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 /* Number of siblings per CPU package */
66 int smp_num_siblings = 1;
67 EXPORT_SYMBOL(smp_num_siblings);
69 /* Last level cache ID of each logical CPU */
70 DEFINE_PER_CPU(u8, cpu_llc_id) = BAD_APICID;
72 /* representing HT siblings of each logical CPU */
73 DEFINE_PER_CPU(cpumask_t, cpu_sibling_map);
74 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
76 /* representing HT and core siblings of each logical CPU */
77 DEFINE_PER_CPU(cpumask_t, cpu_core_map);
78 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
80 /* bitmap of online cpus */
81 cpumask_t cpu_online_map __read_mostly;
82 EXPORT_SYMBOL(cpu_online_map);
84 cpumask_t cpu_callin_map;
85 cpumask_t cpu_callout_map;
86 cpumask_t cpu_possible_map;
87 EXPORT_SYMBOL(cpu_possible_map);
88 static cpumask_t smp_commenced_mask;
90 /* Per CPU bogomips and other parameters */
91 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
92 EXPORT_PER_CPU_SYMBOL(cpu_info);
94 /* which logical CPU number maps to which CPU (physical APIC ID) */
95 u8 x86_cpu_to_apicid_init[NR_CPUS] __initdata =
96 { [0 ... NR_CPUS-1] = BAD_APICID };
97 void *x86_cpu_to_apicid_early_ptr;
98 DEFINE_PER_CPU(u8, x86_cpu_to_apicid) = BAD_APICID;
99 EXPORT_PER_CPU_SYMBOL(x86_cpu_to_apicid);
101 u8 apicid_2_node[MAX_APICID];
104 * Trampoline 80x86 program as an array.
107 extern const unsigned char trampoline_data [];
108 extern const unsigned char trampoline_end [];
109 static unsigned char *trampoline_base;
110 static int trampoline_exec;
112 static void map_cpu_to_logical_apicid(void);
114 /* State of each CPU. */
115 DEFINE_PER_CPU(int, cpu_state) = { 0 };
118 * Currently trivial. Write the real->protected mode
119 * bootstrap into the page concerned. The caller
120 * has made sure it's suitably aligned.
123 static unsigned long __cpuinit setup_trampoline(void)
125 memcpy(trampoline_base, trampoline_data, trampoline_end - trampoline_data);
126 return virt_to_phys(trampoline_base);
130 * We are called very early to get the low memory for the
131 * SMP bootup trampoline page.
133 void __init smp_alloc_memory(void)
135 trampoline_base = (void *) alloc_bootmem_low_pages(PAGE_SIZE);
137 * Has to be in very low memory so we can execute
140 if (__pa(trampoline_base) >= 0x9F000)
143 * Make the SMP trampoline executable:
145 trampoline_exec = set_kernel_exec((unsigned long)trampoline_base, 1);
149 * The bootstrap kernel entry code has set these up. Save them for
153 void __cpuinit smp_store_cpu_info(int id)
155 struct cpuinfo_x86 *c = &cpu_data(id);
160 identify_secondary_cpu(c);
162 * Mask B, Pentium, but not Pentium MMX
164 if (c->x86_vendor == X86_VENDOR_INTEL &&
166 c->x86_mask >= 1 && c->x86_mask <= 4 &&
169 * Remember we have B step Pentia with bugs
174 * Certain Athlons might work (for various values of 'work') in SMP
175 * but they are not certified as MP capable.
177 if ((c->x86_vendor == X86_VENDOR_AMD) && (c->x86 == 6)) {
179 if (num_possible_cpus() == 1)
182 /* Athlon 660/661 is valid. */
183 if ((c->x86_model==6) && ((c->x86_mask==0) || (c->x86_mask==1)))
186 /* Duron 670 is valid */
187 if ((c->x86_model==7) && (c->x86_mask==0))
191 * Athlon 662, Duron 671, and Athlon >model 7 have capability bit.
192 * It's worth noting that the A5 stepping (662) of some Athlon XP's
193 * have the MP bit set.
194 * See http://www.heise.de/newsticker/data/jow-18.10.01-000 for more.
196 if (((c->x86_model==6) && (c->x86_mask>=2)) ||
197 ((c->x86_model==7) && (c->x86_mask>=1)) ||
202 /* If we get here, it's not a certified SMP capable AMD system. */
203 add_taint(TAINT_UNSAFE_SMP);
210 extern void calibrate_delay(void);
212 static atomic_t init_deasserted;
214 static void __cpuinit smp_callin(void)
217 unsigned long timeout;
220 * If waken up by an INIT in an 82489DX configuration
221 * we may get here before an INIT-deassert IPI reaches
222 * our local APIC. We have to wait for the IPI or we'll
223 * lock up on an APIC access.
225 wait_for_init_deassert(&init_deasserted);
228 * (This works even if the APIC is not enabled.)
230 phys_id = GET_APIC_ID(apic_read(APIC_ID));
231 cpuid = smp_processor_id();
232 if (cpu_isset(cpuid, cpu_callin_map)) {
233 printk("huh, phys CPU#%d, CPU#%d already present??\n",
237 Dprintk("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
240 * STARTUP IPIs are fragile beasts as they might sometimes
241 * trigger some glue motherboard logic. Complete APIC bus
242 * silence for 1 second, this overestimates the time the
243 * boot CPU is spending to send the up to 2 STARTUP IPIs
244 * by a factor of two. This should be enough.
248 * Waiting 2s total for startup (udelay is not yet working)
250 timeout = jiffies + 2*HZ;
251 while (time_before(jiffies, timeout)) {
253 * Has the boot CPU finished it's STARTUP sequence?
255 if (cpu_isset(cpuid, cpu_callout_map))
260 if (!time_before(jiffies, timeout)) {
261 printk("BUG: CPU%d started up but did not get a callout!\n",
267 * the boot CPU has finished the init stage and is spinning
268 * on callin_map until we finish. We are free to set up this
269 * CPU, first the APIC. (this is probably redundant on most
273 Dprintk("CALLIN, before setup_local_APIC().\n");
274 smp_callin_clear_local_apic();
276 map_cpu_to_logical_apicid();
282 Dprintk("Stack at about %p\n",&cpuid);
285 * Save our processor parameters
287 smp_store_cpu_info(cpuid);
290 * Allow the master to continue.
292 cpu_set(cpuid, cpu_callin_map);
297 /* maps the cpu to the sched domain representing multi-core */
298 cpumask_t cpu_coregroup_map(int cpu)
300 struct cpuinfo_x86 *c = &cpu_data(cpu);
302 * For perf, we return last level cache shared map.
303 * And for power savings, we return cpu_core_map
305 if (sched_mc_power_savings || sched_smt_power_savings)
306 return per_cpu(cpu_core_map, cpu);
308 return c->llc_shared_map;
311 /* representing cpus for which sibling maps can be computed */
312 static cpumask_t cpu_sibling_setup_map;
314 void __cpuinit set_cpu_sibling_map(int cpu)
317 struct cpuinfo_x86 *c = &cpu_data(cpu);
319 cpu_set(cpu, cpu_sibling_setup_map);
321 if (smp_num_siblings > 1) {
322 for_each_cpu_mask(i, cpu_sibling_setup_map) {
323 if (c->phys_proc_id == cpu_data(i).phys_proc_id &&
324 c->cpu_core_id == cpu_data(i).cpu_core_id) {
325 cpu_set(i, per_cpu(cpu_sibling_map, cpu));
326 cpu_set(cpu, per_cpu(cpu_sibling_map, i));
327 cpu_set(i, per_cpu(cpu_core_map, cpu));
328 cpu_set(cpu, per_cpu(cpu_core_map, i));
329 cpu_set(i, c->llc_shared_map);
330 cpu_set(cpu, cpu_data(i).llc_shared_map);
334 cpu_set(cpu, per_cpu(cpu_sibling_map, cpu));
337 cpu_set(cpu, c->llc_shared_map);
339 if (current_cpu_data.x86_max_cores == 1) {
340 per_cpu(cpu_core_map, cpu) = per_cpu(cpu_sibling_map, cpu);
345 for_each_cpu_mask(i, cpu_sibling_setup_map) {
346 if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
347 per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
348 cpu_set(i, c->llc_shared_map);
349 cpu_set(cpu, cpu_data(i).llc_shared_map);
351 if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
352 cpu_set(i, per_cpu(cpu_core_map, cpu));
353 cpu_set(cpu, per_cpu(cpu_core_map, i));
355 * Does this new cpu bringup a new core?
357 if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1) {
359 * for each core in package, increment
360 * the booted_cores for this new cpu
362 if (first_cpu(per_cpu(cpu_sibling_map, i)) == i)
365 * increment the core count for all
366 * the other cpus in this package
369 cpu_data(i).booted_cores++;
370 } else if (i != cpu && !c->booted_cores)
371 c->booted_cores = cpu_data(i).booted_cores;
377 * Activate a secondary processor.
379 static void __cpuinit start_secondary(void *unused)
382 * Don't put *anything* before cpu_init(), SMP booting is too
383 * fragile that we want to limit the things done here to the
384 * most necessary things.
392 while (!cpu_isset(smp_processor_id(), smp_commenced_mask))
395 * Check TSC synchronization with the BP:
397 check_tsc_sync_target();
399 setup_secondary_clock();
400 if (nmi_watchdog == NMI_IO_APIC) {
401 disable_8259A_irq(0);
402 enable_NMI_through_LVT0();
406 * low-memory mappings have been cleared, flush them from
407 * the local TLBs too.
411 /* This must be done before setting cpu_online_map */
412 set_cpu_sibling_map(raw_smp_processor_id());
416 * We need to hold call_lock, so there is no inconsistency
417 * between the time smp_call_function() determines number of
418 * IPI recipients, and the time when the determination is made
419 * for which cpus receive the IPI. Holding this
420 * lock helps us to not include this cpu in a currently in progress
421 * smp_call_function().
423 lock_ipi_call_lock();
424 cpu_set(smp_processor_id(), cpu_online_map);
425 unlock_ipi_call_lock();
426 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
428 /* We can take interrupts now: we're officially "up". */
436 * Everything has been set up for the secondary
437 * CPUs - they just need to reload everything
438 * from the task structure
439 * This function must not return.
441 void __devinit initialize_secondary(void)
444 * We don't actually need to load the full TSS,
445 * basically just the stack pointer and the ip.
452 :"m" (current->thread.sp),"m" (current->thread.ip));
455 /* Static state in head.S used to set up a CPU */
463 /* which logical CPUs are on which nodes */
464 cpumask_t node_to_cpumask_map[MAX_NUMNODES] __read_mostly =
465 { [0 ... MAX_NUMNODES-1] = CPU_MASK_NONE };
466 EXPORT_SYMBOL(node_to_cpumask_map);
467 /* which node each logical CPU is on */
468 int cpu_to_node_map[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 };
469 EXPORT_SYMBOL(cpu_to_node_map);
471 /* set up a mapping between cpu and node. */
472 static inline void map_cpu_to_node(int cpu, int node)
474 printk("Mapping cpu %d to node %d\n", cpu, node);
475 cpu_set(cpu, node_to_cpumask_map[node]);
476 cpu_to_node_map[cpu] = node;
479 /* undo a mapping between cpu and node. */
480 static inline void unmap_cpu_to_node(int cpu)
484 printk("Unmapping cpu %d from all nodes\n", cpu);
485 for (node = 0; node < MAX_NUMNODES; node ++)
486 cpu_clear(cpu, node_to_cpumask_map[node]);
487 cpu_to_node_map[cpu] = 0;
489 #else /* !CONFIG_NUMA */
491 #define map_cpu_to_node(cpu, node) ({})
492 #define unmap_cpu_to_node(cpu) ({})
494 #endif /* CONFIG_NUMA */
496 u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = BAD_APICID };
498 static void map_cpu_to_logical_apicid(void)
500 int cpu = smp_processor_id();
501 int apicid = logical_smp_processor_id();
502 int node = apicid_to_node(apicid);
504 if (!node_online(node))
505 node = first_online_node;
507 cpu_2_logical_apicid[cpu] = apicid;
508 map_cpu_to_node(cpu, node);
511 static void unmap_cpu_to_logical_apicid(int cpu)
513 cpu_2_logical_apicid[cpu] = BAD_APICID;
514 unmap_cpu_to_node(cpu);
517 static inline void __inquire_remote_apic(int apicid)
519 int i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
520 char *names[] = { "ID", "VERSION", "SPIV" };
522 unsigned long status;
524 printk("Inquiring remote APIC #%d...\n", apicid);
526 for (i = 0; i < ARRAY_SIZE(regs); i++) {
527 printk("... APIC #%d %s: ", apicid, names[i]);
532 status = safe_apic_wait_icr_idle();
534 printk("a previous APIC delivery may have failed\n");
536 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(apicid));
537 apic_write_around(APIC_ICR, APIC_DM_REMRD | regs[i]);
542 status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
543 } while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
546 case APIC_ICR_RR_VALID:
547 status = apic_read(APIC_RRR);
548 printk("%lx\n", status);
556 #ifdef WAKE_SECONDARY_VIA_NMI
558 * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
559 * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
560 * won't ... remember to clear down the APIC, etc later.
563 wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
565 unsigned long send_status, accept_status = 0;
569 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(logical_apicid));
571 /* Boot on the stack */
572 /* Kick the second */
573 apic_write_around(APIC_ICR, APIC_DM_NMI | APIC_DEST_LOGICAL);
575 Dprintk("Waiting for send to finish...\n");
576 send_status = safe_apic_wait_icr_idle();
579 * Give the other CPU some time to accept the IPI.
583 * Due to the Pentium erratum 3AP.
585 maxlvt = lapic_get_maxlvt();
587 apic_read_around(APIC_SPIV);
588 apic_write(APIC_ESR, 0);
590 accept_status = (apic_read(APIC_ESR) & 0xEF);
591 Dprintk("NMI sent.\n");
594 printk("APIC never delivered???\n");
596 printk("APIC delivery error (%lx).\n", accept_status);
598 return (send_status | accept_status);
600 #endif /* WAKE_SECONDARY_VIA_NMI */
602 #ifdef WAKE_SECONDARY_VIA_INIT
604 wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
606 unsigned long send_status, accept_status = 0;
607 int maxlvt, num_starts, j;
610 * Be paranoid about clearing APIC errors.
612 if (APIC_INTEGRATED(apic_version[phys_apicid])) {
613 apic_read_around(APIC_SPIV);
614 apic_write(APIC_ESR, 0);
618 Dprintk("Asserting INIT.\n");
621 * Turn INIT on target chip
623 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
628 apic_write_around(APIC_ICR, APIC_INT_LEVELTRIG | APIC_INT_ASSERT
631 Dprintk("Waiting for send to finish...\n");
632 send_status = safe_apic_wait_icr_idle();
636 Dprintk("Deasserting INIT.\n");
639 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
642 apic_write_around(APIC_ICR, APIC_INT_LEVELTRIG | APIC_DM_INIT);
644 Dprintk("Waiting for send to finish...\n");
645 send_status = safe_apic_wait_icr_idle();
647 atomic_set(&init_deasserted, 1);
650 * Should we send STARTUP IPIs ?
652 * Determine this based on the APIC version.
653 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
655 if (APIC_INTEGRATED(apic_version[phys_apicid]))
661 * Paravirt / VMI wants a startup IPI hook here to set up the
662 * target processor state.
664 startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
665 (unsigned long) stack_start.sp);
668 * Run STARTUP IPI loop.
670 Dprintk("#startup loops: %d.\n", num_starts);
672 maxlvt = lapic_get_maxlvt();
674 for (j = 1; j <= num_starts; j++) {
675 Dprintk("Sending STARTUP #%d.\n",j);
676 apic_read_around(APIC_SPIV);
677 apic_write(APIC_ESR, 0);
679 Dprintk("After apic_write.\n");
686 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
688 /* Boot on the stack */
689 /* Kick the second */
690 apic_write_around(APIC_ICR, APIC_DM_STARTUP
691 | (start_eip >> 12));
694 * Give the other CPU some time to accept the IPI.
698 Dprintk("Startup point 1.\n");
700 Dprintk("Waiting for send to finish...\n");
701 send_status = safe_apic_wait_icr_idle();
704 * Give the other CPU some time to accept the IPI.
708 * Due to the Pentium erratum 3AP.
711 apic_read_around(APIC_SPIV);
712 apic_write(APIC_ESR, 0);
714 accept_status = (apic_read(APIC_ESR) & 0xEF);
715 if (send_status || accept_status)
718 Dprintk("After Startup.\n");
721 printk("APIC never delivered???\n");
723 printk("APIC delivery error (%lx).\n", accept_status);
725 return (send_status | accept_status);
727 #endif /* WAKE_SECONDARY_VIA_INIT */
729 extern cpumask_t cpu_initialized;
730 static inline int alloc_cpu_id(void)
734 cpus_complement(tmp_map, cpu_present_map);
735 cpu = first_cpu(tmp_map);
741 #ifdef CONFIG_HOTPLUG_CPU
742 static struct task_struct * __cpuinitdata cpu_idle_tasks[NR_CPUS];
743 static inline struct task_struct * __cpuinit alloc_idle_task(int cpu)
745 struct task_struct *idle;
747 if ((idle = cpu_idle_tasks[cpu]) != NULL) {
748 /* initialize thread_struct. we really want to avoid destroy
751 idle->thread.sp = (unsigned long)task_pt_regs(idle);
752 init_idle(idle, cpu);
755 idle = fork_idle(cpu);
758 cpu_idle_tasks[cpu] = idle;
762 #define alloc_idle_task(cpu) fork_idle(cpu)
765 static int __cpuinit do_boot_cpu(int apicid, int cpu)
767 * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
768 * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
769 * Returns zero if CPU booted OK, else error code from wakeup_secondary_cpu.
772 struct task_struct *idle;
773 unsigned long boot_error;
775 unsigned long start_eip;
776 unsigned short nmi_high = 0, nmi_low = 0;
779 * Save current MTRR state in case it was changed since early boot
780 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
785 * We can't use kernel_thread since we must avoid to
786 * reschedule the child.
788 idle = alloc_idle_task(cpu);
790 panic("failed fork for CPU %d", cpu);
793 per_cpu(current_task, cpu) = idle;
794 early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
796 idle->thread.ip = (unsigned long) start_secondary;
797 /* start_eip had better be page-aligned! */
798 start_eip = setup_trampoline();
801 alternatives_smp_switch(1);
803 /* So we see what's up */
804 printk("Booting processor %d/%d ip %lx\n", cpu, apicid, start_eip);
805 /* Stack for startup_32 can be just as for start_secondary onwards */
806 stack_start.sp = (void *) idle->thread.sp;
810 per_cpu(x86_cpu_to_apicid, cpu) = apicid;
812 * This grunge runs the startup process for
813 * the targeted processor.
816 atomic_set(&init_deasserted, 0);
818 Dprintk("Setting warm reset code and vector.\n");
820 store_NMI_vector(&nmi_high, &nmi_low);
822 smpboot_setup_warm_reset_vector(start_eip);
825 * Starting actual IPI sequence...
827 boot_error = wakeup_secondary_cpu(apicid, start_eip);
831 * allow APs to start initializing.
833 Dprintk("Before Callout %d.\n", cpu);
834 cpu_set(cpu, cpu_callout_map);
835 Dprintk("After Callout %d.\n", cpu);
838 * Wait 5s total for a response
840 for (timeout = 0; timeout < 50000; timeout++) {
841 if (cpu_isset(cpu, cpu_callin_map))
842 break; /* It has booted */
846 if (cpu_isset(cpu, cpu_callin_map)) {
847 /* number CPUs logically, starting from 1 (BSP is 0) */
849 printk("CPU%d: ", cpu);
850 print_cpu_info(&cpu_data(cpu));
851 Dprintk("CPU has booted.\n");
854 if (*((volatile unsigned char *)trampoline_base)
856 /* trampoline started but...? */
857 printk("Stuck ??\n");
859 /* trampoline code not run */
860 printk("Not responding.\n");
861 inquire_remote_apic(apicid);
866 /* Try to put things back the way they were before ... */
867 unmap_cpu_to_logical_apicid(cpu);
868 cpu_clear(cpu, cpu_callout_map); /* was set here (do_boot_cpu()) */
869 cpu_clear(cpu, cpu_initialized); /* was set by cpu_init() */
872 per_cpu(x86_cpu_to_apicid, cpu) = apicid;
873 cpu_set(cpu, cpu_present_map);
876 /* mark "stuck" area as not stuck */
877 *((volatile unsigned long *)trampoline_base) = 0;
882 #ifdef CONFIG_HOTPLUG_CPU
883 void cpu_exit_clear(void)
885 int cpu = raw_smp_processor_id();
893 cpu_clear(cpu, cpu_callout_map);
894 cpu_clear(cpu, cpu_callin_map);
896 cpu_clear(cpu, smp_commenced_mask);
897 unmap_cpu_to_logical_apicid(cpu);
900 struct warm_boot_cpu_info {
901 struct completion *complete;
902 struct work_struct task;
907 static void __cpuinit do_warm_boot_cpu(struct work_struct *work)
909 struct warm_boot_cpu_info *info =
910 container_of(work, struct warm_boot_cpu_info, task);
911 do_boot_cpu(info->apicid, info->cpu);
912 complete(info->complete);
915 static int __cpuinit __smp_prepare_cpu(int cpu)
917 DECLARE_COMPLETION_ONSTACK(done);
918 struct warm_boot_cpu_info info;
921 apicid = per_cpu(x86_cpu_to_apicid, cpu);
922 if (apicid == BAD_APICID) {
927 info.complete = &done;
928 info.apicid = apicid;
930 INIT_WORK(&info.task, do_warm_boot_cpu);
932 /* init low mem mapping */
933 clone_pgd_range(swapper_pg_dir, swapper_pg_dir + USER_PGD_PTRS,
934 min_t(unsigned long, KERNEL_PGD_PTRS, USER_PGD_PTRS));
936 schedule_work(&info.task);
937 wait_for_completion(&done);
947 * Cycle through the processors sending APIC IPIs to boot each.
950 static int boot_cpu_logical_apicid;
951 /* Where the IO area was mapped on multiquad, always 0 otherwise */
953 #ifdef CONFIG_X86_NUMAQ
954 EXPORT_SYMBOL(xquad_portio);
957 static void __init smp_boot_cpus(unsigned int max_cpus)
959 int apicid, cpu, bit, kicked;
960 unsigned long bogosum = 0;
963 * Setup boot CPU information
965 smp_store_cpu_info(0); /* Final full version of the data */
966 printk("CPU%d: ", 0);
967 print_cpu_info(&cpu_data(0));
969 boot_cpu_physical_apicid = GET_APIC_ID(apic_read(APIC_ID));
970 boot_cpu_logical_apicid = logical_smp_processor_id();
971 per_cpu(x86_cpu_to_apicid, 0) = boot_cpu_physical_apicid;
973 current_thread_info()->cpu = 0;
975 set_cpu_sibling_map(0);
978 * If we couldn't find an SMP configuration at boot time,
979 * get out of here now!
981 if (!smp_found_config && !acpi_lapic) {
982 printk(KERN_NOTICE "SMP motherboard not detected.\n");
983 smpboot_clear_io_apic_irqs();
984 phys_cpu_present_map = physid_mask_of_physid(0);
985 if (APIC_init_uniprocessor())
986 printk(KERN_NOTICE "Local APIC not detected."
987 " Using dummy APIC emulation.\n");
988 map_cpu_to_logical_apicid();
989 cpu_set(0, per_cpu(cpu_sibling_map, 0));
990 cpu_set(0, per_cpu(cpu_core_map, 0));
995 * Should not be necessary because the MP table should list the boot
996 * CPU too, but we do it for the sake of robustness anyway.
997 * Makes no sense to do this check in clustered apic mode, so skip it
999 if (!check_phys_apicid_present(boot_cpu_physical_apicid)) {
1000 printk("weird, boot CPU (#%d) not listed by the BIOS.\n",
1001 boot_cpu_physical_apicid);
1002 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1006 * If we couldn't find a local APIC, then get out of here now!
1008 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) && !cpu_has_apic) {
1009 printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
1010 boot_cpu_physical_apicid);
1011 printk(KERN_ERR "... forcing use of dummy APIC emulation. (tell your hw vendor)\n");
1012 smpboot_clear_io_apic_irqs();
1013 phys_cpu_present_map = physid_mask_of_physid(0);
1014 map_cpu_to_logical_apicid();
1015 cpu_set(0, per_cpu(cpu_sibling_map, 0));
1016 cpu_set(0, per_cpu(cpu_core_map, 0));
1020 verify_local_APIC();
1023 * If SMP should be disabled, then really disable it!
1026 smp_found_config = 0;
1027 printk(KERN_INFO "SMP mode deactivated, forcing use of dummy APIC emulation.\n");
1029 if (nmi_watchdog == NMI_LOCAL_APIC) {
1030 printk(KERN_INFO "activating minimal APIC for NMI watchdog use.\n");
1034 smpboot_clear_io_apic_irqs();
1035 phys_cpu_present_map = physid_mask_of_physid(0);
1036 map_cpu_to_logical_apicid();
1037 cpu_set(0, per_cpu(cpu_sibling_map, 0));
1038 cpu_set(0, per_cpu(cpu_core_map, 0));
1044 map_cpu_to_logical_apicid();
1047 setup_portio_remap();
1050 * Scan the CPU present map and fire up the other CPUs via do_boot_cpu
1052 * In clustered apic mode, phys_cpu_present_map is a constructed thus:
1053 * bits 0-3 are quad0, 4-7 are quad1, etc. A perverse twist on the
1054 * clustered apic ID.
1056 Dprintk("CPU present map: %lx\n", physids_coerce(phys_cpu_present_map));
1059 for (bit = 0; kicked < NR_CPUS && bit < MAX_APICS; bit++) {
1060 apicid = cpu_present_to_apicid(bit);
1062 * Don't even attempt to start the boot CPU!
1064 if ((apicid == boot_cpu_apicid) || (apicid == BAD_APICID))
1067 if (!check_apicid_present(bit))
1069 if (max_cpus <= cpucount+1)
1072 if (((cpu = alloc_cpu_id()) <= 0) || do_boot_cpu(apicid, cpu))
1073 printk("CPU #%d not responding - cannot use it.\n",
1080 * Cleanup possible dangling ends...
1082 smpboot_restore_warm_reset_vector();
1085 * Allow the user to impress friends.
1087 Dprintk("Before bogomips.\n");
1088 for_each_possible_cpu(cpu)
1089 if (cpu_isset(cpu, cpu_callout_map))
1090 bogosum += cpu_data(cpu).loops_per_jiffy;
1092 "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
1094 bogosum/(500000/HZ),
1095 (bogosum/(5000/HZ))%100);
1097 Dprintk("Before bogocount - setting activated=1.\n");
1100 printk(KERN_WARNING "WARNING: SMP operation may be unreliable with B stepping processors.\n");
1103 * Don't taint if we are running SMP kernel on a single non-MP
1106 if (tainted & TAINT_UNSAFE_SMP) {
1108 printk (KERN_INFO "WARNING: This combination of AMD processors is not suitable for SMP.\n");
1110 tainted &= ~TAINT_UNSAFE_SMP;
1113 Dprintk("Boot done.\n");
1116 * construct cpu_sibling_map, so that we can tell sibling CPUs
1119 for_each_possible_cpu(cpu) {
1120 cpus_clear(per_cpu(cpu_sibling_map, cpu));
1121 cpus_clear(per_cpu(cpu_core_map, cpu));
1124 cpu_set(0, per_cpu(cpu_sibling_map, 0));
1125 cpu_set(0, per_cpu(cpu_core_map, 0));
1127 smpboot_setup_io_apic();
1132 /* These are wrappers to interface to the new boot process. Someone
1133 who understands all this stuff should rewrite it properly. --RR 15/Jul/02 */
1134 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1136 smp_commenced_mask = cpumask_of_cpu(0);
1137 cpu_callin_map = cpumask_of_cpu(0);
1139 smp_boot_cpus(max_cpus);
1142 void __init native_smp_prepare_boot_cpu(void)
1144 unsigned int cpu = smp_processor_id();
1147 switch_to_new_gdt();
1149 cpu_set(cpu, cpu_online_map);
1150 cpu_set(cpu, cpu_callout_map);
1151 cpu_set(cpu, cpu_present_map);
1152 cpu_set(cpu, cpu_possible_map);
1153 __get_cpu_var(cpu_state) = CPU_ONLINE;
1156 #ifdef CONFIG_HOTPLUG_CPU
1157 void remove_siblinginfo(int cpu)
1160 struct cpuinfo_x86 *c = &cpu_data(cpu);
1162 for_each_cpu_mask(sibling, per_cpu(cpu_core_map, cpu)) {
1163 cpu_clear(cpu, per_cpu(cpu_core_map, sibling));
1165 * last thread sibling in this cpu core going down
1167 if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1)
1168 cpu_data(sibling).booted_cores--;
1171 for_each_cpu_mask(sibling, per_cpu(cpu_sibling_map, cpu))
1172 cpu_clear(cpu, per_cpu(cpu_sibling_map, sibling));
1173 cpus_clear(per_cpu(cpu_sibling_map, cpu));
1174 cpus_clear(per_cpu(cpu_core_map, cpu));
1175 c->phys_proc_id = 0;
1177 cpu_clear(cpu, cpu_sibling_setup_map);
1180 int __cpu_disable(void)
1182 cpumask_t map = cpu_online_map;
1183 int cpu = smp_processor_id();
1186 * Perhaps use cpufreq to drop frequency, but that could go
1187 * into generic code.
1189 * We won't take down the boot processor on i386 due to some
1190 * interrupts only being able to be serviced by the BSP.
1191 * Especially so if we're not using an IOAPIC -zwane
1195 if (nmi_watchdog == NMI_LOCAL_APIC)
1196 stop_apic_nmi_watchdog(NULL);
1198 /* Allow any queued timer interrupts to get serviced */
1201 local_irq_disable();
1203 remove_siblinginfo(cpu);
1205 cpu_clear(cpu, map);
1207 /* It's now safe to remove this processor from the online map */
1208 cpu_clear(cpu, cpu_online_map);
1212 void __cpu_die(unsigned int cpu)
1214 /* We don't do anything here: idle task is faking death itself. */
1217 for (i = 0; i < 10; i++) {
1218 /* They ack this in play_dead by setting CPU_DEAD */
1219 if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1220 printk ("CPU %d is now offline\n", cpu);
1221 if (1 == num_online_cpus())
1222 alternatives_smp_switch(0);
1227 printk(KERN_ERR "CPU %u didn't die...\n", cpu);
1229 #else /* ... !CONFIG_HOTPLUG_CPU */
1230 int __cpu_disable(void)
1235 void __cpu_die(unsigned int cpu)
1237 /* We said "no" in __cpu_disable */
1240 #endif /* CONFIG_HOTPLUG_CPU */
1242 int __cpuinit native_cpu_up(unsigned int cpu)
1244 unsigned long flags;
1245 #ifdef CONFIG_HOTPLUG_CPU
1249 * We do warm boot only on cpus that had booted earlier
1250 * Otherwise cold boot is all handled from smp_boot_cpus().
1251 * cpu_callin_map is set during AP kickstart process. Its reset
1252 * when a cpu is taken offline from cpu_exit_clear().
1254 if (!cpu_isset(cpu, cpu_callin_map))
1255 ret = __smp_prepare_cpu(cpu);
1261 /* In case one didn't come up */
1262 if (!cpu_isset(cpu, cpu_callin_map)) {
1263 printk(KERN_DEBUG "skipping cpu%d, didn't come online\n", cpu);
1267 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
1268 /* Unleash the CPU! */
1269 cpu_set(cpu, smp_commenced_mask);
1272 * Check TSC synchronization with the AP (keep irqs disabled
1275 local_irq_save(flags);
1276 check_tsc_sync_source(cpu);
1277 local_irq_restore(flags);
1279 while (!cpu_isset(cpu, cpu_online_map)) {
1281 touch_nmi_watchdog();
1287 void __init native_smp_cpus_done(unsigned int max_cpus)
1289 #ifdef CONFIG_X86_IO_APIC
1290 setup_ioapic_dest();
1293 #ifndef CONFIG_HOTPLUG_CPU
1295 * Disable executability of the SMP trampoline:
1297 set_kernel_exec((unsigned long)trampoline_base, trampoline_exec);
1301 void __init smp_intr_init(void)
1304 * IRQ0 must be given a fixed assignment and initialized,
1305 * because it's used before the IO-APIC is set up.
1307 set_intr_gate(FIRST_DEVICE_VECTOR, interrupt[0]);
1310 * The reschedule interrupt is a CPU-to-CPU reschedule-helper
1311 * IPI, driven by wakeup.
1313 set_intr_gate(RESCHEDULE_VECTOR, reschedule_interrupt);
1315 /* IPI for invalidation */
1316 set_intr_gate(INVALIDATE_TLB_VECTOR, invalidate_interrupt);
1318 /* IPI for generic function call */
1319 set_intr_gate(CALL_FUNCTION_VECTOR, call_function_interrupt);
1323 * If the BIOS enumerates physical processors before logical,
1324 * maxcpus=N at enumeration-time can be used to disable HT.
1326 static int __init parse_maxcpus(char *arg)
1328 extern unsigned int maxcpus;
1330 maxcpus = simple_strtoul(arg, NULL, 0);
1333 early_param("maxcpus", parse_maxcpus);