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>
7 * Much of the core SMP work is based on previous work by Thomas Radke, to
8 * whom a great many thanks are extended.
10 * Thanks to Intel for making available several different Pentium,
11 * Pentium Pro and Pentium-II/Xeon MP machines.
12 * Original development of Linux SMP code supported by Caldera.
14 * This code is released under the GNU General Public License version 2 or
18 * Felix Koop : NR_CPUS used properly
19 * Jose Renau : Handle single CPU case.
20 * Alan Cox : By repeated request 8) - Total BogoMIPS report.
21 * Greg Wright : Fix for kernel stacks panic.
22 * Erich Boleyn : MP v1.4 and additional changes.
23 * Matthias Sattler : Changes for 2.1 kernel map.
24 * Michel Lespinasse : Changes for 2.1 kernel map.
25 * Michael Chastain : Change trampoline.S to gnu as.
26 * Alan Cox : Dumb bug: 'B' step PPro's are fine
27 * Ingo Molnar : Added APIC timers, based on code
29 * Ingo Molnar : various cleanups and rewrites
30 * Tigran Aivazian : fixed "0.00 in /proc/uptime on SMP" bug.
31 * Maciej W. Rozycki : Bits for genuine 82489DX APICs
32 * Martin J. Bligh : Added support for multi-quad systems
33 * Dave Jones : Report invalid combinations of Athlon CPUs.
34 * Rusty Russell : Hacked into shape for new "hotplug" boot process. */
36 #include <linux/module.h>
37 #include <linux/init.h>
38 #include <linux/kernel.h>
41 #include <linux/sched.h>
42 #include <linux/kernel_stat.h>
43 #include <linux/bootmem.h>
44 #include <linux/notifier.h>
45 #include <linux/cpu.h>
46 #include <linux/percpu.h>
47 #include <linux/nmi.h>
49 #include <linux/delay.h>
50 #include <linux/mc146818rtc.h>
51 #include <asm/tlbflush.h>
53 #include <asm/arch_hooks.h>
56 #include <mach_apic.h>
57 #include <mach_wakecpu.h>
58 #include <smpboot_hooks.h>
62 /* Set if we find a B stepping CPU */
63 static int __cpuinitdata smp_b_stepping;
65 static cpumask_t smp_commenced_mask;
67 /* which logical CPU number maps to which CPU (physical APIC ID) */
68 u16 x86_cpu_to_apicid_init[NR_CPUS] __initdata =
69 { [0 ... NR_CPUS-1] = BAD_APICID };
70 void *x86_cpu_to_apicid_early_ptr;
71 DEFINE_PER_CPU(u16, x86_cpu_to_apicid) = BAD_APICID;
72 EXPORT_PER_CPU_SYMBOL(x86_cpu_to_apicid);
74 u8 apicid_2_node[MAX_APICID];
77 * Trampoline 80x86 program as an array.
80 extern const unsigned char trampoline_data [];
81 extern const unsigned char trampoline_end [];
82 static unsigned char *trampoline_base;
84 static void map_cpu_to_logical_apicid(void);
86 /* State of each CPU. */
87 DEFINE_PER_CPU(int, cpu_state) = { 0 };
90 * Currently trivial. Write the real->protected mode
91 * bootstrap into the page concerned. The caller
92 * has made sure it's suitably aligned.
95 static unsigned long __cpuinit setup_trampoline(void)
97 memcpy(trampoline_base, trampoline_data, trampoline_end - trampoline_data);
98 return virt_to_phys(trampoline_base);
102 * We are called very early to get the low memory for the
103 * SMP bootup trampoline page.
105 void __init smp_alloc_memory(void)
107 trampoline_base = alloc_bootmem_low_pages(PAGE_SIZE);
109 * Has to be in very low memory so we can execute
112 if (__pa(trampoline_base) >= 0x9F000)
117 * The bootstrap kernel entry code has set these up. Save them for
121 void __cpuinit smp_store_cpu_info(int id)
123 struct cpuinfo_x86 *c = &cpu_data(id);
128 identify_secondary_cpu(c);
130 * Mask B, Pentium, but not Pentium MMX
132 if (c->x86_vendor == X86_VENDOR_INTEL &&
134 c->x86_mask >= 1 && c->x86_mask <= 4 &&
137 * Remember we have B step Pentia with bugs
142 * Certain Athlons might work (for various values of 'work') in SMP
143 * but they are not certified as MP capable.
145 if ((c->x86_vendor == X86_VENDOR_AMD) && (c->x86 == 6)) {
147 if (num_possible_cpus() == 1)
150 /* Athlon 660/661 is valid. */
151 if ((c->x86_model==6) && ((c->x86_mask==0) || (c->x86_mask==1)))
154 /* Duron 670 is valid */
155 if ((c->x86_model==7) && (c->x86_mask==0))
159 * Athlon 662, Duron 671, and Athlon >model 7 have capability bit.
160 * It's worth noting that the A5 stepping (662) of some Athlon XP's
161 * have the MP bit set.
162 * See http://www.heise.de/newsticker/data/jow-18.10.01-000 for more.
164 if (((c->x86_model==6) && (c->x86_mask>=2)) ||
165 ((c->x86_model==7) && (c->x86_mask>=1)) ||
170 /* If we get here, it's not a certified SMP capable AMD system. */
171 add_taint(TAINT_UNSAFE_SMP);
178 static atomic_t init_deasserted;
180 static void __cpuinit smp_callin(void)
183 unsigned long timeout;
186 * If waken up by an INIT in an 82489DX configuration
187 * we may get here before an INIT-deassert IPI reaches
188 * our local APIC. We have to wait for the IPI or we'll
189 * lock up on an APIC access.
191 wait_for_init_deassert(&init_deasserted);
194 * (This works even if the APIC is not enabled.)
196 phys_id = GET_APIC_ID(apic_read(APIC_ID));
197 cpuid = smp_processor_id();
198 if (cpu_isset(cpuid, cpu_callin_map)) {
199 printk("huh, phys CPU#%d, CPU#%d already present??\n",
203 Dprintk("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
206 * STARTUP IPIs are fragile beasts as they might sometimes
207 * trigger some glue motherboard logic. Complete APIC bus
208 * silence for 1 second, this overestimates the time the
209 * boot CPU is spending to send the up to 2 STARTUP IPIs
210 * by a factor of two. This should be enough.
214 * Waiting 2s total for startup (udelay is not yet working)
216 timeout = jiffies + 2*HZ;
217 while (time_before(jiffies, timeout)) {
219 * Has the boot CPU finished it's STARTUP sequence?
221 if (cpu_isset(cpuid, cpu_callout_map))
226 if (!time_before(jiffies, timeout)) {
227 printk("BUG: CPU%d started up but did not get a callout!\n",
233 * the boot CPU has finished the init stage and is spinning
234 * on callin_map until we finish. We are free to set up this
235 * CPU, first the APIC. (this is probably redundant on most
239 Dprintk("CALLIN, before setup_local_APIC().\n");
240 smp_callin_clear_local_apic();
242 map_cpu_to_logical_apicid();
248 Dprintk("Stack at about %p\n",&cpuid);
251 * Save our processor parameters
253 smp_store_cpu_info(cpuid);
256 * Allow the master to continue.
258 cpu_set(cpuid, cpu_callin_map);
264 * Activate a secondary processor.
266 static void __cpuinit start_secondary(void *unused)
269 * Don't put *anything* before cpu_init(), SMP booting is too
270 * fragile that we want to limit the things done here to the
271 * most necessary things.
279 while (!cpu_isset(smp_processor_id(), smp_commenced_mask))
282 * Check TSC synchronization with the BP:
284 check_tsc_sync_target();
286 setup_secondary_clock();
287 if (nmi_watchdog == NMI_IO_APIC) {
288 disable_8259A_irq(0);
289 enable_NMI_through_LVT0();
293 * low-memory mappings have been cleared, flush them from
294 * the local TLBs too.
298 /* This must be done before setting cpu_online_map */
299 set_cpu_sibling_map(raw_smp_processor_id());
303 * We need to hold call_lock, so there is no inconsistency
304 * between the time smp_call_function() determines number of
305 * IPI recipients, and the time when the determination is made
306 * for which cpus receive the IPI. Holding this
307 * lock helps us to not include this cpu in a currently in progress
308 * smp_call_function().
310 lock_ipi_call_lock();
311 cpu_set(smp_processor_id(), cpu_online_map);
312 unlock_ipi_call_lock();
313 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
315 /* We can take interrupts now: we're officially "up". */
323 * Everything has been set up for the secondary
324 * CPUs - they just need to reload everything
325 * from the task structure
326 * This function must not return.
328 void __devinit initialize_secondary(void)
331 * We don't actually need to load the full TSS,
332 * basically just the stack pointer and the ip.
339 :"m" (current->thread.sp),"m" (current->thread.ip));
342 /* Static state in head.S used to set up a CPU */
350 /* which logical CPUs are on which nodes */
351 cpumask_t node_to_cpumask_map[MAX_NUMNODES] __read_mostly =
352 { [0 ... MAX_NUMNODES-1] = CPU_MASK_NONE };
353 EXPORT_SYMBOL(node_to_cpumask_map);
354 /* which node each logical CPU is on */
355 int cpu_to_node_map[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 };
356 EXPORT_SYMBOL(cpu_to_node_map);
358 /* set up a mapping between cpu and node. */
359 static inline void map_cpu_to_node(int cpu, int node)
361 printk("Mapping cpu %d to node %d\n", cpu, node);
362 cpu_set(cpu, node_to_cpumask_map[node]);
363 cpu_to_node_map[cpu] = node;
366 /* undo a mapping between cpu and node. */
367 static inline void unmap_cpu_to_node(int cpu)
371 printk("Unmapping cpu %d from all nodes\n", cpu);
372 for (node = 0; node < MAX_NUMNODES; node ++)
373 cpu_clear(cpu, node_to_cpumask_map[node]);
374 cpu_to_node_map[cpu] = 0;
376 #else /* !CONFIG_NUMA */
378 #define map_cpu_to_node(cpu, node) ({})
379 #define unmap_cpu_to_node(cpu) ({})
381 #endif /* CONFIG_NUMA */
383 u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = BAD_APICID };
385 static void map_cpu_to_logical_apicid(void)
387 int cpu = smp_processor_id();
388 int apicid = logical_smp_processor_id();
389 int node = apicid_to_node(apicid);
391 if (!node_online(node))
392 node = first_online_node;
394 cpu_2_logical_apicid[cpu] = apicid;
395 map_cpu_to_node(cpu, node);
398 static void unmap_cpu_to_logical_apicid(int cpu)
400 cpu_2_logical_apicid[cpu] = BAD_APICID;
401 unmap_cpu_to_node(cpu);
404 static inline void __inquire_remote_apic(int apicid)
406 int i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
407 char *names[] = { "ID", "VERSION", "SPIV" };
409 unsigned long status;
411 printk("Inquiring remote APIC #%d...\n", apicid);
413 for (i = 0; i < ARRAY_SIZE(regs); i++) {
414 printk("... APIC #%d %s: ", apicid, names[i]);
419 status = safe_apic_wait_icr_idle();
421 printk("a previous APIC delivery may have failed\n");
423 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(apicid));
424 apic_write_around(APIC_ICR, APIC_DM_REMRD | regs[i]);
429 status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
430 } while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
433 case APIC_ICR_RR_VALID:
434 status = apic_read(APIC_RRR);
435 printk("%lx\n", status);
443 #ifdef WAKE_SECONDARY_VIA_NMI
445 * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
446 * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
447 * won't ... remember to clear down the APIC, etc later.
450 wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
452 unsigned long send_status, accept_status = 0;
456 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(logical_apicid));
458 /* Boot on the stack */
459 /* Kick the second */
460 apic_write_around(APIC_ICR, APIC_DM_NMI | APIC_DEST_LOGICAL);
462 Dprintk("Waiting for send to finish...\n");
463 send_status = safe_apic_wait_icr_idle();
466 * Give the other CPU some time to accept the IPI.
470 * Due to the Pentium erratum 3AP.
472 maxlvt = lapic_get_maxlvt();
474 apic_read_around(APIC_SPIV);
475 apic_write(APIC_ESR, 0);
477 accept_status = (apic_read(APIC_ESR) & 0xEF);
478 Dprintk("NMI sent.\n");
481 printk("APIC never delivered???\n");
483 printk("APIC delivery error (%lx).\n", accept_status);
485 return (send_status | accept_status);
487 #endif /* WAKE_SECONDARY_VIA_NMI */
489 #ifdef WAKE_SECONDARY_VIA_INIT
491 wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
493 unsigned long send_status, accept_status = 0;
494 int maxlvt, num_starts, j;
497 * Be paranoid about clearing APIC errors.
499 if (APIC_INTEGRATED(apic_version[phys_apicid])) {
500 apic_read_around(APIC_SPIV);
501 apic_write(APIC_ESR, 0);
505 Dprintk("Asserting INIT.\n");
508 * Turn INIT on target chip
510 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
515 apic_write_around(APIC_ICR, APIC_INT_LEVELTRIG | APIC_INT_ASSERT
518 Dprintk("Waiting for send to finish...\n");
519 send_status = safe_apic_wait_icr_idle();
523 Dprintk("Deasserting INIT.\n");
526 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
529 apic_write_around(APIC_ICR, APIC_INT_LEVELTRIG | APIC_DM_INIT);
531 Dprintk("Waiting for send to finish...\n");
532 send_status = safe_apic_wait_icr_idle();
534 atomic_set(&init_deasserted, 1);
537 * Should we send STARTUP IPIs ?
539 * Determine this based on the APIC version.
540 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
542 if (APIC_INTEGRATED(apic_version[phys_apicid]))
548 * Paravirt / VMI wants a startup IPI hook here to set up the
549 * target processor state.
551 startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
552 (unsigned long) stack_start.sp);
555 * Run STARTUP IPI loop.
557 Dprintk("#startup loops: %d.\n", num_starts);
559 maxlvt = lapic_get_maxlvt();
561 for (j = 1; j <= num_starts; j++) {
562 Dprintk("Sending STARTUP #%d.\n",j);
563 apic_read_around(APIC_SPIV);
564 apic_write(APIC_ESR, 0);
566 Dprintk("After apic_write.\n");
573 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
575 /* Boot on the stack */
576 /* Kick the second */
577 apic_write_around(APIC_ICR, APIC_DM_STARTUP
578 | (start_eip >> 12));
581 * Give the other CPU some time to accept the IPI.
585 Dprintk("Startup point 1.\n");
587 Dprintk("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.
595 * Due to the Pentium erratum 3AP.
598 apic_read_around(APIC_SPIV);
599 apic_write(APIC_ESR, 0);
601 accept_status = (apic_read(APIC_ESR) & 0xEF);
602 if (send_status || accept_status)
605 Dprintk("After Startup.\n");
608 printk("APIC never delivered???\n");
610 printk("APIC delivery error (%lx).\n", accept_status);
612 return (send_status | accept_status);
614 #endif /* WAKE_SECONDARY_VIA_INIT */
616 extern cpumask_t cpu_initialized;
617 static inline int alloc_cpu_id(void)
621 cpus_complement(tmp_map, cpu_present_map);
622 cpu = first_cpu(tmp_map);
628 #ifdef CONFIG_HOTPLUG_CPU
629 static struct task_struct * __cpuinitdata cpu_idle_tasks[NR_CPUS];
630 static inline struct task_struct * __cpuinit alloc_idle_task(int cpu)
632 struct task_struct *idle;
634 if ((idle = cpu_idle_tasks[cpu]) != NULL) {
635 /* initialize thread_struct. we really want to avoid destroy
638 idle->thread.sp = (unsigned long)task_pt_regs(idle);
639 init_idle(idle, cpu);
642 idle = fork_idle(cpu);
645 cpu_idle_tasks[cpu] = idle;
649 #define alloc_idle_task(cpu) fork_idle(cpu)
652 static int __cpuinit do_boot_cpu(int apicid, int cpu)
654 * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
655 * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
656 * Returns zero if CPU booted OK, else error code from wakeup_secondary_cpu.
659 struct task_struct *idle;
660 unsigned long boot_error;
662 unsigned long start_eip;
663 unsigned short nmi_high = 0, nmi_low = 0;
666 * Save current MTRR state in case it was changed since early boot
667 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
672 * We can't use kernel_thread since we must avoid to
673 * reschedule the child.
675 idle = alloc_idle_task(cpu);
677 panic("failed fork for CPU %d", cpu);
680 per_cpu(current_task, cpu) = idle;
681 early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
683 idle->thread.ip = (unsigned long) start_secondary;
684 /* start_eip had better be page-aligned! */
685 start_eip = setup_trampoline();
688 alternatives_smp_switch(1);
690 /* So we see what's up */
691 printk("Booting processor %d/%d ip %lx\n", cpu, apicid, start_eip);
692 /* Stack for startup_32 can be just as for start_secondary onwards */
693 stack_start.sp = (void *) idle->thread.sp;
697 per_cpu(x86_cpu_to_apicid, cpu) = apicid;
699 * This grunge runs the startup process for
700 * the targeted processor.
703 atomic_set(&init_deasserted, 0);
705 Dprintk("Setting warm reset code and vector.\n");
707 store_NMI_vector(&nmi_high, &nmi_low);
709 smpboot_setup_warm_reset_vector(start_eip);
712 * Starting actual IPI sequence...
714 boot_error = wakeup_secondary_cpu(apicid, start_eip);
718 * allow APs to start initializing.
720 Dprintk("Before Callout %d.\n", cpu);
721 cpu_set(cpu, cpu_callout_map);
722 Dprintk("After Callout %d.\n", cpu);
725 * Wait 5s total for a response
727 for (timeout = 0; timeout < 50000; timeout++) {
728 if (cpu_isset(cpu, cpu_callin_map))
729 break; /* It has booted */
733 if (cpu_isset(cpu, cpu_callin_map)) {
734 /* number CPUs logically, starting from 1 (BSP is 0) */
736 printk("CPU%d: ", cpu);
737 print_cpu_info(&cpu_data(cpu));
738 Dprintk("CPU has booted.\n");
741 if (*((volatile unsigned char *)trampoline_base)
743 /* trampoline started but...? */
744 printk("Stuck ??\n");
746 /* trampoline code not run */
747 printk("Not responding.\n");
748 inquire_remote_apic(apicid);
753 /* Try to put things back the way they were before ... */
754 unmap_cpu_to_logical_apicid(cpu);
755 cpu_clear(cpu, cpu_callout_map); /* was set here (do_boot_cpu()) */
756 cpu_clear(cpu, cpu_initialized); /* was set by cpu_init() */
759 per_cpu(x86_cpu_to_apicid, cpu) = apicid;
760 cpu_set(cpu, cpu_present_map);
763 /* mark "stuck" area as not stuck */
764 *((volatile unsigned long *)trampoline_base) = 0;
769 #ifdef CONFIG_HOTPLUG_CPU
770 void cpu_exit_clear(void)
772 int cpu = raw_smp_processor_id();
780 cpu_clear(cpu, cpu_callout_map);
781 cpu_clear(cpu, cpu_callin_map);
783 cpu_clear(cpu, smp_commenced_mask);
784 unmap_cpu_to_logical_apicid(cpu);
787 struct warm_boot_cpu_info {
788 struct completion *complete;
789 struct work_struct task;
794 static void __cpuinit do_warm_boot_cpu(struct work_struct *work)
796 struct warm_boot_cpu_info *info =
797 container_of(work, struct warm_boot_cpu_info, task);
798 do_boot_cpu(info->apicid, info->cpu);
799 complete(info->complete);
802 static int __cpuinit __smp_prepare_cpu(int cpu)
804 DECLARE_COMPLETION_ONSTACK(done);
805 struct warm_boot_cpu_info info;
808 apicid = per_cpu(x86_cpu_to_apicid, cpu);
809 if (apicid == BAD_APICID) {
814 info.complete = &done;
815 info.apicid = apicid;
817 INIT_WORK(&info.task, do_warm_boot_cpu);
819 /* init low mem mapping */
820 clone_pgd_range(swapper_pg_dir, swapper_pg_dir + USER_PGD_PTRS,
821 min_t(unsigned long, KERNEL_PGD_PTRS, USER_PGD_PTRS));
823 schedule_work(&info.task);
824 wait_for_completion(&done);
834 * Cycle through the processors sending APIC IPIs to boot each.
837 static int boot_cpu_logical_apicid;
838 /* Where the IO area was mapped on multiquad, always 0 otherwise */
840 #ifdef CONFIG_X86_NUMAQ
841 EXPORT_SYMBOL(xquad_portio);
844 static void __init smp_boot_cpus(unsigned int max_cpus)
846 int apicid, cpu, bit, kicked;
847 unsigned long bogosum = 0;
850 * Setup boot CPU information
852 smp_store_cpu_info(0); /* Final full version of the data */
853 printk("CPU%d: ", 0);
854 print_cpu_info(&cpu_data(0));
856 boot_cpu_physical_apicid = GET_APIC_ID(apic_read(APIC_ID));
857 boot_cpu_logical_apicid = logical_smp_processor_id();
858 per_cpu(x86_cpu_to_apicid, 0) = boot_cpu_physical_apicid;
860 current_thread_info()->cpu = 0;
862 set_cpu_sibling_map(0);
865 * If we couldn't find an SMP configuration at boot time,
866 * get out of here now!
868 if (!smp_found_config && !acpi_lapic) {
869 printk(KERN_NOTICE "SMP motherboard not detected.\n");
870 smpboot_clear_io_apic_irqs();
871 phys_cpu_present_map = physid_mask_of_physid(0);
872 if (APIC_init_uniprocessor())
873 printk(KERN_NOTICE "Local APIC not detected."
874 " Using dummy APIC emulation.\n");
875 map_cpu_to_logical_apicid();
876 cpu_set(0, per_cpu(cpu_sibling_map, 0));
877 cpu_set(0, per_cpu(cpu_core_map, 0));
882 * Should not be necessary because the MP table should list the boot
883 * CPU too, but we do it for the sake of robustness anyway.
884 * Makes no sense to do this check in clustered apic mode, so skip it
886 if (!check_phys_apicid_present(boot_cpu_physical_apicid)) {
887 printk("weird, boot CPU (#%d) not listed by the BIOS.\n",
888 boot_cpu_physical_apicid);
889 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
893 * If we couldn't find a local APIC, then get out of here now!
895 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) && !cpu_has_apic) {
896 printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
897 boot_cpu_physical_apicid);
898 printk(KERN_ERR "... forcing use of dummy APIC emulation. (tell your hw vendor)\n");
899 smpboot_clear_io_apic_irqs();
900 phys_cpu_present_map = physid_mask_of_physid(0);
901 map_cpu_to_logical_apicid();
902 cpu_set(0, per_cpu(cpu_sibling_map, 0));
903 cpu_set(0, per_cpu(cpu_core_map, 0));
910 * If SMP should be disabled, then really disable it!
913 smp_found_config = 0;
914 printk(KERN_INFO "SMP mode deactivated, forcing use of dummy APIC emulation.\n");
916 if (nmi_watchdog == NMI_LOCAL_APIC) {
917 printk(KERN_INFO "activating minimal APIC for NMI watchdog use.\n");
921 smpboot_clear_io_apic_irqs();
922 phys_cpu_present_map = physid_mask_of_physid(0);
923 map_cpu_to_logical_apicid();
924 cpu_set(0, per_cpu(cpu_sibling_map, 0));
925 cpu_set(0, per_cpu(cpu_core_map, 0));
931 map_cpu_to_logical_apicid();
934 setup_portio_remap();
937 * Scan the CPU present map and fire up the other CPUs via do_boot_cpu
939 * In clustered apic mode, phys_cpu_present_map is a constructed thus:
940 * bits 0-3 are quad0, 4-7 are quad1, etc. A perverse twist on the
943 Dprintk("CPU present map: %lx\n", physids_coerce(phys_cpu_present_map));
946 for (bit = 0; kicked < NR_CPUS && bit < MAX_APICS; bit++) {
947 apicid = cpu_present_to_apicid(bit);
949 * Don't even attempt to start the boot CPU!
951 if ((apicid == boot_cpu_apicid) || (apicid == BAD_APICID))
954 if (!check_apicid_present(bit))
956 if (max_cpus <= cpucount+1)
959 if (((cpu = alloc_cpu_id()) <= 0) || do_boot_cpu(apicid, cpu))
960 printk("CPU #%d not responding - cannot use it.\n",
967 * Cleanup possible dangling ends...
969 smpboot_restore_warm_reset_vector();
972 * Allow the user to impress friends.
974 Dprintk("Before bogomips.\n");
975 for_each_possible_cpu(cpu)
976 if (cpu_isset(cpu, cpu_callout_map))
977 bogosum += cpu_data(cpu).loops_per_jiffy;
979 "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
982 (bogosum/(5000/HZ))%100);
984 Dprintk("Before bogocount - setting activated=1.\n");
987 printk(KERN_WARNING "WARNING: SMP operation may be unreliable with B stepping processors.\n");
990 * Don't taint if we are running SMP kernel on a single non-MP
993 if (tainted & TAINT_UNSAFE_SMP) {
995 printk (KERN_INFO "WARNING: This combination of AMD processors is not suitable for SMP.\n");
997 tainted &= ~TAINT_UNSAFE_SMP;
1000 Dprintk("Boot done.\n");
1003 * construct cpu_sibling_map, so that we can tell sibling CPUs
1006 for_each_possible_cpu(cpu) {
1007 cpus_clear(per_cpu(cpu_sibling_map, cpu));
1008 cpus_clear(per_cpu(cpu_core_map, cpu));
1011 cpu_set(0, per_cpu(cpu_sibling_map, 0));
1012 cpu_set(0, per_cpu(cpu_core_map, 0));
1014 smpboot_setup_io_apic();
1019 /* These are wrappers to interface to the new boot process. Someone
1020 who understands all this stuff should rewrite it properly. --RR 15/Jul/02 */
1021 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1023 smp_commenced_mask = cpumask_of_cpu(0);
1024 cpu_callin_map = cpumask_of_cpu(0);
1026 smp_boot_cpus(max_cpus);
1029 void __init native_smp_prepare_boot_cpu(void)
1031 unsigned int cpu = smp_processor_id();
1034 switch_to_new_gdt();
1036 cpu_set(cpu, cpu_online_map);
1037 cpu_set(cpu, cpu_callout_map);
1038 cpu_set(cpu, cpu_present_map);
1039 cpu_set(cpu, cpu_possible_map);
1040 __get_cpu_var(cpu_state) = CPU_ONLINE;
1043 #ifdef CONFIG_HOTPLUG_CPU
1044 int __cpu_disable(void)
1046 cpumask_t map = cpu_online_map;
1047 int cpu = smp_processor_id();
1050 * Perhaps use cpufreq to drop frequency, but that could go
1051 * into generic code.
1053 * We won't take down the boot processor on i386 due to some
1054 * interrupts only being able to be serviced by the BSP.
1055 * Especially so if we're not using an IOAPIC -zwane
1059 if (nmi_watchdog == NMI_LOCAL_APIC)
1060 stop_apic_nmi_watchdog(NULL);
1062 /* Allow any queued timer interrupts to get serviced */
1065 local_irq_disable();
1067 remove_siblinginfo(cpu);
1069 cpu_clear(cpu, map);
1071 /* It's now safe to remove this processor from the online map */
1072 cpu_clear(cpu, cpu_online_map);
1076 void __cpu_die(unsigned int cpu)
1078 /* We don't do anything here: idle task is faking death itself. */
1081 for (i = 0; i < 10; i++) {
1082 /* They ack this in play_dead by setting CPU_DEAD */
1083 if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1084 printk ("CPU %d is now offline\n", cpu);
1085 if (1 == num_online_cpus())
1086 alternatives_smp_switch(0);
1091 printk(KERN_ERR "CPU %u didn't die...\n", cpu);
1093 #else /* ... !CONFIG_HOTPLUG_CPU */
1094 int __cpu_disable(void)
1099 void __cpu_die(unsigned int cpu)
1101 /* We said "no" in __cpu_disable */
1104 #endif /* CONFIG_HOTPLUG_CPU */
1106 int __cpuinit native_cpu_up(unsigned int cpu)
1108 unsigned long flags;
1109 #ifdef CONFIG_HOTPLUG_CPU
1113 * We do warm boot only on cpus that had booted earlier
1114 * Otherwise cold boot is all handled from smp_boot_cpus().
1115 * cpu_callin_map is set during AP kickstart process. Its reset
1116 * when a cpu is taken offline from cpu_exit_clear().
1118 if (!cpu_isset(cpu, cpu_callin_map))
1119 ret = __smp_prepare_cpu(cpu);
1125 /* In case one didn't come up */
1126 if (!cpu_isset(cpu, cpu_callin_map)) {
1127 printk(KERN_DEBUG "skipping cpu%d, didn't come online\n", cpu);
1131 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
1132 /* Unleash the CPU! */
1133 cpu_set(cpu, smp_commenced_mask);
1136 * Check TSC synchronization with the AP (keep irqs disabled
1139 local_irq_save(flags);
1140 check_tsc_sync_source(cpu);
1141 local_irq_restore(flags);
1143 while (!cpu_isset(cpu, cpu_online_map)) {
1145 touch_nmi_watchdog();
1151 void __init native_smp_cpus_done(unsigned int max_cpus)
1153 #ifdef CONFIG_X86_IO_APIC
1154 setup_ioapic_dest();
1159 void __init smp_intr_init(void)
1162 * IRQ0 must be given a fixed assignment and initialized,
1163 * because it's used before the IO-APIC is set up.
1165 set_intr_gate(FIRST_DEVICE_VECTOR, interrupt[0]);
1168 * The reschedule interrupt is a CPU-to-CPU reschedule-helper
1169 * IPI, driven by wakeup.
1171 set_intr_gate(RESCHEDULE_VECTOR, reschedule_interrupt);
1173 /* IPI for invalidation */
1174 set_intr_gate(INVALIDATE_TLB_VECTOR, invalidate_interrupt);
1176 /* IPI for generic function call */
1177 set_intr_gate(CALL_FUNCTION_VECTOR, call_function_interrupt);
1181 * If the BIOS enumerates physical processors before logical,
1182 * maxcpus=N at enumeration-time can be used to disable HT.
1184 static int __init parse_maxcpus(char *arg)
1186 extern unsigned int maxcpus;
1188 maxcpus = simple_strtoul(arg, NULL, 0);
1191 early_param("maxcpus", parse_maxcpus);