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 static 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;
312 set_cpu_sibling_map(int cpu)
315 struct cpuinfo_x86 *c = cpu_data;
317 cpu_set(cpu, cpu_sibling_setup_map);
319 if (smp_num_siblings > 1) {
320 for_each_cpu_mask(i, cpu_sibling_setup_map) {
321 if (c[cpu].phys_proc_id == c[i].phys_proc_id &&
322 c[cpu].cpu_core_id == c[i].cpu_core_id) {
323 cpu_set(i, cpu_sibling_map[cpu]);
324 cpu_set(cpu, cpu_sibling_map[i]);
325 cpu_set(i, cpu_core_map[cpu]);
326 cpu_set(cpu, cpu_core_map[i]);
327 cpu_set(i, c[cpu].llc_shared_map);
328 cpu_set(cpu, c[i].llc_shared_map);
332 cpu_set(cpu, cpu_sibling_map[cpu]);
335 cpu_set(cpu, c[cpu].llc_shared_map);
337 if (current_cpu_data.x86_max_cores == 1) {
338 cpu_core_map[cpu] = cpu_sibling_map[cpu];
339 c[cpu].booted_cores = 1;
343 for_each_cpu_mask(i, cpu_sibling_setup_map) {
344 if (cpu_llc_id[cpu] != BAD_APICID &&
345 cpu_llc_id[cpu] == cpu_llc_id[i]) {
346 cpu_set(i, c[cpu].llc_shared_map);
347 cpu_set(cpu, c[i].llc_shared_map);
349 if (c[cpu].phys_proc_id == c[i].phys_proc_id) {
350 cpu_set(i, cpu_core_map[cpu]);
351 cpu_set(cpu, cpu_core_map[i]);
353 * Does this new cpu bringup a new core?
355 if (cpus_weight(cpu_sibling_map[cpu]) == 1) {
357 * for each core in package, increment
358 * the booted_cores for this new cpu
360 if (first_cpu(cpu_sibling_map[i]) == i)
361 c[cpu].booted_cores++;
363 * increment the core count for all
364 * the other cpus in this package
368 } else if (i != cpu && !c[cpu].booted_cores)
369 c[cpu].booted_cores = c[i].booted_cores;
375 * Activate a secondary processor.
377 static void __cpuinit start_secondary(void *unused)
380 * Don't put *anything* before cpu_init(), SMP booting is too
381 * fragile that we want to limit the things done here to the
382 * most necessary things.
390 while (!cpu_isset(smp_processor_id(), smp_commenced_mask))
393 * Check TSC synchronization with the BP:
395 check_tsc_sync_target();
397 setup_secondary_clock();
398 if (nmi_watchdog == NMI_IO_APIC) {
399 disable_8259A_irq(0);
400 enable_NMI_through_LVT0(NULL);
404 * low-memory mappings have been cleared, flush them from
405 * the local TLBs too.
409 /* This must be done before setting cpu_online_map */
410 set_cpu_sibling_map(raw_smp_processor_id());
414 * We need to hold call_lock, so there is no inconsistency
415 * between the time smp_call_function() determines number of
416 * IPI receipients, and the time when the determination is made
417 * for which cpus receive the IPI. Holding this
418 * lock helps us to not include this cpu in a currently in progress
419 * smp_call_function().
421 lock_ipi_call_lock();
422 cpu_set(smp_processor_id(), cpu_online_map);
423 unlock_ipi_call_lock();
424 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
426 /* We can take interrupts now: we're officially "up". */
434 * Everything has been set up for the secondary
435 * CPUs - they just need to reload everything
436 * from the task structure
437 * This function must not return.
439 void __devinit initialize_secondary(void)
442 * We don't actually need to load the full TSS,
443 * basically just the stack pointer and the eip.
450 :"m" (current->thread.esp),"m" (current->thread.eip));
453 /* Static state in head.S used to set up a CPU */
461 /* which logical CPUs are on which nodes */
462 cpumask_t node_2_cpu_mask[MAX_NUMNODES] __read_mostly =
463 { [0 ... MAX_NUMNODES-1] = CPU_MASK_NONE };
464 EXPORT_SYMBOL(node_2_cpu_mask);
465 /* which node each logical CPU is on */
466 int cpu_2_node[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 };
467 EXPORT_SYMBOL(cpu_2_node);
469 /* set up a mapping between cpu and node. */
470 static inline void map_cpu_to_node(int cpu, int node)
472 printk("Mapping cpu %d to node %d\n", cpu, node);
473 cpu_set(cpu, node_2_cpu_mask[node]);
474 cpu_2_node[cpu] = node;
477 /* undo a mapping between cpu and node. */
478 static inline void unmap_cpu_to_node(int cpu)
482 printk("Unmapping cpu %d from all nodes\n", cpu);
483 for (node = 0; node < MAX_NUMNODES; node ++)
484 cpu_clear(cpu, node_2_cpu_mask[node]);
487 #else /* !CONFIG_NUMA */
489 #define map_cpu_to_node(cpu, node) ({})
490 #define unmap_cpu_to_node(cpu) ({})
492 #endif /* CONFIG_NUMA */
494 u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = BAD_APICID };
496 static void map_cpu_to_logical_apicid(void)
498 int cpu = smp_processor_id();
499 int apicid = logical_smp_processor_id();
500 int node = apicid_to_node(apicid);
502 if (!node_online(node))
503 node = first_online_node;
505 cpu_2_logical_apicid[cpu] = apicid;
506 map_cpu_to_node(cpu, node);
509 static void unmap_cpu_to_logical_apicid(int cpu)
511 cpu_2_logical_apicid[cpu] = BAD_APICID;
512 unmap_cpu_to_node(cpu);
515 static inline void __inquire_remote_apic(int apicid)
517 int i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
518 char *names[] = { "ID", "VERSION", "SPIV" };
520 unsigned long status;
522 printk("Inquiring remote APIC #%d...\n", apicid);
524 for (i = 0; i < ARRAY_SIZE(regs); i++) {
525 printk("... APIC #%d %s: ", apicid, names[i]);
530 status = safe_apic_wait_icr_idle();
532 printk("a previous APIC delivery may have failed\n");
534 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(apicid));
535 apic_write_around(APIC_ICR, APIC_DM_REMRD | regs[i]);
540 status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
541 } while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
544 case APIC_ICR_RR_VALID:
545 status = apic_read(APIC_RRR);
546 printk("%lx\n", status);
554 #ifdef WAKE_SECONDARY_VIA_NMI
556 * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
557 * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
558 * won't ... remember to clear down the APIC, etc later.
561 wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
563 unsigned long send_status, accept_status = 0;
567 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(logical_apicid));
569 /* Boot on the stack */
570 /* Kick the second */
571 apic_write_around(APIC_ICR, APIC_DM_NMI | APIC_DEST_LOGICAL);
573 Dprintk("Waiting for send to finish...\n");
574 send_status = safe_apic_wait_icr_idle();
577 * Give the other CPU some time to accept the IPI.
581 * Due to the Pentium erratum 3AP.
583 maxlvt = lapic_get_maxlvt();
585 apic_read_around(APIC_SPIV);
586 apic_write(APIC_ESR, 0);
588 accept_status = (apic_read(APIC_ESR) & 0xEF);
589 Dprintk("NMI sent.\n");
592 printk("APIC never delivered???\n");
594 printk("APIC delivery error (%lx).\n", accept_status);
596 return (send_status | accept_status);
598 #endif /* WAKE_SECONDARY_VIA_NMI */
600 #ifdef WAKE_SECONDARY_VIA_INIT
602 wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
604 unsigned long send_status, accept_status = 0;
605 int maxlvt, num_starts, j;
608 * Be paranoid about clearing APIC errors.
610 if (APIC_INTEGRATED(apic_version[phys_apicid])) {
611 apic_read_around(APIC_SPIV);
612 apic_write(APIC_ESR, 0);
616 Dprintk("Asserting INIT.\n");
619 * Turn INIT on target chip
621 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
626 apic_write_around(APIC_ICR, APIC_INT_LEVELTRIG | APIC_INT_ASSERT
629 Dprintk("Waiting for send to finish...\n");
630 send_status = safe_apic_wait_icr_idle();
634 Dprintk("Deasserting INIT.\n");
637 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
640 apic_write_around(APIC_ICR, APIC_INT_LEVELTRIG | APIC_DM_INIT);
642 Dprintk("Waiting for send to finish...\n");
643 send_status = safe_apic_wait_icr_idle();
645 atomic_set(&init_deasserted, 1);
648 * Should we send STARTUP IPIs ?
650 * Determine this based on the APIC version.
651 * If we don't have an integrated APIC, don't send the STARTUP IPIs.
653 if (APIC_INTEGRATED(apic_version[phys_apicid]))
659 * Paravirt / VMI wants a startup IPI hook here to set up the
660 * target processor state.
662 startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
663 (unsigned long) stack_start.esp);
666 * Run STARTUP IPI loop.
668 Dprintk("#startup loops: %d.\n", num_starts);
670 maxlvt = lapic_get_maxlvt();
672 for (j = 1; j <= num_starts; j++) {
673 Dprintk("Sending STARTUP #%d.\n",j);
674 apic_read_around(APIC_SPIV);
675 apic_write(APIC_ESR, 0);
677 Dprintk("After apic_write.\n");
684 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
686 /* Boot on the stack */
687 /* Kick the second */
688 apic_write_around(APIC_ICR, APIC_DM_STARTUP
689 | (start_eip >> 12));
692 * Give the other CPU some time to accept the IPI.
696 Dprintk("Startup point 1.\n");
698 Dprintk("Waiting for send to finish...\n");
699 send_status = safe_apic_wait_icr_idle();
702 * Give the other CPU some time to accept the IPI.
706 * Due to the Pentium erratum 3AP.
709 apic_read_around(APIC_SPIV);
710 apic_write(APIC_ESR, 0);
712 accept_status = (apic_read(APIC_ESR) & 0xEF);
713 if (send_status || accept_status)
716 Dprintk("After Startup.\n");
719 printk("APIC never delivered???\n");
721 printk("APIC delivery error (%lx).\n", accept_status);
723 return (send_status | accept_status);
725 #endif /* WAKE_SECONDARY_VIA_INIT */
727 extern cpumask_t cpu_initialized;
728 static inline int alloc_cpu_id(void)
732 cpus_complement(tmp_map, cpu_present_map);
733 cpu = first_cpu(tmp_map);
739 #ifdef CONFIG_HOTPLUG_CPU
740 static struct task_struct * __devinitdata cpu_idle_tasks[NR_CPUS];
741 static inline struct task_struct * alloc_idle_task(int cpu)
743 struct task_struct *idle;
745 if ((idle = cpu_idle_tasks[cpu]) != NULL) {
746 /* initialize thread_struct. we really want to avoid destroy
749 idle->thread.esp = (unsigned long)task_pt_regs(idle);
750 init_idle(idle, cpu);
753 idle = fork_idle(cpu);
756 cpu_idle_tasks[cpu] = idle;
760 #define alloc_idle_task(cpu) fork_idle(cpu)
763 static int __cpuinit do_boot_cpu(int apicid, int cpu)
765 * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
766 * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
767 * Returns zero if CPU booted OK, else error code from wakeup_secondary_cpu.
770 struct task_struct *idle;
771 unsigned long boot_error;
773 unsigned long start_eip;
774 unsigned short nmi_high = 0, nmi_low = 0;
777 * Save current MTRR state in case it was changed since early boot
778 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
783 * We can't use kernel_thread since we must avoid to
784 * reschedule the child.
786 idle = alloc_idle_task(cpu);
788 panic("failed fork for CPU %d", cpu);
791 per_cpu(current_task, cpu) = idle;
792 early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
794 idle->thread.eip = (unsigned long) start_secondary;
795 /* start_eip had better be page-aligned! */
796 start_eip = setup_trampoline();
799 alternatives_smp_switch(1);
801 /* So we see what's up */
802 printk("Booting processor %d/%d eip %lx\n", cpu, apicid, start_eip);
803 /* Stack for startup_32 can be just as for start_secondary onwards */
804 stack_start.esp = (void *) idle->thread.esp;
808 x86_cpu_to_apicid[cpu] = apicid;
810 * This grunge runs the startup process for
811 * the targeted processor.
814 atomic_set(&init_deasserted, 0);
816 Dprintk("Setting warm reset code and vector.\n");
818 store_NMI_vector(&nmi_high, &nmi_low);
820 smpboot_setup_warm_reset_vector(start_eip);
823 * Starting actual IPI sequence...
825 boot_error = wakeup_secondary_cpu(apicid, start_eip);
829 * allow APs to start initializing.
831 Dprintk("Before Callout %d.\n", cpu);
832 cpu_set(cpu, cpu_callout_map);
833 Dprintk("After Callout %d.\n", cpu);
836 * Wait 5s total for a response
838 for (timeout = 0; timeout < 50000; timeout++) {
839 if (cpu_isset(cpu, cpu_callin_map))
840 break; /* It has booted */
844 if (cpu_isset(cpu, cpu_callin_map)) {
845 /* number CPUs logically, starting from 1 (BSP is 0) */
847 printk("CPU%d: ", cpu);
848 print_cpu_info(&cpu_data[cpu]);
849 Dprintk("CPU has booted.\n");
852 if (*((volatile unsigned char *)trampoline_base)
854 /* trampoline started but...? */
855 printk("Stuck ??\n");
857 /* trampoline code not run */
858 printk("Not responding.\n");
859 inquire_remote_apic(apicid);
864 /* Try to put things back the way they were before ... */
865 unmap_cpu_to_logical_apicid(cpu);
866 cpu_clear(cpu, cpu_callout_map); /* was set here (do_boot_cpu()) */
867 cpu_clear(cpu, cpu_initialized); /* was set by cpu_init() */
870 x86_cpu_to_apicid[cpu] = apicid;
871 cpu_set(cpu, cpu_present_map);
874 /* mark "stuck" area as not stuck */
875 *((volatile unsigned long *)trampoline_base) = 0;
880 #ifdef CONFIG_HOTPLUG_CPU
881 void cpu_exit_clear(void)
883 int cpu = raw_smp_processor_id();
891 cpu_clear(cpu, cpu_callout_map);
892 cpu_clear(cpu, cpu_callin_map);
894 cpu_clear(cpu, smp_commenced_mask);
895 unmap_cpu_to_logical_apicid(cpu);
898 struct warm_boot_cpu_info {
899 struct completion *complete;
900 struct work_struct task;
905 static void __cpuinit do_warm_boot_cpu(struct work_struct *work)
907 struct warm_boot_cpu_info *info =
908 container_of(work, struct warm_boot_cpu_info, task);
909 do_boot_cpu(info->apicid, info->cpu);
910 complete(info->complete);
913 static int __cpuinit __smp_prepare_cpu(int cpu)
915 DECLARE_COMPLETION_ONSTACK(done);
916 struct warm_boot_cpu_info info;
919 apicid = x86_cpu_to_apicid[cpu];
920 if (apicid == BAD_APICID) {
925 info.complete = &done;
926 info.apicid = apicid;
928 INIT_WORK(&info.task, do_warm_boot_cpu);
930 /* init low mem mapping */
931 clone_pgd_range(swapper_pg_dir, swapper_pg_dir + USER_PGD_PTRS,
932 min_t(unsigned long, KERNEL_PGD_PTRS, USER_PGD_PTRS));
934 schedule_work(&info.task);
935 wait_for_completion(&done);
944 static void smp_tune_scheduling(void)
947 /* cache size in kB */
948 long cachesize = boot_cpu_data.x86_cache_size;
951 max_cache_size = cachesize * 1024;
956 * Cycle through the processors sending APIC IPIs to boot each.
959 static int boot_cpu_logical_apicid;
960 /* Where the IO area was mapped on multiquad, always 0 otherwise */
962 #ifdef CONFIG_X86_NUMAQ
963 EXPORT_SYMBOL(xquad_portio);
966 static void __init smp_boot_cpus(unsigned int max_cpus)
968 int apicid, cpu, bit, kicked;
969 unsigned long bogosum = 0;
972 * Setup boot CPU information
974 smp_store_cpu_info(0); /* Final full version of the data */
975 printk("CPU%d: ", 0);
976 print_cpu_info(&cpu_data[0]);
978 boot_cpu_physical_apicid = GET_APIC_ID(apic_read(APIC_ID));
979 boot_cpu_logical_apicid = logical_smp_processor_id();
980 x86_cpu_to_apicid[0] = boot_cpu_physical_apicid;
982 current_thread_info()->cpu = 0;
983 smp_tune_scheduling();
985 set_cpu_sibling_map(0);
988 * If we couldn't find an SMP configuration at boot time,
989 * get out of here now!
991 if (!smp_found_config && !acpi_lapic) {
992 printk(KERN_NOTICE "SMP motherboard not detected.\n");
993 smpboot_clear_io_apic_irqs();
994 phys_cpu_present_map = physid_mask_of_physid(0);
995 if (APIC_init_uniprocessor())
996 printk(KERN_NOTICE "Local APIC not detected."
997 " Using dummy APIC emulation.\n");
998 map_cpu_to_logical_apicid();
999 cpu_set(0, cpu_sibling_map[0]);
1000 cpu_set(0, cpu_core_map[0]);
1005 * Should not be necessary because the MP table should list the boot
1006 * CPU too, but we do it for the sake of robustness anyway.
1007 * Makes no sense to do this check in clustered apic mode, so skip it
1009 if (!check_phys_apicid_present(boot_cpu_physical_apicid)) {
1010 printk("weird, boot CPU (#%d) not listed by the BIOS.\n",
1011 boot_cpu_physical_apicid);
1012 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1016 * If we couldn't find a local APIC, then get out of here now!
1018 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) && !cpu_has_apic) {
1019 printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
1020 boot_cpu_physical_apicid);
1021 printk(KERN_ERR "... forcing use of dummy APIC emulation. (tell your hw vendor)\n");
1022 smpboot_clear_io_apic_irqs();
1023 phys_cpu_present_map = physid_mask_of_physid(0);
1024 cpu_set(0, cpu_sibling_map[0]);
1025 cpu_set(0, cpu_core_map[0]);
1029 verify_local_APIC();
1032 * If SMP should be disabled, then really disable it!
1035 smp_found_config = 0;
1036 printk(KERN_INFO "SMP mode deactivated, forcing use of dummy APIC emulation.\n");
1037 smpboot_clear_io_apic_irqs();
1038 phys_cpu_present_map = physid_mask_of_physid(0);
1039 cpu_set(0, cpu_sibling_map[0]);
1040 cpu_set(0, cpu_core_map[0]);
1046 map_cpu_to_logical_apicid();
1049 setup_portio_remap();
1052 * Scan the CPU present map and fire up the other CPUs via do_boot_cpu
1054 * In clustered apic mode, phys_cpu_present_map is a constructed thus:
1055 * bits 0-3 are quad0, 4-7 are quad1, etc. A perverse twist on the
1056 * clustered apic ID.
1058 Dprintk("CPU present map: %lx\n", physids_coerce(phys_cpu_present_map));
1061 for (bit = 0; kicked < NR_CPUS && bit < MAX_APICS; bit++) {
1062 apicid = cpu_present_to_apicid(bit);
1064 * Don't even attempt to start the boot CPU!
1066 if ((apicid == boot_cpu_apicid) || (apicid == BAD_APICID))
1069 if (!check_apicid_present(bit))
1071 if (max_cpus <= cpucount+1)
1074 if (((cpu = alloc_cpu_id()) <= 0) || do_boot_cpu(apicid, cpu))
1075 printk("CPU #%d not responding - cannot use it.\n",
1082 * Cleanup possible dangling ends...
1084 smpboot_restore_warm_reset_vector();
1087 * Allow the user to impress friends.
1089 Dprintk("Before bogomips.\n");
1090 for (cpu = 0; cpu < NR_CPUS; cpu++)
1091 if (cpu_isset(cpu, cpu_callout_map))
1092 bogosum += cpu_data[cpu].loops_per_jiffy;
1094 "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
1096 bogosum/(500000/HZ),
1097 (bogosum/(5000/HZ))%100);
1099 Dprintk("Before bogocount - setting activated=1.\n");
1102 printk(KERN_WARNING "WARNING: SMP operation may be unreliable with B stepping processors.\n");
1105 * Don't taint if we are running SMP kernel on a single non-MP
1108 if (tainted & TAINT_UNSAFE_SMP) {
1110 printk (KERN_INFO "WARNING: This combination of AMD processors is not suitable for SMP.\n");
1112 tainted &= ~TAINT_UNSAFE_SMP;
1115 Dprintk("Boot done.\n");
1118 * construct cpu_sibling_map[], so that we can tell sibling CPUs
1121 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1122 cpus_clear(cpu_sibling_map[cpu]);
1123 cpus_clear(cpu_core_map[cpu]);
1126 cpu_set(0, cpu_sibling_map[0]);
1127 cpu_set(0, cpu_core_map[0]);
1129 smpboot_setup_io_apic();
1134 /* These are wrappers to interface to the new boot process. Someone
1135 who understands all this stuff should rewrite it properly. --RR 15/Jul/02 */
1136 void __init native_smp_prepare_cpus(unsigned int max_cpus)
1138 smp_commenced_mask = cpumask_of_cpu(0);
1139 cpu_callin_map = cpumask_of_cpu(0);
1141 smp_boot_cpus(max_cpus);
1144 void __init native_smp_prepare_boot_cpu(void)
1146 unsigned int cpu = smp_processor_id();
1149 switch_to_new_gdt();
1151 cpu_set(cpu, cpu_online_map);
1152 cpu_set(cpu, cpu_callout_map);
1153 cpu_set(cpu, cpu_present_map);
1154 cpu_set(cpu, cpu_possible_map);
1155 __get_cpu_var(cpu_state) = CPU_ONLINE;
1158 #ifdef CONFIG_HOTPLUG_CPU
1160 remove_siblinginfo(int cpu)
1163 struct cpuinfo_x86 *c = cpu_data;
1165 for_each_cpu_mask(sibling, cpu_core_map[cpu]) {
1166 cpu_clear(cpu, cpu_core_map[sibling]);
1168 * last thread sibling in this cpu core going down
1170 if (cpus_weight(cpu_sibling_map[cpu]) == 1)
1171 c[sibling].booted_cores--;
1174 for_each_cpu_mask(sibling, cpu_sibling_map[cpu])
1175 cpu_clear(cpu, cpu_sibling_map[sibling]);
1176 cpus_clear(cpu_sibling_map[cpu]);
1177 cpus_clear(cpu_core_map[cpu]);
1178 c[cpu].phys_proc_id = 0;
1179 c[cpu].cpu_core_id = 0;
1180 cpu_clear(cpu, cpu_sibling_setup_map);
1183 int __cpu_disable(void)
1185 cpumask_t map = cpu_online_map;
1186 int cpu = smp_processor_id();
1189 * Perhaps use cpufreq to drop frequency, but that could go
1190 * into generic code.
1192 * We won't take down the boot processor on i386 due to some
1193 * interrupts only being able to be serviced by the BSP.
1194 * Especially so if we're not using an IOAPIC -zwane
1198 if (nmi_watchdog == NMI_LOCAL_APIC)
1199 stop_apic_nmi_watchdog(NULL);
1201 /* Allow any queued timer interrupts to get serviced */
1204 local_irq_disable();
1206 remove_siblinginfo(cpu);
1208 cpu_clear(cpu, map);
1210 /* It's now safe to remove this processor from the online map */
1211 cpu_clear(cpu, cpu_online_map);
1215 void __cpu_die(unsigned int cpu)
1217 /* We don't do anything here: idle task is faking death itself. */
1220 for (i = 0; i < 10; i++) {
1221 /* They ack this in play_dead by setting CPU_DEAD */
1222 if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
1223 printk ("CPU %d is now offline\n", cpu);
1224 if (1 == num_online_cpus())
1225 alternatives_smp_switch(0);
1230 printk(KERN_ERR "CPU %u didn't die...\n", cpu);
1232 #else /* ... !CONFIG_HOTPLUG_CPU */
1233 int __cpu_disable(void)
1238 void __cpu_die(unsigned int cpu)
1240 /* We said "no" in __cpu_disable */
1243 #endif /* CONFIG_HOTPLUG_CPU */
1245 int __cpuinit native_cpu_up(unsigned int cpu)
1247 unsigned long flags;
1248 #ifdef CONFIG_HOTPLUG_CPU
1252 * We do warm boot only on cpus that had booted earlier
1253 * Otherwise cold boot is all handled from smp_boot_cpus().
1254 * cpu_callin_map is set during AP kickstart process. Its reset
1255 * when a cpu is taken offline from cpu_exit_clear().
1257 if (!cpu_isset(cpu, cpu_callin_map))
1258 ret = __smp_prepare_cpu(cpu);
1264 /* In case one didn't come up */
1265 if (!cpu_isset(cpu, cpu_callin_map)) {
1266 printk(KERN_DEBUG "skipping cpu%d, didn't come online\n", cpu);
1270 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
1271 /* Unleash the CPU! */
1272 cpu_set(cpu, smp_commenced_mask);
1275 * Check TSC synchronization with the AP (keep irqs disabled
1278 local_irq_save(flags);
1279 check_tsc_sync_source(cpu);
1280 local_irq_restore(flags);
1282 while (!cpu_isset(cpu, cpu_online_map)) {
1284 touch_nmi_watchdog();
1290 void __init native_smp_cpus_done(unsigned int max_cpus)
1292 #ifdef CONFIG_X86_IO_APIC
1293 setup_ioapic_dest();
1296 #ifndef CONFIG_HOTPLUG_CPU
1298 * Disable executability of the SMP trampoline:
1300 set_kernel_exec((unsigned long)trampoline_base, trampoline_exec);
1304 void __init smp_intr_init(void)
1307 * IRQ0 must be given a fixed assignment and initialized,
1308 * because it's used before the IO-APIC is set up.
1310 set_intr_gate(FIRST_DEVICE_VECTOR, interrupt[0]);
1313 * The reschedule interrupt is a CPU-to-CPU reschedule-helper
1314 * IPI, driven by wakeup.
1316 set_intr_gate(RESCHEDULE_VECTOR, reschedule_interrupt);
1318 /* IPI for invalidation */
1319 set_intr_gate(INVALIDATE_TLB_VECTOR, invalidate_interrupt);
1321 /* IPI for generic function call */
1322 set_intr_gate(CALL_FUNCTION_VECTOR, call_function_interrupt);
1326 * If the BIOS enumerates physical processors before logical,
1327 * maxcpus=N at enumeration-time can be used to disable HT.
1329 static int __init parse_maxcpus(char *arg)
1331 extern unsigned int maxcpus;
1333 maxcpus = simple_strtoul(arg, NULL, 0);
1336 early_param("maxcpus", parse_maxcpus);