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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
44 #include <linux/init.h>
45 #include <linux/smp.h>
46 #include <linux/module.h>
47 #include <linux/sched.h>
48 #include <linux/percpu.h>
49 #include <linux/bootmem.h>
50 #include <linux/err.h>
51 #include <linux/nmi.h>
52 #include <linux/tboot.h>
53 #include <linux/stackprotector.h>
54 #include <linux/gfp.h>
55 #include <linux/cpuidle.h>
62 #include <asm/realmode.h>
65 #include <asm/pgtable.h>
66 #include <asm/tlbflush.h>
68 #include <asm/mwait.h>
70 #include <asm/io_apic.h>
72 #include <asm/fpu-internal.h>
73 #include <asm/setup.h>
74 #include <asm/uv/uv.h>
75 #include <linux/mc146818rtc.h>
76 #include <asm/smpboot_hooks.h>
77 #include <asm/i8259.h>
78 #include <asm/realmode.h>
81 /* State of each CPU */
82 DEFINE_PER_CPU(int, cpu_state) = { 0 };
84 /* Number of siblings per CPU package */
85 int smp_num_siblings = 1;
86 EXPORT_SYMBOL(smp_num_siblings);
88 /* Last level cache ID of each logical CPU */
89 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
91 /* representing HT siblings of each logical CPU */
92 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
93 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
95 /* representing HT and core siblings of each logical CPU */
96 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_map);
97 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
99 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_llc_shared_map);
101 /* Per CPU bogomips and other parameters */
102 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
103 EXPORT_PER_CPU_SYMBOL(cpu_info);
105 atomic_t init_deasserted;
108 * Report back to the Boot Processor during boot time or to the caller processor
111 static void smp_callin(void)
114 unsigned long timeout;
117 * If waken up by an INIT in an 82489DX configuration
118 * we may get here before an INIT-deassert IPI reaches
119 * our local APIC. We have to wait for the IPI or we'll
120 * lock up on an APIC access.
122 * Since CPU0 is not wakened up by INIT, it doesn't wait for the IPI.
124 cpuid = smp_processor_id();
125 if (apic->wait_for_init_deassert && cpuid != 0)
126 apic->wait_for_init_deassert(&init_deasserted);
129 * (This works even if the APIC is not enabled.)
131 phys_id = read_apic_id();
132 if (cpumask_test_cpu(cpuid, cpu_callin_mask)) {
133 panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
136 pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
139 * STARTUP IPIs are fragile beasts as they might sometimes
140 * trigger some glue motherboard logic. Complete APIC bus
141 * silence for 1 second, this overestimates the time the
142 * boot CPU is spending to send the up to 2 STARTUP IPIs
143 * by a factor of two. This should be enough.
147 * Waiting 2s total for startup (udelay is not yet working)
149 timeout = jiffies + 2*HZ;
150 while (time_before(jiffies, timeout)) {
152 * Has the boot CPU finished it's STARTUP sequence?
154 if (cpumask_test_cpu(cpuid, cpu_callout_mask))
159 if (!time_before(jiffies, timeout)) {
160 panic("%s: CPU%d started up but did not get a callout!\n",
165 * the boot CPU has finished the init stage and is spinning
166 * on callin_map until we finish. We are free to set up this
167 * CPU, first the APIC. (this is probably redundant on most
171 pr_debug("CALLIN, before setup_local_APIC()\n");
172 if (apic->smp_callin_clear_local_apic)
173 apic->smp_callin_clear_local_apic();
175 end_local_APIC_setup();
178 * Need to setup vector mappings before we enable interrupts.
180 setup_vector_irq(smp_processor_id());
183 * Save our processor parameters. Note: this information
184 * is needed for clock calibration.
186 smp_store_cpu_info(cpuid);
190 * Update loops_per_jiffy in cpu_data. Previous call to
191 * smp_store_cpu_info() stored a value that is close but not as
192 * accurate as the value just calculated.
195 cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
196 pr_debug("Stack at about %p\n", &cpuid);
199 * This must be done before setting cpu_online_mask
200 * or calling notify_cpu_starting.
202 set_cpu_sibling_map(raw_smp_processor_id());
205 notify_cpu_starting(cpuid);
208 * Allow the master to continue.
210 cpumask_set_cpu(cpuid, cpu_callin_mask);
213 static int cpu0_logical_apicid;
214 static int enable_start_cpu0;
216 * Activate a secondary processor.
218 static void notrace start_secondary(void *unused)
221 * Don't put *anything* before cpu_init(), SMP booting is too
222 * fragile that we want to limit the things done here to the
223 * most necessary things.
226 x86_cpuinit.early_percpu_clock_init();
230 enable_start_cpu0 = 0;
233 /* switch away from the initial page table */
234 load_cr3(swapper_pg_dir);
238 /* otherwise gcc will move up smp_processor_id before the cpu_init */
241 * Check TSC synchronization with the BP:
243 check_tsc_sync_target();
246 * Enable the espfix hack for this CPU
248 #ifdef CONFIG_X86_ESPFIX64
253 * We need to hold vector_lock so there the set of online cpus
254 * does not change while we are assigning vectors to cpus. Holding
255 * this lock ensures we don't half assign or remove an irq from a cpu.
258 set_cpu_online(smp_processor_id(), true);
259 unlock_vector_lock();
260 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
261 x86_platform.nmi_init();
263 /* enable local interrupts */
266 /* to prevent fake stack check failure in clock setup */
267 boot_init_stack_canary();
269 x86_cpuinit.setup_percpu_clockev();
272 cpu_startup_entry(CPUHP_ONLINE);
275 void __init smp_store_boot_cpu_info(void)
277 int id = 0; /* CPU 0 */
278 struct cpuinfo_x86 *c = &cpu_data(id);
285 * The bootstrap kernel entry code has set these up. Save them for
288 void smp_store_cpu_info(int id)
290 struct cpuinfo_x86 *c = &cpu_data(id);
295 * During boot time, CPU0 has this setup already. Save the info when
296 * bringing up AP or offlined CPU0.
298 identify_secondary_cpu(c);
302 topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
304 int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
306 return !WARN_ONCE(cpu_to_node(cpu1) != cpu_to_node(cpu2),
307 "sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
308 "[node: %d != %d]. Ignoring dependency.\n",
309 cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
312 #define link_mask(_m, c1, c2) \
314 cpumask_set_cpu((c1), cpu_##_m##_mask(c2)); \
315 cpumask_set_cpu((c2), cpu_##_m##_mask(c1)); \
318 static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
320 if (cpu_has_topoext) {
321 int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
323 if (c->phys_proc_id == o->phys_proc_id &&
324 per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2) &&
325 c->compute_unit_id == o->compute_unit_id)
326 return topology_sane(c, o, "smt");
328 } else if (c->phys_proc_id == o->phys_proc_id &&
329 c->cpu_core_id == o->cpu_core_id) {
330 return topology_sane(c, o, "smt");
336 static bool match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
338 int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
340 if (per_cpu(cpu_llc_id, cpu1) != BAD_APICID &&
341 per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2))
342 return topology_sane(c, o, "llc");
347 static bool match_mc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
349 if (c->phys_proc_id == o->phys_proc_id) {
350 if (cpu_has(c, X86_FEATURE_AMD_DCM))
353 return topology_sane(c, o, "mc");
358 void set_cpu_sibling_map(int cpu)
360 bool has_smt = smp_num_siblings > 1;
361 bool has_mp = has_smt || boot_cpu_data.x86_max_cores > 1;
362 struct cpuinfo_x86 *c = &cpu_data(cpu);
363 struct cpuinfo_x86 *o;
366 cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
369 cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
370 cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
371 cpumask_set_cpu(cpu, cpu_core_mask(cpu));
376 for_each_cpu(i, cpu_sibling_setup_mask) {
379 if ((i == cpu) || (has_smt && match_smt(c, o)))
380 link_mask(sibling, cpu, i);
382 if ((i == cpu) || (has_mp && match_llc(c, o)))
383 link_mask(llc_shared, cpu, i);
388 * This needs a separate iteration over the cpus because we rely on all
389 * cpu_sibling_mask links to be set-up.
391 for_each_cpu(i, cpu_sibling_setup_mask) {
394 if ((i == cpu) || (has_mp && match_mc(c, o))) {
395 link_mask(core, cpu, i);
398 * Does this new cpu bringup a new core?
400 if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
402 * for each core in package, increment
403 * the booted_cores for this new cpu
405 if (cpumask_first(cpu_sibling_mask(i)) == i)
408 * increment the core count for all
409 * the other cpus in this package
412 cpu_data(i).booted_cores++;
413 } else if (i != cpu && !c->booted_cores)
414 c->booted_cores = cpu_data(i).booted_cores;
419 /* maps the cpu to the sched domain representing multi-core */
420 const struct cpumask *cpu_coregroup_mask(int cpu)
422 return cpu_llc_shared_mask(cpu);
425 static void impress_friends(void)
428 unsigned long bogosum = 0;
430 * Allow the user to impress friends.
432 pr_debug("Before bogomips\n");
433 for_each_possible_cpu(cpu)
434 if (cpumask_test_cpu(cpu, cpu_callout_mask))
435 bogosum += cpu_data(cpu).loops_per_jiffy;
436 pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
439 (bogosum/(5000/HZ))%100);
441 pr_debug("Before bogocount - setting activated=1\n");
444 void __inquire_remote_apic(int apicid)
446 unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
447 const char * const names[] = { "ID", "VERSION", "SPIV" };
451 pr_info("Inquiring remote APIC 0x%x...\n", apicid);
453 for (i = 0; i < ARRAY_SIZE(regs); i++) {
454 pr_info("... APIC 0x%x %s: ", apicid, names[i]);
459 status = safe_apic_wait_icr_idle();
461 pr_cont("a previous APIC delivery may have failed\n");
463 apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
468 status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
469 } while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
472 case APIC_ICR_RR_VALID:
473 status = apic_read(APIC_RRR);
474 pr_cont("%08x\n", status);
483 * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
484 * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
485 * won't ... remember to clear down the APIC, etc later.
488 wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
490 unsigned long send_status, accept_status = 0;
494 /* Boot on the stack */
495 /* Kick the second */
496 apic_icr_write(APIC_DM_NMI | apic->dest_logical, apicid);
498 pr_debug("Waiting for send to finish...\n");
499 send_status = safe_apic_wait_icr_idle();
502 * Give the other CPU some time to accept the IPI.
505 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
506 maxlvt = lapic_get_maxlvt();
507 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
508 apic_write(APIC_ESR, 0);
509 accept_status = (apic_read(APIC_ESR) & 0xEF);
511 pr_debug("NMI sent\n");
514 pr_err("APIC never delivered???\n");
516 pr_err("APIC delivery error (%lx)\n", accept_status);
518 return (send_status | accept_status);
522 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
524 unsigned long send_status, accept_status = 0;
525 int maxlvt, num_starts, j;
527 maxlvt = lapic_get_maxlvt();
530 * Be paranoid about clearing APIC errors.
532 if (APIC_INTEGRATED(apic_version[phys_apicid])) {
533 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
534 apic_write(APIC_ESR, 0);
538 pr_debug("Asserting INIT\n");
541 * Turn INIT on target chip
546 apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
549 pr_debug("Waiting for send to finish...\n");
550 send_status = safe_apic_wait_icr_idle();
554 pr_debug("Deasserting INIT\n");
558 apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
560 pr_debug("Waiting for send to finish...\n");
561 send_status = safe_apic_wait_icr_idle();
564 atomic_set(&init_deasserted, 1);
567 * Should we send STARTUP IPIs ?
569 * Determine this based on the APIC version.
570 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
572 if (APIC_INTEGRATED(apic_version[phys_apicid]))
578 * Paravirt / VMI wants a startup IPI hook here to set up the
579 * target processor state.
581 startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
585 * Run STARTUP IPI loop.
587 pr_debug("#startup loops: %d\n", num_starts);
589 for (j = 1; j <= num_starts; j++) {
590 pr_debug("Sending STARTUP #%d\n", j);
591 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
592 apic_write(APIC_ESR, 0);
594 pr_debug("After apic_write\n");
601 /* Boot on the stack */
602 /* Kick the second */
603 apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
607 * Give the other CPU some time to accept the IPI.
611 pr_debug("Startup point 1\n");
613 pr_debug("Waiting for send to finish...\n");
614 send_status = safe_apic_wait_icr_idle();
617 * Give the other CPU some time to accept the IPI.
620 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
621 apic_write(APIC_ESR, 0);
622 accept_status = (apic_read(APIC_ESR) & 0xEF);
623 if (send_status || accept_status)
626 pr_debug("After Startup\n");
629 pr_err("APIC never delivered???\n");
631 pr_err("APIC delivery error (%lx)\n", accept_status);
633 return (send_status | accept_status);
636 void smp_announce(void)
638 int num_nodes = num_online_nodes();
640 printk(KERN_INFO "x86: Booted up %d node%s, %d CPUs\n",
641 num_nodes, (num_nodes > 1 ? "s" : ""), num_online_cpus());
644 /* reduce the number of lines printed when booting a large cpu count system */
645 static void announce_cpu(int cpu, int apicid)
647 static int current_node = -1;
648 int node = early_cpu_to_node(cpu);
649 static int width, node_width;
652 width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
655 node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
658 printk(KERN_INFO "x86: Booting SMP configuration:\n");
660 if (system_state == SYSTEM_BOOTING) {
661 if (node != current_node) {
662 if (current_node > (-1))
666 printk(KERN_INFO ".... node %*s#%d, CPUs: ",
667 node_width - num_digits(node), " ", node);
670 /* Add padding for the BSP */
672 pr_cont("%*s", width + 1, " ");
674 pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
677 pr_info("Booting Node %d Processor %d APIC 0x%x\n",
681 static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
685 cpu = smp_processor_id();
686 if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
693 * Wake up AP by INIT, INIT, STARTUP sequence.
695 * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
696 * boot-strap code which is not a desired behavior for waking up BSP. To
697 * void the boot-strap code, wake up CPU0 by NMI instead.
699 * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
700 * (i.e. physically hot removed and then hot added), NMI won't wake it up.
701 * We'll change this code in the future to wake up hard offlined CPU0 if
702 * real platform and request are available.
705 wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
706 int *cpu0_nmi_registered)
712 * Wake up AP by INIT, INIT, STARTUP sequence.
715 return wakeup_secondary_cpu_via_init(apicid, start_ip);
718 * Wake up BSP by nmi.
720 * Register a NMI handler to help wake up CPU0.
722 boot_error = register_nmi_handler(NMI_LOCAL,
723 wakeup_cpu0_nmi, 0, "wake_cpu0");
726 enable_start_cpu0 = 1;
727 *cpu0_nmi_registered = 1;
728 if (apic->dest_logical == APIC_DEST_LOGICAL)
729 id = cpu0_logical_apicid;
732 boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
739 * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
740 * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
741 * Returns zero if CPU booted OK, else error code from
742 * ->wakeup_secondary_cpu.
744 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
746 volatile u32 *trampoline_status =
747 (volatile u32 *) __va(real_mode_header->trampoline_status);
748 /* start_ip had better be page-aligned! */
749 unsigned long start_ip = real_mode_header->trampoline_start;
751 unsigned long boot_error = 0;
753 int cpu0_nmi_registered = 0;
755 /* Just in case we booted with a single CPU. */
756 alternatives_enable_smp();
758 idle->thread.sp = (unsigned long) (((struct pt_regs *)
759 (THREAD_SIZE + task_stack_page(idle))) - 1);
760 per_cpu(current_task, cpu) = idle;
763 /* Stack for startup_32 can be just as for start_secondary onwards */
766 clear_tsk_thread_flag(idle, TIF_FORK);
767 initial_gs = per_cpu_offset(cpu);
768 per_cpu(kernel_stack, cpu) =
769 (unsigned long)task_stack_page(idle) -
770 KERNEL_STACK_OFFSET + THREAD_SIZE;
772 early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
773 initial_code = (unsigned long)start_secondary;
774 stack_start = idle->thread.sp;
776 /* So we see what's up */
777 announce_cpu(cpu, apicid);
780 * This grunge runs the startup process for
781 * the targeted processor.
784 atomic_set(&init_deasserted, 0);
786 if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
788 pr_debug("Setting warm reset code and vector.\n");
790 smpboot_setup_warm_reset_vector(start_ip);
792 * Be paranoid about clearing APIC errors.
794 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
795 apic_write(APIC_ESR, 0);
801 * Wake up a CPU in difference cases:
802 * - Use the method in the APIC driver if it's defined
804 * - Use an INIT boot APIC message for APs or NMI for BSP.
806 if (apic->wakeup_secondary_cpu)
807 boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
809 boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
810 &cpu0_nmi_registered);
814 * allow APs to start initializing.
816 pr_debug("Before Callout %d\n", cpu);
817 cpumask_set_cpu(cpu, cpu_callout_mask);
818 pr_debug("After Callout %d\n", cpu);
821 * Wait 5s total for a response
823 for (timeout = 0; timeout < 50000; timeout++) {
824 if (cpumask_test_cpu(cpu, cpu_callin_mask))
825 break; /* It has booted */
828 * Allow other tasks to run while we wait for the
829 * AP to come online. This also gives a chance
830 * for the MTRR work(triggered by the AP coming online)
831 * to be completed in the stop machine context.
836 if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
837 print_cpu_msr(&cpu_data(cpu));
838 pr_debug("CPU%d: has booted.\n", cpu);
841 if (*trampoline_status == 0xA5A5A5A5)
842 /* trampoline started but...? */
843 pr_err("CPU%d: Stuck ??\n", cpu);
845 /* trampoline code not run */
846 pr_err("CPU%d: Not responding\n", cpu);
847 if (apic->inquire_remote_apic)
848 apic->inquire_remote_apic(apicid);
853 /* Try to put things back the way they were before ... */
854 numa_remove_cpu(cpu); /* was set by numa_add_cpu */
856 /* was set by do_boot_cpu() */
857 cpumask_clear_cpu(cpu, cpu_callout_mask);
859 /* was set by cpu_init() */
860 cpumask_clear_cpu(cpu, cpu_initialized_mask);
862 set_cpu_present(cpu, false);
863 per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
866 /* mark "stuck" area as not stuck */
867 *trampoline_status = 0;
869 if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
871 * Cleanup possible dangling ends...
873 smpboot_restore_warm_reset_vector();
876 * Clean up the nmi handler. Do this after the callin and callout sync
877 * to avoid impact of possible long unregister time.
879 if (cpu0_nmi_registered)
880 unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
885 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
887 int apicid = apic->cpu_present_to_apicid(cpu);
891 WARN_ON(irqs_disabled());
893 pr_debug("++++++++++++++++++++=_---CPU UP %u\n", cpu);
895 if (apicid == BAD_APICID ||
896 !physid_isset(apicid, phys_cpu_present_map) ||
897 !apic->apic_id_valid(apicid)) {
898 pr_err("%s: bad cpu %d\n", __func__, cpu);
903 * Already booted CPU?
905 if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
906 pr_debug("do_boot_cpu %d Already started\n", cpu);
911 * Save current MTRR state in case it was changed since early boot
912 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
916 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
918 /* the FPU context is blank, nobody can own it */
919 __cpu_disable_lazy_restore(cpu);
921 err = do_boot_cpu(apicid, cpu, tidle);
923 pr_debug("do_boot_cpu failed %d\n", err);
928 * Check TSC synchronization with the AP (keep irqs disabled
931 local_irq_save(flags);
932 check_tsc_sync_source(cpu);
933 local_irq_restore(flags);
935 while (!cpu_online(cpu)) {
937 touch_nmi_watchdog();
944 * arch_disable_smp_support() - disables SMP support for x86 at runtime
946 void arch_disable_smp_support(void)
948 disable_ioapic_support();
952 * Fall back to non SMP mode after errors.
954 * RED-PEN audit/test this more. I bet there is more state messed up here.
956 static __init void disable_smp(void)
958 init_cpu_present(cpumask_of(0));
959 init_cpu_possible(cpumask_of(0));
960 smpboot_clear_io_apic_irqs();
962 if (smp_found_config)
963 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
965 physid_set_mask_of_physid(0, &phys_cpu_present_map);
966 cpumask_set_cpu(0, cpu_sibling_mask(0));
967 cpumask_set_cpu(0, cpu_core_mask(0));
971 * Various sanity checks.
973 static int __init smp_sanity_check(unsigned max_cpus)
977 #if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
978 if (def_to_bigsmp && nr_cpu_ids > 8) {
982 pr_warn("More than 8 CPUs detected - skipping them\n"
983 "Use CONFIG_X86_BIGSMP\n");
986 for_each_present_cpu(cpu) {
988 set_cpu_present(cpu, false);
993 for_each_possible_cpu(cpu) {
995 set_cpu_possible(cpu, false);
1003 if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1004 pr_warn("weird, boot CPU (#%d) not listed by the BIOS\n",
1005 hard_smp_processor_id());
1007 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1011 * If we couldn't find an SMP configuration at boot time,
1012 * get out of here now!
1014 if (!smp_found_config && !acpi_lapic) {
1016 pr_notice("SMP motherboard not detected\n");
1018 if (APIC_init_uniprocessor())
1019 pr_notice("Local APIC not detected. Using dummy APIC emulation.\n");
1024 * Should not be necessary because the MP table should list the boot
1025 * CPU too, but we do it for the sake of robustness anyway.
1027 if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
1028 pr_notice("weird, boot CPU (#%d) not listed by the BIOS\n",
1029 boot_cpu_physical_apicid);
1030 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1035 * If we couldn't find a local APIC, then get out of here now!
1037 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1039 if (!disable_apic) {
1040 pr_err("BIOS bug, local APIC #%d not detected!...\n",
1041 boot_cpu_physical_apicid);
1042 pr_err("... forcing use of dummy APIC emulation (tell your hw vendor)\n");
1044 smpboot_clear_io_apic();
1045 disable_ioapic_support();
1049 verify_local_APIC();
1052 * If SMP should be disabled, then really disable it!
1055 pr_info("SMP mode deactivated\n");
1056 smpboot_clear_io_apic();
1060 bsp_end_local_APIC_setup();
1067 static void __init smp_cpu_index_default(void)
1070 struct cpuinfo_x86 *c;
1072 for_each_possible_cpu(i) {
1074 /* mark all to hotplug */
1075 c->cpu_index = nr_cpu_ids;
1080 * Prepare for SMP bootup. The MP table or ACPI has been read
1081 * earlier. Just do some sanity checking here and enable APIC mode.
1083 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1088 smp_cpu_index_default();
1091 * Setup boot CPU information
1093 smp_store_boot_cpu_info(); /* Final full version of the data */
1094 cpumask_copy(cpu_callin_mask, cpumask_of(0));
1097 current_thread_info()->cpu = 0; /* needed? */
1098 for_each_possible_cpu(i) {
1099 zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
1100 zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
1101 zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
1103 set_cpu_sibling_map(0);
1106 if (smp_sanity_check(max_cpus) < 0) {
1107 pr_info("SMP disabled\n");
1112 default_setup_apic_routing();
1115 if (read_apic_id() != boot_cpu_physical_apicid) {
1116 panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1117 read_apic_id(), boot_cpu_physical_apicid);
1118 /* Or can we switch back to PIC here? */
1125 * Switch from PIC to APIC mode.
1130 cpu0_logical_apicid = apic_read(APIC_LDR);
1132 cpu0_logical_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1135 * Enable IO APIC before setting up error vector
1137 if (!skip_ioapic_setup && nr_ioapics)
1140 bsp_end_local_APIC_setup();
1142 if (apic->setup_portio_remap)
1143 apic->setup_portio_remap();
1145 smpboot_setup_io_apic();
1147 * Set up local APIC timer on boot CPU.
1150 pr_info("CPU%d: ", 0);
1151 print_cpu_info(&cpu_data(0));
1152 x86_init.timers.setup_percpu_clockev();
1157 set_mtrr_aps_delayed_init();
1162 void arch_enable_nonboot_cpus_begin(void)
1164 set_mtrr_aps_delayed_init();
1167 void arch_enable_nonboot_cpus_end(void)
1173 * Early setup to make printk work.
1175 void __init native_smp_prepare_boot_cpu(void)
1177 int me = smp_processor_id();
1178 switch_to_new_gdt(me);
1179 /* already set me in cpu_online_mask in boot_cpu_init() */
1180 cpumask_set_cpu(me, cpu_callout_mask);
1181 per_cpu(cpu_state, me) = CPU_ONLINE;
1184 void __init native_smp_cpus_done(unsigned int max_cpus)
1186 pr_debug("Boot done\n");
1190 #ifdef CONFIG_X86_IO_APIC
1191 setup_ioapic_dest();
1196 static int __initdata setup_possible_cpus = -1;
1197 static int __init _setup_possible_cpus(char *str)
1199 get_option(&str, &setup_possible_cpus);
1202 early_param("possible_cpus", _setup_possible_cpus);
1206 * cpu_possible_mask should be static, it cannot change as cpu's
1207 * are onlined, or offlined. The reason is per-cpu data-structures
1208 * are allocated by some modules at init time, and dont expect to
1209 * do this dynamically on cpu arrival/departure.
1210 * cpu_present_mask on the other hand can change dynamically.
1211 * In case when cpu_hotplug is not compiled, then we resort to current
1212 * behaviour, which is cpu_possible == cpu_present.
1215 * Three ways to find out the number of additional hotplug CPUs:
1216 * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1217 * - The user can overwrite it with possible_cpus=NUM
1218 * - Otherwise don't reserve additional CPUs.
1219 * We do this because additional CPUs waste a lot of memory.
1222 __init void prefill_possible_map(void)
1226 /* no processor from mptable or madt */
1227 if (!num_processors)
1230 i = setup_max_cpus ?: 1;
1231 if (setup_possible_cpus == -1) {
1232 possible = num_processors;
1233 #ifdef CONFIG_HOTPLUG_CPU
1235 possible += disabled_cpus;
1241 possible = setup_possible_cpus;
1243 total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1245 /* nr_cpu_ids could be reduced via nr_cpus= */
1246 if (possible > nr_cpu_ids) {
1247 pr_warn("%d Processors exceeds NR_CPUS limit of %d\n",
1248 possible, nr_cpu_ids);
1249 possible = nr_cpu_ids;
1252 #ifdef CONFIG_HOTPLUG_CPU
1253 if (!setup_max_cpus)
1256 pr_warn("%d Processors exceeds max_cpus limit of %u\n",
1257 possible, setup_max_cpus);
1261 pr_info("Allowing %d CPUs, %d hotplug CPUs\n",
1262 possible, max_t(int, possible - num_processors, 0));
1264 for (i = 0; i < possible; i++)
1265 set_cpu_possible(i, true);
1266 for (; i < NR_CPUS; i++)
1267 set_cpu_possible(i, false);
1269 nr_cpu_ids = possible;
1272 #ifdef CONFIG_HOTPLUG_CPU
1274 static void remove_siblinginfo(int cpu)
1277 struct cpuinfo_x86 *c = &cpu_data(cpu);
1279 for_each_cpu(sibling, cpu_core_mask(cpu)) {
1280 cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1282 * last thread sibling in this cpu core going down
1284 if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1285 cpu_data(sibling).booted_cores--;
1288 for_each_cpu(sibling, cpu_sibling_mask(cpu))
1289 cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1290 for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
1291 cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
1292 cpumask_clear(cpu_llc_shared_mask(cpu));
1293 cpumask_clear(cpu_sibling_mask(cpu));
1294 cpumask_clear(cpu_core_mask(cpu));
1295 c->phys_proc_id = 0;
1297 cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1300 static void __ref remove_cpu_from_maps(int cpu)
1302 set_cpu_online(cpu, false);
1303 cpumask_clear_cpu(cpu, cpu_callout_mask);
1304 cpumask_clear_cpu(cpu, cpu_callin_mask);
1305 /* was set by cpu_init() */
1306 cpumask_clear_cpu(cpu, cpu_initialized_mask);
1307 numa_remove_cpu(cpu);
1310 void cpu_disable_common(void)
1312 int cpu = smp_processor_id();
1314 remove_siblinginfo(cpu);
1316 /* It's now safe to remove this processor from the online map */
1318 remove_cpu_from_maps(cpu);
1319 unlock_vector_lock();
1323 int native_cpu_disable(void)
1327 ret = check_irq_vectors_for_cpu_disable();
1333 cpu_disable_common();
1337 void native_cpu_die(unsigned int cpu)
1339 /* We don't do anything here: idle task is faking death itself. */
1342 for (i = 0; i < 10; i++) {
1343 /* They ack this in play_dead by setting CPU_DEAD */
1344 if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1345 if (system_state == SYSTEM_RUNNING)
1346 pr_info("CPU %u is now offline\n", cpu);
1351 pr_err("CPU %u didn't die...\n", cpu);
1354 void play_dead_common(void)
1357 reset_lazy_tlbstate();
1358 amd_e400_remove_cpu(raw_smp_processor_id());
1362 __this_cpu_write(cpu_state, CPU_DEAD);
1365 * With physical CPU hotplug, we should halt the cpu
1367 local_irq_disable();
1370 static bool wakeup_cpu0(void)
1372 if (smp_processor_id() == 0 && enable_start_cpu0)
1379 * We need to flush the caches before going to sleep, lest we have
1380 * dirty data in our caches when we come back up.
1382 static inline void mwait_play_dead(void)
1384 unsigned int eax, ebx, ecx, edx;
1385 unsigned int highest_cstate = 0;
1386 unsigned int highest_subcstate = 0;
1390 if (!this_cpu_has(X86_FEATURE_MWAIT))
1392 if (!this_cpu_has(X86_FEATURE_CLFLSH))
1394 if (__this_cpu_read(cpu_info.cpuid_level) < CPUID_MWAIT_LEAF)
1397 eax = CPUID_MWAIT_LEAF;
1399 native_cpuid(&eax, &ebx, &ecx, &edx);
1402 * eax will be 0 if EDX enumeration is not valid.
1403 * Initialized below to cstate, sub_cstate value when EDX is valid.
1405 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED)) {
1408 edx >>= MWAIT_SUBSTATE_SIZE;
1409 for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
1410 if (edx & MWAIT_SUBSTATE_MASK) {
1412 highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
1415 eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
1416 (highest_subcstate - 1);
1420 * This should be a memory location in a cache line which is
1421 * unlikely to be touched by other processors. The actual
1422 * content is immaterial as it is not actually modified in any way.
1424 mwait_ptr = ¤t_thread_info()->flags;
1430 * The CLFLUSH is a workaround for erratum AAI65 for
1431 * the Xeon 7400 series. It's not clear it is actually
1432 * needed, but it should be harmless in either case.
1433 * The WBINVD is insufficient due to the spurious-wakeup
1434 * case where we return around the loop.
1439 __monitor(mwait_ptr, 0, 0);
1443 * If NMI wants to wake up CPU0, start CPU0.
1450 static inline void hlt_play_dead(void)
1452 if (__this_cpu_read(cpu_info.x86) >= 4)
1458 * If NMI wants to wake up CPU0, start CPU0.
1465 void native_play_dead(void)
1468 tboot_shutdown(TB_SHUTDOWN_WFS);
1470 mwait_play_dead(); /* Only returns on failure */
1471 if (cpuidle_play_dead())
1475 #else /* ... !CONFIG_HOTPLUG_CPU */
1476 int native_cpu_disable(void)
1481 void native_cpu_die(unsigned int cpu)
1483 /* We said "no" in __cpu_disable */
1487 void native_play_dead(void)