2 * x86 SMP booting functions
4 * (c) 1995 Alan Cox, Building #3 <alan@redhat.com>
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 <linux/mc146818rtc.h>
66 #include <mach_apic.h>
67 #include <mach_wakecpu.h>
68 #include <smpboot_hooks.h>
71 u8 apicid_2_node[MAX_APICID];
72 static int low_mappings;
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))
91 static struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
92 #define get_idle_for_cpu(x) (idle_thread_array[(x)])
93 #define set_idle_for_cpu(x, p) (idle_thread_array[(x)] = (p))
96 /* Number of siblings per CPU package */
97 int smp_num_siblings = 1;
98 EXPORT_SYMBOL(smp_num_siblings);
100 /* Last level cache ID of each logical CPU */
101 DEFINE_PER_CPU(u16, cpu_llc_id) = BAD_APICID;
103 /* bitmap of online cpus */
104 cpumask_t cpu_online_map __read_mostly;
105 EXPORT_SYMBOL(cpu_online_map);
107 cpumask_t cpu_callin_map;
108 cpumask_t cpu_callout_map;
109 cpumask_t cpu_possible_map;
110 EXPORT_SYMBOL(cpu_possible_map);
112 /* representing HT siblings of each logical CPU */
113 DEFINE_PER_CPU(cpumask_t, cpu_sibling_map);
114 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
116 /* representing HT and core siblings of each logical CPU */
117 DEFINE_PER_CPU(cpumask_t, cpu_core_map);
118 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
120 /* Per CPU bogomips and other parameters */
121 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
122 EXPORT_PER_CPU_SYMBOL(cpu_info);
124 static atomic_t init_deasserted;
126 static int boot_cpu_logical_apicid;
128 /* representing cpus for which sibling maps can be computed */
129 static cpumask_t cpu_sibling_setup_map;
131 /* Set if we find a B stepping CPU */
132 static int __cpuinitdata smp_b_stepping;
134 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_32)
136 /* which logical CPUs are on which nodes */
137 cpumask_t node_to_cpumask_map[MAX_NUMNODES] __read_mostly =
138 { [0 ... MAX_NUMNODES-1] = CPU_MASK_NONE };
139 EXPORT_SYMBOL(node_to_cpumask_map);
140 /* which node each logical CPU is on */
141 int cpu_to_node_map[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 };
142 EXPORT_SYMBOL(cpu_to_node_map);
144 /* set up a mapping between cpu and node. */
145 static void map_cpu_to_node(int cpu, int node)
147 printk(KERN_INFO "Mapping cpu %d to node %d\n", cpu, node);
148 cpu_set(cpu, node_to_cpumask_map[node]);
149 cpu_to_node_map[cpu] = node;
152 /* undo a mapping between cpu and node. */
153 static void unmap_cpu_to_node(int cpu)
157 printk(KERN_INFO "Unmapping cpu %d from all nodes\n", cpu);
158 for (node = 0; node < MAX_NUMNODES; node++)
159 cpu_clear(cpu, node_to_cpumask_map[node]);
160 cpu_to_node_map[cpu] = 0;
162 #else /* !(CONFIG_NUMA && CONFIG_X86_32) */
163 #define map_cpu_to_node(cpu, node) ({})
164 #define unmap_cpu_to_node(cpu) ({})
168 u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly =
169 { [0 ... NR_CPUS-1] = BAD_APICID };
171 static void map_cpu_to_logical_apicid(void)
173 int cpu = smp_processor_id();
174 int apicid = logical_smp_processor_id();
175 int node = apicid_to_node(apicid);
177 if (!node_online(node))
178 node = first_online_node;
180 cpu_2_logical_apicid[cpu] = apicid;
181 map_cpu_to_node(cpu, node);
184 void numa_remove_cpu(int cpu)
186 cpu_2_logical_apicid[cpu] = BAD_APICID;
187 unmap_cpu_to_node(cpu);
190 #define map_cpu_to_logical_apicid() do {} while (0)
194 * Report back to the Boot Processor.
197 static void __cpuinit smp_callin(void)
200 unsigned long timeout;
203 * If waken up by an INIT in an 82489DX configuration
204 * we may get here before an INIT-deassert IPI reaches
205 * our local APIC. We have to wait for the IPI or we'll
206 * lock up on an APIC access.
208 wait_for_init_deassert(&init_deasserted);
211 * (This works even if the APIC is not enabled.)
213 phys_id = GET_APIC_ID(read_apic_id());
214 cpuid = smp_processor_id();
215 if (cpu_isset(cpuid, cpu_callin_map)) {
216 panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
219 pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
222 * STARTUP IPIs are fragile beasts as they might sometimes
223 * trigger some glue motherboard logic. Complete APIC bus
224 * silence for 1 second, this overestimates the time the
225 * boot CPU is spending to send the up to 2 STARTUP IPIs
226 * by a factor of two. This should be enough.
230 * Waiting 2s total for startup (udelay is not yet working)
232 timeout = jiffies + 2*HZ;
233 while (time_before(jiffies, timeout)) {
235 * Has the boot CPU finished it's STARTUP sequence?
237 if (cpu_isset(cpuid, cpu_callout_map))
242 if (!time_before(jiffies, timeout)) {
243 panic("%s: CPU%d started up but did not get a callout!\n",
248 * the boot CPU has finished the init stage and is spinning
249 * on callin_map until we finish. We are free to set up this
250 * CPU, first the APIC. (this is probably redundant on most
254 pr_debug("CALLIN, before setup_local_APIC().\n");
255 smp_callin_clear_local_apic();
257 end_local_APIC_setup();
258 map_cpu_to_logical_apicid();
263 * Need to enable IRQs because it can take longer and then
264 * the NMI watchdog might kill us.
269 pr_debug("Stack at about %p\n", &cpuid);
272 * Save our processor parameters
274 smp_store_cpu_info(cpuid);
277 * Allow the master to continue.
279 cpu_set(cpuid, cpu_callin_map);
283 * Activate a secondary processor.
285 static void __cpuinit start_secondary(void *unused)
288 * Don't put *anything* before cpu_init(), SMP booting is too
289 * fragile that we want to limit the things done here to the
290 * most necessary things.
299 /* otherwise gcc will move up smp_processor_id before the cpu_init */
302 * Check TSC synchronization with the BP:
304 check_tsc_sync_target();
306 if (nmi_watchdog == NMI_IO_APIC) {
307 disable_8259A_irq(0);
308 enable_NMI_through_LVT0();
318 /* This must be done before setting cpu_online_map */
319 set_cpu_sibling_map(raw_smp_processor_id());
323 * We need to hold call_lock, so there is no inconsistency
324 * between the time smp_call_function() determines number of
325 * IPI recipients, and the time when the determination is made
326 * for which cpus receive the IPI. Holding this
327 * lock helps us to not include this cpu in a currently in progress
328 * smp_call_function().
330 * We need to hold vector_lock so there the set of online cpus
331 * does not change while we are assigning vectors to cpus. Holding
332 * this lock ensures we don't half assign or remove an irq from a cpu.
336 __setup_vector_irq(smp_processor_id());
337 cpu_set(smp_processor_id(), cpu_online_map);
338 unlock_vector_lock();
339 ipi_call_unlock_irq();
340 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
342 setup_secondary_clock();
348 static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c)
351 * Mask B, Pentium, but not Pentium MMX
353 if (c->x86_vendor == X86_VENDOR_INTEL &&
355 c->x86_mask >= 1 && c->x86_mask <= 4 &&
358 * Remember we have B step Pentia with bugs
363 * Certain Athlons might work (for various values of 'work') in SMP
364 * but they are not certified as MP capable.
366 if ((c->x86_vendor == X86_VENDOR_AMD) && (c->x86 == 6)) {
368 if (num_possible_cpus() == 1)
371 /* Athlon 660/661 is valid. */
372 if ((c->x86_model == 6) && ((c->x86_mask == 0) ||
376 /* Duron 670 is valid */
377 if ((c->x86_model == 7) && (c->x86_mask == 0))
381 * Athlon 662, Duron 671, and Athlon >model 7 have capability
382 * bit. It's worth noting that the A5 stepping (662) of some
383 * Athlon XP's have the MP bit set.
384 * See http://www.heise.de/newsticker/data/jow-18.10.01-000 for
387 if (((c->x86_model == 6) && (c->x86_mask >= 2)) ||
388 ((c->x86_model == 7) && (c->x86_mask >= 1)) ||
393 /* If we get here, not a certified SMP capable AMD system. */
394 add_taint(TAINT_UNSAFE_SMP);
401 static void __cpuinit smp_checks(void)
404 printk(KERN_WARNING "WARNING: SMP operation may be unreliable"
405 "with B stepping processors.\n");
408 * Don't taint if we are running SMP kernel on a single non-MP
411 if (tainted & TAINT_UNSAFE_SMP) {
412 if (num_online_cpus())
413 printk(KERN_INFO "WARNING: This combination of AMD"
414 "processors is not suitable for SMP.\n");
416 tainted &= ~TAINT_UNSAFE_SMP;
421 * The bootstrap kernel entry code has set these up. Save them for
425 void __cpuinit smp_store_cpu_info(int id)
427 struct cpuinfo_x86 *c = &cpu_data(id);
432 identify_secondary_cpu(c);
437 void __cpuinit set_cpu_sibling_map(int cpu)
440 struct cpuinfo_x86 *c = &cpu_data(cpu);
442 cpu_set(cpu, cpu_sibling_setup_map);
444 if (smp_num_siblings > 1) {
445 for_each_cpu_mask_nr(i, cpu_sibling_setup_map) {
446 if (c->phys_proc_id == cpu_data(i).phys_proc_id &&
447 c->cpu_core_id == cpu_data(i).cpu_core_id) {
448 cpu_set(i, per_cpu(cpu_sibling_map, cpu));
449 cpu_set(cpu, per_cpu(cpu_sibling_map, i));
450 cpu_set(i, per_cpu(cpu_core_map, cpu));
451 cpu_set(cpu, per_cpu(cpu_core_map, i));
452 cpu_set(i, c->llc_shared_map);
453 cpu_set(cpu, cpu_data(i).llc_shared_map);
457 cpu_set(cpu, per_cpu(cpu_sibling_map, cpu));
460 cpu_set(cpu, c->llc_shared_map);
462 if (current_cpu_data.x86_max_cores == 1) {
463 per_cpu(cpu_core_map, cpu) = per_cpu(cpu_sibling_map, cpu);
468 for_each_cpu_mask_nr(i, cpu_sibling_setup_map) {
469 if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
470 per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
471 cpu_set(i, c->llc_shared_map);
472 cpu_set(cpu, cpu_data(i).llc_shared_map);
474 if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
475 cpu_set(i, per_cpu(cpu_core_map, cpu));
476 cpu_set(cpu, per_cpu(cpu_core_map, i));
478 * Does this new cpu bringup a new core?
480 if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1) {
482 * for each core in package, increment
483 * the booted_cores for this new cpu
485 if (first_cpu(per_cpu(cpu_sibling_map, i)) == i)
488 * increment the core count for all
489 * the other cpus in this package
492 cpu_data(i).booted_cores++;
493 } else if (i != cpu && !c->booted_cores)
494 c->booted_cores = cpu_data(i).booted_cores;
499 /* maps the cpu to the sched domain representing multi-core */
500 cpumask_t cpu_coregroup_map(int cpu)
502 struct cpuinfo_x86 *c = &cpu_data(cpu);
504 * For perf, we return last level cache shared map.
505 * And for power savings, we return cpu_core_map
507 if (sched_mc_power_savings || sched_smt_power_savings)
508 return per_cpu(cpu_core_map, cpu);
510 return c->llc_shared_map;
513 static void impress_friends(void)
516 unsigned long bogosum = 0;
518 * Allow the user to impress friends.
520 pr_debug("Before bogomips.\n");
521 for_each_possible_cpu(cpu)
522 if (cpu_isset(cpu, cpu_callout_map))
523 bogosum += cpu_data(cpu).loops_per_jiffy;
525 "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
528 (bogosum/(5000/HZ))%100);
530 pr_debug("Before bogocount - setting activated=1.\n");
533 static inline void __inquire_remote_apic(int apicid)
535 unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
536 char *names[] = { "ID", "VERSION", "SPIV" };
540 printk(KERN_INFO "Inquiring remote APIC #%d...\n", apicid);
542 for (i = 0; i < ARRAY_SIZE(regs); i++) {
543 printk(KERN_INFO "... APIC #%d %s: ", apicid, names[i]);
548 status = safe_apic_wait_icr_idle();
551 "a previous APIC delivery may have failed\n");
553 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(apicid));
554 apic_write(APIC_ICR, APIC_DM_REMRD | regs[i]);
559 status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
560 } while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
563 case APIC_ICR_RR_VALID:
564 status = apic_read(APIC_RRR);
565 printk(KERN_CONT "%08x\n", status);
568 printk(KERN_CONT "failed\n");
573 #ifdef WAKE_SECONDARY_VIA_NMI
575 * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
576 * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
577 * won't ... remember to clear down the APIC, etc later.
580 wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
582 unsigned long send_status, accept_status = 0;
586 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(logical_apicid));
588 /* Boot on the stack */
589 /* Kick the second */
590 apic_write(APIC_ICR, APIC_DM_NMI | APIC_DEST_LOGICAL);
592 pr_debug("Waiting for send to finish...\n");
593 send_status = safe_apic_wait_icr_idle();
596 * Give the other CPU some time to accept the IPI.
599 maxlvt = lapic_get_maxlvt();
600 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
601 apic_write(APIC_ESR, 0);
602 accept_status = (apic_read(APIC_ESR) & 0xEF);
603 pr_debug("NMI sent.\n");
606 printk(KERN_ERR "APIC never delivered???\n");
608 printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
610 return (send_status | accept_status);
612 #endif /* WAKE_SECONDARY_VIA_NMI */
614 #ifdef WAKE_SECONDARY_VIA_INIT
616 wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
618 unsigned long send_status, accept_status = 0;
619 int maxlvt, num_starts, j;
621 if (get_uv_system_type() == UV_NON_UNIQUE_APIC) {
622 send_status = uv_wakeup_secondary(phys_apicid, start_eip);
623 atomic_set(&init_deasserted, 1);
627 maxlvt = lapic_get_maxlvt();
630 * Be paranoid about clearing APIC errors.
632 if (APIC_INTEGRATED(apic_version[phys_apicid])) {
633 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
634 apic_write(APIC_ESR, 0);
638 pr_debug("Asserting INIT.\n");
641 * Turn INIT on target chip
643 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
649 APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT);
651 pr_debug("Waiting for send to finish...\n");
652 send_status = safe_apic_wait_icr_idle();
656 pr_debug("Deasserting INIT.\n");
659 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
662 apic_write(APIC_ICR, APIC_INT_LEVELTRIG | APIC_DM_INIT);
664 pr_debug("Waiting for send to finish...\n");
665 send_status = safe_apic_wait_icr_idle();
668 atomic_set(&init_deasserted, 1);
671 * Should we send STARTUP IPIs ?
673 * Determine this based on the APIC version.
674 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
676 if (APIC_INTEGRATED(apic_version[phys_apicid]))
682 * Paravirt / VMI wants a startup IPI hook here to set up the
683 * target processor state.
685 startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
686 (unsigned long)stack_start.sp);
689 * Run STARTUP IPI loop.
691 pr_debug("#startup loops: %d.\n", num_starts);
693 for (j = 1; j <= num_starts; j++) {
694 pr_debug("Sending STARTUP #%d.\n", j);
695 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
696 apic_write(APIC_ESR, 0);
698 pr_debug("After apic_write.\n");
705 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
707 /* Boot on the stack */
708 /* Kick the second */
709 apic_write(APIC_ICR, APIC_DM_STARTUP | (start_eip >> 12));
712 * Give the other CPU some time to accept the IPI.
716 pr_debug("Startup point 1.\n");
718 pr_debug("Waiting for send to finish...\n");
719 send_status = safe_apic_wait_icr_idle();
722 * Give the other CPU some time to accept the IPI.
725 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
726 apic_write(APIC_ESR, 0);
727 accept_status = (apic_read(APIC_ESR) & 0xEF);
728 if (send_status || accept_status)
731 pr_debug("After Startup.\n");
734 printk(KERN_ERR "APIC never delivered???\n");
736 printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
738 return (send_status | accept_status);
740 #endif /* WAKE_SECONDARY_VIA_INIT */
743 struct work_struct work;
744 struct task_struct *idle;
745 struct completion done;
749 static void __cpuinit do_fork_idle(struct work_struct *work)
751 struct create_idle *c_idle =
752 container_of(work, struct create_idle, work);
754 c_idle->idle = fork_idle(c_idle->cpu);
755 complete(&c_idle->done);
760 /* __ref because it's safe to call free_bootmem when after_bootmem == 0. */
761 static void __ref free_bootmem_pda(struct x8664_pda *oldpda)
764 free_bootmem((unsigned long)oldpda, sizeof(*oldpda));
768 * Allocate node local memory for the AP pda.
770 * Must be called after the _cpu_pda pointer table is initialized.
772 int __cpuinit get_local_pda(int cpu)
774 struct x8664_pda *oldpda, *newpda;
775 unsigned long size = sizeof(struct x8664_pda);
776 int node = cpu_to_node(cpu);
778 if (cpu_pda(cpu) && !cpu_pda(cpu)->in_bootmem)
781 oldpda = cpu_pda(cpu);
782 newpda = kmalloc_node(size, GFP_ATOMIC, node);
784 printk(KERN_ERR "Could not allocate node local PDA "
785 "for CPU %d on node %d\n", cpu, node);
788 return 0; /* have a usable pda */
794 memcpy(newpda, oldpda, size);
795 free_bootmem_pda(oldpda);
798 newpda->in_bootmem = 0;
799 cpu_pda(cpu) = newpda;
802 #endif /* CONFIG_X86_64 */
804 static int __cpuinit do_boot_cpu(int apicid, int cpu)
806 * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
807 * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
808 * Returns zero if CPU booted OK, else error code from wakeup_secondary_cpu.
811 unsigned long boot_error = 0;
813 unsigned long start_ip;
814 unsigned short nmi_high = 0, nmi_low = 0;
815 struct create_idle c_idle = {
817 .done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
819 INIT_WORK(&c_idle.work, do_fork_idle);
822 /* Allocate node local memory for AP pdas */
824 boot_error = get_local_pda(cpu);
827 /* if can't get pda memory, can't start cpu */
831 alternatives_smp_switch(1);
833 c_idle.idle = get_idle_for_cpu(cpu);
836 * We can't use kernel_thread since we must avoid to
837 * reschedule the child.
840 c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
841 (THREAD_SIZE + task_stack_page(c_idle.idle))) - 1);
842 init_idle(c_idle.idle, cpu);
846 if (!keventd_up() || current_is_keventd())
847 c_idle.work.func(&c_idle.work);
849 schedule_work(&c_idle.work);
850 wait_for_completion(&c_idle.done);
853 if (IS_ERR(c_idle.idle)) {
854 printk("failed fork for CPU %d\n", cpu);
855 return PTR_ERR(c_idle.idle);
858 set_idle_for_cpu(cpu, c_idle.idle);
861 per_cpu(current_task, cpu) = c_idle.idle;
863 /* Stack for startup_32 can be just as for start_secondary onwards */
866 cpu_pda(cpu)->pcurrent = c_idle.idle;
867 clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
869 early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
870 initial_code = (unsigned long)start_secondary;
871 stack_start.sp = (void *) c_idle.idle->thread.sp;
873 /* start_ip had better be page-aligned! */
874 start_ip = setup_trampoline();
876 /* So we see what's up */
877 printk(KERN_INFO "Booting processor %d/%d ip %lx\n",
878 cpu, apicid, start_ip);
881 * This grunge runs the startup process for
882 * the targeted processor.
885 atomic_set(&init_deasserted, 0);
887 if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
889 pr_debug("Setting warm reset code and vector.\n");
891 store_NMI_vector(&nmi_high, &nmi_low);
893 smpboot_setup_warm_reset_vector(start_ip);
895 * Be paranoid about clearing APIC errors.
897 apic_write(APIC_ESR, 0);
902 * Starting actual IPI sequence...
904 boot_error = wakeup_secondary_cpu(apicid, start_ip);
908 * allow APs to start initializing.
910 pr_debug("Before Callout %d.\n", cpu);
911 cpu_set(cpu, cpu_callout_map);
912 pr_debug("After Callout %d.\n", cpu);
915 * Wait 5s total for a response
917 for (timeout = 0; timeout < 50000; timeout++) {
918 if (cpu_isset(cpu, cpu_callin_map))
919 break; /* It has booted */
923 if (cpu_isset(cpu, cpu_callin_map)) {
924 /* number CPUs logically, starting from 1 (BSP is 0) */
926 printk(KERN_INFO "CPU%d: ", cpu);
927 print_cpu_info(&cpu_data(cpu));
928 pr_debug("CPU has booted.\n");
931 if (*((volatile unsigned char *)trampoline_base)
933 /* trampoline started but...? */
934 printk(KERN_ERR "Stuck ??\n");
936 /* trampoline code not run */
937 printk(KERN_ERR "Not responding.\n");
938 if (get_uv_system_type() != UV_NON_UNIQUE_APIC)
939 inquire_remote_apic(apicid);
946 /* Try to put things back the way they were before ... */
947 numa_remove_cpu(cpu); /* was set by numa_add_cpu */
948 cpu_clear(cpu, cpu_callout_map); /* was set by do_boot_cpu() */
949 cpu_clear(cpu, cpu_initialized); /* was set by cpu_init() */
950 cpu_clear(cpu, cpu_present_map);
951 per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
954 /* mark "stuck" area as not stuck */
955 *((volatile unsigned long *)trampoline_base) = 0;
958 * Cleanup possible dangling ends...
960 smpboot_restore_warm_reset_vector();
965 int __cpuinit native_cpu_up(unsigned int cpu)
967 int apicid = cpu_present_to_apicid(cpu);
971 WARN_ON(irqs_disabled());
973 pr_debug("++++++++++++++++++++=_---CPU UP %u\n", cpu);
975 if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
976 !physid_isset(apicid, phys_cpu_present_map)) {
977 printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
982 * Already booted CPU?
984 if (cpu_isset(cpu, cpu_callin_map)) {
985 pr_debug("do_boot_cpu %d Already started\n", cpu);
990 * Save current MTRR state in case it was changed since early boot
991 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
995 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
998 /* init low mem mapping */
999 clone_pgd_range(swapper_pg_dir, swapper_pg_dir + KERNEL_PGD_BOUNDARY,
1000 min_t(unsigned long, KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
1004 err = do_boot_cpu(apicid, cpu);
1009 err = do_boot_cpu(apicid, cpu);
1012 pr_debug("do_boot_cpu failed %d\n", err);
1017 * Check TSC synchronization with the AP (keep irqs disabled
1020 local_irq_save(flags);
1021 check_tsc_sync_source(cpu);
1022 local_irq_restore(flags);
1024 while (!cpu_online(cpu)) {
1026 touch_nmi_watchdog();
1033 * Fall back to non SMP mode after errors.
1035 * RED-PEN audit/test this more. I bet there is more state messed up here.
1037 static __init void disable_smp(void)
1039 cpu_present_map = cpumask_of_cpu(0);
1040 cpu_possible_map = cpumask_of_cpu(0);
1041 smpboot_clear_io_apic_irqs();
1043 if (smp_found_config)
1044 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1046 physid_set_mask_of_physid(0, &phys_cpu_present_map);
1047 map_cpu_to_logical_apicid();
1048 cpu_set(0, per_cpu(cpu_sibling_map, 0));
1049 cpu_set(0, per_cpu(cpu_core_map, 0));
1053 * Various sanity checks.
1055 static int __init smp_sanity_check(unsigned max_cpus)
1059 #if defined(CONFIG_X86_PC) && defined(CONFIG_X86_32)
1060 if (def_to_bigsmp && nr_cpu_ids > 8) {
1065 "More than 8 CPUs detected - skipping them.\n"
1066 "Use CONFIG_X86_GENERICARCH and CONFIG_X86_BIGSMP.\n");
1069 for_each_present_cpu(cpu) {
1071 cpu_clear(cpu, cpu_present_map);
1076 for_each_possible_cpu(cpu) {
1078 cpu_clear(cpu, cpu_possible_map);
1086 if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1087 printk(KERN_WARNING "weird, boot CPU (#%d) not listed"
1088 "by the BIOS.\n", hard_smp_processor_id());
1089 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1093 * If we couldn't find an SMP configuration at boot time,
1094 * get out of here now!
1096 if (!smp_found_config && !acpi_lapic) {
1098 printk(KERN_NOTICE "SMP motherboard not detected.\n");
1100 if (APIC_init_uniprocessor())
1101 printk(KERN_NOTICE "Local APIC not detected."
1102 " Using dummy APIC emulation.\n");
1107 * Should not be necessary because the MP table should list the boot
1108 * CPU too, but we do it for the sake of robustness anyway.
1110 if (!check_phys_apicid_present(boot_cpu_physical_apicid)) {
1112 "weird, boot CPU (#%d) not listed by the BIOS.\n",
1113 boot_cpu_physical_apicid);
1114 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1119 * If we couldn't find a local APIC, then get out of here now!
1121 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
1123 printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
1124 boot_cpu_physical_apicid);
1125 printk(KERN_ERR "... forcing use of dummy APIC emulation."
1126 "(tell your hw vendor)\n");
1127 smpboot_clear_io_apic();
1131 verify_local_APIC();
1134 * If SMP should be disabled, then really disable it!
1137 printk(KERN_INFO "SMP mode deactivated.\n");
1138 smpboot_clear_io_apic();
1140 localise_nmi_watchdog();
1144 end_local_APIC_setup();
1151 static void __init smp_cpu_index_default(void)
1154 struct cpuinfo_x86 *c;
1156 for_each_possible_cpu(i) {
1158 /* mark all to hotplug */
1159 c->cpu_index = NR_CPUS;
1164 * Prepare for SMP bootup. The MP table or ACPI has been read
1165 * earlier. Just do some sanity checking here and enable APIC mode.
1167 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1170 smp_cpu_index_default();
1171 current_cpu_data = boot_cpu_data;
1172 cpu_callin_map = cpumask_of_cpu(0);
1175 * Setup boot CPU information
1177 smp_store_cpu_info(0); /* Final full version of the data */
1178 boot_cpu_logical_apicid = logical_smp_processor_id();
1179 current_thread_info()->cpu = 0; /* needed? */
1180 set_cpu_sibling_map(0);
1182 if (smp_sanity_check(max_cpus) < 0) {
1183 printk(KERN_INFO "SMP disabled\n");
1189 if (GET_APIC_ID(read_apic_id()) != boot_cpu_physical_apicid) {
1190 panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1191 GET_APIC_ID(read_apic_id()), boot_cpu_physical_apicid);
1192 /* Or can we switch back to PIC here? */
1199 * Switch from PIC to APIC mode.
1203 #ifdef CONFIG_X86_64
1205 * Enable IO APIC before setting up error vector
1207 if (!skip_ioapic_setup && nr_ioapics)
1210 end_local_APIC_setup();
1212 map_cpu_to_logical_apicid();
1214 setup_portio_remap();
1216 smpboot_setup_io_apic();
1218 * Set up local APIC timer on boot CPU.
1221 printk(KERN_INFO "CPU%d: ", 0);
1222 print_cpu_info(&cpu_data(0));
1231 * Early setup to make printk work.
1233 void __init native_smp_prepare_boot_cpu(void)
1235 int me = smp_processor_id();
1236 #ifdef CONFIG_X86_32
1239 switch_to_new_gdt();
1240 /* already set me in cpu_online_map in boot_cpu_init() */
1241 cpu_set(me, cpu_callout_map);
1242 per_cpu(cpu_state, me) = CPU_ONLINE;
1245 void __init native_smp_cpus_done(unsigned int max_cpus)
1247 pr_debug("Boot done.\n");
1251 #ifdef CONFIG_X86_IO_APIC
1252 setup_ioapic_dest();
1254 check_nmi_watchdog();
1257 #ifdef CONFIG_HOTPLUG_CPU
1259 static void remove_siblinginfo(int cpu)
1262 struct cpuinfo_x86 *c = &cpu_data(cpu);
1264 for_each_cpu_mask_nr(sibling, per_cpu(cpu_core_map, cpu)) {
1265 cpu_clear(cpu, per_cpu(cpu_core_map, sibling));
1267 * last thread sibling in this cpu core going down
1269 if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1)
1270 cpu_data(sibling).booted_cores--;
1273 for_each_cpu_mask_nr(sibling, per_cpu(cpu_sibling_map, cpu))
1274 cpu_clear(cpu, per_cpu(cpu_sibling_map, sibling));
1275 cpus_clear(per_cpu(cpu_sibling_map, cpu));
1276 cpus_clear(per_cpu(cpu_core_map, cpu));
1277 c->phys_proc_id = 0;
1279 cpu_clear(cpu, cpu_sibling_setup_map);
1282 static int additional_cpus __initdata = -1;
1284 static __init int setup_additional_cpus(char *s)
1286 return s && get_option(&s, &additional_cpus) ? 0 : -EINVAL;
1288 early_param("additional_cpus", setup_additional_cpus);
1291 * cpu_possible_map should be static, it cannot change as cpu's
1292 * are onlined, or offlined. The reason is per-cpu data-structures
1293 * are allocated by some modules at init time, and dont expect to
1294 * do this dynamically on cpu arrival/departure.
1295 * cpu_present_map on the other hand can change dynamically.
1296 * In case when cpu_hotplug is not compiled, then we resort to current
1297 * behaviour, which is cpu_possible == cpu_present.
1300 * Three ways to find out the number of additional hotplug CPUs:
1301 * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1302 * - The user can overwrite it with additional_cpus=NUM
1303 * - Otherwise don't reserve additional CPUs.
1304 * We do this because additional CPUs waste a lot of memory.
1307 __init void prefill_possible_map(void)
1312 /* no processor from mptable or madt */
1313 if (!num_processors)
1316 if (additional_cpus == -1) {
1317 if (disabled_cpus > 0)
1318 additional_cpus = disabled_cpus;
1320 additional_cpus = 0;
1323 possible = num_processors + additional_cpus;
1324 if (possible > NR_CPUS)
1327 printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1328 possible, max_t(int, possible - num_processors, 0));
1330 for (i = 0; i < possible; i++)
1331 cpu_set(i, cpu_possible_map);
1333 nr_cpu_ids = possible;
1336 static void __ref remove_cpu_from_maps(int cpu)
1338 cpu_clear(cpu, cpu_online_map);
1339 cpu_clear(cpu, cpu_callout_map);
1340 cpu_clear(cpu, cpu_callin_map);
1341 /* was set by cpu_init() */
1342 cpu_clear(cpu, cpu_initialized);
1343 numa_remove_cpu(cpu);
1346 int __cpu_disable(void)
1348 int cpu = smp_processor_id();
1351 * Perhaps use cpufreq to drop frequency, but that could go
1352 * into generic code.
1354 * We won't take down the boot processor on i386 due to some
1355 * interrupts only being able to be serviced by the BSP.
1356 * Especially so if we're not using an IOAPIC -zwane
1361 if (nmi_watchdog == NMI_LOCAL_APIC)
1362 stop_apic_nmi_watchdog(NULL);
1367 * Allow any queued timer interrupts to get serviced
1368 * This is only a temporary solution until we cleanup
1369 * fixup_irqs as we do for IA64.
1374 local_irq_disable();
1375 remove_siblinginfo(cpu);
1377 /* It's now safe to remove this processor from the online map */
1379 remove_cpu_from_maps(cpu);
1380 unlock_vector_lock();
1381 fixup_irqs(cpu_online_map);
1385 void __cpu_die(unsigned int cpu)
1387 /* We don't do anything here: idle task is faking death itself. */
1390 for (i = 0; i < 10; i++) {
1391 /* They ack this in play_dead by setting CPU_DEAD */
1392 if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1393 printk(KERN_INFO "CPU %d is now offline\n", cpu);
1394 if (1 == num_online_cpus())
1395 alternatives_smp_switch(0);
1400 printk(KERN_ERR "CPU %u didn't die...\n", cpu);
1402 #else /* ... !CONFIG_HOTPLUG_CPU */
1403 int __cpu_disable(void)
1408 void __cpu_die(unsigned int cpu)
1410 /* We said "no" in __cpu_disable */