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 __devinitdata 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 int cpu_llc_id[NR_CPUS] __cpuinitdata = {[0 ... NR_CPUS-1] = BAD_APICID};
72 /* representing HT siblings of each logical CPU */
73 cpumask_t cpu_sibling_map[NR_CPUS] __read_mostly;
74 EXPORT_SYMBOL(cpu_sibling_map);
76 /* representing HT and core siblings of each logical CPU */
77 cpumask_t cpu_core_map[NR_CPUS] __read_mostly;
78 EXPORT_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 EXPORT_SYMBOL(cpu_callout_map);
87 cpumask_t cpu_possible_map;
88 EXPORT_SYMBOL(cpu_possible_map);
89 static cpumask_t smp_commenced_mask;
91 /* Per CPU bogomips and other parameters */
92 struct cpuinfo_x86 cpu_data[NR_CPUS] __cacheline_aligned;
93 EXPORT_SYMBOL(cpu_data);
95 u8 x86_cpu_to_apicid[NR_CPUS] __read_mostly =
96 { [0 ... NR_CPUS-1] = 0xff };
97 EXPORT_SYMBOL(x86_cpu_to_apicid);
99 u8 apicid_2_node[MAX_APICID];
102 * Trampoline 80x86 program as an array.
105 extern unsigned char trampoline_data [];
106 extern unsigned char trampoline_end [];
107 static unsigned char *trampoline_base;
108 static int trampoline_exec;
110 static void map_cpu_to_logical_apicid(void);
112 /* State of each CPU. */
113 DEFINE_PER_CPU(int, cpu_state) = { 0 };
116 * Currently trivial. Write the real->protected mode
117 * bootstrap into the page concerned. The caller
118 * has made sure it's suitably aligned.
121 static unsigned long __devinit setup_trampoline(void)
123 memcpy(trampoline_base, trampoline_data, trampoline_end - trampoline_data);
124 return virt_to_phys(trampoline_base);
128 * We are called very early to get the low memory for the
129 * SMP bootup trampoline page.
131 void __init smp_alloc_memory(void)
133 trampoline_base = (void *) alloc_bootmem_low_pages(PAGE_SIZE);
135 * Has to be in very low memory so we can execute
138 if (__pa(trampoline_base) >= 0x9F000)
141 * Make the SMP trampoline executable:
143 trampoline_exec = set_kernel_exec((unsigned long)trampoline_base, 1);
147 * The bootstrap kernel entry code has set these up. Save them for
151 void __cpuinit smp_store_cpu_info(int id)
153 struct cpuinfo_x86 *c = cpu_data + id;
157 identify_secondary_cpu(c);
159 * Mask B, Pentium, but not Pentium MMX
161 if (c->x86_vendor == X86_VENDOR_INTEL &&
163 c->x86_mask >= 1 && c->x86_mask <= 4 &&
166 * Remember we have B step Pentia with bugs
171 * Certain Athlons might work (for various values of 'work') in SMP
172 * but they are not certified as MP capable.
174 if ((c->x86_vendor == X86_VENDOR_AMD) && (c->x86 == 6)) {
176 if (num_possible_cpus() == 1)
179 /* Athlon 660/661 is valid. */
180 if ((c->x86_model==6) && ((c->x86_mask==0) || (c->x86_mask==1)))
183 /* Duron 670 is valid */
184 if ((c->x86_model==7) && (c->x86_mask==0))
188 * Athlon 662, Duron 671, and Athlon >model 7 have capability bit.
189 * It's worth noting that the A5 stepping (662) of some Athlon XP's
190 * have the MP bit set.
191 * See http://www.heise.de/newsticker/data/jow-18.10.01-000 for more.
193 if (((c->x86_model==6) && (c->x86_mask>=2)) ||
194 ((c->x86_model==7) && (c->x86_mask>=1)) ||
199 /* If we get here, it's not a certified SMP capable AMD system. */
200 add_taint(TAINT_UNSAFE_SMP);
207 extern void calibrate_delay(void);
209 static atomic_t init_deasserted;
211 static void __cpuinit smp_callin(void)
214 unsigned long timeout;
217 * If waken up by an INIT in an 82489DX configuration
218 * we may get here before an INIT-deassert IPI reaches
219 * our local APIC. We have to wait for the IPI or we'll
220 * lock up on an APIC access.
222 wait_for_init_deassert(&init_deasserted);
225 * (This works even if the APIC is not enabled.)
227 phys_id = GET_APIC_ID(apic_read(APIC_ID));
228 cpuid = smp_processor_id();
229 if (cpu_isset(cpuid, cpu_callin_map)) {
230 printk("huh, phys CPU#%d, CPU#%d already present??\n",
234 Dprintk("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
237 * STARTUP IPIs are fragile beasts as they might sometimes
238 * trigger some glue motherboard logic. Complete APIC bus
239 * silence for 1 second, this overestimates the time the
240 * boot CPU is spending to send the up to 2 STARTUP IPIs
241 * by a factor of two. This should be enough.
245 * Waiting 2s total for startup (udelay is not yet working)
247 timeout = jiffies + 2*HZ;
248 while (time_before(jiffies, timeout)) {
250 * Has the boot CPU finished it's STARTUP sequence?
252 if (cpu_isset(cpuid, cpu_callout_map))
257 if (!time_before(jiffies, timeout)) {
258 printk("BUG: CPU%d started up but did not get a callout!\n",
264 * the boot CPU has finished the init stage and is spinning
265 * on callin_map until we finish. We are free to set up this
266 * CPU, first the APIC. (this is probably redundant on most
270 Dprintk("CALLIN, before setup_local_APIC().\n");
271 smp_callin_clear_local_apic();
273 map_cpu_to_logical_apicid();
279 Dprintk("Stack at about %p\n",&cpuid);
282 * Save our processor parameters
284 smp_store_cpu_info(cpuid);
287 * Allow the master to continue.
289 cpu_set(cpuid, cpu_callin_map);
294 /* maps the cpu to the sched domain representing multi-core */
295 cpumask_t cpu_coregroup_map(int cpu)
297 struct cpuinfo_x86 *c = cpu_data + cpu;
299 * For perf, we return last level cache shared map.
300 * And for power savings, we return cpu_core_map
302 if (sched_mc_power_savings || sched_smt_power_savings)
303 return cpu_core_map[cpu];
305 return c->llc_shared_map;
308 /* representing cpus for which sibling maps can be computed */
309 static cpumask_t cpu_sibling_setup_map;
311 void set_cpu_sibling_map(int cpu)
314 struct cpuinfo_x86 *c = cpu_data;
316 cpu_set(cpu, cpu_sibling_setup_map);
318 if (smp_num_siblings > 1) {
319 for_each_cpu_mask(i, cpu_sibling_setup_map) {
320 if (c[cpu].phys_proc_id == c[i].phys_proc_id &&
321 c[cpu].cpu_core_id == c[i].cpu_core_id) {
322 cpu_set(i, cpu_sibling_map[cpu]);
323 cpu_set(cpu, cpu_sibling_map[i]);
324 cpu_set(i, cpu_core_map[cpu]);
325 cpu_set(cpu, cpu_core_map[i]);
326 cpu_set(i, c[cpu].llc_shared_map);
327 cpu_set(cpu, c[i].llc_shared_map);
331 cpu_set(cpu, cpu_sibling_map[cpu]);
334 cpu_set(cpu, c[cpu].llc_shared_map);
336 if (current_cpu_data.x86_max_cores == 1) {
337 cpu_core_map[cpu] = cpu_sibling_map[cpu];
338 c[cpu].booted_cores = 1;
342 for_each_cpu_mask(i, cpu_sibling_setup_map) {
343 if (cpu_llc_id[cpu] != BAD_APICID &&
344 cpu_llc_id[cpu] == cpu_llc_id[i]) {
345 cpu_set(i, c[cpu].llc_shared_map);
346 cpu_set(cpu, c[i].llc_shared_map);
348 if (c[cpu].phys_proc_id == c[i].phys_proc_id) {
349 cpu_set(i, cpu_core_map[cpu]);
350 cpu_set(cpu, cpu_core_map[i]);
352 * Does this new cpu bringup a new core?
354 if (cpus_weight(cpu_sibling_map[cpu]) == 1) {
356 * for each core in package, increment
357 * the booted_cores for this new cpu
359 if (first_cpu(cpu_sibling_map[i]) == i)
360 c[cpu].booted_cores++;
362 * increment the core count for all
363 * the other cpus in this package
367 } else if (i != cpu && !c[cpu].booted_cores)
368 c[cpu].booted_cores = c[i].booted_cores;
374 * Activate a secondary processor.
376 static void __cpuinit start_secondary(void *unused)
379 * Don't put *anything* before cpu_init(), SMP booting is too
380 * fragile that we want to limit the things done here to the
381 * most necessary things.
389 while (!cpu_isset(smp_processor_id(), smp_commenced_mask))
392 * Check TSC synchronization with the BP:
394 check_tsc_sync_target();
396 setup_secondary_clock();
397 if (nmi_watchdog == NMI_IO_APIC) {
398 disable_8259A_irq(0);
399 enable_NMI_through_LVT0(NULL);
403 * low-memory mappings have been cleared, flush them from
404 * the local TLBs too.
408 /* This must be done before setting cpu_online_map */
409 set_cpu_sibling_map(raw_smp_processor_id());
413 * We need to hold call_lock, so there is no inconsistency
414 * between the time smp_call_function() determines number of
415 * IPI receipients, and the time when the determination is made
416 * for which cpus receive the IPI. Holding this
417 * lock helps us to not include this cpu in a currently in progress
418 * smp_call_function().
420 lock_ipi_call_lock();
421 cpu_set(smp_processor_id(), cpu_online_map);
422 unlock_ipi_call_lock();
423 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
425 /* We can take interrupts now: we're officially "up". */
433 * Everything has been set up for the secondary
434 * CPUs - they just need to reload everything
435 * from the task structure
436 * This function must not return.
438 void __devinit initialize_secondary(void)
441 * We don't actually need to load the full TSS,
442 * basically just the stack pointer and the eip.
449 :"m" (current->thread.esp),"m" (current->thread.eip));
452 /* Static state in head.S used to set up a CPU */
460 /* which logical CPUs are on which nodes */
461 cpumask_t node_2_cpu_mask[MAX_NUMNODES] __read_mostly =
462 { [0 ... MAX_NUMNODES-1] = CPU_MASK_NONE };
463 EXPORT_SYMBOL(node_2_cpu_mask);
464 /* which node each logical CPU is on */
465 int cpu_2_node[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 };
466 EXPORT_SYMBOL(cpu_2_node);
468 /* set up a mapping between cpu and node. */
469 static inline void map_cpu_to_node(int cpu, int node)
471 printk("Mapping cpu %d to node %d\n", cpu, node);
472 cpu_set(cpu, node_2_cpu_mask[node]);
473 cpu_2_node[cpu] = node;
476 /* undo a mapping between cpu and node. */
477 static inline void unmap_cpu_to_node(int cpu)
481 printk("Unmapping cpu %d from all nodes\n", cpu);
482 for (node = 0; node < MAX_NUMNODES; node ++)
483 cpu_clear(cpu, node_2_cpu_mask[node]);
486 #else /* !CONFIG_NUMA */
488 #define map_cpu_to_node(cpu, node) ({})
489 #define unmap_cpu_to_node(cpu) ({})
491 #endif /* CONFIG_NUMA */
493 u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = BAD_APICID };
495 static void map_cpu_to_logical_apicid(void)
497 int cpu = smp_processor_id();
498 int apicid = logical_smp_processor_id();
499 int node = apicid_to_node(apicid);
501 if (!node_online(node))
502 node = first_online_node;
504 cpu_2_logical_apicid[cpu] = apicid;
505 map_cpu_to_node(cpu, node);
508 static void unmap_cpu_to_logical_apicid(int cpu)
510 cpu_2_logical_apicid[cpu] = BAD_APICID;
511 unmap_cpu_to_node(cpu);
514 static inline void __inquire_remote_apic(int apicid)
516 int i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
517 char *names[] = { "ID", "VERSION", "SPIV" };
519 unsigned long status;
521 printk("Inquiring remote APIC #%d...\n", apicid);
523 for (i = 0; i < ARRAY_SIZE(regs); i++) {
524 printk("... APIC #%d %s: ", apicid, names[i]);
529 status = safe_apic_wait_icr_idle();
531 printk("a previous APIC delivery may have failed\n");
533 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(apicid));
534 apic_write_around(APIC_ICR, APIC_DM_REMRD | regs[i]);
539 status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
540 } while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
543 case APIC_ICR_RR_VALID:
544 status = apic_read(APIC_RRR);
545 printk("%lx\n", status);
553 #ifdef WAKE_SECONDARY_VIA_NMI
555 * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
556 * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
557 * won't ... remember to clear down the APIC, etc later.
560 wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
562 unsigned long send_status, accept_status = 0;
566 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(logical_apicid));
568 /* Boot on the stack */
569 /* Kick the second */
570 apic_write_around(APIC_ICR, APIC_DM_NMI | APIC_DEST_LOGICAL);
572 Dprintk("Waiting for send to finish...\n");
573 send_status = safe_apic_wait_icr_idle();
576 * Give the other CPU some time to accept the IPI.
580 * Due to the Pentium erratum 3AP.
582 maxlvt = lapic_get_maxlvt();
584 apic_read_around(APIC_SPIV);
585 apic_write(APIC_ESR, 0);
587 accept_status = (apic_read(APIC_ESR) & 0xEF);
588 Dprintk("NMI sent.\n");
591 printk("APIC never delivered???\n");
593 printk("APIC delivery error (%lx).\n", accept_status);
595 return (send_status | accept_status);
597 #endif /* WAKE_SECONDARY_VIA_NMI */
599 #ifdef WAKE_SECONDARY_VIA_INIT
601 wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
603 unsigned long send_status, accept_status = 0;
604 int maxlvt, num_starts, j;
607 * Be paranoid about clearing APIC errors.
609 if (APIC_INTEGRATED(apic_version[phys_apicid])) {
610 apic_read_around(APIC_SPIV);
611 apic_write(APIC_ESR, 0);
615 Dprintk("Asserting INIT.\n");
618 * Turn INIT on target chip
620 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
625 apic_write_around(APIC_ICR, APIC_INT_LEVELTRIG | APIC_INT_ASSERT
628 Dprintk("Waiting for send to finish...\n");
629 send_status = safe_apic_wait_icr_idle();
633 Dprintk("Deasserting INIT.\n");
636 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
639 apic_write_around(APIC_ICR, APIC_INT_LEVELTRIG | APIC_DM_INIT);
641 Dprintk("Waiting for send to finish...\n");
642 send_status = safe_apic_wait_icr_idle();
644 atomic_set(&init_deasserted, 1);
647 * Should we send STARTUP IPIs ?
649 * Determine this based on the APIC version.
650 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
652 if (APIC_INTEGRATED(apic_version[phys_apicid]))
658 * Paravirt / VMI wants a startup IPI hook here to set up the
659 * target processor state.
661 startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
662 (unsigned long) stack_start.esp);
665 * Run STARTUP IPI loop.
667 Dprintk("#startup loops: %d.\n", num_starts);
669 maxlvt = lapic_get_maxlvt();
671 for (j = 1; j <= num_starts; j++) {
672 Dprintk("Sending STARTUP #%d.\n",j);
673 apic_read_around(APIC_SPIV);
674 apic_write(APIC_ESR, 0);
676 Dprintk("After apic_write.\n");
683 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
685 /* Boot on the stack */
686 /* Kick the second */
687 apic_write_around(APIC_ICR, APIC_DM_STARTUP
688 | (start_eip >> 12));
691 * Give the other CPU some time to accept the IPI.
695 Dprintk("Startup point 1.\n");
697 Dprintk("Waiting for send to finish...\n");
698 send_status = safe_apic_wait_icr_idle();
701 * Give the other CPU some time to accept the IPI.
705 * Due to the Pentium erratum 3AP.
708 apic_read_around(APIC_SPIV);
709 apic_write(APIC_ESR, 0);
711 accept_status = (apic_read(APIC_ESR) & 0xEF);
712 if (send_status || accept_status)
715 Dprintk("After Startup.\n");
718 printk("APIC never delivered???\n");
720 printk("APIC delivery error (%lx).\n", accept_status);
722 return (send_status | accept_status);
724 #endif /* WAKE_SECONDARY_VIA_INIT */
726 extern cpumask_t cpu_initialized;
727 static inline int alloc_cpu_id(void)
731 cpus_complement(tmp_map, cpu_present_map);
732 cpu = first_cpu(tmp_map);
738 #ifdef CONFIG_HOTPLUG_CPU
739 static struct task_struct * __devinitdata cpu_idle_tasks[NR_CPUS];
740 static inline struct task_struct * alloc_idle_task(int cpu)
742 struct task_struct *idle;
744 if ((idle = cpu_idle_tasks[cpu]) != NULL) {
745 /* initialize thread_struct. we really want to avoid destroy
748 idle->thread.esp = (unsigned long)task_pt_regs(idle);
749 init_idle(idle, cpu);
752 idle = fork_idle(cpu);
755 cpu_idle_tasks[cpu] = idle;
759 #define alloc_idle_task(cpu) fork_idle(cpu)
762 static int __cpuinit do_boot_cpu(int apicid, int cpu)
764 * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
765 * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
766 * Returns zero if CPU booted OK, else error code from wakeup_secondary_cpu.
769 struct task_struct *idle;
770 unsigned long boot_error;
772 unsigned long start_eip;
773 unsigned short nmi_high = 0, nmi_low = 0;
776 * Save current MTRR state in case it was changed since early boot
777 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
782 * We can't use kernel_thread since we must avoid to
783 * reschedule the child.
785 idle = alloc_idle_task(cpu);
787 panic("failed fork for CPU %d", cpu);
790 per_cpu(current_task, cpu) = idle;
791 early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
793 idle->thread.eip = (unsigned long) start_secondary;
794 /* start_eip had better be page-aligned! */
795 start_eip = setup_trampoline();
798 alternatives_smp_switch(1);
800 /* So we see what's up */
801 printk("Booting processor %d/%d eip %lx\n", cpu, apicid, start_eip);
802 /* Stack for startup_32 can be just as for start_secondary onwards */
803 stack_start.esp = (void *) idle->thread.esp;
807 x86_cpu_to_apicid[cpu] = apicid;
809 * This grunge runs the startup process for
810 * the targeted processor.
813 atomic_set(&init_deasserted, 0);
815 Dprintk("Setting warm reset code and vector.\n");
817 store_NMI_vector(&nmi_high, &nmi_low);
819 smpboot_setup_warm_reset_vector(start_eip);
822 * Starting actual IPI sequence...
824 boot_error = wakeup_secondary_cpu(apicid, start_eip);
828 * allow APs to start initializing.
830 Dprintk("Before Callout %d.\n", cpu);
831 cpu_set(cpu, cpu_callout_map);
832 Dprintk("After Callout %d.\n", cpu);
835 * Wait 5s total for a response
837 for (timeout = 0; timeout < 50000; timeout++) {
838 if (cpu_isset(cpu, cpu_callin_map))
839 break; /* It has booted */
843 if (cpu_isset(cpu, cpu_callin_map)) {
844 /* number CPUs logically, starting from 1 (BSP is 0) */
846 printk("CPU%d: ", cpu);
847 print_cpu_info(&cpu_data[cpu]);
848 Dprintk("CPU has booted.\n");
851 if (*((volatile unsigned char *)trampoline_base)
853 /* trampoline started but...? */
854 printk("Stuck ??\n");
856 /* trampoline code not run */
857 printk("Not responding.\n");
858 inquire_remote_apic(apicid);
863 /* Try to put things back the way they were before ... */
864 unmap_cpu_to_logical_apicid(cpu);
865 cpu_clear(cpu, cpu_callout_map); /* was set here (do_boot_cpu()) */
866 cpu_clear(cpu, cpu_initialized); /* was set by cpu_init() */
869 x86_cpu_to_apicid[cpu] = apicid;
870 cpu_set(cpu, cpu_present_map);
873 /* mark "stuck" area as not stuck */
874 *((volatile unsigned long *)trampoline_base) = 0;
879 #ifdef CONFIG_HOTPLUG_CPU
880 void cpu_exit_clear(void)
882 int cpu = raw_smp_processor_id();
890 cpu_clear(cpu, cpu_callout_map);
891 cpu_clear(cpu, cpu_callin_map);
893 cpu_clear(cpu, smp_commenced_mask);
894 unmap_cpu_to_logical_apicid(cpu);
897 struct warm_boot_cpu_info {
898 struct completion *complete;
899 struct work_struct task;
904 static void __cpuinit do_warm_boot_cpu(struct work_struct *work)
906 struct warm_boot_cpu_info *info =
907 container_of(work, struct warm_boot_cpu_info, task);
908 do_boot_cpu(info->apicid, info->cpu);
909 complete(info->complete);
912 static int __cpuinit __smp_prepare_cpu(int cpu)
914 DECLARE_COMPLETION_ONSTACK(done);
915 struct warm_boot_cpu_info info;
918 apicid = x86_cpu_to_apicid[cpu];
919 if (apicid == BAD_APICID) {
924 info.complete = &done;
925 info.apicid = apicid;
927 INIT_WORK(&info.task, do_warm_boot_cpu);
929 /* init low mem mapping */
930 clone_pgd_range(swapper_pg_dir, swapper_pg_dir + USER_PGD_PTRS,
931 min_t(unsigned long, KERNEL_PGD_PTRS, USER_PGD_PTRS));
933 schedule_work(&info.task);
934 wait_for_completion(&done);
944 * Cycle through the processors sending APIC IPIs to boot each.
947 static int boot_cpu_logical_apicid;
948 /* Where the IO area was mapped on multiquad, always 0 otherwise */
950 #ifdef CONFIG_X86_NUMAQ
951 EXPORT_SYMBOL(xquad_portio);
954 static void __init smp_boot_cpus(unsigned int max_cpus)
956 int apicid, cpu, bit, kicked;
957 unsigned long bogosum = 0;
960 * Setup boot CPU information
962 smp_store_cpu_info(0); /* Final full version of the data */
963 printk("CPU%d: ", 0);
964 print_cpu_info(&cpu_data[0]);
966 boot_cpu_physical_apicid = GET_APIC_ID(apic_read(APIC_ID));
967 boot_cpu_logical_apicid = logical_smp_processor_id();
968 x86_cpu_to_apicid[0] = boot_cpu_physical_apicid;
970 current_thread_info()->cpu = 0;
972 set_cpu_sibling_map(0);
975 * If we couldn't find an SMP configuration at boot time,
976 * get out of here now!
978 if (!smp_found_config && !acpi_lapic) {
979 printk(KERN_NOTICE "SMP motherboard not detected.\n");
980 smpboot_clear_io_apic_irqs();
981 phys_cpu_present_map = physid_mask_of_physid(0);
982 if (APIC_init_uniprocessor())
983 printk(KERN_NOTICE "Local APIC not detected."
984 " Using dummy APIC emulation.\n");
985 map_cpu_to_logical_apicid();
986 cpu_set(0, cpu_sibling_map[0]);
987 cpu_set(0, cpu_core_map[0]);
992 * Should not be necessary because the MP table should list the boot
993 * CPU too, but we do it for the sake of robustness anyway.
994 * Makes no sense to do this check in clustered apic mode, so skip it
996 if (!check_phys_apicid_present(boot_cpu_physical_apicid)) {
997 printk("weird, boot CPU (#%d) not listed by the BIOS.\n",
998 boot_cpu_physical_apicid);
999 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1003 * If we couldn't find a local APIC, then get out of here now!
1005 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) && !cpu_has_apic) {
1006 printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
1007 boot_cpu_physical_apicid);
1008 printk(KERN_ERR "... forcing use of dummy APIC emulation. (tell your hw vendor)\n");
1009 smpboot_clear_io_apic_irqs();
1010 phys_cpu_present_map = physid_mask_of_physid(0);
1011 cpu_set(0, cpu_sibling_map[0]);
1012 cpu_set(0, cpu_core_map[0]);
1016 verify_local_APIC();
1019 * If SMP should be disabled, then really disable it!
1022 smp_found_config = 0;
1023 printk(KERN_INFO "SMP mode deactivated, forcing use of dummy APIC emulation.\n");
1024 smpboot_clear_io_apic_irqs();
1025 phys_cpu_present_map = physid_mask_of_physid(0);
1026 cpu_set(0, cpu_sibling_map[0]);
1027 cpu_set(0, cpu_core_map[0]);
1033 map_cpu_to_logical_apicid();
1036 setup_portio_remap();
1039 * Scan the CPU present map and fire up the other CPUs via do_boot_cpu
1041 * In clustered apic mode, phys_cpu_present_map is a constructed thus:
1042 * bits 0-3 are quad0, 4-7 are quad1, etc. A perverse twist on the
1043 * clustered apic ID.
1045 Dprintk("CPU present map: %lx\n", physids_coerce(phys_cpu_present_map));
1048 for (bit = 0; kicked < NR_CPUS && bit < MAX_APICS; bit++) {
1049 apicid = cpu_present_to_apicid(bit);
1051 * Don't even attempt to start the boot CPU!
1053 if ((apicid == boot_cpu_apicid) || (apicid == BAD_APICID))
1056 if (!check_apicid_present(bit))
1058 if (max_cpus <= cpucount+1)
1061 if (((cpu = alloc_cpu_id()) <= 0) || do_boot_cpu(apicid, cpu))
1062 printk("CPU #%d not responding - cannot use it.\n",
1069 * Cleanup possible dangling ends...
1071 smpboot_restore_warm_reset_vector();
1074 * Allow the user to impress friends.
1076 Dprintk("Before bogomips.\n");
1077 for (cpu = 0; cpu < NR_CPUS; cpu++)
1078 if (cpu_isset(cpu, cpu_callout_map))
1079 bogosum += cpu_data[cpu].loops_per_jiffy;
1081 "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
1083 bogosum/(500000/HZ),
1084 (bogosum/(5000/HZ))%100);
1086 Dprintk("Before bogocount - setting activated=1.\n");
1089 printk(KERN_WARNING "WARNING: SMP operation may be unreliable with B stepping processors.\n");
1092 * Don't taint if we are running SMP kernel on a single non-MP
1095 if (tainted & TAINT_UNSAFE_SMP) {
1097 printk (KERN_INFO "WARNING: This combination of AMD processors is not suitable for SMP.\n");
1099 tainted &= ~TAINT_UNSAFE_SMP;
1102 Dprintk("Boot done.\n");
1105 * construct cpu_sibling_map[], so that we can tell sibling CPUs
1108 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1109 cpus_clear(cpu_sibling_map[cpu]);
1110 cpus_clear(cpu_core_map[cpu]);
1113 cpu_set(0, cpu_sibling_map[0]);
1114 cpu_set(0, cpu_core_map[0]);
1116 smpboot_setup_io_apic();
1121 /* These are wrappers to interface to the new boot process. Someone
1122 who understands all this stuff should rewrite it properly. --RR 15/Jul/02 */
1123 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1125 smp_commenced_mask = cpumask_of_cpu(0);
1126 cpu_callin_map = cpumask_of_cpu(0);
1128 smp_boot_cpus(max_cpus);
1131 void __init native_smp_prepare_boot_cpu(void)
1133 unsigned int cpu = smp_processor_id();
1136 switch_to_new_gdt();
1138 cpu_set(cpu, cpu_online_map);
1139 cpu_set(cpu, cpu_callout_map);
1140 cpu_set(cpu, cpu_present_map);
1141 cpu_set(cpu, cpu_possible_map);
1142 __get_cpu_var(cpu_state) = CPU_ONLINE;
1145 #ifdef CONFIG_HOTPLUG_CPU
1146 void remove_siblinginfo(int cpu)
1149 struct cpuinfo_x86 *c = cpu_data;
1151 for_each_cpu_mask(sibling, cpu_core_map[cpu]) {
1152 cpu_clear(cpu, cpu_core_map[sibling]);
1154 * last thread sibling in this cpu core going down
1156 if (cpus_weight(cpu_sibling_map[cpu]) == 1)
1157 c[sibling].booted_cores--;
1160 for_each_cpu_mask(sibling, cpu_sibling_map[cpu])
1161 cpu_clear(cpu, cpu_sibling_map[sibling]);
1162 cpus_clear(cpu_sibling_map[cpu]);
1163 cpus_clear(cpu_core_map[cpu]);
1164 c[cpu].phys_proc_id = 0;
1165 c[cpu].cpu_core_id = 0;
1166 cpu_clear(cpu, cpu_sibling_setup_map);
1169 int __cpu_disable(void)
1171 cpumask_t map = cpu_online_map;
1172 int cpu = smp_processor_id();
1175 * Perhaps use cpufreq to drop frequency, but that could go
1176 * into generic code.
1178 * We won't take down the boot processor on i386 due to some
1179 * interrupts only being able to be serviced by the BSP.
1180 * Especially so if we're not using an IOAPIC -zwane
1184 if (nmi_watchdog == NMI_LOCAL_APIC)
1185 stop_apic_nmi_watchdog(NULL);
1187 /* Allow any queued timer interrupts to get serviced */
1190 local_irq_disable();
1192 remove_siblinginfo(cpu);
1194 cpu_clear(cpu, map);
1196 /* It's now safe to remove this processor from the online map */
1197 cpu_clear(cpu, cpu_online_map);
1201 void __cpu_die(unsigned int cpu)
1203 /* We don't do anything here: idle task is faking death itself. */
1206 for (i = 0; i < 10; i++) {
1207 /* They ack this in play_dead by setting CPU_DEAD */
1208 if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1209 printk ("CPU %d is now offline\n", cpu);
1210 if (1 == num_online_cpus())
1211 alternatives_smp_switch(0);
1216 printk(KERN_ERR "CPU %u didn't die...\n", cpu);
1218 #else /* ... !CONFIG_HOTPLUG_CPU */
1219 int __cpu_disable(void)
1224 void __cpu_die(unsigned int cpu)
1226 /* We said "no" in __cpu_disable */
1229 #endif /* CONFIG_HOTPLUG_CPU */
1231 int __cpuinit native_cpu_up(unsigned int cpu)
1233 unsigned long flags;
1234 #ifdef CONFIG_HOTPLUG_CPU
1238 * We do warm boot only on cpus that had booted earlier
1239 * Otherwise cold boot is all handled from smp_boot_cpus().
1240 * cpu_callin_map is set during AP kickstart process. Its reset
1241 * when a cpu is taken offline from cpu_exit_clear().
1243 if (!cpu_isset(cpu, cpu_callin_map))
1244 ret = __smp_prepare_cpu(cpu);
1250 /* In case one didn't come up */
1251 if (!cpu_isset(cpu, cpu_callin_map)) {
1252 printk(KERN_DEBUG "skipping cpu%d, didn't come online\n", cpu);
1256 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
1257 /* Unleash the CPU! */
1258 cpu_set(cpu, smp_commenced_mask);
1261 * Check TSC synchronization with the AP (keep irqs disabled
1264 local_irq_save(flags);
1265 check_tsc_sync_source(cpu);
1266 local_irq_restore(flags);
1268 while (!cpu_isset(cpu, cpu_online_map)) {
1270 touch_nmi_watchdog();
1276 void __init native_smp_cpus_done(unsigned int max_cpus)
1278 #ifdef CONFIG_X86_IO_APIC
1279 setup_ioapic_dest();
1282 #ifndef CONFIG_HOTPLUG_CPU
1284 * Disable executability of the SMP trampoline:
1286 set_kernel_exec((unsigned long)trampoline_base, trampoline_exec);
1290 void __init smp_intr_init(void)
1293 * IRQ0 must be given a fixed assignment and initialized,
1294 * because it's used before the IO-APIC is set up.
1296 set_intr_gate(FIRST_DEVICE_VECTOR, interrupt[0]);
1299 * The reschedule interrupt is a CPU-to-CPU reschedule-helper
1300 * IPI, driven by wakeup.
1302 set_intr_gate(RESCHEDULE_VECTOR, reschedule_interrupt);
1304 /* IPI for invalidation */
1305 set_intr_gate(INVALIDATE_TLB_VECTOR, invalidate_interrupt);
1307 /* IPI for generic function call */
1308 set_intr_gate(CALL_FUNCTION_VECTOR, call_function_interrupt);
1312 * If the BIOS enumerates physical processors before logical,
1313 * maxcpus=N at enumeration-time can be used to disable HT.
1315 static int __init parse_maxcpus(char *arg)
1317 extern unsigned int maxcpus;
1319 maxcpus = simple_strtoul(arg, NULL, 0);
1322 early_param("maxcpus", parse_maxcpus);