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>
57 #include <asm/trampoline.h>
60 #include <asm/pgtable.h>
61 #include <asm/tlbflush.h>
64 #include <asm/genapic.h>
65 #include <asm/setup.h>
66 #include <linux/mc146818rtc.h>
68 #include <mach_apic.h>
69 #include <mach_wakecpu.h>
70 #include <smpboot_hooks.h>
73 u8 apicid_2_node[MAX_APICID];
74 static int low_mappings;
77 /* State of each CPU */
78 DEFINE_PER_CPU(int, cpu_state) = { 0 };
80 /* Store all idle threads, this can be reused instead of creating
81 * a new thread. Also avoids complicated thread destroy functionality
84 #ifdef CONFIG_HOTPLUG_CPU
86 * Needed only for CONFIG_HOTPLUG_CPU because __cpuinitdata is
87 * removed after init for !CONFIG_HOTPLUG_CPU.
89 static DEFINE_PER_CPU(struct task_struct *, idle_thread_array);
90 #define get_idle_for_cpu(x) (per_cpu(idle_thread_array, x))
91 #define set_idle_for_cpu(x, p) (per_cpu(idle_thread_array, x) = (p))
93 static struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
94 #define get_idle_for_cpu(x) (idle_thread_array[(x)])
95 #define set_idle_for_cpu(x, p) (idle_thread_array[(x)] = (p))
98 /* Number of siblings per CPU package */
99 int smp_num_siblings = 1;
100 EXPORT_SYMBOL(smp_num_siblings);
102 /* Last level cache ID of each logical CPU */
103 DEFINE_PER_CPU(u16, cpu_llc_id) = BAD_APICID;
105 /* representing HT siblings of each logical CPU */
106 DEFINE_PER_CPU(cpumask_t, cpu_sibling_map);
107 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
109 /* representing HT and core siblings of each logical CPU */
110 DEFINE_PER_CPU(cpumask_t, cpu_core_map);
111 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
113 /* Per CPU bogomips and other parameters */
114 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
115 EXPORT_PER_CPU_SYMBOL(cpu_info);
117 static atomic_t init_deasserted;
120 /* Set if we find a B stepping CPU */
121 static int __cpuinitdata smp_b_stepping;
123 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_32)
125 /* which logical CPUs are on which nodes */
126 cpumask_t node_to_cpumask_map[MAX_NUMNODES] __read_mostly =
127 { [0 ... MAX_NUMNODES-1] = CPU_MASK_NONE };
128 EXPORT_SYMBOL(node_to_cpumask_map);
129 /* which node each logical CPU is on */
130 int cpu_to_node_map[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 };
131 EXPORT_SYMBOL(cpu_to_node_map);
133 /* set up a mapping between cpu and node. */
134 static void map_cpu_to_node(int cpu, int node)
136 printk(KERN_INFO "Mapping cpu %d to node %d\n", cpu, node);
137 cpumask_set_cpu(cpu, &node_to_cpumask_map[node]);
138 cpu_to_node_map[cpu] = node;
141 /* undo a mapping between cpu and node. */
142 static void unmap_cpu_to_node(int cpu)
146 printk(KERN_INFO "Unmapping cpu %d from all nodes\n", cpu);
147 for (node = 0; node < MAX_NUMNODES; node++)
148 cpumask_clear_cpu(cpu, &node_to_cpumask_map[node]);
149 cpu_to_node_map[cpu] = 0;
151 #else /* !(CONFIG_NUMA && CONFIG_X86_32) */
152 #define map_cpu_to_node(cpu, node) ({})
153 #define unmap_cpu_to_node(cpu) ({})
157 static int boot_cpu_logical_apicid;
159 u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly =
160 { [0 ... NR_CPUS-1] = BAD_APICID };
162 static void map_cpu_to_logical_apicid(void)
164 int cpu = smp_processor_id();
165 int apicid = logical_smp_processor_id();
166 int node = apicid_to_node(apicid);
168 if (!node_online(node))
169 node = first_online_node;
171 cpu_2_logical_apicid[cpu] = apicid;
172 map_cpu_to_node(cpu, node);
175 void numa_remove_cpu(int cpu)
177 cpu_2_logical_apicid[cpu] = BAD_APICID;
178 unmap_cpu_to_node(cpu);
181 #define map_cpu_to_logical_apicid() do {} while (0)
185 * Report back to the Boot Processor.
188 static void __cpuinit smp_callin(void)
191 unsigned long timeout;
194 * If waken up by an INIT in an 82489DX configuration
195 * we may get here before an INIT-deassert IPI reaches
196 * our local APIC. We have to wait for the IPI or we'll
197 * lock up on an APIC access.
199 wait_for_init_deassert(&init_deasserted);
202 * (This works even if the APIC is not enabled.)
204 phys_id = read_apic_id();
205 cpuid = smp_processor_id();
206 if (cpumask_test_cpu(cpuid, cpu_callin_mask)) {
207 panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
210 pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
213 * STARTUP IPIs are fragile beasts as they might sometimes
214 * trigger some glue motherboard logic. Complete APIC bus
215 * silence for 1 second, this overestimates the time the
216 * boot CPU is spending to send the up to 2 STARTUP IPIs
217 * by a factor of two. This should be enough.
221 * Waiting 2s total for startup (udelay is not yet working)
223 timeout = jiffies + 2*HZ;
224 while (time_before(jiffies, timeout)) {
226 * Has the boot CPU finished it's STARTUP sequence?
228 if (cpumask_test_cpu(cpuid, cpu_callout_mask))
233 if (!time_before(jiffies, timeout)) {
234 panic("%s: CPU%d started up but did not get a callout!\n",
239 * the boot CPU has finished the init stage and is spinning
240 * on callin_map until we finish. We are free to set up this
241 * CPU, first the APIC. (this is probably redundant on most
245 pr_debug("CALLIN, before setup_local_APIC().\n");
246 smp_callin_clear_local_apic();
248 end_local_APIC_setup();
249 map_cpu_to_logical_apicid();
251 notify_cpu_starting(cpuid);
255 * Need to enable IRQs because it can take longer and then
256 * the NMI watchdog might kill us.
261 pr_debug("Stack at about %p\n", &cpuid);
264 * Save our processor parameters
266 smp_store_cpu_info(cpuid);
269 * Allow the master to continue.
271 cpumask_set_cpu(cpuid, cpu_callin_mask);
274 static int __cpuinitdata unsafe_smp;
277 * Activate a secondary processor.
279 notrace static void __cpuinit start_secondary(void *unused)
282 * Don't put *anything* before cpu_init(), SMP booting is too
283 * fragile that we want to limit the things done here to the
284 * most necessary things.
291 /* otherwise gcc will move up smp_processor_id before the cpu_init */
294 * Check TSC synchronization with the BP:
296 check_tsc_sync_target();
298 if (nmi_watchdog == NMI_IO_APIC) {
299 disable_8259A_irq(0);
300 enable_NMI_through_LVT0();
310 /* This must be done before setting cpu_online_map */
311 set_cpu_sibling_map(raw_smp_processor_id());
315 * We need to hold call_lock, so there is no inconsistency
316 * between the time smp_call_function() determines number of
317 * IPI recipients, and the time when the determination is made
318 * for which cpus receive the IPI. Holding this
319 * lock helps us to not include this cpu in a currently in progress
320 * smp_call_function().
322 * We need to hold vector_lock so there the set of online cpus
323 * does not change while we are assigning vectors to cpus. Holding
324 * this lock ensures we don't half assign or remove an irq from a cpu.
328 __setup_vector_irq(smp_processor_id());
329 set_cpu_online(smp_processor_id(), true);
330 unlock_vector_lock();
332 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
334 /* enable local interrupts */
337 setup_secondary_clock();
343 static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c)
346 * Mask B, Pentium, but not Pentium MMX
348 if (c->x86_vendor == X86_VENDOR_INTEL &&
350 c->x86_mask >= 1 && c->x86_mask <= 4 &&
353 * Remember we have B step Pentia with bugs
358 * Certain Athlons might work (for various values of 'work') in SMP
359 * but they are not certified as MP capable.
361 if ((c->x86_vendor == X86_VENDOR_AMD) && (c->x86 == 6)) {
363 if (num_possible_cpus() == 1)
366 /* Athlon 660/661 is valid. */
367 if ((c->x86_model == 6) && ((c->x86_mask == 0) ||
371 /* Duron 670 is valid */
372 if ((c->x86_model == 7) && (c->x86_mask == 0))
376 * Athlon 662, Duron 671, and Athlon >model 7 have capability
377 * bit. It's worth noting that the A5 stepping (662) of some
378 * Athlon XP's have the MP bit set.
379 * See http://www.heise.de/newsticker/data/jow-18.10.01-000 for
382 if (((c->x86_model == 6) && (c->x86_mask >= 2)) ||
383 ((c->x86_model == 7) && (c->x86_mask >= 1)) ||
388 /* If we get here, not a certified SMP capable AMD system. */
396 static void __cpuinit smp_checks(void)
399 printk(KERN_WARNING "WARNING: SMP operation may be unreliable"
400 "with B stepping processors.\n");
403 * Don't taint if we are running SMP kernel on a single non-MP
406 if (unsafe_smp && num_online_cpus() > 1) {
407 printk(KERN_INFO "WARNING: This combination of AMD"
408 "processors is not suitable for SMP.\n");
409 add_taint(TAINT_UNSAFE_SMP);
414 * The bootstrap kernel entry code has set these up. Save them for
418 void __cpuinit smp_store_cpu_info(int id)
420 struct cpuinfo_x86 *c = &cpu_data(id);
425 identify_secondary_cpu(c);
430 void __cpuinit set_cpu_sibling_map(int cpu)
433 struct cpuinfo_x86 *c = &cpu_data(cpu);
435 cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
437 if (smp_num_siblings > 1) {
438 for_each_cpu(i, cpu_sibling_setup_mask) {
439 struct cpuinfo_x86 *o = &cpu_data(i);
441 if (c->phys_proc_id == o->phys_proc_id &&
442 c->cpu_core_id == o->cpu_core_id) {
443 cpumask_set_cpu(i, cpu_sibling_mask(cpu));
444 cpumask_set_cpu(cpu, cpu_sibling_mask(i));
445 cpumask_set_cpu(i, cpu_core_mask(cpu));
446 cpumask_set_cpu(cpu, cpu_core_mask(i));
447 cpumask_set_cpu(i, &c->llc_shared_map);
448 cpumask_set_cpu(cpu, &o->llc_shared_map);
452 cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
455 cpumask_set_cpu(cpu, &c->llc_shared_map);
457 if (current_cpu_data.x86_max_cores == 1) {
458 cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
463 for_each_cpu(i, cpu_sibling_setup_mask) {
464 if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
465 per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
466 cpumask_set_cpu(i, &c->llc_shared_map);
467 cpumask_set_cpu(cpu, &cpu_data(i).llc_shared_map);
469 if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
470 cpumask_set_cpu(i, cpu_core_mask(cpu));
471 cpumask_set_cpu(cpu, cpu_core_mask(i));
473 * Does this new cpu bringup a new core?
475 if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
477 * for each core in package, increment
478 * the booted_cores for this new cpu
480 if (cpumask_first(cpu_sibling_mask(i)) == i)
483 * increment the core count for all
484 * the other cpus in this package
487 cpu_data(i).booted_cores++;
488 } else if (i != cpu && !c->booted_cores)
489 c->booted_cores = cpu_data(i).booted_cores;
494 /* maps the cpu to the sched domain representing multi-core */
495 const struct cpumask *cpu_coregroup_mask(int cpu)
497 struct cpuinfo_x86 *c = &cpu_data(cpu);
499 * For perf, we return last level cache shared map.
500 * And for power savings, we return cpu_core_map
502 if (sched_mc_power_savings || sched_smt_power_savings)
503 return cpu_core_mask(cpu);
505 return &c->llc_shared_map;
508 cpumask_t cpu_coregroup_map(int cpu)
510 return *cpu_coregroup_mask(cpu);
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 (cpumask_test_cpu(cpu, cpu_callout_mask))
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 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 0x%x...\n", apicid);
542 for (i = 0; i < ARRAY_SIZE(regs); i++) {
543 printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
548 status = safe_apic_wait_icr_idle();
551 "a previous APIC delivery may have failed\n");
553 apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
558 status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
559 } while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
562 case APIC_ICR_RR_VALID:
563 status = apic_read(APIC_RRR);
564 printk(KERN_CONT "%08x\n", status);
567 printk(KERN_CONT "failed\n");
573 * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
574 * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
575 * won't ... remember to clear down the APIC, etc later.
578 wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
580 unsigned long send_status, accept_status = 0;
584 /* Boot on the stack */
585 /* Kick the second */
586 apic_icr_write(APIC_DM_NMI | APIC_DEST_LOGICAL, logical_apicid);
588 pr_debug("Waiting for send to finish...\n");
589 send_status = safe_apic_wait_icr_idle();
592 * Give the other CPU some time to accept the IPI.
595 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
596 maxlvt = lapic_get_maxlvt();
597 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
598 apic_write(APIC_ESR, 0);
599 accept_status = (apic_read(APIC_ESR) & 0xEF);
601 pr_debug("NMI sent.\n");
604 printk(KERN_ERR "APIC never delivered???\n");
606 printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
608 return (send_status | accept_status);
612 wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
614 unsigned long send_status, accept_status = 0;
615 int maxlvt, num_starts, j;
617 if (get_uv_system_type() == UV_NON_UNIQUE_APIC) {
618 send_status = uv_wakeup_secondary(phys_apicid, start_eip);
619 atomic_set(&init_deasserted, 1);
623 maxlvt = lapic_get_maxlvt();
626 * Be paranoid about clearing APIC errors.
628 if (APIC_INTEGRATED(apic_version[phys_apicid])) {
629 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
630 apic_write(APIC_ESR, 0);
634 pr_debug("Asserting INIT.\n");
637 * Turn INIT on target chip
642 apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
645 pr_debug("Waiting for send to finish...\n");
646 send_status = safe_apic_wait_icr_idle();
650 pr_debug("Deasserting INIT.\n");
654 apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
656 pr_debug("Waiting for send to finish...\n");
657 send_status = safe_apic_wait_icr_idle();
660 atomic_set(&init_deasserted, 1);
663 * Should we send STARTUP IPIs ?
665 * Determine this based on the APIC version.
666 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
668 if (APIC_INTEGRATED(apic_version[phys_apicid]))
674 * Paravirt / VMI wants a startup IPI hook here to set up the
675 * target processor state.
677 startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
678 (unsigned long)stack_start.sp);
681 * Run STARTUP IPI loop.
683 pr_debug("#startup loops: %d.\n", num_starts);
685 for (j = 1; j <= num_starts; j++) {
686 pr_debug("Sending STARTUP #%d.\n", j);
687 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
688 apic_write(APIC_ESR, 0);
690 pr_debug("After apic_write.\n");
697 /* Boot on the stack */
698 /* Kick the second */
699 apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
703 * Give the other CPU some time to accept the IPI.
707 pr_debug("Startup point 1.\n");
709 pr_debug("Waiting for send to finish...\n");
710 send_status = safe_apic_wait_icr_idle();
713 * Give the other CPU some time to accept the IPI.
716 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
717 apic_write(APIC_ESR, 0);
718 accept_status = (apic_read(APIC_ESR) & 0xEF);
719 if (send_status || accept_status)
722 pr_debug("After Startup.\n");
725 printk(KERN_ERR "APIC never delivered???\n");
727 printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
729 return (send_status | accept_status);
733 struct work_struct work;
734 struct task_struct *idle;
735 struct completion done;
739 static void __cpuinit do_fork_idle(struct work_struct *work)
741 struct create_idle *c_idle =
742 container_of(work, struct create_idle, work);
744 c_idle->idle = fork_idle(c_idle->cpu);
745 complete(&c_idle->done);
750 /* __ref because it's safe to call free_bootmem when after_bootmem == 0. */
751 static void __ref free_bootmem_pda(struct x8664_pda *oldpda)
754 free_bootmem((unsigned long)oldpda, sizeof(*oldpda));
758 * Allocate node local memory for the AP pda.
760 * Must be called after the _cpu_pda pointer table is initialized.
762 int __cpuinit get_local_pda(int cpu)
764 struct x8664_pda *oldpda, *newpda;
765 unsigned long size = sizeof(struct x8664_pda);
766 int node = cpu_to_node(cpu);
768 if (cpu_pda(cpu) && !cpu_pda(cpu)->in_bootmem)
771 oldpda = cpu_pda(cpu);
772 newpda = kmalloc_node(size, GFP_ATOMIC, node);
774 printk(KERN_ERR "Could not allocate node local PDA "
775 "for CPU %d on node %d\n", cpu, node);
778 return 0; /* have a usable pda */
784 memcpy(newpda, oldpda, size);
785 free_bootmem_pda(oldpda);
788 newpda->in_bootmem = 0;
789 cpu_pda(cpu) = newpda;
792 #endif /* CONFIG_X86_64 */
794 static int __cpuinit do_boot_cpu(int apicid, int cpu)
796 * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
797 * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
798 * Returns zero if CPU booted OK, else error code from wakeup_secondary_cpu.
801 unsigned long boot_error = 0;
803 unsigned long start_ip;
804 unsigned short nmi_high = 0, nmi_low = 0;
805 struct create_idle c_idle = {
807 .done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
809 INIT_WORK(&c_idle.work, do_fork_idle);
812 /* Allocate node local memory for AP pdas */
814 boot_error = get_local_pda(cpu);
817 /* if can't get pda memory, can't start cpu */
821 alternatives_smp_switch(1);
823 c_idle.idle = get_idle_for_cpu(cpu);
826 * We can't use kernel_thread since we must avoid to
827 * reschedule the child.
830 c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
831 (THREAD_SIZE + task_stack_page(c_idle.idle))) - 1);
832 init_idle(c_idle.idle, cpu);
836 if (!keventd_up() || current_is_keventd())
837 c_idle.work.func(&c_idle.work);
839 schedule_work(&c_idle.work);
840 wait_for_completion(&c_idle.done);
843 if (IS_ERR(c_idle.idle)) {
844 printk("failed fork for CPU %d\n", cpu);
845 return PTR_ERR(c_idle.idle);
848 set_idle_for_cpu(cpu, c_idle.idle);
851 per_cpu(current_task, cpu) = c_idle.idle;
853 /* Stack for startup_32 can be just as for start_secondary onwards */
856 cpu_pda(cpu)->pcurrent = c_idle.idle;
857 clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
859 early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
860 initial_code = (unsigned long)start_secondary;
861 stack_start.sp = (void *) c_idle.idle->thread.sp;
863 /* start_ip had better be page-aligned! */
864 start_ip = setup_trampoline();
866 /* So we see what's up */
867 printk(KERN_INFO "Booting processor %d APIC 0x%x ip 0x%lx\n",
868 cpu, apicid, start_ip);
871 * This grunge runs the startup process for
872 * the targeted processor.
875 atomic_set(&init_deasserted, 0);
877 if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
879 pr_debug("Setting warm reset code and vector.\n");
881 store_NMI_vector(&nmi_high, &nmi_low);
883 smpboot_setup_warm_reset_vector(start_ip);
885 * Be paranoid about clearing APIC errors.
887 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
888 apic_write(APIC_ESR, 0);
894 * Starting actual IPI sequence...
896 boot_error = wakeup_secondary_cpu(apicid, start_ip);
900 * allow APs to start initializing.
902 pr_debug("Before Callout %d.\n", cpu);
903 cpumask_set_cpu(cpu, cpu_callout_mask);
904 pr_debug("After Callout %d.\n", cpu);
907 * Wait 5s total for a response
909 for (timeout = 0; timeout < 50000; timeout++) {
910 if (cpumask_test_cpu(cpu, cpu_callin_mask))
911 break; /* It has booted */
915 if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
916 /* number CPUs logically, starting from 1 (BSP is 0) */
918 printk(KERN_INFO "CPU%d: ", cpu);
919 print_cpu_info(&cpu_data(cpu));
920 pr_debug("CPU has booted.\n");
923 if (*((volatile unsigned char *)trampoline_base)
925 /* trampoline started but...? */
926 printk(KERN_ERR "Stuck ??\n");
928 /* trampoline code not run */
929 printk(KERN_ERR "Not responding.\n");
930 if (get_uv_system_type() != UV_NON_UNIQUE_APIC)
931 inquire_remote_apic(apicid);
938 /* Try to put things back the way they were before ... */
939 numa_remove_cpu(cpu); /* was set by numa_add_cpu */
941 /* was set by do_boot_cpu() */
942 cpumask_clear_cpu(cpu, cpu_callout_mask);
944 /* was set by cpu_init() */
945 cpumask_clear_cpu(cpu, cpu_initialized_mask);
947 set_cpu_present(cpu, false);
948 per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
951 /* mark "stuck" area as not stuck */
952 *((volatile unsigned long *)trampoline_base) = 0;
955 * Cleanup possible dangling ends...
957 smpboot_restore_warm_reset_vector();
962 int __cpuinit native_cpu_up(unsigned int cpu)
964 int apicid = cpu_present_to_apicid(cpu);
968 WARN_ON(irqs_disabled());
970 pr_debug("++++++++++++++++++++=_---CPU UP %u\n", cpu);
972 if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
973 !physid_isset(apicid, phys_cpu_present_map)) {
974 printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
979 * Already booted CPU?
981 if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
982 pr_debug("do_boot_cpu %d Already started\n", cpu);
987 * Save current MTRR state in case it was changed since early boot
988 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
992 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
995 /* init low mem mapping */
996 clone_pgd_range(swapper_pg_dir, swapper_pg_dir + KERNEL_PGD_BOUNDARY,
997 min_t(unsigned long, KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
1001 err = do_boot_cpu(apicid, cpu);
1006 err = do_boot_cpu(apicid, cpu);
1009 pr_debug("do_boot_cpu failed %d\n", err);
1014 * Check TSC synchronization with the AP (keep irqs disabled
1017 local_irq_save(flags);
1018 check_tsc_sync_source(cpu);
1019 local_irq_restore(flags);
1021 while (!cpu_online(cpu)) {
1023 touch_nmi_watchdog();
1030 * Fall back to non SMP mode after errors.
1032 * RED-PEN audit/test this more. I bet there is more state messed up here.
1034 static __init void disable_smp(void)
1036 /* use the read/write pointers to the present and possible maps */
1037 cpumask_copy(&cpu_present_map, cpumask_of(0));
1038 cpumask_copy(&cpu_possible_map, cpumask_of(0));
1039 smpboot_clear_io_apic_irqs();
1041 if (smp_found_config)
1042 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1044 physid_set_mask_of_physid(0, &phys_cpu_present_map);
1045 map_cpu_to_logical_apicid();
1046 cpumask_set_cpu(0, cpu_sibling_mask(0));
1047 cpumask_set_cpu(0, cpu_core_mask(0));
1051 * Various sanity checks.
1053 static int __init smp_sanity_check(unsigned max_cpus)
1057 #if defined(CONFIG_X86_PC) && defined(CONFIG_X86_32)
1058 if (def_to_bigsmp && nr_cpu_ids > 8) {
1063 "More than 8 CPUs detected - skipping them.\n"
1064 "Use CONFIG_X86_GENERICARCH and CONFIG_X86_BIGSMP.\n");
1067 for_each_present_cpu(cpu) {
1069 set_cpu_present(cpu, false);
1074 for_each_possible_cpu(cpu) {
1076 set_cpu_possible(cpu, false);
1084 if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1086 "weird, boot CPU (#%d) not listed by the BIOS.\n",
1087 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_cpu_ids;
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 cpumask_copy(cpu_callin_mask, cpumask_of(0));
1175 * Setup boot CPU information
1177 smp_store_cpu_info(0); /* Final full version of the data */
1178 #ifdef CONFIG_X86_32
1179 boot_cpu_logical_apicid = logical_smp_processor_id();
1181 current_thread_info()->cpu = 0; /* needed? */
1182 set_cpu_sibling_map(0);
1184 #ifdef CONFIG_X86_64
1186 setup_apic_routing();
1189 if (smp_sanity_check(max_cpus) < 0) {
1190 printk(KERN_INFO "SMP disabled\n");
1196 if (read_apic_id() != boot_cpu_physical_apicid) {
1197 panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1198 read_apic_id(), boot_cpu_physical_apicid);
1199 /* Or can we switch back to PIC here? */
1206 * Switch from PIC to APIC mode.
1210 #ifdef CONFIG_X86_64
1212 * Enable IO APIC before setting up error vector
1214 if (!skip_ioapic_setup && nr_ioapics)
1217 end_local_APIC_setup();
1219 map_cpu_to_logical_apicid();
1221 setup_portio_remap();
1223 smpboot_setup_io_apic();
1225 * Set up local APIC timer on boot CPU.
1228 printk(KERN_INFO "CPU%d: ", 0);
1229 print_cpu_info(&cpu_data(0));
1238 * Early setup to make printk work.
1240 void __init native_smp_prepare_boot_cpu(void)
1242 int me = smp_processor_id();
1243 #ifdef CONFIG_X86_32
1246 switch_to_new_gdt();
1247 /* already set me in cpu_online_mask in boot_cpu_init() */
1248 cpumask_set_cpu(me, cpu_callout_mask);
1249 per_cpu(cpu_state, me) = CPU_ONLINE;
1252 void __init native_smp_cpus_done(unsigned int max_cpus)
1254 pr_debug("Boot done.\n");
1258 #ifdef CONFIG_X86_IO_APIC
1259 setup_ioapic_dest();
1261 check_nmi_watchdog();
1264 static int __initdata setup_possible_cpus = -1;
1265 static int __init _setup_possible_cpus(char *str)
1267 get_option(&str, &setup_possible_cpus);
1270 early_param("possible_cpus", _setup_possible_cpus);
1274 * cpu_possible_map should be static, it cannot change as cpu's
1275 * are onlined, or offlined. The reason is per-cpu data-structures
1276 * are allocated by some modules at init time, and dont expect to
1277 * do this dynamically on cpu arrival/departure.
1278 * cpu_present_map on the other hand can change dynamically.
1279 * In case when cpu_hotplug is not compiled, then we resort to current
1280 * behaviour, which is cpu_possible == cpu_present.
1283 * Three ways to find out the number of additional hotplug CPUs:
1284 * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
1285 * - The user can overwrite it with possible_cpus=NUM
1286 * - Otherwise don't reserve additional CPUs.
1287 * We do this because additional CPUs waste a lot of memory.
1290 __init void prefill_possible_map(void)
1294 /* no processor from mptable or madt */
1295 if (!num_processors)
1298 if (setup_possible_cpus == -1)
1299 possible = num_processors + disabled_cpus;
1301 possible = setup_possible_cpus;
1303 total_cpus = max_t(int, possible, num_processors + disabled_cpus);
1305 if (possible > CONFIG_NR_CPUS) {
1307 "%d Processors exceeds NR_CPUS limit of %d\n",
1308 possible, CONFIG_NR_CPUS);
1309 possible = CONFIG_NR_CPUS;
1312 printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
1313 possible, max_t(int, possible - num_processors, 0));
1315 for (i = 0; i < possible; i++)
1316 set_cpu_possible(i, true);
1318 nr_cpu_ids = possible;
1321 #ifdef CONFIG_HOTPLUG_CPU
1323 static void remove_siblinginfo(int cpu)
1326 struct cpuinfo_x86 *c = &cpu_data(cpu);
1328 for_each_cpu(sibling, cpu_core_mask(cpu)) {
1329 cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
1331 * last thread sibling in this cpu core going down
1333 if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
1334 cpu_data(sibling).booted_cores--;
1337 for_each_cpu(sibling, cpu_sibling_mask(cpu))
1338 cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
1339 cpumask_clear(cpu_sibling_mask(cpu));
1340 cpumask_clear(cpu_core_mask(cpu));
1341 c->phys_proc_id = 0;
1343 cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
1346 static void __ref remove_cpu_from_maps(int cpu)
1348 set_cpu_online(cpu, false);
1349 cpumask_clear_cpu(cpu, cpu_callout_mask);
1350 cpumask_clear_cpu(cpu, cpu_callin_mask);
1351 /* was set by cpu_init() */
1352 cpumask_clear_cpu(cpu, cpu_initialized_mask);
1353 numa_remove_cpu(cpu);
1356 void cpu_disable_common(void)
1358 int cpu = smp_processor_id();
1361 * Allow any queued timer interrupts to get serviced
1362 * This is only a temporary solution until we cleanup
1363 * fixup_irqs as we do for IA64.
1368 local_irq_disable();
1369 remove_siblinginfo(cpu);
1371 /* It's now safe to remove this processor from the online map */
1373 remove_cpu_from_maps(cpu);
1374 unlock_vector_lock();
1378 int native_cpu_disable(void)
1380 int cpu = smp_processor_id();
1383 * Perhaps use cpufreq to drop frequency, but that could go
1384 * into generic code.
1386 * We won't take down the boot processor on i386 due to some
1387 * interrupts only being able to be serviced by the BSP.
1388 * Especially so if we're not using an IOAPIC -zwane
1393 if (nmi_watchdog == NMI_LOCAL_APIC)
1394 stop_apic_nmi_watchdog(NULL);
1397 cpu_disable_common();
1401 void native_cpu_die(unsigned int cpu)
1403 /* We don't do anything here: idle task is faking death itself. */
1406 for (i = 0; i < 10; i++) {
1407 /* They ack this in play_dead by setting CPU_DEAD */
1408 if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1409 printk(KERN_INFO "CPU %d is now offline\n", cpu);
1410 if (1 == num_online_cpus())
1411 alternatives_smp_switch(0);
1416 printk(KERN_ERR "CPU %u didn't die...\n", cpu);
1419 void play_dead_common(void)
1422 reset_lazy_tlbstate();
1423 irq_ctx_exit(raw_smp_processor_id());
1424 c1e_remove_cpu(raw_smp_processor_id());
1428 __get_cpu_var(cpu_state) = CPU_DEAD;
1431 * With physical CPU hotplug, we should halt the cpu
1433 local_irq_disable();
1436 void native_play_dead(void)
1442 #else /* ... !CONFIG_HOTPLUG_CPU */
1443 int native_cpu_disable(void)
1448 void native_cpu_die(unsigned int cpu)
1450 /* We said "no" in __cpu_disable */
1454 void native_play_dead(void)