2 * x86 SMP booting functions
4 * (c) 1995 Alan Cox, Building #3 <alan@lxorguk.ukuu.org.uk>
5 * (c) 1998, 1999, 2000 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>
56 #include <asm/trampoline.h>
59 #include <asm/pgtable.h>
60 #include <asm/tlbflush.h>
63 #include <asm/genapic.h>
64 #include <asm/setup.h>
65 #include <linux/mc146818rtc.h>
67 #include <mach_apic.h>
68 #include <mach_wakecpu.h>
69 #include <smpboot_hooks.h>
72 u8 apicid_2_node[MAX_APICID];
73 static int low_mappings;
76 /* State of each CPU */
77 DEFINE_PER_CPU(int, cpu_state) = { 0 };
79 /* Store all idle threads, this can be reused instead of creating
80 * a new thread. Also avoids complicated thread destroy functionality
83 #ifdef CONFIG_HOTPLUG_CPU
85 * Needed only for CONFIG_HOTPLUG_CPU because __cpuinitdata is
86 * removed after init for !CONFIG_HOTPLUG_CPU.
88 static DEFINE_PER_CPU(struct task_struct *, idle_thread_array);
89 #define get_idle_for_cpu(x) (per_cpu(idle_thread_array, x))
90 #define set_idle_for_cpu(x, p) (per_cpu(idle_thread_array, x) = (p))
92 static struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
93 #define get_idle_for_cpu(x) (idle_thread_array[(x)])
94 #define set_idle_for_cpu(x, p) (idle_thread_array[(x)] = (p))
97 /* Number of siblings per CPU package */
98 int smp_num_siblings = 1;
99 EXPORT_SYMBOL(smp_num_siblings);
101 /* Last level cache ID of each logical CPU */
102 DEFINE_PER_CPU(u16, cpu_llc_id) = BAD_APICID;
104 /* representing HT siblings of each logical CPU */
105 DEFINE_PER_CPU(cpumask_t, cpu_sibling_map);
106 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
108 /* representing HT and core siblings of each logical CPU */
109 DEFINE_PER_CPU(cpumask_t, cpu_core_map);
110 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
112 /* Per CPU bogomips and other parameters */
113 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
114 EXPORT_PER_CPU_SYMBOL(cpu_info);
116 static atomic_t init_deasserted;
119 /* Set if we find a B stepping CPU */
120 static int __cpuinitdata smp_b_stepping;
122 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_32)
124 /* which logical CPUs are on which nodes */
125 cpumask_t node_to_cpumask_map[MAX_NUMNODES] __read_mostly =
126 { [0 ... MAX_NUMNODES-1] = CPU_MASK_NONE };
127 EXPORT_SYMBOL(node_to_cpumask_map);
128 /* which node each logical CPU is on */
129 int cpu_to_node_map[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 };
130 EXPORT_SYMBOL(cpu_to_node_map);
132 /* set up a mapping between cpu and node. */
133 static void map_cpu_to_node(int cpu, int node)
135 printk(KERN_INFO "Mapping cpu %d to node %d\n", cpu, node);
136 cpumask_set_cpu(cpu, &node_to_cpumask_map[node]);
137 cpu_to_node_map[cpu] = node;
140 /* undo a mapping between cpu and node. */
141 static void unmap_cpu_to_node(int cpu)
145 printk(KERN_INFO "Unmapping cpu %d from all nodes\n", cpu);
146 for (node = 0; node < MAX_NUMNODES; node++)
147 cpumask_clear_cpu(cpu, &node_to_cpumask_map[node]);
148 cpu_to_node_map[cpu] = 0;
150 #else /* !(CONFIG_NUMA && CONFIG_X86_32) */
151 #define map_cpu_to_node(cpu, node) ({})
152 #define unmap_cpu_to_node(cpu) ({})
156 static int boot_cpu_logical_apicid;
158 u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly =
159 { [0 ... NR_CPUS-1] = BAD_APICID };
161 static void map_cpu_to_logical_apicid(void)
163 int cpu = smp_processor_id();
164 int apicid = logical_smp_processor_id();
165 int node = apicid_to_node(apicid);
167 if (!node_online(node))
168 node = first_online_node;
170 cpu_2_logical_apicid[cpu] = apicid;
171 map_cpu_to_node(cpu, node);
174 void numa_remove_cpu(int cpu)
176 cpu_2_logical_apicid[cpu] = BAD_APICID;
177 unmap_cpu_to_node(cpu);
180 #define map_cpu_to_logical_apicid() do {} while (0)
184 * Report back to the Boot Processor.
187 static void __cpuinit smp_callin(void)
190 unsigned long timeout;
193 * If waken up by an INIT in an 82489DX configuration
194 * we may get here before an INIT-deassert IPI reaches
195 * our local APIC. We have to wait for the IPI or we'll
196 * lock up on an APIC access.
198 wait_for_init_deassert(&init_deasserted);
201 * (This works even if the APIC is not enabled.)
203 phys_id = read_apic_id();
204 cpuid = smp_processor_id();
205 if (cpumask_test_cpu(cpuid, cpu_callin_mask)) {
206 panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
209 pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
212 * STARTUP IPIs are fragile beasts as they might sometimes
213 * trigger some glue motherboard logic. Complete APIC bus
214 * silence for 1 second, this overestimates the time the
215 * boot CPU is spending to send the up to 2 STARTUP IPIs
216 * by a factor of two. This should be enough.
220 * Waiting 2s total for startup (udelay is not yet working)
222 timeout = jiffies + 2*HZ;
223 while (time_before(jiffies, timeout)) {
225 * Has the boot CPU finished it's STARTUP sequence?
227 if (cpumask_test_cpu(cpuid, cpu_callout_mask))
232 if (!time_before(jiffies, timeout)) {
233 panic("%s: CPU%d started up but did not get a callout!\n",
238 * the boot CPU has finished the init stage and is spinning
239 * on callin_map until we finish. We are free to set up this
240 * CPU, first the APIC. (this is probably redundant on most
244 pr_debug("CALLIN, before setup_local_APIC().\n");
245 smp_callin_clear_local_apic();
247 end_local_APIC_setup();
248 map_cpu_to_logical_apicid();
250 notify_cpu_starting(cpuid);
254 * Need to enable IRQs because it can take longer and then
255 * the NMI watchdog might kill us.
260 pr_debug("Stack at about %p\n", &cpuid);
263 * Save our processor parameters
265 smp_store_cpu_info(cpuid);
268 * Allow the master to continue.
270 cpumask_set_cpu(cpuid, cpu_callin_mask);
273 static int __cpuinitdata unsafe_smp;
276 * Activate a secondary processor.
278 notrace static void __cpuinit start_secondary(void *unused)
281 * Don't put *anything* before cpu_init(), SMP booting is too
282 * fragile that we want to limit the things done here to the
283 * most necessary things.
290 /* otherwise gcc will move up smp_processor_id before the cpu_init */
293 * Check TSC synchronization with the BP:
295 check_tsc_sync_target();
297 if (nmi_watchdog == NMI_IO_APIC) {
298 disable_8259A_irq(0);
299 enable_NMI_through_LVT0();
309 /* This must be done before setting cpu_online_map */
310 set_cpu_sibling_map(raw_smp_processor_id());
314 * We need to hold call_lock, so there is no inconsistency
315 * between the time smp_call_function() determines number of
316 * IPI recipients, and the time when the determination is made
317 * for which cpus receive the IPI. Holding this
318 * lock helps us to not include this cpu in a currently in progress
319 * smp_call_function().
321 * We need to hold vector_lock so there the set of online cpus
322 * does not change while we are assigning vectors to cpus. Holding
323 * this lock ensures we don't half assign or remove an irq from a cpu.
327 __setup_vector_irq(smp_processor_id());
328 set_cpu_online(smp_processor_id(), true);
329 unlock_vector_lock();
331 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
333 /* enable local interrupts */
336 setup_secondary_clock();
342 static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c)
345 * Mask B, Pentium, but not Pentium MMX
347 if (c->x86_vendor == X86_VENDOR_INTEL &&
349 c->x86_mask >= 1 && c->x86_mask <= 4 &&
352 * Remember we have B step Pentia with bugs
357 * Certain Athlons might work (for various values of 'work') in SMP
358 * but they are not certified as MP capable.
360 if ((c->x86_vendor == X86_VENDOR_AMD) && (c->x86 == 6)) {
362 if (num_possible_cpus() == 1)
365 /* Athlon 660/661 is valid. */
366 if ((c->x86_model == 6) && ((c->x86_mask == 0) ||
370 /* Duron 670 is valid */
371 if ((c->x86_model == 7) && (c->x86_mask == 0))
375 * Athlon 662, Duron 671, and Athlon >model 7 have capability
376 * bit. It's worth noting that the A5 stepping (662) of some
377 * Athlon XP's have the MP bit set.
378 * See http://www.heise.de/newsticker/data/jow-18.10.01-000 for
381 if (((c->x86_model == 6) && (c->x86_mask >= 2)) ||
382 ((c->x86_model == 7) && (c->x86_mask >= 1)) ||
387 /* If we get here, not a certified SMP capable AMD system. */
395 static void __cpuinit smp_checks(void)
398 printk(KERN_WARNING "WARNING: SMP operation may be unreliable"
399 "with B stepping processors.\n");
402 * Don't taint if we are running SMP kernel on a single non-MP
405 if (unsafe_smp && num_online_cpus() > 1) {
406 printk(KERN_INFO "WARNING: This combination of AMD"
407 "processors is not suitable for SMP.\n");
408 add_taint(TAINT_UNSAFE_SMP);
413 * The bootstrap kernel entry code has set these up. Save them for
417 void __cpuinit smp_store_cpu_info(int id)
419 struct cpuinfo_x86 *c = &cpu_data(id);
424 identify_secondary_cpu(c);
429 void __cpuinit set_cpu_sibling_map(int cpu)
432 struct cpuinfo_x86 *c = &cpu_data(cpu);
434 cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
436 if (smp_num_siblings > 1) {
437 for_each_cpu(i, cpu_sibling_setup_mask) {
438 struct cpuinfo_x86 *o = &cpu_data(i);
440 if (c->phys_proc_id == o->phys_proc_id &&
441 c->cpu_core_id == o->cpu_core_id) {
442 cpumask_set_cpu(i, cpu_sibling_mask(cpu));
443 cpumask_set_cpu(cpu, cpu_sibling_mask(i));
444 cpumask_set_cpu(i, cpu_core_mask(cpu));
445 cpumask_set_cpu(cpu, cpu_core_mask(i));
446 cpumask_set_cpu(i, &c->llc_shared_map);
447 cpumask_set_cpu(cpu, &o->llc_shared_map);
451 cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
454 cpumask_set_cpu(cpu, &c->llc_shared_map);
456 if (current_cpu_data.x86_max_cores == 1) {
457 cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
462 for_each_cpu(i, cpu_sibling_setup_mask) {
463 if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
464 per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
465 cpumask_set_cpu(i, &c->llc_shared_map);
466 cpumask_set_cpu(cpu, &cpu_data(i).llc_shared_map);
468 if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
469 cpumask_set_cpu(i, cpu_core_mask(cpu));
470 cpumask_set_cpu(cpu, cpu_core_mask(i));
472 * Does this new cpu bringup a new core?
474 if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
476 * for each core in package, increment
477 * the booted_cores for this new cpu
479 if (cpumask_first(cpu_sibling_mask(i)) == i)
482 * increment the core count for all
483 * the other cpus in this package
486 cpu_data(i).booted_cores++;
487 } else if (i != cpu && !c->booted_cores)
488 c->booted_cores = cpu_data(i).booted_cores;
493 /* maps the cpu to the sched domain representing multi-core */
494 const struct cpumask *cpu_coregroup_mask(int cpu)
496 struct cpuinfo_x86 *c = &cpu_data(cpu);
498 * For perf, we return last level cache shared map.
499 * And for power savings, we return cpu_core_map
501 if (sched_mc_power_savings || sched_smt_power_savings)
502 return cpu_core_mask(cpu);
504 return &c->llc_shared_map;
507 cpumask_t cpu_coregroup_map(int cpu)
509 return *cpu_coregroup_mask(cpu);
512 static void impress_friends(void)
515 unsigned long bogosum = 0;
517 * Allow the user to impress friends.
519 pr_debug("Before bogomips.\n");
520 for_each_possible_cpu(cpu)
521 if (cpumask_test_cpu(cpu, cpu_callout_mask))
522 bogosum += cpu_data(cpu).loops_per_jiffy;
524 "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
527 (bogosum/(5000/HZ))%100);
529 pr_debug("Before bogocount - setting activated=1.\n");
532 void __inquire_remote_apic(int apicid)
534 unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
535 char *names[] = { "ID", "VERSION", "SPIV" };
539 printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
541 for (i = 0; i < ARRAY_SIZE(regs); i++) {
542 printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
547 status = safe_apic_wait_icr_idle();
550 "a previous APIC delivery may have failed\n");
552 apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
557 status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
558 } while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
561 case APIC_ICR_RR_VALID:
562 status = apic_read(APIC_RRR);
563 printk(KERN_CONT "%08x\n", status);
566 printk(KERN_CONT "failed\n");
572 * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
573 * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
574 * won't ... remember to clear down the APIC, etc later.
577 wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
579 unsigned long send_status, accept_status = 0;
583 /* Boot on the stack */
584 /* Kick the second */
585 apic_icr_write(APIC_DM_NMI | APIC_DEST_LOGICAL, logical_apicid);
587 pr_debug("Waiting for send to finish...\n");
588 send_status = safe_apic_wait_icr_idle();
591 * Give the other CPU some time to accept the IPI.
594 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
595 maxlvt = lapic_get_maxlvt();
596 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
597 apic_write(APIC_ESR, 0);
598 accept_status = (apic_read(APIC_ESR) & 0xEF);
600 pr_debug("NMI sent.\n");
603 printk(KERN_ERR "APIC never delivered???\n");
605 printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
607 return (send_status | accept_status);
611 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
613 unsigned long send_status, accept_status = 0;
614 int maxlvt, num_starts, j;
616 if (get_uv_system_type() == UV_NON_UNIQUE_APIC) {
617 send_status = uv_wakeup_secondary(phys_apicid, start_eip);
618 atomic_set(&init_deasserted, 1);
622 maxlvt = lapic_get_maxlvt();
625 * Be paranoid about clearing APIC errors.
627 if (APIC_INTEGRATED(apic_version[phys_apicid])) {
628 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
629 apic_write(APIC_ESR, 0);
633 pr_debug("Asserting INIT.\n");
636 * Turn INIT on target chip
641 apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
644 pr_debug("Waiting for send to finish...\n");
645 send_status = safe_apic_wait_icr_idle();
649 pr_debug("Deasserting INIT.\n");
653 apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
655 pr_debug("Waiting for send to finish...\n");
656 send_status = safe_apic_wait_icr_idle();
659 atomic_set(&init_deasserted, 1);
662 * Should we send STARTUP IPIs ?
664 * Determine this based on the APIC version.
665 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
667 if (APIC_INTEGRATED(apic_version[phys_apicid]))
673 * Paravirt / VMI wants a startup IPI hook here to set up the
674 * target processor state.
676 startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
677 (unsigned long)stack_start.sp);
680 * Run STARTUP IPI loop.
682 pr_debug("#startup loops: %d.\n", num_starts);
684 for (j = 1; j <= num_starts; j++) {
685 pr_debug("Sending STARTUP #%d.\n", j);
686 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
687 apic_write(APIC_ESR, 0);
689 pr_debug("After apic_write.\n");
696 /* Boot on the stack */
697 /* Kick the second */
698 apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
702 * Give the other CPU some time to accept the IPI.
706 pr_debug("Startup point 1.\n");
708 pr_debug("Waiting for send to finish...\n");
709 send_status = safe_apic_wait_icr_idle();
712 * Give the other CPU some time to accept the IPI.
715 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
716 apic_write(APIC_ESR, 0);
717 accept_status = (apic_read(APIC_ESR) & 0xEF);
718 if (send_status || accept_status)
721 pr_debug("After Startup.\n");
724 printk(KERN_ERR "APIC never delivered???\n");
726 printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
728 return (send_status | accept_status);
732 struct work_struct work;
733 struct task_struct *idle;
734 struct completion done;
738 static void __cpuinit do_fork_idle(struct work_struct *work)
740 struct create_idle *c_idle =
741 container_of(work, struct create_idle, work);
743 c_idle->idle = fork_idle(c_idle->cpu);
744 complete(&c_idle->done);
749 /* __ref because it's safe to call free_bootmem when after_bootmem == 0. */
750 static void __ref free_bootmem_pda(struct x8664_pda *oldpda)
753 free_bootmem((unsigned long)oldpda, sizeof(*oldpda));
757 * Allocate node local memory for the AP pda.
759 * Must be called after the _cpu_pda pointer table is initialized.
761 int __cpuinit get_local_pda(int cpu)
763 struct x8664_pda *oldpda, *newpda;
764 unsigned long size = sizeof(struct x8664_pda);
765 int node = cpu_to_node(cpu);
767 if (cpu_pda(cpu) && !cpu_pda(cpu)->in_bootmem)
770 oldpda = cpu_pda(cpu);
771 newpda = kmalloc_node(size, GFP_ATOMIC, node);
773 printk(KERN_ERR "Could not allocate node local PDA "
774 "for CPU %d on node %d\n", cpu, node);
777 return 0; /* have a usable pda */
783 memcpy(newpda, oldpda, size);
784 free_bootmem_pda(oldpda);
787 newpda->in_bootmem = 0;
788 cpu_pda(cpu) = newpda;
791 #endif /* CONFIG_X86_64 */
793 static int __cpuinit do_boot_cpu(int apicid, int cpu)
795 * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
796 * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
797 * Returns zero if CPU booted OK, else error code from wakeup_secondary_cpu.
800 unsigned long boot_error = 0;
802 unsigned long start_ip;
803 unsigned short nmi_high = 0, nmi_low = 0;
804 struct create_idle c_idle = {
806 .done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
808 INIT_WORK(&c_idle.work, do_fork_idle);
811 /* Allocate node local memory for AP pdas */
813 boot_error = get_local_pda(cpu);
816 /* if can't get pda memory, can't start cpu */
820 alternatives_smp_switch(1);
822 c_idle.idle = get_idle_for_cpu(cpu);
825 * We can't use kernel_thread since we must avoid to
826 * reschedule the child.
829 c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
830 (THREAD_SIZE + task_stack_page(c_idle.idle))) - 1);
831 init_idle(c_idle.idle, cpu);
835 if (!keventd_up() || current_is_keventd())
836 c_idle.work.func(&c_idle.work);
838 schedule_work(&c_idle.work);
839 wait_for_completion(&c_idle.done);
842 if (IS_ERR(c_idle.idle)) {
843 printk("failed fork for CPU %d\n", cpu);
844 return PTR_ERR(c_idle.idle);
847 set_idle_for_cpu(cpu, c_idle.idle);
850 per_cpu(current_task, cpu) = c_idle.idle;
852 /* Stack for startup_32 can be just as for start_secondary onwards */
855 cpu_pda(cpu)->pcurrent = c_idle.idle;
856 clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
858 early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
859 initial_code = (unsigned long)start_secondary;
860 stack_start.sp = (void *) c_idle.idle->thread.sp;
862 /* start_ip had better be page-aligned! */
863 start_ip = setup_trampoline();
865 /* So we see what's up */
866 printk(KERN_INFO "Booting processor %d APIC 0x%x ip 0x%lx\n",
867 cpu, apicid, start_ip);
870 * This grunge runs the startup process for
871 * the targeted processor.
874 atomic_set(&init_deasserted, 0);
876 if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
878 pr_debug("Setting warm reset code and vector.\n");
880 store_NMI_vector(&nmi_high, &nmi_low);
882 smpboot_setup_warm_reset_vector(start_ip);
884 * Be paranoid about clearing APIC errors.
886 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
887 apic_write(APIC_ESR, 0);
893 * Starting actual IPI sequence...
895 boot_error = wakeup_secondary_cpu(apicid, start_ip);
899 * allow APs to start initializing.
901 pr_debug("Before Callout %d.\n", cpu);
902 cpumask_set_cpu(cpu, cpu_callout_mask);
903 pr_debug("After Callout %d.\n", cpu);
906 * Wait 5s total for a response
908 for (timeout = 0; timeout < 50000; timeout++) {
909 if (cpumask_test_cpu(cpu, cpu_callin_mask))
910 break; /* It has booted */
914 if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
915 /* number CPUs logically, starting from 1 (BSP is 0) */
917 printk(KERN_INFO "CPU%d: ", cpu);
918 print_cpu_info(&cpu_data(cpu));
919 pr_debug("CPU has booted.\n");
922 if (*((volatile unsigned char *)trampoline_base)
924 /* trampoline started but...? */
925 printk(KERN_ERR "Stuck ??\n");
927 /* trampoline code not run */
928 printk(KERN_ERR "Not responding.\n");
929 if (get_uv_system_type() != UV_NON_UNIQUE_APIC)
930 inquire_remote_apic(apicid);
937 /* Try to put things back the way they were before ... */
938 numa_remove_cpu(cpu); /* was set by numa_add_cpu */
940 /* was set by do_boot_cpu() */
941 cpumask_clear_cpu(cpu, cpu_callout_mask);
943 /* was set by cpu_init() */
944 cpumask_clear_cpu(cpu, cpu_initialized_mask);
946 set_cpu_present(cpu, false);
947 per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
950 /* mark "stuck" area as not stuck */
951 *((volatile unsigned long *)trampoline_base) = 0;
954 * Cleanup possible dangling ends...
956 smpboot_restore_warm_reset_vector();
961 int __cpuinit native_cpu_up(unsigned int cpu)
963 int apicid = cpu_present_to_apicid(cpu);
967 WARN_ON(irqs_disabled());
969 pr_debug("++++++++++++++++++++=_---CPU UP %u\n", cpu);
971 if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
972 !physid_isset(apicid, phys_cpu_present_map)) {
973 printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
978 * Already booted CPU?
980 if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
981 pr_debug("do_boot_cpu %d Already started\n", cpu);
986 * Save current MTRR state in case it was changed since early boot
987 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
991 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
994 /* init low mem mapping */
995 clone_pgd_range(swapper_pg_dir, swapper_pg_dir + KERNEL_PGD_BOUNDARY,
996 min_t(unsigned long, KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
1000 err = do_boot_cpu(apicid, cpu);
1005 err = do_boot_cpu(apicid, cpu);
1008 pr_debug("do_boot_cpu failed %d\n", err);
1013 * Check TSC synchronization with the AP (keep irqs disabled
1016 local_irq_save(flags);
1017 check_tsc_sync_source(cpu);
1018 local_irq_restore(flags);
1020 while (!cpu_online(cpu)) {
1022 touch_nmi_watchdog();
1029 * Fall back to non SMP mode after errors.
1031 * RED-PEN audit/test this more. I bet there is more state messed up here.
1033 static __init void disable_smp(void)
1035 /* use the read/write pointers to the present and possible maps */
1036 cpumask_copy(&cpu_present_map, cpumask_of(0));
1037 cpumask_copy(&cpu_possible_map, cpumask_of(0));
1038 smpboot_clear_io_apic_irqs();
1040 if (smp_found_config)
1041 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1043 physid_set_mask_of_physid(0, &phys_cpu_present_map);
1044 map_cpu_to_logical_apicid();
1045 cpumask_set_cpu(0, cpu_sibling_mask(0));
1046 cpumask_set_cpu(0, cpu_core_mask(0));
1050 * Various sanity checks.
1052 static int __init smp_sanity_check(unsigned max_cpus)
1056 #if defined(CONFIG_X86_PC) && defined(CONFIG_X86_32)
1057 if (def_to_bigsmp && nr_cpu_ids > 8) {
1062 "More than 8 CPUs detected - skipping them.\n"
1063 "Use CONFIG_X86_GENERICARCH and CONFIG_X86_BIGSMP.\n");
1066 for_each_present_cpu(cpu) {
1068 set_cpu_present(cpu, false);
1073 for_each_possible_cpu(cpu) {
1075 set_cpu_possible(cpu, false);
1083 if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1085 "weird, boot CPU (#%d) not listed by the BIOS.\n",
1086 hard_smp_processor_id());
1088 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1092 * If we couldn't find an SMP configuration at boot time,
1093 * get out of here now!
1095 if (!smp_found_config && !acpi_lapic) {
1097 printk(KERN_NOTICE "SMP motherboard not detected.\n");
1099 if (APIC_init_uniprocessor())
1100 printk(KERN_NOTICE "Local APIC not detected."
1101 " Using dummy APIC emulation.\n");
1106 * Should not be necessary because the MP table should list the boot
1107 * CPU too, but we do it for the sake of robustness anyway.
1109 if (!check_phys_apicid_present(boot_cpu_physical_apicid)) {
1111 "weird, boot CPU (#%d) not listed by the BIOS.\n",
1112 boot_cpu_physical_apicid);
1113 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1118 * If we couldn't find a local APIC, then get out of here now!
1120 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1122 printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
1123 boot_cpu_physical_apicid);
1124 printk(KERN_ERR "... forcing use of dummy APIC emulation."
1125 "(tell your hw vendor)\n");
1126 smpboot_clear_io_apic();
1130 verify_local_APIC();
1133 * If SMP should be disabled, then really disable it!
1136 printk(KERN_INFO "SMP mode deactivated.\n");
1137 smpboot_clear_io_apic();
1139 localise_nmi_watchdog();
1143 end_local_APIC_setup();
1150 static void __init smp_cpu_index_default(void)
1153 struct cpuinfo_x86 *c;
1155 for_each_possible_cpu(i) {
1157 /* mark all to hotplug */
1158 c->cpu_index = nr_cpu_ids;
1163 * Prepare for SMP bootup. The MP table or ACPI has been read
1164 * earlier. Just do some sanity checking here and enable APIC mode.
1166 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1169 smp_cpu_index_default();
1170 current_cpu_data = boot_cpu_data;
1171 cpumask_copy(cpu_callin_mask, cpumask_of(0));
1174 * Setup boot CPU information
1176 smp_store_cpu_info(0); /* Final full version of the data */
1177 #ifdef CONFIG_X86_32
1178 boot_cpu_logical_apicid = logical_smp_processor_id();
1180 current_thread_info()->cpu = 0; /* needed? */
1181 set_cpu_sibling_map(0);
1183 #ifdef CONFIG_X86_64
1185 setup_apic_routing();
1188 if (smp_sanity_check(max_cpus) < 0) {
1189 printk(KERN_INFO "SMP disabled\n");
1195 if (read_apic_id() != boot_cpu_physical_apicid) {
1196 panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1197 read_apic_id(), boot_cpu_physical_apicid);
1198 /* Or can we switch back to PIC here? */
1205 * Switch from PIC to APIC mode.
1209 #ifdef CONFIG_X86_64
1211 * Enable IO APIC before setting up error vector
1213 if (!skip_ioapic_setup && nr_ioapics)
1216 end_local_APIC_setup();
1218 map_cpu_to_logical_apicid();
1220 setup_portio_remap();
1222 smpboot_setup_io_apic();
1224 * Set up local APIC timer on boot CPU.
1227 printk(KERN_INFO "CPU%d: ", 0);
1228 print_cpu_info(&cpu_data(0));
1237 * Early setup to make printk work.
1239 void __init native_smp_prepare_boot_cpu(void)
1241 int me = smp_processor_id();
1242 #ifdef CONFIG_X86_32
1245 switch_to_new_gdt();
1246 /* already set me in cpu_online_mask in boot_cpu_init() */
1247 cpumask_set_cpu(me, cpu_callout_mask);
1248 per_cpu(cpu_state, me) = CPU_ONLINE;
1251 void __init native_smp_cpus_done(unsigned int max_cpus)
1253 pr_debug("Boot done.\n");
1257 #ifdef CONFIG_X86_IO_APIC
1258 setup_ioapic_dest();
1260 check_nmi_watchdog();
1263 static int __initdata setup_possible_cpus = -1;
1264 static int __init _setup_possible_cpus(char *str)
1266 get_option(&str, &setup_possible_cpus);
1269 early_param("possible_cpus", _setup_possible_cpus);
1273 * cpu_possible_map should be static, it cannot change as cpu's
1274 * are onlined, or offlined. The reason is per-cpu data-structures
1275 * are allocated by some modules at init time, and dont expect to
1276 * do this dynamically on cpu arrival/departure.
1277 * cpu_present_map on the other hand can change dynamically.
1278 * In case when cpu_hotplug is not compiled, then we resort to current
1279 * behaviour, which is cpu_possible == cpu_present.
1282 * Three ways to find out the number of additional hotplug CPUs:
1283 * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1284 * - The user can overwrite it with possible_cpus=NUM
1285 * - Otherwise don't reserve additional CPUs.
1286 * We do this because additional CPUs waste a lot of memory.
1289 __init void prefill_possible_map(void)
1293 /* no processor from mptable or madt */
1294 if (!num_processors)
1297 if (setup_possible_cpus == -1)
1298 possible = num_processors + disabled_cpus;
1300 possible = setup_possible_cpus;
1302 total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1304 if (possible > CONFIG_NR_CPUS) {
1306 "%d Processors exceeds NR_CPUS limit of %d\n",
1307 possible, CONFIG_NR_CPUS);
1308 possible = CONFIG_NR_CPUS;
1311 printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1312 possible, max_t(int, possible - num_processors, 0));
1314 for (i = 0; i < possible; i++)
1315 set_cpu_possible(i, true);
1317 nr_cpu_ids = possible;
1320 #ifdef CONFIG_HOTPLUG_CPU
1322 static void remove_siblinginfo(int cpu)
1325 struct cpuinfo_x86 *c = &cpu_data(cpu);
1327 for_each_cpu(sibling, cpu_core_mask(cpu)) {
1328 cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1330 * last thread sibling in this cpu core going down
1332 if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1333 cpu_data(sibling).booted_cores--;
1336 for_each_cpu(sibling, cpu_sibling_mask(cpu))
1337 cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1338 cpumask_clear(cpu_sibling_mask(cpu));
1339 cpumask_clear(cpu_core_mask(cpu));
1340 c->phys_proc_id = 0;
1342 cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1345 static void __ref remove_cpu_from_maps(int cpu)
1347 set_cpu_online(cpu, false);
1348 cpumask_clear_cpu(cpu, cpu_callout_mask);
1349 cpumask_clear_cpu(cpu, cpu_callin_mask);
1350 /* was set by cpu_init() */
1351 cpumask_clear_cpu(cpu, cpu_initialized_mask);
1352 numa_remove_cpu(cpu);
1355 void cpu_disable_common(void)
1357 int cpu = smp_processor_id();
1360 * Allow any queued timer interrupts to get serviced
1361 * This is only a temporary solution until we cleanup
1362 * fixup_irqs as we do for IA64.
1367 local_irq_disable();
1368 remove_siblinginfo(cpu);
1370 /* It's now safe to remove this processor from the online map */
1372 remove_cpu_from_maps(cpu);
1373 unlock_vector_lock();
1377 int native_cpu_disable(void)
1379 int cpu = smp_processor_id();
1382 * Perhaps use cpufreq to drop frequency, but that could go
1383 * into generic code.
1385 * We won't take down the boot processor on i386 due to some
1386 * interrupts only being able to be serviced by the BSP.
1387 * Especially so if we're not using an IOAPIC -zwane
1392 if (nmi_watchdog == NMI_LOCAL_APIC)
1393 stop_apic_nmi_watchdog(NULL);
1396 cpu_disable_common();
1400 void native_cpu_die(unsigned int cpu)
1402 /* We don't do anything here: idle task is faking death itself. */
1405 for (i = 0; i < 10; i++) {
1406 /* They ack this in play_dead by setting CPU_DEAD */
1407 if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1408 printk(KERN_INFO "CPU %d is now offline\n", cpu);
1409 if (1 == num_online_cpus())
1410 alternatives_smp_switch(0);
1415 printk(KERN_ERR "CPU %u didn't die...\n", cpu);
1418 void play_dead_common(void)
1421 reset_lazy_tlbstate();
1422 irq_ctx_exit(raw_smp_processor_id());
1423 c1e_remove_cpu(raw_smp_processor_id());
1427 __get_cpu_var(cpu_state) = CPU_DEAD;
1430 * With physical CPU hotplug, we should halt the cpu
1432 local_irq_disable();
1435 void native_play_dead(void)
1441 #else /* ... !CONFIG_HOTPLUG_CPU */
1442 int native_cpu_disable(void)
1447 void native_cpu_die(unsigned int cpu)
1449 /* We said "no" in __cpu_disable */
1453 void native_play_dead(void)