2 * Local APIC handling, local APIC timers
4 * (c) 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
7 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
8 * thanks to Eric Gilmore
10 * for testing these extensively.
11 * Maciej W. Rozycki : Various updates and fixes.
12 * Mikael Pettersson : Power Management for UP-APIC.
14 * Mikael Pettersson : PM converted to driver model.
17 #include <linux/perf_event.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/mc146818rtc.h>
20 #include <linux/acpi_pmtmr.h>
21 #include <linux/clockchips.h>
22 #include <linux/interrupt.h>
23 #include <linux/bootmem.h>
24 #include <linux/ftrace.h>
25 #include <linux/ioport.h>
26 #include <linux/module.h>
27 #include <linux/syscore_ops.h>
28 #include <linux/delay.h>
29 #include <linux/timex.h>
30 #include <linux/i8253.h>
31 #include <linux/dmar.h>
32 #include <linux/init.h>
33 #include <linux/cpu.h>
34 #include <linux/dmi.h>
35 #include <linux/smp.h>
38 #include <asm/irq_remapping.h>
39 #include <asm/perf_event.h>
40 #include <asm/x86_init.h>
41 #include <asm/pgalloc.h>
42 #include <linux/atomic.h>
43 #include <asm/mpspec.h>
44 #include <asm/i8259.h>
45 #include <asm/proto.h>
47 #include <asm/io_apic.h>
56 #include <asm/hypervisor.h>
58 #define CREATE_TRACE_POINTS
59 #include <asm/trace/irq_vectors.h>
61 unsigned int num_processors;
63 unsigned disabled_cpus __cpuinitdata;
65 /* Processor that is doing the boot up */
66 unsigned int boot_cpu_physical_apicid = -1U;
69 * The highest APIC ID seen during enumeration.
71 unsigned int max_physical_apicid;
74 * Bitmask of physically existing CPUs:
76 physid_mask_t phys_cpu_present_map;
79 * Map cpu index to physical APIC ID
81 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
82 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
83 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
84 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
89 * On x86_32, the mapping between cpu and logical apicid may vary
90 * depending on apic in use. The following early percpu variable is
91 * used for the mapping. This is where the behaviors of x86_64 and 32
92 * actually diverge. Let's keep it ugly for now.
94 DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
96 /* Local APIC was disabled by the BIOS and enabled by the kernel */
97 static int enabled_via_apicbase;
100 * Handle interrupt mode configuration register (IMCR).
101 * This register controls whether the interrupt signals
102 * that reach the BSP come from the master PIC or from the
103 * local APIC. Before entering Symmetric I/O Mode, either
104 * the BIOS or the operating system must switch out of
105 * PIC Mode by changing the IMCR.
107 static inline void imcr_pic_to_apic(void)
109 /* select IMCR register */
111 /* NMI and 8259 INTR go through APIC */
115 static inline void imcr_apic_to_pic(void)
117 /* select IMCR register */
119 /* NMI and 8259 INTR go directly to BSP */
125 * Knob to control our willingness to enable the local APIC.
129 static int force_enable_local_apic __initdata;
131 * APIC command line parameters
133 static int __init parse_lapic(char *arg)
135 if (config_enabled(CONFIG_X86_32) && !arg)
136 force_enable_local_apic = 1;
137 else if (arg && !strncmp(arg, "notscdeadline", 13))
138 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
141 early_param("lapic", parse_lapic);
144 static int apic_calibrate_pmtmr __initdata;
145 static __init int setup_apicpmtimer(char *s)
147 apic_calibrate_pmtmr = 1;
151 __setup("apicpmtimer", setup_apicpmtimer);
155 #ifdef CONFIG_X86_X2APIC
156 /* x2apic enabled before OS handover */
157 int x2apic_preenabled;
158 static int x2apic_disabled;
160 static __init int setup_nox2apic(char *str)
162 if (x2apic_enabled()) {
163 int apicid = native_apic_msr_read(APIC_ID);
166 pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
171 pr_warning("x2apic already enabled. will disable it\n");
173 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
179 early_param("nox2apic", setup_nox2apic);
182 unsigned long mp_lapic_addr;
184 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
185 static int disable_apic_timer __initdata;
186 /* Local APIC timer works in C2 */
187 int local_apic_timer_c2_ok;
188 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
190 int first_system_vector = 0xfe;
193 * Debug level, exported for io_apic.c
195 unsigned int apic_verbosity;
199 /* Have we found an MP table */
200 int smp_found_config;
202 static struct resource lapic_resource = {
203 .name = "Local APIC",
204 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
207 unsigned int lapic_timer_frequency = 0;
209 static void apic_pm_activate(void);
211 static unsigned long apic_phys;
214 * Get the LAPIC version
216 static inline int lapic_get_version(void)
218 return GET_APIC_VERSION(apic_read(APIC_LVR));
222 * Check, if the APIC is integrated or a separate chip
224 static inline int lapic_is_integrated(void)
229 return APIC_INTEGRATED(lapic_get_version());
234 * Check, whether this is a modern or a first generation APIC
236 static int modern_apic(void)
238 /* AMD systems use old APIC versions, so check the CPU */
239 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
240 boot_cpu_data.x86 >= 0xf)
242 return lapic_get_version() >= 0x14;
246 * right after this call apic become NOOP driven
247 * so apic->write/read doesn't do anything
249 static void __init apic_disable(void)
251 pr_info("APIC: switched to apic NOOP\n");
255 void native_apic_wait_icr_idle(void)
257 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
261 u32 native_safe_apic_wait_icr_idle(void)
268 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
271 inc_irq_stat(icr_read_retry_count);
273 } while (timeout++ < 1000);
278 void native_apic_icr_write(u32 low, u32 id)
280 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
281 apic_write(APIC_ICR, low);
284 u64 native_apic_icr_read(void)
288 icr2 = apic_read(APIC_ICR2);
289 icr1 = apic_read(APIC_ICR);
291 return icr1 | ((u64)icr2 << 32);
296 * get_physical_broadcast - Get number of physical broadcast IDs
298 int get_physical_broadcast(void)
300 return modern_apic() ? 0xff : 0xf;
305 * lapic_get_maxlvt - get the maximum number of local vector table entries
307 int lapic_get_maxlvt(void)
311 v = apic_read(APIC_LVR);
313 * - we always have APIC integrated on 64bit mode
314 * - 82489DXs do not report # of LVT entries
316 return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
324 #define APIC_DIVISOR 16
325 #define TSC_DIVISOR 32
328 * This function sets up the local APIC timer, with a timeout of
329 * 'clocks' APIC bus clock. During calibration we actually call
330 * this function twice on the boot CPU, once with a bogus timeout
331 * value, second time for real. The other (noncalibrating) CPUs
332 * call this function only once, with the real, calibrated value.
334 * We do reads before writes even if unnecessary, to get around the
335 * P5 APIC double write bug.
337 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
339 unsigned int lvtt_value, tmp_value;
341 lvtt_value = LOCAL_TIMER_VECTOR;
343 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
344 else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
345 lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
347 if (!lapic_is_integrated())
348 lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
351 lvtt_value |= APIC_LVT_MASKED;
353 apic_write(APIC_LVTT, lvtt_value);
355 if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
356 printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
363 tmp_value = apic_read(APIC_TDCR);
364 apic_write(APIC_TDCR,
365 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
369 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
373 * Setup extended LVT, AMD specific
375 * Software should use the LVT offsets the BIOS provides. The offsets
376 * are determined by the subsystems using it like those for MCE
377 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts
378 * are supported. Beginning with family 10h at least 4 offsets are
381 * Since the offsets must be consistent for all cores, we keep track
382 * of the LVT offsets in software and reserve the offset for the same
383 * vector also to be used on other cores. An offset is freed by
384 * setting the entry to APIC_EILVT_MASKED.
386 * If the BIOS is right, there should be no conflicts. Otherwise a
387 * "[Firmware Bug]: ..." error message is generated. However, if
388 * software does not properly determines the offsets, it is not
389 * necessarily a BIOS bug.
392 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
394 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
396 return (old & APIC_EILVT_MASKED)
397 || (new == APIC_EILVT_MASKED)
398 || ((new & ~APIC_EILVT_MASKED) == old);
401 static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
403 unsigned int rsvd, vector;
405 if (offset >= APIC_EILVT_NR_MAX)
408 rsvd = atomic_read(&eilvt_offsets[offset]);
410 vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */
411 if (vector && !eilvt_entry_is_changeable(vector, new))
412 /* may not change if vectors are different */
414 rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
415 } while (rsvd != new);
417 rsvd &= ~APIC_EILVT_MASKED;
418 if (rsvd && rsvd != vector)
419 pr_info("LVT offset %d assigned for vector 0x%02x\n",
426 * If mask=1, the LVT entry does not generate interrupts while mask=0
427 * enables the vector. See also the BKDGs. Must be called with
428 * preemption disabled.
431 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
433 unsigned long reg = APIC_EILVTn(offset);
434 unsigned int new, old, reserved;
436 new = (mask << 16) | (msg_type << 8) | vector;
437 old = apic_read(reg);
438 reserved = reserve_eilvt_offset(offset, new);
440 if (reserved != new) {
441 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
442 "vector 0x%x, but the register is already in use for "
443 "vector 0x%x on another cpu\n",
444 smp_processor_id(), reg, offset, new, reserved);
448 if (!eilvt_entry_is_changeable(old, new)) {
449 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
450 "vector 0x%x, but the register is already in use for "
451 "vector 0x%x on this cpu\n",
452 smp_processor_id(), reg, offset, new, old);
456 apic_write(reg, new);
460 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
463 * Program the next event, relative to now
465 static int lapic_next_event(unsigned long delta,
466 struct clock_event_device *evt)
468 apic_write(APIC_TMICT, delta);
472 static int lapic_next_deadline(unsigned long delta,
473 struct clock_event_device *evt)
478 wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
483 * Setup the lapic timer in periodic or oneshot mode
485 static void lapic_timer_setup(enum clock_event_mode mode,
486 struct clock_event_device *evt)
491 /* Lapic used as dummy for broadcast ? */
492 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
495 local_irq_save(flags);
498 case CLOCK_EVT_MODE_PERIODIC:
499 case CLOCK_EVT_MODE_ONESHOT:
500 __setup_APIC_LVTT(lapic_timer_frequency,
501 mode != CLOCK_EVT_MODE_PERIODIC, 1);
503 case CLOCK_EVT_MODE_UNUSED:
504 case CLOCK_EVT_MODE_SHUTDOWN:
505 v = apic_read(APIC_LVTT);
506 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
507 apic_write(APIC_LVTT, v);
508 apic_write(APIC_TMICT, 0);
510 case CLOCK_EVT_MODE_RESUME:
511 /* Nothing to do here */
515 local_irq_restore(flags);
519 * Local APIC timer broadcast function
521 static void lapic_timer_broadcast(const struct cpumask *mask)
524 apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
530 * The local apic timer can be used for any function which is CPU local.
532 static struct clock_event_device lapic_clockevent = {
534 .features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
535 | CLOCK_EVT_FEAT_C3STOP | CLOCK_EVT_FEAT_DUMMY,
537 .set_mode = lapic_timer_setup,
538 .set_next_event = lapic_next_event,
539 .broadcast = lapic_timer_broadcast,
543 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
546 * Setup the local APIC timer for this CPU. Copy the initialized values
547 * of the boot CPU and register the clock event in the framework.
549 static void __cpuinit setup_APIC_timer(void)
551 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
553 if (this_cpu_has(X86_FEATURE_ARAT)) {
554 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
555 /* Make LAPIC timer preferrable over percpu HPET */
556 lapic_clockevent.rating = 150;
559 memcpy(levt, &lapic_clockevent, sizeof(*levt));
560 levt->cpumask = cpumask_of(smp_processor_id());
562 if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
563 levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
564 CLOCK_EVT_FEAT_DUMMY);
565 levt->set_next_event = lapic_next_deadline;
566 clockevents_config_and_register(levt,
567 (tsc_khz / TSC_DIVISOR) * 1000,
570 clockevents_register_device(levt);
574 * In this functions we calibrate APIC bus clocks to the external timer.
576 * We want to do the calibration only once since we want to have local timer
577 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
580 * This was previously done by reading the PIT/HPET and waiting for a wrap
581 * around to find out, that a tick has elapsed. I have a box, where the PIT
582 * readout is broken, so it never gets out of the wait loop again. This was
583 * also reported by others.
585 * Monitoring the jiffies value is inaccurate and the clockevents
586 * infrastructure allows us to do a simple substitution of the interrupt
589 * The calibration routine also uses the pm_timer when possible, as the PIT
590 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
591 * back to normal later in the boot process).
594 #define LAPIC_CAL_LOOPS (HZ/10)
596 static __initdata int lapic_cal_loops = -1;
597 static __initdata long lapic_cal_t1, lapic_cal_t2;
598 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
599 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
600 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
603 * Temporary interrupt handler.
605 static void __init lapic_cal_handler(struct clock_event_device *dev)
607 unsigned long long tsc = 0;
608 long tapic = apic_read(APIC_TMCCT);
609 unsigned long pm = acpi_pm_read_early();
614 switch (lapic_cal_loops++) {
616 lapic_cal_t1 = tapic;
617 lapic_cal_tsc1 = tsc;
619 lapic_cal_j1 = jiffies;
622 case LAPIC_CAL_LOOPS:
623 lapic_cal_t2 = tapic;
624 lapic_cal_tsc2 = tsc;
625 if (pm < lapic_cal_pm1)
626 pm += ACPI_PM_OVRRUN;
628 lapic_cal_j2 = jiffies;
634 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
636 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
637 const long pm_thresh = pm_100ms / 100;
641 #ifndef CONFIG_X86_PM_TIMER
645 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
647 /* Check, if the PM timer is available */
651 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
653 if (deltapm > (pm_100ms - pm_thresh) &&
654 deltapm < (pm_100ms + pm_thresh)) {
655 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
659 res = (((u64)deltapm) * mult) >> 22;
660 do_div(res, 1000000);
661 pr_warning("APIC calibration not consistent "
662 "with PM-Timer: %ldms instead of 100ms\n",(long)res);
664 /* Correct the lapic counter value */
665 res = (((u64)(*delta)) * pm_100ms);
666 do_div(res, deltapm);
667 pr_info("APIC delta adjusted to PM-Timer: "
668 "%lu (%ld)\n", (unsigned long)res, *delta);
671 /* Correct the tsc counter value */
673 res = (((u64)(*deltatsc)) * pm_100ms);
674 do_div(res, deltapm);
675 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
676 "PM-Timer: %lu (%ld)\n",
677 (unsigned long)res, *deltatsc);
678 *deltatsc = (long)res;
684 static int __init calibrate_APIC_clock(void)
686 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
687 void (*real_handler)(struct clock_event_device *dev);
688 unsigned long deltaj;
689 long delta, deltatsc;
690 int pm_referenced = 0;
693 * check if lapic timer has already been calibrated by platform
694 * specific routine, such as tsc calibration code. if so, we just fill
695 * in the clockevent structure and return.
698 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
700 } else if (lapic_timer_frequency) {
701 apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
702 lapic_timer_frequency);
703 lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
704 TICK_NSEC, lapic_clockevent.shift);
705 lapic_clockevent.max_delta_ns =
706 clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
707 lapic_clockevent.min_delta_ns =
708 clockevent_delta2ns(0xF, &lapic_clockevent);
709 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
713 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
714 "calibrating APIC timer ...\n");
718 /* Replace the global interrupt handler */
719 real_handler = global_clock_event->event_handler;
720 global_clock_event->event_handler = lapic_cal_handler;
723 * Setup the APIC counter to maximum. There is no way the lapic
724 * can underflow in the 100ms detection time frame
726 __setup_APIC_LVTT(0xffffffff, 0, 0);
728 /* Let the interrupts run */
731 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
736 /* Restore the real event handler */
737 global_clock_event->event_handler = real_handler;
739 /* Build delta t1-t2 as apic timer counts down */
740 delta = lapic_cal_t1 - lapic_cal_t2;
741 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
743 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
745 /* we trust the PM based calibration if possible */
746 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
749 /* Calculate the scaled math multiplication factor */
750 lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
751 lapic_clockevent.shift);
752 lapic_clockevent.max_delta_ns =
753 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
754 lapic_clockevent.min_delta_ns =
755 clockevent_delta2ns(0xF, &lapic_clockevent);
757 lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
759 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
760 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
761 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
762 lapic_timer_frequency);
765 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
767 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
768 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
771 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
773 lapic_timer_frequency / (1000000 / HZ),
774 lapic_timer_frequency % (1000000 / HZ));
777 * Do a sanity check on the APIC calibration result
779 if (lapic_timer_frequency < (1000000 / HZ)) {
781 pr_warning("APIC frequency too slow, disabling apic timer\n");
785 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
788 * PM timer calibration failed or not turned on
789 * so lets try APIC timer based calibration
791 if (!pm_referenced) {
792 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
795 * Setup the apic timer manually
797 levt->event_handler = lapic_cal_handler;
798 lapic_timer_setup(CLOCK_EVT_MODE_PERIODIC, levt);
799 lapic_cal_loops = -1;
801 /* Let the interrupts run */
804 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
807 /* Stop the lapic timer */
808 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt);
811 deltaj = lapic_cal_j2 - lapic_cal_j1;
812 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
814 /* Check, if the jiffies result is consistent */
815 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
816 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
818 levt->features |= CLOCK_EVT_FEAT_DUMMY;
822 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
823 pr_warning("APIC timer disabled due to verification failure\n");
831 * Setup the boot APIC
833 * Calibrate and verify the result.
835 void __init setup_boot_APIC_clock(void)
838 * The local apic timer can be disabled via the kernel
839 * commandline or from the CPU detection code. Register the lapic
840 * timer as a dummy clock event source on SMP systems, so the
841 * broadcast mechanism is used. On UP systems simply ignore it.
843 if (disable_apic_timer) {
844 pr_info("Disabling APIC timer\n");
845 /* No broadcast on UP ! */
846 if (num_possible_cpus() > 1) {
847 lapic_clockevent.mult = 1;
853 if (calibrate_APIC_clock()) {
854 /* No broadcast on UP ! */
855 if (num_possible_cpus() > 1)
861 * If nmi_watchdog is set to IO_APIC, we need the
862 * PIT/HPET going. Otherwise register lapic as a dummy
865 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
867 /* Setup the lapic or request the broadcast */
871 void __cpuinit setup_secondary_APIC_clock(void)
877 * The guts of the apic timer interrupt
879 static void local_apic_timer_interrupt(void)
881 int cpu = smp_processor_id();
882 struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
885 * Normally we should not be here till LAPIC has been initialized but
886 * in some cases like kdump, its possible that there is a pending LAPIC
887 * timer interrupt from previous kernel's context and is delivered in
888 * new kernel the moment interrupts are enabled.
890 * Interrupts are enabled early and LAPIC is setup much later, hence
891 * its possible that when we get here evt->event_handler is NULL.
892 * Check for event_handler being NULL and discard the interrupt as
895 if (!evt->event_handler) {
896 pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
898 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
903 * the NMI deadlock-detector uses this.
905 inc_irq_stat(apic_timer_irqs);
907 evt->event_handler(evt);
911 * Local APIC timer interrupt. This is the most natural way for doing
912 * local interrupts, but local timer interrupts can be emulated by
913 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
915 * [ if a single-CPU system runs an SMP kernel then we call the local
916 * interrupt as well. Thus we cannot inline the local irq ... ]
918 void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
920 struct pt_regs *old_regs = set_irq_regs(regs);
923 * NOTE! We'd better ACK the irq immediately,
924 * because timer handling can be slow.
926 * update_process_times() expects us to have done irq_enter().
927 * Besides, if we don't timer interrupts ignore the global
928 * interrupt lock, which is the WrongThing (tm) to do.
931 local_apic_timer_interrupt();
934 set_irq_regs(old_regs);
937 void __irq_entry smp_trace_apic_timer_interrupt(struct pt_regs *regs)
939 struct pt_regs *old_regs = set_irq_regs(regs);
942 * NOTE! We'd better ACK the irq immediately,
943 * because timer handling can be slow.
945 * update_process_times() expects us to have done irq_enter().
946 * Besides, if we don't timer interrupts ignore the global
947 * interrupt lock, which is the WrongThing (tm) to do.
950 trace_local_timer_entry(LOCAL_TIMER_VECTOR);
951 local_apic_timer_interrupt();
952 trace_local_timer_exit(LOCAL_TIMER_VECTOR);
955 set_irq_regs(old_regs);
958 int setup_profiling_timer(unsigned int multiplier)
964 * Local APIC start and shutdown
968 * clear_local_APIC - shutdown the local APIC
970 * This is called, when a CPU is disabled and before rebooting, so the state of
971 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
972 * leftovers during boot.
974 void clear_local_APIC(void)
979 /* APIC hasn't been mapped yet */
980 if (!x2apic_mode && !apic_phys)
983 maxlvt = lapic_get_maxlvt();
985 * Masking an LVT entry can trigger a local APIC error
986 * if the vector is zero. Mask LVTERR first to prevent this.
989 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
990 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
993 * Careful: we have to set masks only first to deassert
994 * any level-triggered sources.
996 v = apic_read(APIC_LVTT);
997 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
998 v = apic_read(APIC_LVT0);
999 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1000 v = apic_read(APIC_LVT1);
1001 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1003 v = apic_read(APIC_LVTPC);
1004 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1007 /* lets not touch this if we didn't frob it */
1008 #ifdef CONFIG_X86_THERMAL_VECTOR
1010 v = apic_read(APIC_LVTTHMR);
1011 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1014 #ifdef CONFIG_X86_MCE_INTEL
1016 v = apic_read(APIC_LVTCMCI);
1017 if (!(v & APIC_LVT_MASKED))
1018 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1023 * Clean APIC state for other OSs:
1025 apic_write(APIC_LVTT, APIC_LVT_MASKED);
1026 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1027 apic_write(APIC_LVT1, APIC_LVT_MASKED);
1029 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1031 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1033 /* Integrated APIC (!82489DX) ? */
1034 if (lapic_is_integrated()) {
1036 /* Clear ESR due to Pentium errata 3AP and 11AP */
1037 apic_write(APIC_ESR, 0);
1038 apic_read(APIC_ESR);
1043 * disable_local_APIC - clear and disable the local APIC
1045 void disable_local_APIC(void)
1049 /* APIC hasn't been mapped yet */
1050 if (!x2apic_mode && !apic_phys)
1056 * Disable APIC (implies clearing of registers
1059 value = apic_read(APIC_SPIV);
1060 value &= ~APIC_SPIV_APIC_ENABLED;
1061 apic_write(APIC_SPIV, value);
1063 #ifdef CONFIG_X86_32
1065 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1066 * restore the disabled state.
1068 if (enabled_via_apicbase) {
1071 rdmsr(MSR_IA32_APICBASE, l, h);
1072 l &= ~MSR_IA32_APICBASE_ENABLE;
1073 wrmsr(MSR_IA32_APICBASE, l, h);
1079 * If Linux enabled the LAPIC against the BIOS default disable it down before
1080 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1081 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1082 * for the case where Linux didn't enable the LAPIC.
1084 void lapic_shutdown(void)
1086 unsigned long flags;
1088 if (!cpu_has_apic && !apic_from_smp_config())
1091 local_irq_save(flags);
1093 #ifdef CONFIG_X86_32
1094 if (!enabled_via_apicbase)
1098 disable_local_APIC();
1101 local_irq_restore(flags);
1105 * This is to verify that we're looking at a real local APIC.
1106 * Check these against your board if the CPUs aren't getting
1107 * started for no apparent reason.
1109 int __init verify_local_APIC(void)
1111 unsigned int reg0, reg1;
1114 * The version register is read-only in a real APIC.
1116 reg0 = apic_read(APIC_LVR);
1117 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
1118 apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
1119 reg1 = apic_read(APIC_LVR);
1120 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
1123 * The two version reads above should print the same
1124 * numbers. If the second one is different, then we
1125 * poke at a non-APIC.
1131 * Check if the version looks reasonably.
1133 reg1 = GET_APIC_VERSION(reg0);
1134 if (reg1 == 0x00 || reg1 == 0xff)
1136 reg1 = lapic_get_maxlvt();
1137 if (reg1 < 0x02 || reg1 == 0xff)
1141 * The ID register is read/write in a real APIC.
1143 reg0 = apic_read(APIC_ID);
1144 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
1145 apic_write(APIC_ID, reg0 ^ apic->apic_id_mask);
1146 reg1 = apic_read(APIC_ID);
1147 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
1148 apic_write(APIC_ID, reg0);
1149 if (reg1 != (reg0 ^ apic->apic_id_mask))
1153 * The next two are just to see if we have sane values.
1154 * They're only really relevant if we're in Virtual Wire
1155 * compatibility mode, but most boxes are anymore.
1157 reg0 = apic_read(APIC_LVT0);
1158 apic_printk(APIC_DEBUG, "Getting LVT0: %x\n", reg0);
1159 reg1 = apic_read(APIC_LVT1);
1160 apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
1166 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1168 void __init sync_Arb_IDs(void)
1171 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1174 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1180 apic_wait_icr_idle();
1182 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1183 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1184 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1188 * An initial setup of the virtual wire mode.
1190 void __init init_bsp_APIC(void)
1195 * Don't do the setup now if we have a SMP BIOS as the
1196 * through-I/O-APIC virtual wire mode might be active.
1198 if (smp_found_config || !cpu_has_apic)
1202 * Do not trust the local APIC being empty at bootup.
1209 value = apic_read(APIC_SPIV);
1210 value &= ~APIC_VECTOR_MASK;
1211 value |= APIC_SPIV_APIC_ENABLED;
1213 #ifdef CONFIG_X86_32
1214 /* This bit is reserved on P4/Xeon and should be cleared */
1215 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1216 (boot_cpu_data.x86 == 15))
1217 value &= ~APIC_SPIV_FOCUS_DISABLED;
1220 value |= APIC_SPIV_FOCUS_DISABLED;
1221 value |= SPURIOUS_APIC_VECTOR;
1222 apic_write(APIC_SPIV, value);
1225 * Set up the virtual wire mode.
1227 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1228 value = APIC_DM_NMI;
1229 if (!lapic_is_integrated()) /* 82489DX */
1230 value |= APIC_LVT_LEVEL_TRIGGER;
1231 apic_write(APIC_LVT1, value);
1234 static void __cpuinit lapic_setup_esr(void)
1236 unsigned int oldvalue, value, maxlvt;
1238 if (!lapic_is_integrated()) {
1239 pr_info("No ESR for 82489DX.\n");
1243 if (apic->disable_esr) {
1245 * Something untraceable is creating bad interrupts on
1246 * secondary quads ... for the moment, just leave the
1247 * ESR disabled - we can't do anything useful with the
1248 * errors anyway - mbligh
1250 pr_info("Leaving ESR disabled.\n");
1254 maxlvt = lapic_get_maxlvt();
1255 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1256 apic_write(APIC_ESR, 0);
1257 oldvalue = apic_read(APIC_ESR);
1259 /* enables sending errors */
1260 value = ERROR_APIC_VECTOR;
1261 apic_write(APIC_LVTERR, value);
1264 * spec says clear errors after enabling vector.
1267 apic_write(APIC_ESR, 0);
1268 value = apic_read(APIC_ESR);
1269 if (value != oldvalue)
1270 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1271 "vector: 0x%08x after: 0x%08x\n",
1276 * setup_local_APIC - setup the local APIC
1278 * Used to setup local APIC while initializing BSP or bringin up APs.
1279 * Always called with preemption disabled.
1281 void __cpuinit setup_local_APIC(void)
1283 int cpu = smp_processor_id();
1284 unsigned int value, queued;
1285 int i, j, acked = 0;
1286 unsigned long long tsc = 0, ntsc;
1287 long long max_loops = cpu_khz;
1293 disable_ioapic_support();
1297 #ifdef CONFIG_X86_32
1298 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1299 if (lapic_is_integrated() && apic->disable_esr) {
1300 apic_write(APIC_ESR, 0);
1301 apic_write(APIC_ESR, 0);
1302 apic_write(APIC_ESR, 0);
1303 apic_write(APIC_ESR, 0);
1306 perf_events_lapic_init();
1309 * Double-check whether this APIC is really registered.
1310 * This is meaningless in clustered apic mode, so we skip it.
1312 BUG_ON(!apic->apic_id_registered());
1315 * Intel recommends to set DFR, LDR and TPR before enabling
1316 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1317 * document number 292116). So here it goes...
1319 apic->init_apic_ldr();
1321 #ifdef CONFIG_X86_32
1323 * APIC LDR is initialized. If logical_apicid mapping was
1324 * initialized during get_smp_config(), make sure it matches the
1327 i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1328 WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1329 /* always use the value from LDR */
1330 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1331 logical_smp_processor_id();
1334 * Some NUMA implementations (NUMAQ) don't initialize apicid to
1335 * node mapping during NUMA init. Now that logical apicid is
1336 * guaranteed to be known, give it another chance. This is already
1337 * a bit too late - percpu allocation has already happened without
1338 * proper NUMA affinity.
1340 if (apic->x86_32_numa_cpu_node)
1341 set_apicid_to_node(early_per_cpu(x86_cpu_to_apicid, cpu),
1342 apic->x86_32_numa_cpu_node(cpu));
1346 * Set Task Priority to 'accept all'. We never change this
1349 value = apic_read(APIC_TASKPRI);
1350 value &= ~APIC_TPRI_MASK;
1351 apic_write(APIC_TASKPRI, value);
1354 * After a crash, we no longer service the interrupts and a pending
1355 * interrupt from previous kernel might still have ISR bit set.
1357 * Most probably by now CPU has serviced that pending interrupt and
1358 * it might not have done the ack_APIC_irq() because it thought,
1359 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1360 * does not clear the ISR bit and cpu thinks it has already serivced
1361 * the interrupt. Hence a vector might get locked. It was noticed
1362 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1366 for (i = APIC_ISR_NR - 1; i >= 0; i--)
1367 queued |= apic_read(APIC_IRR + i*0x10);
1369 for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1370 value = apic_read(APIC_ISR + i*0x10);
1371 for (j = 31; j >= 0; j--) {
1372 if (value & (1<<j)) {
1379 printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1386 max_loops = (cpu_khz << 10) - (ntsc - tsc);
1390 } while (queued && max_loops > 0);
1391 WARN_ON(max_loops <= 0);
1394 * Now that we are all set up, enable the APIC
1396 value = apic_read(APIC_SPIV);
1397 value &= ~APIC_VECTOR_MASK;
1401 value |= APIC_SPIV_APIC_ENABLED;
1403 #ifdef CONFIG_X86_32
1405 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1406 * certain networking cards. If high frequency interrupts are
1407 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1408 * entry is masked/unmasked at a high rate as well then sooner or
1409 * later IOAPIC line gets 'stuck', no more interrupts are received
1410 * from the device. If focus CPU is disabled then the hang goes
1413 * [ This bug can be reproduced easily with a level-triggered
1414 * PCI Ne2000 networking cards and PII/PIII processors, dual
1418 * Actually disabling the focus CPU check just makes the hang less
1419 * frequent as it makes the interrupt distributon model be more
1420 * like LRU than MRU (the short-term load is more even across CPUs).
1421 * See also the comment in end_level_ioapic_irq(). --macro
1425 * - enable focus processor (bit==0)
1426 * - 64bit mode always use processor focus
1427 * so no need to set it
1429 value &= ~APIC_SPIV_FOCUS_DISABLED;
1433 * Set spurious IRQ vector
1435 value |= SPURIOUS_APIC_VECTOR;
1436 apic_write(APIC_SPIV, value);
1439 * Set up LVT0, LVT1:
1441 * set up through-local-APIC on the BP's LINT0. This is not
1442 * strictly necessary in pure symmetric-IO mode, but sometimes
1443 * we delegate interrupts to the 8259A.
1446 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1448 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1449 if (!cpu && (pic_mode || !value)) {
1450 value = APIC_DM_EXTINT;
1451 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1453 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1454 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1456 apic_write(APIC_LVT0, value);
1459 * only the BP should see the LINT1 NMI signal, obviously.
1462 value = APIC_DM_NMI;
1464 value = APIC_DM_NMI | APIC_LVT_MASKED;
1465 if (!lapic_is_integrated()) /* 82489DX */
1466 value |= APIC_LVT_LEVEL_TRIGGER;
1467 apic_write(APIC_LVT1, value);
1469 #ifdef CONFIG_X86_MCE_INTEL
1470 /* Recheck CMCI information after local APIC is up on CPU #0 */
1476 void __cpuinit end_local_APIC_setup(void)
1480 #ifdef CONFIG_X86_32
1483 /* Disable the local apic timer */
1484 value = apic_read(APIC_LVTT);
1485 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1486 apic_write(APIC_LVTT, value);
1493 void __init bsp_end_local_APIC_setup(void)
1495 end_local_APIC_setup();
1498 * Now that local APIC setup is completed for BP, configure the fault
1499 * handling for interrupt remapping.
1501 irq_remap_enable_fault_handling();
1505 #ifdef CONFIG_X86_X2APIC
1507 * Need to disable xapic and x2apic at the same time and then enable xapic mode
1509 static inline void __disable_x2apic(u64 msr)
1511 wrmsrl(MSR_IA32_APICBASE,
1512 msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1513 wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1516 static __init void disable_x2apic(void)
1520 if (!cpu_has_x2apic)
1523 rdmsrl(MSR_IA32_APICBASE, msr);
1524 if (msr & X2APIC_ENABLE) {
1525 u32 x2apic_id = read_apic_id();
1527 if (x2apic_id >= 255)
1528 panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1530 pr_info("Disabling x2apic\n");
1531 __disable_x2apic(msr);
1534 clear_cpu_cap(&cpu_data(0), X86_FEATURE_X2APIC);
1535 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1538 x2apic_disabled = 1;
1541 register_lapic_address(mp_lapic_addr);
1545 void check_x2apic(void)
1547 if (x2apic_enabled()) {
1548 pr_info("x2apic enabled by BIOS, switching to x2apic ops\n");
1549 x2apic_preenabled = x2apic_mode = 1;
1553 void enable_x2apic(void)
1557 rdmsrl(MSR_IA32_APICBASE, msr);
1558 if (x2apic_disabled) {
1559 __disable_x2apic(msr);
1566 if (!(msr & X2APIC_ENABLE)) {
1567 printk_once(KERN_INFO "Enabling x2apic\n");
1568 wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1571 #endif /* CONFIG_X86_X2APIC */
1573 int __init enable_IR(void)
1575 #ifdef CONFIG_IRQ_REMAP
1576 if (!irq_remapping_supported()) {
1577 pr_debug("intr-remapping not supported\n");
1581 if (!x2apic_preenabled && skip_ioapic_setup) {
1582 pr_info("Skipped enabling intr-remap because of skipping "
1587 return irq_remapping_enable();
1592 void __init enable_IR_x2apic(void)
1594 unsigned long flags;
1595 int ret, x2apic_enabled = 0;
1596 int hardware_init_ret;
1598 /* Make sure irq_remap_ops are initialized */
1599 setup_irq_remapping_ops();
1601 hardware_init_ret = irq_remapping_prepare();
1602 if (hardware_init_ret && !x2apic_supported())
1605 ret = save_ioapic_entries();
1607 pr_info("Saving IO-APIC state failed: %d\n", ret);
1611 local_irq_save(flags);
1612 legacy_pic->mask_all();
1613 mask_ioapic_entries();
1615 if (x2apic_preenabled && nox2apic)
1618 if (hardware_init_ret)
1623 if (!x2apic_supported())
1627 /* IR is required if there is APIC ID > 255 even when running
1630 if (max_physical_apicid > 255 ||
1631 !hypervisor_x2apic_available()) {
1632 if (x2apic_preenabled)
1637 * without IR all CPUs can be addressed by IOAPIC/MSI
1638 * only in physical mode
1640 x2apic_force_phys();
1643 if (ret == IRQ_REMAP_XAPIC_MODE) {
1644 pr_info("x2apic not enabled, IRQ remapping is in xapic mode\n");
1650 if (x2apic_supported() && !x2apic_mode) {
1653 pr_info("Enabled x2apic\n");
1657 if (ret < 0) /* IR enabling failed */
1658 restore_ioapic_entries();
1659 legacy_pic->restore_mask();
1660 local_irq_restore(flags);
1663 #ifdef CONFIG_X86_64
1665 * Detect and enable local APICs on non-SMP boards.
1666 * Original code written by Keir Fraser.
1667 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1668 * not correctly set up (usually the APIC timer won't work etc.)
1670 static int __init detect_init_APIC(void)
1672 if (!cpu_has_apic) {
1673 pr_info("No local APIC present\n");
1677 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1682 static int __init apic_verify(void)
1687 * The APIC feature bit should now be enabled
1690 features = cpuid_edx(1);
1691 if (!(features & (1 << X86_FEATURE_APIC))) {
1692 pr_warning("Could not enable APIC!\n");
1695 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1696 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1698 /* The BIOS may have set up the APIC at some other address */
1699 if (boot_cpu_data.x86 >= 6) {
1700 rdmsr(MSR_IA32_APICBASE, l, h);
1701 if (l & MSR_IA32_APICBASE_ENABLE)
1702 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1705 pr_info("Found and enabled local APIC!\n");
1709 int __init apic_force_enable(unsigned long addr)
1717 * Some BIOSes disable the local APIC in the APIC_BASE
1718 * MSR. This can only be done in software for Intel P6 or later
1719 * and AMD K7 (Model > 1) or later.
1721 if (boot_cpu_data.x86 >= 6) {
1722 rdmsr(MSR_IA32_APICBASE, l, h);
1723 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1724 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1725 l &= ~MSR_IA32_APICBASE_BASE;
1726 l |= MSR_IA32_APICBASE_ENABLE | addr;
1727 wrmsr(MSR_IA32_APICBASE, l, h);
1728 enabled_via_apicbase = 1;
1731 return apic_verify();
1735 * Detect and initialize APIC
1737 static int __init detect_init_APIC(void)
1739 /* Disabled by kernel option? */
1743 switch (boot_cpu_data.x86_vendor) {
1744 case X86_VENDOR_AMD:
1745 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1746 (boot_cpu_data.x86 >= 15))
1749 case X86_VENDOR_INTEL:
1750 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1751 (boot_cpu_data.x86 == 5 && cpu_has_apic))
1758 if (!cpu_has_apic) {
1760 * Over-ride BIOS and try to enable the local APIC only if
1761 * "lapic" specified.
1763 if (!force_enable_local_apic) {
1764 pr_info("Local APIC disabled by BIOS -- "
1765 "you can enable it with \"lapic\"\n");
1768 if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1780 pr_info("No local APIC present or hardware disabled\n");
1786 * init_apic_mappings - initialize APIC mappings
1788 void __init init_apic_mappings(void)
1790 unsigned int new_apicid;
1793 boot_cpu_physical_apicid = read_apic_id();
1797 /* If no local APIC can be found return early */
1798 if (!smp_found_config && detect_init_APIC()) {
1799 /* lets NOP'ify apic operations */
1800 pr_info("APIC: disable apic facility\n");
1803 apic_phys = mp_lapic_addr;
1806 * acpi lapic path already maps that address in
1807 * acpi_register_lapic_address()
1809 if (!acpi_lapic && !smp_found_config)
1810 register_lapic_address(apic_phys);
1814 * Fetch the APIC ID of the BSP in case we have a
1815 * default configuration (or the MP table is broken).
1817 new_apicid = read_apic_id();
1818 if (boot_cpu_physical_apicid != new_apicid) {
1819 boot_cpu_physical_apicid = new_apicid;
1821 * yeah -- we lie about apic_version
1822 * in case if apic was disabled via boot option
1823 * but it's not a problem for SMP compiled kernel
1824 * since smp_sanity_check is prepared for such a case
1825 * and disable smp mode
1827 apic_version[new_apicid] =
1828 GET_APIC_VERSION(apic_read(APIC_LVR));
1832 void __init register_lapic_address(unsigned long address)
1834 mp_lapic_addr = address;
1837 set_fixmap_nocache(FIX_APIC_BASE, address);
1838 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1839 APIC_BASE, mp_lapic_addr);
1841 if (boot_cpu_physical_apicid == -1U) {
1842 boot_cpu_physical_apicid = read_apic_id();
1843 apic_version[boot_cpu_physical_apicid] =
1844 GET_APIC_VERSION(apic_read(APIC_LVR));
1849 * This initializes the IO-APIC and APIC hardware if this is
1852 int apic_version[MAX_LOCAL_APIC];
1854 int __init APIC_init_uniprocessor(void)
1857 pr_info("Apic disabled\n");
1860 #ifdef CONFIG_X86_64
1861 if (!cpu_has_apic) {
1863 pr_info("Apic disabled by BIOS\n");
1867 if (!smp_found_config && !cpu_has_apic)
1871 * Complain if the BIOS pretends there is one.
1873 if (!cpu_has_apic &&
1874 APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
1875 pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
1876 boot_cpu_physical_apicid);
1881 default_setup_apic_routing();
1883 verify_local_APIC();
1886 #ifdef CONFIG_X86_64
1887 apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
1890 * Hack: In case of kdump, after a crash, kernel might be booting
1891 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
1892 * might be zero if read from MP tables. Get it from LAPIC.
1894 # ifdef CONFIG_CRASH_DUMP
1895 boot_cpu_physical_apicid = read_apic_id();
1898 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1901 #ifdef CONFIG_X86_IO_APIC
1903 * Now enable IO-APICs, actually call clear_IO_APIC
1904 * We need clear_IO_APIC before enabling error vector
1906 if (!skip_ioapic_setup && nr_ioapics)
1910 bsp_end_local_APIC_setup();
1912 #ifdef CONFIG_X86_IO_APIC
1913 if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
1920 x86_init.timers.setup_percpu_clockev();
1925 * Local APIC interrupts
1929 * This interrupt should _never_ happen with our APIC/SMP architecture
1931 static inline void __smp_spurious_interrupt(void)
1936 * Check if this really is a spurious interrupt and ACK it
1937 * if it is a vectored one. Just in case...
1938 * Spurious interrupts should not be ACKed.
1940 v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
1941 if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
1944 inc_irq_stat(irq_spurious_count);
1946 /* see sw-dev-man vol 3, chapter 7.4.13.5 */
1947 pr_info("spurious APIC interrupt on CPU#%d, "
1948 "should never happen.\n", smp_processor_id());
1951 void smp_spurious_interrupt(struct pt_regs *regs)
1954 __smp_spurious_interrupt();
1958 void smp_trace_spurious_interrupt(struct pt_regs *regs)
1961 trace_spurious_apic_entry(SPURIOUS_APIC_VECTOR);
1962 __smp_spurious_interrupt();
1963 trace_spurious_apic_exit(SPURIOUS_APIC_VECTOR);
1968 * This interrupt should never happen with our APIC/SMP architecture
1970 static inline void __smp_error_interrupt(struct pt_regs *regs)
1974 static const char * const error_interrupt_reason[] = {
1975 "Send CS error", /* APIC Error Bit 0 */
1976 "Receive CS error", /* APIC Error Bit 1 */
1977 "Send accept error", /* APIC Error Bit 2 */
1978 "Receive accept error", /* APIC Error Bit 3 */
1979 "Redirectable IPI", /* APIC Error Bit 4 */
1980 "Send illegal vector", /* APIC Error Bit 5 */
1981 "Received illegal vector", /* APIC Error Bit 6 */
1982 "Illegal register address", /* APIC Error Bit 7 */
1985 /* First tickle the hardware, only then report what went on. -- REW */
1986 v0 = apic_read(APIC_ESR);
1987 apic_write(APIC_ESR, 0);
1988 v1 = apic_read(APIC_ESR);
1990 atomic_inc(&irq_err_count);
1992 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x(%02x)",
1993 smp_processor_id(), v0 , v1);
1998 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2003 apic_printk(APIC_DEBUG, KERN_CONT "\n");
2007 void smp_error_interrupt(struct pt_regs *regs)
2010 __smp_error_interrupt(regs);
2014 void smp_trace_error_interrupt(struct pt_regs *regs)
2017 trace_error_apic_entry(ERROR_APIC_VECTOR);
2018 __smp_error_interrupt(regs);
2019 trace_error_apic_exit(ERROR_APIC_VECTOR);
2024 * connect_bsp_APIC - attach the APIC to the interrupt system
2026 void __init connect_bsp_APIC(void)
2028 #ifdef CONFIG_X86_32
2031 * Do not trust the local APIC being empty at bootup.
2035 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
2036 * local APIC to INT and NMI lines.
2038 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2039 "enabling APIC mode.\n");
2043 if (apic->enable_apic_mode)
2044 apic->enable_apic_mode();
2048 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2049 * @virt_wire_setup: indicates, whether virtual wire mode is selected
2051 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2054 void disconnect_bsp_APIC(int virt_wire_setup)
2058 #ifdef CONFIG_X86_32
2061 * Put the board back into PIC mode (has an effect only on
2062 * certain older boards). Note that APIC interrupts, including
2063 * IPIs, won't work beyond this point! The only exception are
2066 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2067 "entering PIC mode.\n");
2073 /* Go back to Virtual Wire compatibility mode */
2075 /* For the spurious interrupt use vector F, and enable it */
2076 value = apic_read(APIC_SPIV);
2077 value &= ~APIC_VECTOR_MASK;
2078 value |= APIC_SPIV_APIC_ENABLED;
2080 apic_write(APIC_SPIV, value);
2082 if (!virt_wire_setup) {
2084 * For LVT0 make it edge triggered, active high,
2085 * external and enabled
2087 value = apic_read(APIC_LVT0);
2088 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2089 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2090 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2091 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2092 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2093 apic_write(APIC_LVT0, value);
2096 apic_write(APIC_LVT0, APIC_LVT_MASKED);
2100 * For LVT1 make it edge triggered, active high,
2103 value = apic_read(APIC_LVT1);
2104 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2105 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2106 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2107 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2108 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2109 apic_write(APIC_LVT1, value);
2112 void __cpuinit generic_processor_info(int apicid, int version)
2114 int cpu, max = nr_cpu_ids;
2115 bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2116 phys_cpu_present_map);
2119 * If boot cpu has not been detected yet, then only allow upto
2120 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2122 if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2123 apicid != boot_cpu_physical_apicid) {
2124 int thiscpu = max + disabled_cpus - 1;
2127 "ACPI: NR_CPUS/possible_cpus limit of %i almost"
2128 " reached. Keeping one slot for boot cpu."
2129 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2135 if (num_processors >= nr_cpu_ids) {
2136 int thiscpu = max + disabled_cpus;
2139 "ACPI: NR_CPUS/possible_cpus limit of %i reached."
2140 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2147 if (apicid == boot_cpu_physical_apicid) {
2149 * x86_bios_cpu_apicid is required to have processors listed
2150 * in same order as logical cpu numbers. Hence the first
2151 * entry is BSP, and so on.
2152 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2157 cpu = cpumask_next_zero(-1, cpu_present_mask);
2162 if (version == 0x0) {
2163 pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2167 apic_version[apicid] = version;
2169 if (version != apic_version[boot_cpu_physical_apicid]) {
2170 pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2171 apic_version[boot_cpu_physical_apicid], cpu, version);
2174 physid_set(apicid, phys_cpu_present_map);
2175 if (apicid > max_physical_apicid)
2176 max_physical_apicid = apicid;
2178 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2179 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2180 early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2182 #ifdef CONFIG_X86_32
2183 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2184 apic->x86_32_early_logical_apicid(cpu);
2186 set_cpu_possible(cpu, true);
2187 set_cpu_present(cpu, true);
2190 int hard_smp_processor_id(void)
2192 return read_apic_id();
2195 void default_init_apic_ldr(void)
2199 apic_write(APIC_DFR, APIC_DFR_VALUE);
2200 val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
2201 val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
2202 apic_write(APIC_LDR, val);
2205 int default_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
2206 const struct cpumask *andmask,
2207 unsigned int *apicid)
2211 for_each_cpu_and(cpu, cpumask, andmask) {
2212 if (cpumask_test_cpu(cpu, cpu_online_mask))
2216 if (likely(cpu < nr_cpu_ids)) {
2217 *apicid = per_cpu(x86_cpu_to_apicid, cpu);
2225 * Override the generic EOI implementation with an optimized version.
2226 * Only called during early boot when only one CPU is active and with
2227 * interrupts disabled, so we know this does not race with actual APIC driver
2230 void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2234 for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2235 /* Should happen once for each apic */
2236 WARN_ON((*drv)->eoi_write == eoi_write);
2237 (*drv)->eoi_write = eoi_write;
2248 * 'active' is true if the local APIC was enabled by us and
2249 * not the BIOS; this signifies that we are also responsible
2250 * for disabling it before entering apm/acpi suspend
2253 /* r/w apic fields */
2254 unsigned int apic_id;
2255 unsigned int apic_taskpri;
2256 unsigned int apic_ldr;
2257 unsigned int apic_dfr;
2258 unsigned int apic_spiv;
2259 unsigned int apic_lvtt;
2260 unsigned int apic_lvtpc;
2261 unsigned int apic_lvt0;
2262 unsigned int apic_lvt1;
2263 unsigned int apic_lvterr;
2264 unsigned int apic_tmict;
2265 unsigned int apic_tdcr;
2266 unsigned int apic_thmr;
2269 static int lapic_suspend(void)
2271 unsigned long flags;
2274 if (!apic_pm_state.active)
2277 maxlvt = lapic_get_maxlvt();
2279 apic_pm_state.apic_id = apic_read(APIC_ID);
2280 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2281 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2282 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2283 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2284 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2286 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2287 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2288 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2289 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2290 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2291 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2292 #ifdef CONFIG_X86_THERMAL_VECTOR
2294 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2297 local_irq_save(flags);
2298 disable_local_APIC();
2300 irq_remapping_disable();
2302 local_irq_restore(flags);
2306 static void lapic_resume(void)
2309 unsigned long flags;
2312 if (!apic_pm_state.active)
2315 local_irq_save(flags);
2318 * IO-APIC and PIC have their own resume routines.
2319 * We just mask them here to make sure the interrupt
2320 * subsystem is completely quiet while we enable x2apic
2321 * and interrupt-remapping.
2323 mask_ioapic_entries();
2324 legacy_pic->mask_all();
2330 * Make sure the APICBASE points to the right address
2332 * FIXME! This will be wrong if we ever support suspend on
2333 * SMP! We'll need to do this as part of the CPU restore!
2335 if (boot_cpu_data.x86 >= 6) {
2336 rdmsr(MSR_IA32_APICBASE, l, h);
2337 l &= ~MSR_IA32_APICBASE_BASE;
2338 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2339 wrmsr(MSR_IA32_APICBASE, l, h);
2343 maxlvt = lapic_get_maxlvt();
2344 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2345 apic_write(APIC_ID, apic_pm_state.apic_id);
2346 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2347 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2348 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2349 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2350 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2351 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2352 #if defined(CONFIG_X86_MCE_P4THERMAL) || defined(CONFIG_X86_MCE_INTEL)
2354 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2357 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2358 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2359 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2360 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2361 apic_write(APIC_ESR, 0);
2362 apic_read(APIC_ESR);
2363 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2364 apic_write(APIC_ESR, 0);
2365 apic_read(APIC_ESR);
2367 irq_remapping_reenable(x2apic_mode);
2369 local_irq_restore(flags);
2373 * This device has no shutdown method - fully functioning local APICs
2374 * are needed on every CPU up until machine_halt/restart/poweroff.
2377 static struct syscore_ops lapic_syscore_ops = {
2378 .resume = lapic_resume,
2379 .suspend = lapic_suspend,
2382 static void __cpuinit apic_pm_activate(void)
2384 apic_pm_state.active = 1;
2387 static int __init init_lapic_sysfs(void)
2389 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2391 register_syscore_ops(&lapic_syscore_ops);
2396 /* local apic needs to resume before other devices access its registers. */
2397 core_initcall(init_lapic_sysfs);
2399 #else /* CONFIG_PM */
2401 static void apic_pm_activate(void) { }
2403 #endif /* CONFIG_PM */
2405 #ifdef CONFIG_X86_64
2407 static int __cpuinit apic_cluster_num(void)
2409 int i, clusters, zeros;
2411 u16 *bios_cpu_apicid;
2412 DECLARE_BITMAP(clustermap, NUM_APIC_CLUSTERS);
2414 bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
2415 bitmap_zero(clustermap, NUM_APIC_CLUSTERS);
2417 for (i = 0; i < nr_cpu_ids; i++) {
2418 /* are we being called early in kernel startup? */
2419 if (bios_cpu_apicid) {
2420 id = bios_cpu_apicid[i];
2421 } else if (i < nr_cpu_ids) {
2423 id = per_cpu(x86_bios_cpu_apicid, i);
2429 if (id != BAD_APICID)
2430 __set_bit(APIC_CLUSTERID(id), clustermap);
2433 /* Problem: Partially populated chassis may not have CPUs in some of
2434 * the APIC clusters they have been allocated. Only present CPUs have
2435 * x86_bios_cpu_apicid entries, thus causing zeroes in the bitmap.
2436 * Since clusters are allocated sequentially, count zeros only if
2437 * they are bounded by ones.
2441 for (i = 0; i < NUM_APIC_CLUSTERS; i++) {
2442 if (test_bit(i, clustermap)) {
2443 clusters += 1 + zeros;
2452 static int __cpuinitdata multi_checked;
2453 static int __cpuinitdata multi;
2455 static int __cpuinit set_multi(const struct dmi_system_id *d)
2459 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2464 static const __cpuinitconst struct dmi_system_id multi_dmi_table[] = {
2466 .callback = set_multi,
2467 .ident = "IBM System Summit2",
2469 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2470 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2476 static void __cpuinit dmi_check_multi(void)
2481 dmi_check_system(multi_dmi_table);
2486 * apic_is_clustered_box() -- Check if we can expect good TSC
2488 * Thus far, the major user of this is IBM's Summit2 series:
2489 * Clustered boxes may have unsynced TSC problems if they are
2491 * Use DMI to check them
2493 __cpuinit int apic_is_clustered_box(void)
2503 * ScaleMP vSMPowered boxes have one cluster per board and TSCs are
2504 * not guaranteed to be synced between boards
2506 if (apic_cluster_num() > 1)
2514 * APIC command line parameters
2516 static int __init setup_disableapic(char *arg)
2519 setup_clear_cpu_cap(X86_FEATURE_APIC);
2522 early_param("disableapic", setup_disableapic);
2524 /* same as disableapic, for compatibility */
2525 static int __init setup_nolapic(char *arg)
2527 return setup_disableapic(arg);
2529 early_param("nolapic", setup_nolapic);
2531 static int __init parse_lapic_timer_c2_ok(char *arg)
2533 local_apic_timer_c2_ok = 1;
2536 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2538 static int __init parse_disable_apic_timer(char *arg)
2540 disable_apic_timer = 1;
2543 early_param("noapictimer", parse_disable_apic_timer);
2545 static int __init parse_nolapic_timer(char *arg)
2547 disable_apic_timer = 1;
2550 early_param("nolapic_timer", parse_nolapic_timer);
2552 static int __init apic_set_verbosity(char *arg)
2555 #ifdef CONFIG_X86_64
2556 skip_ioapic_setup = 0;
2562 if (strcmp("debug", arg) == 0)
2563 apic_verbosity = APIC_DEBUG;
2564 else if (strcmp("verbose", arg) == 0)
2565 apic_verbosity = APIC_VERBOSE;
2567 pr_warning("APIC Verbosity level %s not recognised"
2568 " use apic=verbose or apic=debug\n", arg);
2574 early_param("apic", apic_set_verbosity);
2576 static int __init lapic_insert_resource(void)
2581 /* Put local APIC into the resource map. */
2582 lapic_resource.start = apic_phys;
2583 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2584 insert_resource(&iomem_resource, &lapic_resource);
2590 * need call insert after e820_reserve_resources()
2591 * that is using request_resource
2593 late_initcall(lapic_insert_resource);