu)
1311 /* We don't do anything here: idle task is faking death itself. */
1314 for (i = 0; i < 10; i++) {
1315 /* They ack this in play_dead by setting CPU_DEAD */
1316 if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1317 if (system_state == SYSTEM_RUNNING)
1318 pr_info("CPU %u is now offline\n", cpu);
1320 if (1 == num_online_cpus())
1321 alternatives_smp_switch(0);
1326 pr_err("CPU %u didn't die...\n", cpu);
1329 void play_dead_common(void)
1332 reset_lazy_tlbstate();
1333 c1e_remove_cpu(raw_smp_processor_id());
1337 __this_cpu_write(cpu_state, CPU_DEAD);
1340 * With physical CPU hotplug, we should halt the cpu
1342 local_irq_disable();
1346 * We need to flush the caches before going to sleep, lest we have
1347 * dirty data in our caches when we come back up.
1349 static inline void mwait_play_dead(void)
1351 unsigned int eax, ebx, ecx, edx;
1352 unsigned int highest_cstate = 0;
1353 unsigned int highest_subcstate = 0;
1356 struct cpuinfo_x86 *c = __this_cpu_ptr(&cpu_info);
1358 if (!(cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)))
1360 if (!cpu_has(__this_cpu_ptr(&cpu_info), X86_FEATURE_CLFLSH))
1362 if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1365 eax = CPUID_MWAIT_LEAF;
1367 native_cpuid(&eax, &ebx, &ecx, &edx);
1370 * eax will be 0 if EDX enumeration is not valid.
1371 * Initialized below to cstate, sub_cstate value when EDX is valid.
1373 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1376 edx >>= MWAIT_SUBSTATE_SIZE;
1377 for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1378 if (edx & MWAIT_SUBSTATE_MASK) {
1380 highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1383 eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1384 (highest_subcstate - 1);
1388 * This should be a memory location in a cache line which is
1389 * unlikely to be touched by other processors. The actual
1390 * content is immaterial as it is not actually modified in any way.
1392 mwait_ptr = ¤t_thread_info()->flags;
1398 * The CLFLUSH is a workaround for erratum AAI65 for
1399 * the Xeon 7400 series. It's not clear it is actually
1400 * needed, but it should be harmless in either case.
1401 * The WBINVD is insufficient due to the spurious-wakeup
1402 * case where we return around the loop.
1405 __monitor(mwait_ptr, 0, 0);
1411 static inline void hlt_play_dead(void)
1413 if (__this_cpu_read(cpu_info.x86) >= 4)
1421 void native_play_dead(void)
1424 tboot_shutdown(TB_SHUTDOWN_WFS);
1426 mwait_play_dead(); /* Only returns on failure */
1430 #else /* ... !CONFIG_HOTPLUG_CPU */
1431 int native_cpu_disable(void)
1436 void native_cpu_die(unsigned int cpu)
1438 /* We said "no" in __cpu_disable */
1442 void native_play_dead(void)