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/sysdev.h>
28 #include <linux/delay.h>
29 #include <linux/timex.h>
30 #include <linux/dmar.h>
31 #include <linux/init.h>
32 #include <linux/cpu.h>
33 #include <linux/dmi.h>
34 #include <linux/smp.h>
37 #include <asm/perf_event.h>
38 #include <asm/x86_init.h>
39 #include <asm/pgalloc.h>
40 #include <asm/atomic.h>
41 #include <asm/mpspec.h>
42 #include <asm/i8253.h>
43 #include <asm/i8259.h>
44 #include <asm/proto.h>
46 #include <asm/io_apic.h>
54 #include <asm/hypervisor.h>
56 unsigned int num_processors;
58 unsigned disabled_cpus __cpuinitdata;
60 /* Processor that is doing the boot up */
61 unsigned int boot_cpu_physical_apicid = -1U;
64 * The highest APIC ID seen during enumeration.
66 unsigned int max_physical_apicid;
69 * Bitmask of physically existing CPUs:
71 physid_mask_t phys_cpu_present_map;
74 * Map cpu index to physical APIC ID
76 DEFINE_EARLY_PER_CPU(u16, x86_cpu_to_apicid, BAD_APICID);
77 DEFINE_EARLY_PER_CPU(u16, x86_bios_cpu_apicid, BAD_APICID);
78 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
79 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
83 * Knob to control our willingness to enable the local APIC.
87 static int force_enable_local_apic;
89 * APIC command line parameters
91 static int __init parse_lapic(char *arg)
93 force_enable_local_apic = 1;
96 early_param("lapic", parse_lapic);
97 /* Local APIC was disabled by the BIOS and enabled by the kernel */
98 static int enabled_via_apicbase;
101 * Handle interrupt mode configuration register (IMCR).
102 * This register controls whether the interrupt signals
103 * that reach the BSP come from the master PIC or from the
104 * local APIC. Before entering Symmetric I/O Mode, either
105 * the BIOS or the operating system must switch out of
106 * PIC Mode by changing the IMCR.
108 static inline void imcr_pic_to_apic(void)
110 /* select IMCR register */
112 /* NMI and 8259 INTR go through APIC */
116 static inline void imcr_apic_to_pic(void)
118 /* select IMCR register */
120 /* NMI and 8259 INTR go directly to BSP */
126 static int apic_calibrate_pmtmr __initdata;
127 static __init int setup_apicpmtimer(char *s)
129 apic_calibrate_pmtmr = 1;
133 __setup("apicpmtimer", setup_apicpmtimer);
137 #ifdef CONFIG_X86_X2APIC
138 /* x2apic enabled before OS handover */
139 static int x2apic_preenabled;
140 static __init int setup_nox2apic(char *str)
142 if (x2apic_enabled()) {
143 pr_warning("Bios already enabled x2apic, "
144 "can't enforce nox2apic");
148 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
151 early_param("nox2apic", setup_nox2apic);
154 unsigned long mp_lapic_addr;
156 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
157 static int disable_apic_timer __cpuinitdata;
158 /* Local APIC timer works in C2 */
159 int local_apic_timer_c2_ok;
160 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
162 int first_system_vector = 0xfe;
165 * Debug level, exported for io_apic.c
167 unsigned int apic_verbosity;
171 /* Have we found an MP table */
172 int smp_found_config;
174 static struct resource lapic_resource = {
175 .name = "Local APIC",
176 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
179 static unsigned int calibration_result;
181 static int lapic_next_event(unsigned long delta,
182 struct clock_event_device *evt);
183 static void lapic_timer_setup(enum clock_event_mode mode,
184 struct clock_event_device *evt);
185 static void lapic_timer_broadcast(const struct cpumask *mask);
186 static void apic_pm_activate(void);
189 * The local apic timer can be used for any function which is CPU local.
191 static struct clock_event_device lapic_clockevent = {
193 .features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
194 | CLOCK_EVT_FEAT_C3STOP | CLOCK_EVT_FEAT_DUMMY,
196 .set_mode = lapic_timer_setup,
197 .set_next_event = lapic_next_event,
198 .broadcast = lapic_timer_broadcast,
202 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
204 static unsigned long apic_phys;
207 * Get the LAPIC version
209 static inline int lapic_get_version(void)
211 return GET_APIC_VERSION(apic_read(APIC_LVR));
215 * Check, if the APIC is integrated or a separate chip
217 static inline int lapic_is_integrated(void)
222 return APIC_INTEGRATED(lapic_get_version());
227 * Check, whether this is a modern or a first generation APIC
229 static int modern_apic(void)
231 /* AMD systems use old APIC versions, so check the CPU */
232 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
233 boot_cpu_data.x86 >= 0xf)
235 return lapic_get_version() >= 0x14;
239 * right after this call apic become NOOP driven
240 * so apic->write/read doesn't do anything
242 void apic_disable(void)
244 pr_info("APIC: switched to apic NOOP\n");
248 void native_apic_wait_icr_idle(void)
250 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
254 u32 native_safe_apic_wait_icr_idle(void)
261 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
265 } while (timeout++ < 1000);
270 void native_apic_icr_write(u32 low, u32 id)
272 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
273 apic_write(APIC_ICR, low);
276 u64 native_apic_icr_read(void)
280 icr2 = apic_read(APIC_ICR2);
281 icr1 = apic_read(APIC_ICR);
283 return icr1 | ((u64)icr2 << 32);
287 * enable_NMI_through_LVT0 - enable NMI through local vector table 0
289 void __cpuinit enable_NMI_through_LVT0(void)
293 /* unmask and set to NMI */
296 /* Level triggered for 82489DX (32bit mode) */
297 if (!lapic_is_integrated())
298 v |= APIC_LVT_LEVEL_TRIGGER;
300 apic_write(APIC_LVT0, v);
305 * get_physical_broadcast - Get number of physical broadcast IDs
307 int get_physical_broadcast(void)
309 return modern_apic() ? 0xff : 0xf;
314 * lapic_get_maxlvt - get the maximum number of local vector table entries
316 int lapic_get_maxlvt(void)
320 v = apic_read(APIC_LVR);
322 * - we always have APIC integrated on 64bit mode
323 * - 82489DXs do not report # of LVT entries
325 return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
333 #define APIC_DIVISOR 16
336 * This function sets up the local APIC timer, with a timeout of
337 * 'clocks' APIC bus clock. During calibration we actually call
338 * this function twice on the boot CPU, once with a bogus timeout
339 * value, second time for real. The other (noncalibrating) CPUs
340 * call this function only once, with the real, calibrated value.
342 * We do reads before writes even if unnecessary, to get around the
343 * P5 APIC double write bug.
345 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
347 unsigned int lvtt_value, tmp_value;
349 lvtt_value = LOCAL_TIMER_VECTOR;
351 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
352 if (!lapic_is_integrated())
353 lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
356 lvtt_value |= APIC_LVT_MASKED;
358 apic_write(APIC_LVTT, lvtt_value);
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; /* 0: uninitialized */
405 if (offset >= APIC_EILVT_NR_MAX)
408 rsvd = atomic_read(&eilvt_offsets[offset]) & ~APIC_EILVT_MASKED;
411 !eilvt_entry_is_changeable(rsvd, new))
412 /* may not change if vectors are different */
414 rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
415 } while (rsvd != new);
421 * If mask=1, the LVT entry does not generate interrupts while mask=0
422 * enables the vector. See also the BKDGs.
425 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
427 unsigned long reg = APIC_EILVTn(offset);
428 unsigned int new, old, reserved;
430 new = (mask << 16) | (msg_type << 8) | vector;
431 old = apic_read(reg);
432 reserved = reserve_eilvt_offset(offset, new);
434 if (reserved != new) {
435 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
436 "vector 0x%x, but the register is already in use for "
437 "vector 0x%x on another cpu\n",
438 smp_processor_id(), reg, offset, new, reserved);
442 if (!eilvt_entry_is_changeable(old, new)) {
443 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
444 "vector 0x%x, but the register is already in use for "
445 "vector 0x%x on this cpu\n",
446 smp_processor_id(), reg, offset, new, old);
450 apic_write(reg, new);
454 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
457 * Program the next event, relative to now
459 static int lapic_next_event(unsigned long delta,
460 struct clock_event_device *evt)
462 apic_write(APIC_TMICT, delta);
467 * Setup the lapic timer in periodic or oneshot mode
469 static void lapic_timer_setup(enum clock_event_mode mode,
470 struct clock_event_device *evt)
475 /* Lapic used as dummy for broadcast ? */
476 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
479 local_irq_save(flags);
482 case CLOCK_EVT_MODE_PERIODIC:
483 case CLOCK_EVT_MODE_ONESHOT:
484 __setup_APIC_LVTT(calibration_result,
485 mode != CLOCK_EVT_MODE_PERIODIC, 1);
487 case CLOCK_EVT_MODE_UNUSED:
488 case CLOCK_EVT_MODE_SHUTDOWN:
489 v = apic_read(APIC_LVTT);
490 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
491 apic_write(APIC_LVTT, v);
492 apic_write(APIC_TMICT, 0);
494 case CLOCK_EVT_MODE_RESUME:
495 /* Nothing to do here */
499 local_irq_restore(flags);
503 * Local APIC timer broadcast function
505 static void lapic_timer_broadcast(const struct cpumask *mask)
508 apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
513 * Setup the local APIC timer for this CPU. Copy the initialized values
514 * of the boot CPU and register the clock event in the framework.
516 static void __cpuinit setup_APIC_timer(void)
518 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
520 if (cpu_has(__this_cpu_ptr(&cpu_info), X86_FEATURE_ARAT)) {
521 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
522 /* Make LAPIC timer preferrable over percpu HPET */
523 lapic_clockevent.rating = 150;
526 memcpy(levt, &lapic_clockevent, sizeof(*levt));
527 levt->cpumask = cpumask_of(smp_processor_id());
529 clockevents_register_device(levt);
533 * In this functions we calibrate APIC bus clocks to the external timer.
535 * We want to do the calibration only once since we want to have local timer
536 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
539 * This was previously done by reading the PIT/HPET and waiting for a wrap
540 * around to find out, that a tick has elapsed. I have a box, where the PIT
541 * readout is broken, so it never gets out of the wait loop again. This was
542 * also reported by others.
544 * Monitoring the jiffies value is inaccurate and the clockevents
545 * infrastructure allows us to do a simple substitution of the interrupt
548 * The calibration routine also uses the pm_timer when possible, as the PIT
549 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
550 * back to normal later in the boot process).
553 #define LAPIC_CAL_LOOPS (HZ/10)
555 static __initdata int lapic_cal_loops = -1;
556 static __initdata long lapic_cal_t1, lapic_cal_t2;
557 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
558 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
559 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
562 * Temporary interrupt handler.
564 static void __init lapic_cal_handler(struct clock_event_device *dev)
566 unsigned long long tsc = 0;
567 long tapic = apic_read(APIC_TMCCT);
568 unsigned long pm = acpi_pm_read_early();
573 switch (lapic_cal_loops++) {
575 lapic_cal_t1 = tapic;
576 lapic_cal_tsc1 = tsc;
578 lapic_cal_j1 = jiffies;
581 case LAPIC_CAL_LOOPS:
582 lapic_cal_t2 = tapic;
583 lapic_cal_tsc2 = tsc;
584 if (pm < lapic_cal_pm1)
585 pm += ACPI_PM_OVRRUN;
587 lapic_cal_j2 = jiffies;
593 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
595 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
596 const long pm_thresh = pm_100ms / 100;
600 #ifndef CONFIG_X86_PM_TIMER
604 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
606 /* Check, if the PM timer is available */
610 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
612 if (deltapm > (pm_100ms - pm_thresh) &&
613 deltapm < (pm_100ms + pm_thresh)) {
614 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
618 res = (((u64)deltapm) * mult) >> 22;
619 do_div(res, 1000000);
620 pr_warning("APIC calibration not consistent "
621 "with PM-Timer: %ldms instead of 100ms\n",(long)res);
623 /* Correct the lapic counter value */
624 res = (((u64)(*delta)) * pm_100ms);
625 do_div(res, deltapm);
626 pr_info("APIC delta adjusted to PM-Timer: "
627 "%lu (%ld)\n", (unsigned long)res, *delta);
630 /* Correct the tsc counter value */
632 res = (((u64)(*deltatsc)) * pm_100ms);
633 do_div(res, deltapm);
634 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
635 "PM-Timer: %lu (%ld)\n",
636 (unsigned long)res, *deltatsc);
637 *deltatsc = (long)res;
643 static int __init calibrate_APIC_clock(void)
645 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
646 void (*real_handler)(struct clock_event_device *dev);
647 unsigned long deltaj;
648 long delta, deltatsc;
649 int pm_referenced = 0;
653 /* Replace the global interrupt handler */
654 real_handler = global_clock_event->event_handler;
655 global_clock_event->event_handler = lapic_cal_handler;
658 * Setup the APIC counter to maximum. There is no way the lapic
659 * can underflow in the 100ms detection time frame
661 __setup_APIC_LVTT(0xffffffff, 0, 0);
663 /* Let the interrupts run */
666 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
671 /* Restore the real event handler */
672 global_clock_event->event_handler = real_handler;
674 /* Build delta t1-t2 as apic timer counts down */
675 delta = lapic_cal_t1 - lapic_cal_t2;
676 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
678 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
680 /* we trust the PM based calibration if possible */
681 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
684 /* Calculate the scaled math multiplication factor */
685 lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
686 lapic_clockevent.shift);
687 lapic_clockevent.max_delta_ns =
688 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
689 lapic_clockevent.min_delta_ns =
690 clockevent_delta2ns(0xF, &lapic_clockevent);
692 calibration_result = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
694 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
695 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
696 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
700 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
702 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
703 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
706 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
708 calibration_result / (1000000 / HZ),
709 calibration_result % (1000000 / HZ));
712 * Do a sanity check on the APIC calibration result
714 if (calibration_result < (1000000 / HZ)) {
716 pr_warning("APIC frequency too slow, disabling apic timer\n");
720 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
723 * PM timer calibration failed or not turned on
724 * so lets try APIC timer based calibration
726 if (!pm_referenced) {
727 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
730 * Setup the apic timer manually
732 levt->event_handler = lapic_cal_handler;
733 lapic_timer_setup(CLOCK_EVT_MODE_PERIODIC, levt);
734 lapic_cal_loops = -1;
736 /* Let the interrupts run */
739 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
742 /* Stop the lapic timer */
743 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt);
746 deltaj = lapic_cal_j2 - lapic_cal_j1;
747 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
749 /* Check, if the jiffies result is consistent */
750 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
751 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
753 levt->features |= CLOCK_EVT_FEAT_DUMMY;
757 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
758 pr_warning("APIC timer disabled due to verification failure\n");
766 * Setup the boot APIC
768 * Calibrate and verify the result.
770 void __init setup_boot_APIC_clock(void)
773 * The local apic timer can be disabled via the kernel
774 * commandline or from the CPU detection code. Register the lapic
775 * timer as a dummy clock event source on SMP systems, so the
776 * broadcast mechanism is used. On UP systems simply ignore it.
778 if (disable_apic_timer) {
779 pr_info("Disabling APIC timer\n");
780 /* No broadcast on UP ! */
781 if (num_possible_cpus() > 1) {
782 lapic_clockevent.mult = 1;
788 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
789 "calibrating APIC timer ...\n");
791 if (calibrate_APIC_clock()) {
792 /* No broadcast on UP ! */
793 if (num_possible_cpus() > 1)
799 * If nmi_watchdog is set to IO_APIC, we need the
800 * PIT/HPET going. Otherwise register lapic as a dummy
803 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
805 /* Setup the lapic or request the broadcast */
809 void __cpuinit setup_secondary_APIC_clock(void)
815 * The guts of the apic timer interrupt
817 static void local_apic_timer_interrupt(void)
819 int cpu = smp_processor_id();
820 struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
823 * Normally we should not be here till LAPIC has been initialized but
824 * in some cases like kdump, its possible that there is a pending LAPIC
825 * timer interrupt from previous kernel's context and is delivered in
826 * new kernel the moment interrupts are enabled.
828 * Interrupts are enabled early and LAPIC is setup much later, hence
829 * its possible that when we get here evt->event_handler is NULL.
830 * Check for event_handler being NULL and discard the interrupt as
833 if (!evt->event_handler) {
834 pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
836 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
841 * the NMI deadlock-detector uses this.
843 inc_irq_stat(apic_timer_irqs);
845 evt->event_handler(evt);
849 * Local APIC timer interrupt. This is the most natural way for doing
850 * local interrupts, but local timer interrupts can be emulated by
851 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
853 * [ if a single-CPU system runs an SMP kernel then we call the local
854 * interrupt as well. Thus we cannot inline the local irq ... ]
856 void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
858 struct pt_regs *old_regs = set_irq_regs(regs);
861 * NOTE! We'd better ACK the irq immediately,
862 * because timer handling can be slow.
866 * update_process_times() expects us to have done irq_enter().
867 * Besides, if we don't timer interrupts ignore the global
868 * interrupt lock, which is the WrongThing (tm) to do.
872 local_apic_timer_interrupt();
875 set_irq_regs(old_regs);
878 int setup_profiling_timer(unsigned int multiplier)
884 * Local APIC start and shutdown
888 * clear_local_APIC - shutdown the local APIC
890 * This is called, when a CPU is disabled and before rebooting, so the state of
891 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
892 * leftovers during boot.
894 void clear_local_APIC(void)
899 /* APIC hasn't been mapped yet */
900 if (!x2apic_mode && !apic_phys)
903 maxlvt = lapic_get_maxlvt();
905 * Masking an LVT entry can trigger a local APIC error
906 * if the vector is zero. Mask LVTERR first to prevent this.
909 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
910 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
913 * Careful: we have to set masks only first to deassert
914 * any level-triggered sources.
916 v = apic_read(APIC_LVTT);
917 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
918 v = apic_read(APIC_LVT0);
919 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
920 v = apic_read(APIC_LVT1);
921 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
923 v = apic_read(APIC_LVTPC);
924 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
927 /* lets not touch this if we didn't frob it */
928 #ifdef CONFIG_X86_THERMAL_VECTOR
930 v = apic_read(APIC_LVTTHMR);
931 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
934 #ifdef CONFIG_X86_MCE_INTEL
936 v = apic_read(APIC_LVTCMCI);
937 if (!(v & APIC_LVT_MASKED))
938 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
943 * Clean APIC state for other OSs:
945 apic_write(APIC_LVTT, APIC_LVT_MASKED);
946 apic_write(APIC_LVT0, APIC_LVT_MASKED);
947 apic_write(APIC_LVT1, APIC_LVT_MASKED);
949 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
951 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
953 /* Integrated APIC (!82489DX) ? */
954 if (lapic_is_integrated()) {
956 /* Clear ESR due to Pentium errata 3AP and 11AP */
957 apic_write(APIC_ESR, 0);
963 * disable_local_APIC - clear and disable the local APIC
965 void disable_local_APIC(void)
969 /* APIC hasn't been mapped yet */
970 if (!x2apic_mode && !apic_phys)
976 * Disable APIC (implies clearing of registers
979 value = apic_read(APIC_SPIV);
980 value &= ~APIC_SPIV_APIC_ENABLED;
981 apic_write(APIC_SPIV, value);
985 * When LAPIC was disabled by the BIOS and enabled by the kernel,
986 * restore the disabled state.
988 if (enabled_via_apicbase) {
991 rdmsr(MSR_IA32_APICBASE, l, h);
992 l &= ~MSR_IA32_APICBASE_ENABLE;
993 wrmsr(MSR_IA32_APICBASE, l, h);
999 * If Linux enabled the LAPIC against the BIOS default disable it down before
1000 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1001 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1002 * for the case where Linux didn't enable the LAPIC.
1004 void lapic_shutdown(void)
1006 unsigned long flags;
1008 if (!cpu_has_apic && !apic_from_smp_config())
1011 local_irq_save(flags);
1013 #ifdef CONFIG_X86_32
1014 if (!enabled_via_apicbase)
1018 disable_local_APIC();
1021 local_irq_restore(flags);
1025 * This is to verify that we're looking at a real local APIC.
1026 * Check these against your board if the CPUs aren't getting
1027 * started for no apparent reason.
1029 int __init verify_local_APIC(void)
1031 unsigned int reg0, reg1;
1034 * The version register is read-only in a real APIC.
1036 reg0 = apic_read(APIC_LVR);
1037 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
1038 apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
1039 reg1 = apic_read(APIC_LVR);
1040 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
1043 * The two version reads above should print the same
1044 * numbers. If the second one is different, then we
1045 * poke at a non-APIC.
1051 * Check if the version looks reasonably.
1053 reg1 = GET_APIC_VERSION(reg0);
1054 if (reg1 == 0x00 || reg1 == 0xff)
1056 reg1 = lapic_get_maxlvt();
1057 if (reg1 < 0x02 || reg1 == 0xff)
1061 * The ID register is read/write in a real APIC.
1063 reg0 = apic_read(APIC_ID);
1064 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
1065 apic_write(APIC_ID, reg0 ^ apic->apic_id_mask);
1066 reg1 = apic_read(APIC_ID);
1067 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
1068 apic_write(APIC_ID, reg0);
1069 if (reg1 != (reg0 ^ apic->apic_id_mask))
1073 * The next two are just to see if we have sane values.
1074 * They're only really relevant if we're in Virtual Wire
1075 * compatibility mode, but most boxes are anymore.
1077 reg0 = apic_read(APIC_LVT0);
1078 apic_printk(APIC_DEBUG, "Getting LVT0: %x\n", reg0);
1079 reg1 = apic_read(APIC_LVT1);
1080 apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
1086 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1088 void __init sync_Arb_IDs(void)
1091 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1094 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1100 apic_wait_icr_idle();
1102 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1103 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1104 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1108 * An initial setup of the virtual wire mode.
1110 void __init init_bsp_APIC(void)
1115 * Don't do the setup now if we have a SMP BIOS as the
1116 * through-I/O-APIC virtual wire mode might be active.
1118 if (smp_found_config || !cpu_has_apic)
1122 * Do not trust the local APIC being empty at bootup.
1129 value = apic_read(APIC_SPIV);
1130 value &= ~APIC_VECTOR_MASK;
1131 value |= APIC_SPIV_APIC_ENABLED;
1133 #ifdef CONFIG_X86_32
1134 /* This bit is reserved on P4/Xeon and should be cleared */
1135 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1136 (boot_cpu_data.x86 == 15))
1137 value &= ~APIC_SPIV_FOCUS_DISABLED;
1140 value |= APIC_SPIV_FOCUS_DISABLED;
1141 value |= SPURIOUS_APIC_VECTOR;
1142 apic_write(APIC_SPIV, value);
1145 * Set up the virtual wire mode.
1147 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1148 value = APIC_DM_NMI;
1149 if (!lapic_is_integrated()) /* 82489DX */
1150 value |= APIC_LVT_LEVEL_TRIGGER;
1151 apic_write(APIC_LVT1, value);
1154 static void __cpuinit lapic_setup_esr(void)
1156 unsigned int oldvalue, value, maxlvt;
1158 if (!lapic_is_integrated()) {
1159 pr_info("No ESR for 82489DX.\n");
1163 if (apic->disable_esr) {
1165 * Something untraceable is creating bad interrupts on
1166 * secondary quads ... for the moment, just leave the
1167 * ESR disabled - we can't do anything useful with the
1168 * errors anyway - mbligh
1170 pr_info("Leaving ESR disabled.\n");
1174 maxlvt = lapic_get_maxlvt();
1175 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1176 apic_write(APIC_ESR, 0);
1177 oldvalue = apic_read(APIC_ESR);
1179 /* enables sending errors */
1180 value = ERROR_APIC_VECTOR;
1181 apic_write(APIC_LVTERR, value);
1184 * spec says clear errors after enabling vector.
1187 apic_write(APIC_ESR, 0);
1188 value = apic_read(APIC_ESR);
1189 if (value != oldvalue)
1190 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1191 "vector: 0x%08x after: 0x%08x\n",
1196 * setup_local_APIC - setup the local APIC
1198 * Used to setup local APIC while initializing BSP or bringin up APs.
1199 * Always called with preemption disabled.
1201 void __cpuinit setup_local_APIC(void)
1203 int cpu = smp_processor_id();
1204 unsigned int value, queued;
1205 int i, j, acked = 0;
1206 unsigned long long tsc = 0, ntsc;
1207 long long max_loops = cpu_khz;
1213 disable_ioapic_support();
1217 #ifdef CONFIG_X86_32
1218 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1219 if (lapic_is_integrated() && apic->disable_esr) {
1220 apic_write(APIC_ESR, 0);
1221 apic_write(APIC_ESR, 0);
1222 apic_write(APIC_ESR, 0);
1223 apic_write(APIC_ESR, 0);
1226 perf_events_lapic_init();
1229 * Double-check whether this APIC is really registered.
1230 * This is meaningless in clustered apic mode, so we skip it.
1232 BUG_ON(!apic->apic_id_registered());
1235 * Intel recommends to set DFR, LDR and TPR before enabling
1236 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1237 * document number 292116). So here it goes...
1239 apic->init_apic_ldr();
1242 * Set Task Priority to 'accept all'. We never change this
1245 value = apic_read(APIC_TASKPRI);
1246 value &= ~APIC_TPRI_MASK;
1247 apic_write(APIC_TASKPRI, value);
1250 * After a crash, we no longer service the interrupts and a pending
1251 * interrupt from previous kernel might still have ISR bit set.
1253 * Most probably by now CPU has serviced that pending interrupt and
1254 * it might not have done the ack_APIC_irq() because it thought,
1255 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1256 * does not clear the ISR bit and cpu thinks it has already serivced
1257 * the interrupt. Hence a vector might get locked. It was noticed
1258 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1262 for (i = APIC_ISR_NR - 1; i >= 0; i--)
1263 queued |= apic_read(APIC_IRR + i*0x10);
1265 for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1266 value = apic_read(APIC_ISR + i*0x10);
1267 for (j = 31; j >= 0; j--) {
1268 if (value & (1<<j)) {
1275 printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1281 max_loops = (cpu_khz << 10) - (ntsc - tsc);
1284 } while (queued && max_loops > 0);
1285 WARN_ON(max_loops <= 0);
1288 * Now that we are all set up, enable the APIC
1290 value = apic_read(APIC_SPIV);
1291 value &= ~APIC_VECTOR_MASK;
1295 value |= APIC_SPIV_APIC_ENABLED;
1297 #ifdef CONFIG_X86_32
1299 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1300 * certain networking cards. If high frequency interrupts are
1301 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1302 * entry is masked/unmasked at a high rate as well then sooner or
1303 * later IOAPIC line gets 'stuck', no more interrupts are received
1304 * from the device. If focus CPU is disabled then the hang goes
1307 * [ This bug can be reproduced easily with a level-triggered
1308 * PCI Ne2000 networking cards and PII/PIII processors, dual
1312 * Actually disabling the focus CPU check just makes the hang less
1313 * frequent as it makes the interrupt distributon model be more
1314 * like LRU than MRU (the short-term load is more even across CPUs).
1315 * See also the comment in end_level_ioapic_irq(). --macro
1319 * - enable focus processor (bit==0)
1320 * - 64bit mode always use processor focus
1321 * so no need to set it
1323 value &= ~APIC_SPIV_FOCUS_DISABLED;
1327 * Set spurious IRQ vector
1329 value |= SPURIOUS_APIC_VECTOR;
1330 apic_write(APIC_SPIV, value);
1333 * Set up LVT0, LVT1:
1335 * set up through-local-APIC on the BP's LINT0. This is not
1336 * strictly necessary in pure symmetric-IO mode, but sometimes
1337 * we delegate interrupts to the 8259A.
1340 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1342 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1343 if (!cpu && (pic_mode || !value)) {
1344 value = APIC_DM_EXTINT;
1345 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1347 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1348 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1350 apic_write(APIC_LVT0, value);
1353 * only the BP should see the LINT1 NMI signal, obviously.
1356 value = APIC_DM_NMI;
1358 value = APIC_DM_NMI | APIC_LVT_MASKED;
1359 if (!lapic_is_integrated()) /* 82489DX */
1360 value |= APIC_LVT_LEVEL_TRIGGER;
1361 apic_write(APIC_LVT1, value);
1363 #ifdef CONFIG_X86_MCE_INTEL
1364 /* Recheck CMCI information after local APIC is up on CPU #0 */
1370 void __cpuinit end_local_APIC_setup(void)
1374 #ifdef CONFIG_X86_32
1377 /* Disable the local apic timer */
1378 value = apic_read(APIC_LVTT);
1379 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1380 apic_write(APIC_LVTT, value);
1387 void __init bsp_end_local_APIC_setup(void)
1389 end_local_APIC_setup();
1392 * Now that local APIC setup is completed for BP, configure the fault
1393 * handling for interrupt remapping.
1395 if (intr_remapping_enabled)
1396 enable_drhd_fault_handling();
1400 #ifdef CONFIG_X86_X2APIC
1401 void check_x2apic(void)
1403 if (x2apic_enabled()) {
1404 pr_info("x2apic enabled by BIOS, switching to x2apic ops\n");
1405 x2apic_preenabled = x2apic_mode = 1;
1409 void enable_x2apic(void)
1416 rdmsr(MSR_IA32_APICBASE, msr, msr2);
1417 if (!(msr & X2APIC_ENABLE)) {
1418 printk_once(KERN_INFO "Enabling x2apic\n");
1419 wrmsr(MSR_IA32_APICBASE, msr | X2APIC_ENABLE, 0);
1422 #endif /* CONFIG_X86_X2APIC */
1424 int __init enable_IR(void)
1426 #ifdef CONFIG_INTR_REMAP
1427 if (!intr_remapping_supported()) {
1428 pr_debug("intr-remapping not supported\n");
1432 if (!x2apic_preenabled && skip_ioapic_setup) {
1433 pr_info("Skipped enabling intr-remap because of skipping "
1438 if (enable_intr_remapping(x2apic_supported()))
1441 pr_info("Enabled Interrupt-remapping\n");
1449 void __init enable_IR_x2apic(void)
1451 unsigned long flags;
1452 struct IO_APIC_route_entry **ioapic_entries;
1453 int ret, x2apic_enabled = 0;
1454 int dmar_table_init_ret;
1456 dmar_table_init_ret = dmar_table_init();
1457 if (dmar_table_init_ret && !x2apic_supported())
1460 ioapic_entries = alloc_ioapic_entries();
1461 if (!ioapic_entries) {
1462 pr_err("Allocate ioapic_entries failed\n");
1466 ret = save_IO_APIC_setup(ioapic_entries);
1468 pr_info("Saving IO-APIC state failed: %d\n", ret);
1472 local_irq_save(flags);
1473 legacy_pic->mask_all();
1474 mask_IO_APIC_setup(ioapic_entries);
1476 if (dmar_table_init_ret)
1482 /* IR is required if there is APIC ID > 255 even when running
1485 if (max_physical_apicid > 255 ||
1486 !hypervisor_x2apic_available())
1489 * without IR all CPUs can be addressed by IOAPIC/MSI
1490 * only in physical mode
1492 x2apic_force_phys();
1497 if (x2apic_supported() && !x2apic_mode) {
1500 pr_info("Enabled x2apic\n");
1504 if (!ret) /* IR enabling failed */
1505 restore_IO_APIC_setup(ioapic_entries);
1506 legacy_pic->restore_mask();
1507 local_irq_restore(flags);
1511 free_ioapic_entries(ioapic_entries);
1516 if (x2apic_preenabled)
1517 panic("x2apic: enabled by BIOS but kernel init failed.");
1518 else if (cpu_has_x2apic)
1519 pr_info("Not enabling x2apic, Intr-remapping init failed.\n");
1522 #ifdef CONFIG_X86_64
1524 * Detect and enable local APICs on non-SMP boards.
1525 * Original code written by Keir Fraser.
1526 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1527 * not correctly set up (usually the APIC timer won't work etc.)
1529 static int __init detect_init_APIC(void)
1531 if (!cpu_has_apic) {
1532 pr_info("No local APIC present\n");
1536 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1541 static int apic_verify(void)
1546 * The APIC feature bit should now be enabled
1549 features = cpuid_edx(1);
1550 if (!(features & (1 << X86_FEATURE_APIC))) {
1551 pr_warning("Could not enable APIC!\n");
1554 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1555 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1557 /* The BIOS may have set up the APIC at some other address */
1558 rdmsr(MSR_IA32_APICBASE, l, h);
1559 if (l & MSR_IA32_APICBASE_ENABLE)
1560 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1562 pr_info("Found and enabled local APIC!\n");
1566 int apic_force_enable(void)
1574 * Some BIOSes disable the local APIC in the APIC_BASE
1575 * MSR. This can only be done in software for Intel P6 or later
1576 * and AMD K7 (Model > 1) or later.
1578 rdmsr(MSR_IA32_APICBASE, l, h);
1579 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1580 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1581 l &= ~MSR_IA32_APICBASE_BASE;
1582 l |= MSR_IA32_APICBASE_ENABLE | APIC_DEFAULT_PHYS_BASE;
1583 wrmsr(MSR_IA32_APICBASE, l, h);
1584 enabled_via_apicbase = 1;
1586 return apic_verify();
1590 * Detect and initialize APIC
1592 static int __init detect_init_APIC(void)
1594 /* Disabled by kernel option? */
1598 switch (boot_cpu_data.x86_vendor) {
1599 case X86_VENDOR_AMD:
1600 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1601 (boot_cpu_data.x86 >= 15))
1604 case X86_VENDOR_INTEL:
1605 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1606 (boot_cpu_data.x86 == 5 && cpu_has_apic))
1613 if (!cpu_has_apic) {
1615 * Over-ride BIOS and try to enable the local APIC only if
1616 * "lapic" specified.
1618 if (!force_enable_local_apic) {
1619 pr_info("Local APIC disabled by BIOS -- "
1620 "you can enable it with \"lapic\"\n");
1623 if (apic_force_enable())
1635 pr_info("No local APIC present or hardware disabled\n");
1641 * init_apic_mappings - initialize APIC mappings
1643 void __init init_apic_mappings(void)
1645 unsigned int new_apicid;
1648 boot_cpu_physical_apicid = read_apic_id();
1652 /* If no local APIC can be found return early */
1653 if (!smp_found_config && detect_init_APIC()) {
1654 /* lets NOP'ify apic operations */
1655 pr_info("APIC: disable apic facility\n");
1658 apic_phys = mp_lapic_addr;
1661 * acpi lapic path already maps that address in
1662 * acpi_register_lapic_address()
1664 if (!acpi_lapic && !smp_found_config)
1665 register_lapic_address(apic_phys);
1669 * Fetch the APIC ID of the BSP in case we have a
1670 * default configuration (or the MP table is broken).
1672 new_apicid = read_apic_id();
1673 if (boot_cpu_physical_apicid != new_apicid) {
1674 boot_cpu_physical_apicid = new_apicid;
1676 * yeah -- we lie about apic_version
1677 * in case if apic was disabled via boot option
1678 * but it's not a problem for SMP compiled kernel
1679 * since smp_sanity_check is prepared for such a case
1680 * and disable smp mode
1682 apic_version[new_apicid] =
1683 GET_APIC_VERSION(apic_read(APIC_LVR));
1687 void __init register_lapic_address(unsigned long address)
1689 mp_lapic_addr = address;
1692 set_fixmap_nocache(FIX_APIC_BASE, address);
1693 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1694 APIC_BASE, mp_lapic_addr);
1696 if (boot_cpu_physical_apicid == -1U) {
1697 boot_cpu_physical_apicid = read_apic_id();
1698 apic_version[boot_cpu_physical_apicid] =
1699 GET_APIC_VERSION(apic_read(APIC_LVR));
1704 * This initializes the IO-APIC and APIC hardware if this is
1707 int apic_version[MAX_LOCAL_APIC];
1709 int __init APIC_init_uniprocessor(void)
1712 pr_info("Apic disabled\n");
1715 #ifdef CONFIG_X86_64
1716 if (!cpu_has_apic) {
1718 pr_info("Apic disabled by BIOS\n");
1722 if (!smp_found_config && !cpu_has_apic)
1726 * Complain if the BIOS pretends there is one.
1728 if (!cpu_has_apic &&
1729 APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
1730 pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
1731 boot_cpu_physical_apicid);
1736 default_setup_apic_routing();
1738 verify_local_APIC();
1741 #ifdef CONFIG_X86_64
1742 apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
1745 * Hack: In case of kdump, after a crash, kernel might be booting
1746 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
1747 * might be zero if read from MP tables. Get it from LAPIC.
1749 # ifdef CONFIG_CRASH_DUMP
1750 boot_cpu_physical_apicid = read_apic_id();
1753 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1756 #ifdef CONFIG_X86_IO_APIC
1758 * Now enable IO-APICs, actually call clear_IO_APIC
1759 * We need clear_IO_APIC before enabling error vector
1761 if (!skip_ioapic_setup && nr_ioapics)
1765 bsp_end_local_APIC_setup();
1767 #ifdef CONFIG_X86_IO_APIC
1768 if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
1775 x86_init.timers.setup_percpu_clockev();
1780 * Local APIC interrupts
1784 * This interrupt should _never_ happen with our APIC/SMP architecture
1786 void smp_spurious_interrupt(struct pt_regs *regs)
1793 * Check if this really is a spurious interrupt and ACK it
1794 * if it is a vectored one. Just in case...
1795 * Spurious interrupts should not be ACKed.
1797 v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
1798 if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
1801 inc_irq_stat(irq_spurious_count);
1803 /* see sw-dev-man vol 3, chapter 7.4.13.5 */
1804 pr_info("spurious APIC interrupt on CPU#%d, "
1805 "should never happen.\n", smp_processor_id());
1810 * This interrupt should never happen with our APIC/SMP architecture
1812 void smp_error_interrupt(struct pt_regs *regs)
1818 /* First tickle the hardware, only then report what went on. -- REW */
1819 v = apic_read(APIC_ESR);
1820 apic_write(APIC_ESR, 0);
1821 v1 = apic_read(APIC_ESR);
1823 atomic_inc(&irq_err_count);
1826 * Here is what the APIC error bits mean:
1828 * 1: Receive CS error
1829 * 2: Send accept error
1830 * 3: Receive accept error
1832 * 5: Send illegal vector
1833 * 6: Received illegal vector
1834 * 7: Illegal register address
1836 pr_debug("APIC error on CPU%d: %02x(%02x)\n",
1837 smp_processor_id(), v , v1);
1842 * connect_bsp_APIC - attach the APIC to the interrupt system
1844 void __init connect_bsp_APIC(void)
1846 #ifdef CONFIG_X86_32
1849 * Do not trust the local APIC being empty at bootup.
1853 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
1854 * local APIC to INT and NMI lines.
1856 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
1857 "enabling APIC mode.\n");
1861 if (apic->enable_apic_mode)
1862 apic->enable_apic_mode();
1866 * disconnect_bsp_APIC - detach the APIC from the interrupt system
1867 * @virt_wire_setup: indicates, whether virtual wire mode is selected
1869 * Virtual wire mode is necessary to deliver legacy interrupts even when the
1872 void disconnect_bsp_APIC(int virt_wire_setup)
1876 #ifdef CONFIG_X86_32
1879 * Put the board back into PIC mode (has an effect only on
1880 * certain older boards). Note that APIC interrupts, including
1881 * IPIs, won't work beyond this point! The only exception are
1884 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
1885 "entering PIC mode.\n");
1891 /* Go back to Virtual Wire compatibility mode */
1893 /* For the spurious interrupt use vector F, and enable it */
1894 value = apic_read(APIC_SPIV);
1895 value &= ~APIC_VECTOR_MASK;
1896 value |= APIC_SPIV_APIC_ENABLED;
1898 apic_write(APIC_SPIV, value);
1900 if (!virt_wire_setup) {
1902 * For LVT0 make it edge triggered, active high,
1903 * external and enabled
1905 value = apic_read(APIC_LVT0);
1906 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1907 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1908 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1909 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1910 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
1911 apic_write(APIC_LVT0, value);
1914 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1918 * For LVT1 make it edge triggered, active high,
1921 value = apic_read(APIC_LVT1);
1922 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1923 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1924 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1925 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1926 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
1927 apic_write(APIC_LVT1, value);
1930 void __cpuinit generic_processor_info(int apicid, int version)
1937 if (version == 0x0) {
1938 pr_warning("BIOS bug, APIC version is 0 for CPU#%d! "
1939 "fixing up to 0x10. (tell your hw vendor)\n",
1943 apic_version[apicid] = version;
1945 if (num_processors >= nr_cpu_ids) {
1946 int max = nr_cpu_ids;
1947 int thiscpu = max + disabled_cpus;
1950 "ACPI: NR_CPUS/possible_cpus limit of %i reached."
1951 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
1958 cpu = cpumask_next_zero(-1, cpu_present_mask);
1960 if (version != apic_version[boot_cpu_physical_apicid])
1962 "ACPI: apic version mismatch, bootcpu: %x cpu %d: %x\n",
1963 apic_version[boot_cpu_physical_apicid], cpu, version);
1965 physid_set(apicid, phys_cpu_present_map);
1966 if (apicid == boot_cpu_physical_apicid) {
1968 * x86_bios_cpu_apicid is required to have processors listed
1969 * in same order as logical cpu numbers. Hence the first
1970 * entry is BSP, and so on.
1974 if (apicid > max_physical_apicid)
1975 max_physical_apicid = apicid;
1977 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
1978 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
1979 early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
1982 set_cpu_possible(cpu, true);
1983 set_cpu_present(cpu, true);
1986 int hard_smp_processor_id(void)
1988 return read_apic_id();
1991 void default_init_apic_ldr(void)
1995 apic_write(APIC_DFR, APIC_DFR_VALUE);
1996 val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
1997 val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
1998 apic_write(APIC_LDR, val);
2001 #ifdef CONFIG_X86_32
2002 int default_apicid_to_node(int logical_apicid)
2005 return apicid_2_node[hard_smp_processor_id()];
2019 * 'active' is true if the local APIC was enabled by us and
2020 * not the BIOS; this signifies that we are also responsible
2021 * for disabling it before entering apm/acpi suspend
2024 /* r/w apic fields */
2025 unsigned int apic_id;
2026 unsigned int apic_taskpri;
2027 unsigned int apic_ldr;
2028 unsigned int apic_dfr;
2029 unsigned int apic_spiv;
2030 unsigned int apic_lvtt;
2031 unsigned int apic_lvtpc;
2032 unsigned int apic_lvt0;
2033 unsigned int apic_lvt1;
2034 unsigned int apic_lvterr;
2035 unsigned int apic_tmict;
2036 unsigned int apic_tdcr;
2037 unsigned int apic_thmr;
2040 static int lapic_suspend(struct sys_device *dev, pm_message_t state)
2042 unsigned long flags;
2045 if (!apic_pm_state.active)
2048 maxlvt = lapic_get_maxlvt();
2050 apic_pm_state.apic_id = apic_read(APIC_ID);
2051 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2052 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2053 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2054 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2055 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2057 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2058 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2059 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2060 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2061 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2062 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2063 #ifdef CONFIG_X86_THERMAL_VECTOR
2065 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2068 local_irq_save(flags);
2069 disable_local_APIC();
2071 if (intr_remapping_enabled)
2072 disable_intr_remapping();
2074 local_irq_restore(flags);
2078 static int lapic_resume(struct sys_device *dev)
2081 unsigned long flags;
2084 struct IO_APIC_route_entry **ioapic_entries = NULL;
2086 if (!apic_pm_state.active)
2089 local_irq_save(flags);
2090 if (intr_remapping_enabled) {
2091 ioapic_entries = alloc_ioapic_entries();
2092 if (!ioapic_entries) {
2093 WARN(1, "Alloc ioapic_entries in lapic resume failed.");
2098 ret = save_IO_APIC_setup(ioapic_entries);
2100 WARN(1, "Saving IO-APIC state failed: %d\n", ret);
2101 free_ioapic_entries(ioapic_entries);
2105 mask_IO_APIC_setup(ioapic_entries);
2106 legacy_pic->mask_all();
2113 * Make sure the APICBASE points to the right address
2115 * FIXME! This will be wrong if we ever support suspend on
2116 * SMP! We'll need to do this as part of the CPU restore!
2118 rdmsr(MSR_IA32_APICBASE, l, h);
2119 l &= ~MSR_IA32_APICBASE_BASE;
2120 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2121 wrmsr(MSR_IA32_APICBASE, l, h);
2124 maxlvt = lapic_get_maxlvt();
2125 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2126 apic_write(APIC_ID, apic_pm_state.apic_id);
2127 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2128 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2129 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2130 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2131 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2132 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2133 #if defined(CONFIG_X86_MCE_P4THERMAL) || defined(CONFIG_X86_MCE_INTEL)
2135 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2138 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2139 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2140 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2141 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2142 apic_write(APIC_ESR, 0);
2143 apic_read(APIC_ESR);
2144 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2145 apic_write(APIC_ESR, 0);
2146 apic_read(APIC_ESR);
2148 if (intr_remapping_enabled) {
2149 reenable_intr_remapping(x2apic_mode);
2150 legacy_pic->restore_mask();
2151 restore_IO_APIC_setup(ioapic_entries);
2152 free_ioapic_entries(ioapic_entries);
2155 local_irq_restore(flags);
2161 * This device has no shutdown method - fully functioning local APICs
2162 * are needed on every CPU up until machine_halt/restart/poweroff.
2165 static struct sysdev_class lapic_sysclass = {
2167 .resume = lapic_resume,
2168 .suspend = lapic_suspend,
2171 static struct sys_device device_lapic = {
2173 .cls = &lapic_sysclass,
2176 static void __cpuinit apic_pm_activate(void)
2178 apic_pm_state.active = 1;
2181 static int __init init_lapic_sysfs(void)
2187 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2189 error = sysdev_class_register(&lapic_sysclass);
2191 error = sysdev_register(&device_lapic);
2195 /* local apic needs to resume before other devices access its registers. */
2196 core_initcall(init_lapic_sysfs);
2198 #else /* CONFIG_PM */
2200 static void apic_pm_activate(void) { }
2202 #endif /* CONFIG_PM */
2204 #ifdef CONFIG_X86_64
2206 static int __cpuinit apic_cluster_num(void)
2208 int i, clusters, zeros;
2210 u16 *bios_cpu_apicid;
2211 DECLARE_BITMAP(clustermap, NUM_APIC_CLUSTERS);
2213 bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
2214 bitmap_zero(clustermap, NUM_APIC_CLUSTERS);
2216 for (i = 0; i < nr_cpu_ids; i++) {
2217 /* are we being called early in kernel startup? */
2218 if (bios_cpu_apicid) {
2219 id = bios_cpu_apicid[i];
2220 } else if (i < nr_cpu_ids) {
2222 id = per_cpu(x86_bios_cpu_apicid, i);
2228 if (id != BAD_APICID)
2229 __set_bit(APIC_CLUSTERID(id), clustermap);
2232 /* Problem: Partially populated chassis may not have CPUs in some of
2233 * the APIC clusters they have been allocated. Only present CPUs have
2234 * x86_bios_cpu_apicid entries, thus causing zeroes in the bitmap.
2235 * Since clusters are allocated sequentially, count zeros only if
2236 * they are bounded by ones.
2240 for (i = 0; i < NUM_APIC_CLUSTERS; i++) {
2241 if (test_bit(i, clustermap)) {
2242 clusters += 1 + zeros;
2251 static int __cpuinitdata multi_checked;
2252 static int __cpuinitdata multi;
2254 static int __cpuinit set_multi(const struct dmi_system_id *d)
2258 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2263 static const __cpuinitconst struct dmi_system_id multi_dmi_table[] = {
2265 .callback = set_multi,
2266 .ident = "IBM System Summit2",
2268 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2269 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2275 static void __cpuinit dmi_check_multi(void)
2280 dmi_check_system(multi_dmi_table);
2285 * apic_is_clustered_box() -- Check if we can expect good TSC
2287 * Thus far, the major user of this is IBM's Summit2 series:
2288 * Clustered boxes may have unsynced TSC problems if they are
2290 * Use DMI to check them
2292 __cpuinit int apic_is_clustered_box(void)
2302 * ScaleMP vSMPowered boxes have one cluster per board and TSCs are
2303 * not guaranteed to be synced between boards
2305 if (apic_cluster_num() > 1)
2313 * APIC command line parameters
2315 static int __init setup_disableapic(char *arg)
2318 setup_clear_cpu_cap(X86_FEATURE_APIC);
2321 early_param("disableapic", setup_disableapic);
2323 /* same as disableapic, for compatibility */
2324 static int __init setup_nolapic(char *arg)
2326 return setup_disableapic(arg);
2328 early_param("nolapic", setup_nolapic);
2330 static int __init parse_lapic_timer_c2_ok(char *arg)
2332 local_apic_timer_c2_ok = 1;
2335 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2337 static int __init parse_disable_apic_timer(char *arg)
2339 disable_apic_timer = 1;
2342 early_param("noapictimer", parse_disable_apic_timer);
2344 static int __init parse_nolapic_timer(char *arg)
2346 disable_apic_timer = 1;
2349 early_param("nolapic_timer", parse_nolapic_timer);
2351 static int __init apic_set_verbosity(char *arg)
2354 #ifdef CONFIG_X86_64
2355 skip_ioapic_setup = 0;
2361 if (strcmp("debug", arg) == 0)
2362 apic_verbosity = APIC_DEBUG;
2363 else if (strcmp("verbose", arg) == 0)
2364 apic_verbosity = APIC_VERBOSE;
2366 pr_warning("APIC Verbosity level %s not recognised"
2367 " use apic=verbose or apic=debug\n", arg);
2373 early_param("apic", apic_set_verbosity);
2375 static int __init lapic_insert_resource(void)
2380 /* Put local APIC into the resource map. */
2381 lapic_resource.start = apic_phys;
2382 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2383 insert_resource(&iomem_resource, &lapic_resource);
2389 * need call insert after e820_reserve_resources()
2390 * that is using request_resource
2392 late_initcall(lapic_insert_resource);