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 unsigned int num_processors;
60 unsigned disabled_cpus __cpuinitdata;
62 /* Processor that is doing the boot up */
63 unsigned int boot_cpu_physical_apicid = -1U;
66 * The highest APIC ID seen during enumeration.
68 unsigned int max_physical_apicid;
71 * Bitmask of physically existing CPUs:
73 physid_mask_t phys_cpu_present_map;
76 * Map cpu index to physical APIC ID
78 DEFINE_EARLY_PER_CPU(u16, x86_cpu_to_apicid, BAD_APICID);
79 DEFINE_EARLY_PER_CPU(u16, x86_bios_cpu_apicid, BAD_APICID);
80 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
81 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
86 * On x86_32, the mapping between cpu and logical apicid may vary
87 * depending on apic in use. The following early percpu variable is
88 * used for the mapping. This is where the behaviors of x86_64 and 32
89 * actually diverge. Let's keep it ugly for now.
91 DEFINE_EARLY_PER_CPU(int, x86_cpu_to_logical_apicid, BAD_APICID);
94 * Knob to control our willingness to enable the local APIC.
98 static int force_enable_local_apic __initdata;
100 * APIC command line parameters
102 static int __init parse_lapic(char *arg)
104 force_enable_local_apic = 1;
107 early_param("lapic", parse_lapic);
108 /* Local APIC was disabled by the BIOS and enabled by the kernel */
109 static int enabled_via_apicbase;
112 * Handle interrupt mode configuration register (IMCR).
113 * This register controls whether the interrupt signals
114 * that reach the BSP come from the master PIC or from the
115 * local APIC. Before entering Symmetric I/O Mode, either
116 * the BIOS or the operating system must switch out of
117 * PIC Mode by changing the IMCR.
119 static inline void imcr_pic_to_apic(void)
121 /* select IMCR register */
123 /* NMI and 8259 INTR go through APIC */
127 static inline void imcr_apic_to_pic(void)
129 /* select IMCR register */
131 /* NMI and 8259 INTR go directly to BSP */
137 static int apic_calibrate_pmtmr __initdata;
138 static __init int setup_apicpmtimer(char *s)
140 apic_calibrate_pmtmr = 1;
144 __setup("apicpmtimer", setup_apicpmtimer);
148 #ifdef CONFIG_X86_X2APIC
149 /* x2apic enabled before OS handover */
150 int x2apic_preenabled;
151 static int x2apic_disabled;
153 static __init int setup_nox2apic(char *str)
155 if (x2apic_enabled()) {
156 int apicid = native_apic_msr_read(APIC_ID);
159 pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
164 pr_warning("x2apic already enabled. will disable it\n");
166 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
172 early_param("nox2apic", setup_nox2apic);
175 unsigned long mp_lapic_addr;
177 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
178 static int disable_apic_timer __initdata;
179 /* Local APIC timer works in C2 */
180 int local_apic_timer_c2_ok;
181 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
183 int first_system_vector = 0xfe;
186 * Debug level, exported for io_apic.c
188 unsigned int apic_verbosity;
192 /* Have we found an MP table */
193 int smp_found_config;
195 static struct resource lapic_resource = {
196 .name = "Local APIC",
197 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
200 unsigned int lapic_timer_frequency = 0;
202 static void apic_pm_activate(void);
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 static void __init 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;
264 inc_irq_stat(icr_read_retry_count);
266 } while (timeout++ < 1000);
271 void native_apic_icr_write(u32 low, u32 id)
273 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
274 apic_write(APIC_ICR, low);
277 u64 native_apic_icr_read(void)
281 icr2 = apic_read(APIC_ICR2);
282 icr1 = apic_read(APIC_ICR);
284 return icr1 | ((u64)icr2 << 32);
289 * get_physical_broadcast - Get number of physical broadcast IDs
291 int get_physical_broadcast(void)
293 return modern_apic() ? 0xff : 0xf;
298 * lapic_get_maxlvt - get the maximum number of local vector table entries
300 int lapic_get_maxlvt(void)
304 v = apic_read(APIC_LVR);
306 * - we always have APIC integrated on 64bit mode
307 * - 82489DXs do not report # of LVT entries
309 return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
317 #define APIC_DIVISOR 16
320 * This function sets up the local APIC timer, with a timeout of
321 * 'clocks' APIC bus clock. During calibration we actually call
322 * this function twice on the boot CPU, once with a bogus timeout
323 * value, second time for real. The other (noncalibrating) CPUs
324 * call this function only once, with the real, calibrated value.
326 * We do reads before writes even if unnecessary, to get around the
327 * P5 APIC double write bug.
329 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
331 unsigned int lvtt_value, tmp_value;
333 lvtt_value = LOCAL_TIMER_VECTOR;
335 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
336 if (!lapic_is_integrated())
337 lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
340 lvtt_value |= APIC_LVT_MASKED;
342 apic_write(APIC_LVTT, lvtt_value);
347 tmp_value = apic_read(APIC_TDCR);
348 apic_write(APIC_TDCR,
349 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
353 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
357 * Setup extended LVT, AMD specific
359 * Software should use the LVT offsets the BIOS provides. The offsets
360 * are determined by the subsystems using it like those for MCE
361 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts
362 * are supported. Beginning with family 10h at least 4 offsets are
365 * Since the offsets must be consistent for all cores, we keep track
366 * of the LVT offsets in software and reserve the offset for the same
367 * vector also to be used on other cores. An offset is freed by
368 * setting the entry to APIC_EILVT_MASKED.
370 * If the BIOS is right, there should be no conflicts. Otherwise a
371 * "[Firmware Bug]: ..." error message is generated. However, if
372 * software does not properly determines the offsets, it is not
373 * necessarily a BIOS bug.
376 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
378 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
380 return (old & APIC_EILVT_MASKED)
381 || (new == APIC_EILVT_MASKED)
382 || ((new & ~APIC_EILVT_MASKED) == old);
385 static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
387 unsigned int rsvd, vector;
389 if (offset >= APIC_EILVT_NR_MAX)
392 rsvd = atomic_read(&eilvt_offsets[offset]);
394 vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */
395 if (vector && !eilvt_entry_is_changeable(vector, new))
396 /* may not change if vectors are different */
398 rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
399 } while (rsvd != new);
401 rsvd &= ~APIC_EILVT_MASKED;
402 if (rsvd && rsvd != vector)
403 pr_info("LVT offset %d assigned for vector 0x%02x\n",
410 * If mask=1, the LVT entry does not generate interrupts while mask=0
411 * enables the vector. See also the BKDGs. Must be called with
412 * preemption disabled.
415 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
417 unsigned long reg = APIC_EILVTn(offset);
418 unsigned int new, old, reserved;
420 new = (mask << 16) | (msg_type << 8) | vector;
421 old = apic_read(reg);
422 reserved = reserve_eilvt_offset(offset, new);
424 if (reserved != new) {
425 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
426 "vector 0x%x, but the register is already in use for "
427 "vector 0x%x on another cpu\n",
428 smp_processor_id(), reg, offset, new, reserved);
432 if (!eilvt_entry_is_changeable(old, new)) {
433 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
434 "vector 0x%x, but the register is already in use for "
435 "vector 0x%x on this cpu\n",
436 smp_processor_id(), reg, offset, new, old);
440 apic_write(reg, new);
444 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
447 * Program the next event, relative to now
449 static int lapic_next_event(unsigned long delta,
450 struct clock_event_device *evt)
452 apic_write(APIC_TMICT, delta);
457 * Setup the lapic timer in periodic or oneshot mode
459 static void lapic_timer_setup(enum clock_event_mode mode,
460 struct clock_event_device *evt)
465 /* Lapic used as dummy for broadcast ? */
466 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
469 local_irq_save(flags);
472 case CLOCK_EVT_MODE_PERIODIC:
473 case CLOCK_EVT_MODE_ONESHOT:
474 __setup_APIC_LVTT(lapic_timer_frequency,
475 mode != CLOCK_EVT_MODE_PERIODIC, 1);
477 case CLOCK_EVT_MODE_UNUSED:
478 case CLOCK_EVT_MODE_SHUTDOWN:
479 v = apic_read(APIC_LVTT);
480 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
481 apic_write(APIC_LVTT, v);
482 apic_write(APIC_TMICT, 0);
484 case CLOCK_EVT_MODE_RESUME:
485 /* Nothing to do here */
489 local_irq_restore(flags);
493 * Local APIC timer broadcast function
495 static void lapic_timer_broadcast(const struct cpumask *mask)
498 apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
504 * The local apic timer can be used for any function which is CPU local.
506 static struct clock_event_device lapic_clockevent = {
508 .features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
509 | CLOCK_EVT_FEAT_C3STOP | CLOCK_EVT_FEAT_DUMMY,
511 .set_mode = lapic_timer_setup,
512 .set_next_event = lapic_next_event,
513 .broadcast = lapic_timer_broadcast,
517 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
520 * Setup the local APIC timer for this CPU. Copy the initialized values
521 * of the boot CPU and register the clock event in the framework.
523 static void __cpuinit setup_APIC_timer(void)
525 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
527 if (this_cpu_has(X86_FEATURE_ARAT)) {
528 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
529 /* Make LAPIC timer preferrable over percpu HPET */
530 lapic_clockevent.rating = 150;
533 memcpy(levt, &lapic_clockevent, sizeof(*levt));
534 levt->cpumask = cpumask_of(smp_processor_id());
536 clockevents_register_device(levt);
540 * In this functions we calibrate APIC bus clocks to the external timer.
542 * We want to do the calibration only once since we want to have local timer
543 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
546 * This was previously done by reading the PIT/HPET and waiting for a wrap
547 * around to find out, that a tick has elapsed. I have a box, where the PIT
548 * readout is broken, so it never gets out of the wait loop again. This was
549 * also reported by others.
551 * Monitoring the jiffies value is inaccurate and the clockevents
552 * infrastructure allows us to do a simple substitution of the interrupt
555 * The calibration routine also uses the pm_timer when possible, as the PIT
556 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
557 * back to normal later in the boot process).
560 #define LAPIC_CAL_LOOPS (HZ/10)
562 static __initdata int lapic_cal_loops = -1;
563 static __initdata long lapic_cal_t1, lapic_cal_t2;
564 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
565 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
566 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
569 * Temporary interrupt handler.
571 static void __init lapic_cal_handler(struct clock_event_device *dev)
573 unsigned long long tsc = 0;
574 long tapic = apic_read(APIC_TMCCT);
575 unsigned long pm = acpi_pm_read_early();
580 switch (lapic_cal_loops++) {
582 lapic_cal_t1 = tapic;
583 lapic_cal_tsc1 = tsc;
585 lapic_cal_j1 = jiffies;
588 case LAPIC_CAL_LOOPS:
589 lapic_cal_t2 = tapic;
590 lapic_cal_tsc2 = tsc;
591 if (pm < lapic_cal_pm1)
592 pm += ACPI_PM_OVRRUN;
594 lapic_cal_j2 = jiffies;
600 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
602 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
603 const long pm_thresh = pm_100ms / 100;
607 #ifndef CONFIG_X86_PM_TIMER
611 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
613 /* Check, if the PM timer is available */
617 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
619 if (deltapm > (pm_100ms - pm_thresh) &&
620 deltapm < (pm_100ms + pm_thresh)) {
621 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
625 res = (((u64)deltapm) * mult) >> 22;
626 do_div(res, 1000000);
627 pr_warning("APIC calibration not consistent "
628 "with PM-Timer: %ldms instead of 100ms\n",(long)res);
630 /* Correct the lapic counter value */
631 res = (((u64)(*delta)) * pm_100ms);
632 do_div(res, deltapm);
633 pr_info("APIC delta adjusted to PM-Timer: "
634 "%lu (%ld)\n", (unsigned long)res, *delta);
637 /* Correct the tsc counter value */
639 res = (((u64)(*deltatsc)) * pm_100ms);
640 do_div(res, deltapm);
641 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
642 "PM-Timer: %lu (%ld)\n",
643 (unsigned long)res, *deltatsc);
644 *deltatsc = (long)res;
650 static int __init calibrate_APIC_clock(void)
652 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
653 void (*real_handler)(struct clock_event_device *dev);
654 unsigned long deltaj;
655 long delta, deltatsc;
656 int pm_referenced = 0;
659 * check if lapic timer has already been calibrated by platform
660 * specific routine, such as tsc calibration code. if so, we just fill
661 * in the clockevent structure and return.
664 if (lapic_timer_frequency) {
665 apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
666 lapic_timer_frequency);
667 lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
668 TICK_NSEC, lapic_clockevent.shift);
669 lapic_clockevent.max_delta_ns =
670 clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
671 lapic_clockevent.min_delta_ns =
672 clockevent_delta2ns(0xF, &lapic_clockevent);
673 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
679 /* Replace the global interrupt handler */
680 real_handler = global_clock_event->event_handler;
681 global_clock_event->event_handler = lapic_cal_handler;
684 * Setup the APIC counter to maximum. There is no way the lapic
685 * can underflow in the 100ms detection time frame
687 __setup_APIC_LVTT(0xffffffff, 0, 0);
689 /* Let the interrupts run */
692 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
697 /* Restore the real event handler */
698 global_clock_event->event_handler = real_handler;
700 /* Build delta t1-t2 as apic timer counts down */
701 delta = lapic_cal_t1 - lapic_cal_t2;
702 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
704 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
706 /* we trust the PM based calibration if possible */
707 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
710 /* Calculate the scaled math multiplication factor */
711 lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
712 lapic_clockevent.shift);
713 lapic_clockevent.max_delta_ns =
714 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
715 lapic_clockevent.min_delta_ns =
716 clockevent_delta2ns(0xF, &lapic_clockevent);
718 lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
720 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
721 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
722 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
723 lapic_timer_frequency);
726 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
728 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
729 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
732 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
734 lapic_timer_frequency / (1000000 / HZ),
735 lapic_timer_frequency % (1000000 / HZ));
738 * Do a sanity check on the APIC calibration result
740 if (lapic_timer_frequency < (1000000 / HZ)) {
742 pr_warning("APIC frequency too slow, disabling apic timer\n");
746 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
749 * PM timer calibration failed or not turned on
750 * so lets try APIC timer based calibration
752 if (!pm_referenced) {
753 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
756 * Setup the apic timer manually
758 levt->event_handler = lapic_cal_handler;
759 lapic_timer_setup(CLOCK_EVT_MODE_PERIODIC, levt);
760 lapic_cal_loops = -1;
762 /* Let the interrupts run */
765 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
768 /* Stop the lapic timer */
769 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt);
772 deltaj = lapic_cal_j2 - lapic_cal_j1;
773 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
775 /* Check, if the jiffies result is consistent */
776 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
777 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
779 levt->features |= CLOCK_EVT_FEAT_DUMMY;
783 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
784 pr_warning("APIC timer disabled due to verification failure\n");
792 * Setup the boot APIC
794 * Calibrate and verify the result.
796 void __init setup_boot_APIC_clock(void)
799 * The local apic timer can be disabled via the kernel
800 * commandline or from the CPU detection code. Register the lapic
801 * timer as a dummy clock event source on SMP systems, so the
802 * broadcast mechanism is used. On UP systems simply ignore it.
804 if (disable_apic_timer) {
805 pr_info("Disabling APIC timer\n");
806 /* No broadcast on UP ! */
807 if (num_possible_cpus() > 1) {
808 lapic_clockevent.mult = 1;
814 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
815 "calibrating APIC timer ...\n");
817 if (calibrate_APIC_clock()) {
818 /* No broadcast on UP ! */
819 if (num_possible_cpus() > 1)
825 * If nmi_watchdog is set to IO_APIC, we need the
826 * PIT/HPET going. Otherwise register lapic as a dummy
829 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
831 /* Setup the lapic or request the broadcast */
835 void __cpuinit setup_secondary_APIC_clock(void)
841 * The guts of the apic timer interrupt
843 static void local_apic_timer_interrupt(void)
845 int cpu = smp_processor_id();
846 struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
849 * Normally we should not be here till LAPIC has been initialized but
850 * in some cases like kdump, its possible that there is a pending LAPIC
851 * timer interrupt from previous kernel's context and is delivered in
852 * new kernel the moment interrupts are enabled.
854 * Interrupts are enabled early and LAPIC is setup much later, hence
855 * its possible that when we get here evt->event_handler is NULL.
856 * Check for event_handler being NULL and discard the interrupt as
859 if (!evt->event_handler) {
860 pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
862 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
867 * the NMI deadlock-detector uses this.
869 inc_irq_stat(apic_timer_irqs);
871 evt->event_handler(evt);
875 * Local APIC timer interrupt. This is the most natural way for doing
876 * local interrupts, but local timer interrupts can be emulated by
877 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
879 * [ if a single-CPU system runs an SMP kernel then we call the local
880 * interrupt as well. Thus we cannot inline the local irq ... ]
882 void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
884 struct pt_regs *old_regs = set_irq_regs(regs);
887 * NOTE! We'd better ACK the irq immediately,
888 * because timer handling can be slow.
892 * update_process_times() expects us to have done irq_enter().
893 * Besides, if we don't timer interrupts ignore the global
894 * interrupt lock, which is the WrongThing (tm) to do.
898 local_apic_timer_interrupt();
901 set_irq_regs(old_regs);
904 int setup_profiling_timer(unsigned int multiplier)
910 * Local APIC start and shutdown
914 * clear_local_APIC - shutdown the local APIC
916 * This is called, when a CPU is disabled and before rebooting, so the state of
917 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
918 * leftovers during boot.
920 void clear_local_APIC(void)
925 /* APIC hasn't been mapped yet */
926 if (!x2apic_mode && !apic_phys)
929 maxlvt = lapic_get_maxlvt();
931 * Masking an LVT entry can trigger a local APIC error
932 * if the vector is zero. Mask LVTERR first to prevent this.
935 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
936 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
939 * Careful: we have to set masks only first to deassert
940 * any level-triggered sources.
942 v = apic_read(APIC_LVTT);
943 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
944 v = apic_read(APIC_LVT0);
945 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
946 v = apic_read(APIC_LVT1);
947 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
949 v = apic_read(APIC_LVTPC);
950 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
953 /* lets not touch this if we didn't frob it */
954 #ifdef CONFIG_X86_THERMAL_VECTOR
956 v = apic_read(APIC_LVTTHMR);
957 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
960 #ifdef CONFIG_X86_MCE_INTEL
962 v = apic_read(APIC_LVTCMCI);
963 if (!(v & APIC_LVT_MASKED))
964 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
969 * Clean APIC state for other OSs:
971 apic_write(APIC_LVTT, APIC_LVT_MASKED);
972 apic_write(APIC_LVT0, APIC_LVT_MASKED);
973 apic_write(APIC_LVT1, APIC_LVT_MASKED);
975 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
977 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
979 /* Integrated APIC (!82489DX) ? */
980 if (lapic_is_integrated()) {
982 /* Clear ESR due to Pentium errata 3AP and 11AP */
983 apic_write(APIC_ESR, 0);
989 * disable_local_APIC - clear and disable the local APIC
991 void disable_local_APIC(void)
995 /* APIC hasn't been mapped yet */
996 if (!x2apic_mode && !apic_phys)
1002 * Disable APIC (implies clearing of registers
1005 value = apic_read(APIC_SPIV);
1006 value &= ~APIC_SPIV_APIC_ENABLED;
1007 apic_write(APIC_SPIV, value);
1009 #ifdef CONFIG_X86_32
1011 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1012 * restore the disabled state.
1014 if (enabled_via_apicbase) {
1017 rdmsr(MSR_IA32_APICBASE, l, h);
1018 l &= ~MSR_IA32_APICBASE_ENABLE;
1019 wrmsr(MSR_IA32_APICBASE, l, h);
1025 * If Linux enabled the LAPIC against the BIOS default disable it down before
1026 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1027 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1028 * for the case where Linux didn't enable the LAPIC.
1030 void lapic_shutdown(void)
1032 unsigned long flags;
1034 if (!cpu_has_apic && !apic_from_smp_config())
1037 local_irq_save(flags);
1039 #ifdef CONFIG_X86_32
1040 if (!enabled_via_apicbase)
1044 disable_local_APIC();
1047 local_irq_restore(flags);
1051 * This is to verify that we're looking at a real local APIC.
1052 * Check these against your board if the CPUs aren't getting
1053 * started for no apparent reason.
1055 int __init verify_local_APIC(void)
1057 unsigned int reg0, reg1;
1060 * The version register is read-only in a real APIC.
1062 reg0 = apic_read(APIC_LVR);
1063 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
1064 apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
1065 reg1 = apic_read(APIC_LVR);
1066 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
1069 * The two version reads above should print the same
1070 * numbers. If the second one is different, then we
1071 * poke at a non-APIC.
1077 * Check if the version looks reasonably.
1079 reg1 = GET_APIC_VERSION(reg0);
1080 if (reg1 == 0x00 || reg1 == 0xff)
1082 reg1 = lapic_get_maxlvt();
1083 if (reg1 < 0x02 || reg1 == 0xff)
1087 * The ID register is read/write in a real APIC.
1089 reg0 = apic_read(APIC_ID);
1090 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
1091 apic_write(APIC_ID, reg0 ^ apic->apic_id_mask);
1092 reg1 = apic_read(APIC_ID);
1093 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
1094 apic_write(APIC_ID, reg0);
1095 if (reg1 != (reg0 ^ apic->apic_id_mask))
1099 * The next two are just to see if we have sane values.
1100 * They're only really relevant if we're in Virtual Wire
1101 * compatibility mode, but most boxes are anymore.
1103 reg0 = apic_read(APIC_LVT0);
1104 apic_printk(APIC_DEBUG, "Getting LVT0: %x\n", reg0);
1105 reg1 = apic_read(APIC_LVT1);
1106 apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
1112 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1114 void __init sync_Arb_IDs(void)
1117 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1120 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1126 apic_wait_icr_idle();
1128 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1129 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1130 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1134 * An initial setup of the virtual wire mode.
1136 void __init init_bsp_APIC(void)
1141 * Don't do the setup now if we have a SMP BIOS as the
1142 * through-I/O-APIC virtual wire mode might be active.
1144 if (smp_found_config || !cpu_has_apic)
1148 * Do not trust the local APIC being empty at bootup.
1155 value = apic_read(APIC_SPIV);
1156 value &= ~APIC_VECTOR_MASK;
1157 value |= APIC_SPIV_APIC_ENABLED;
1159 #ifdef CONFIG_X86_32
1160 /* This bit is reserved on P4/Xeon and should be cleared */
1161 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1162 (boot_cpu_data.x86 == 15))
1163 value &= ~APIC_SPIV_FOCUS_DISABLED;
1166 value |= APIC_SPIV_FOCUS_DISABLED;
1167 value |= SPURIOUS_APIC_VECTOR;
1168 apic_write(APIC_SPIV, value);
1171 * Set up the virtual wire mode.
1173 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1174 value = APIC_DM_NMI;
1175 if (!lapic_is_integrated()) /* 82489DX */
1176 value |= APIC_LVT_LEVEL_TRIGGER;
1177 apic_write(APIC_LVT1, value);
1180 static void __cpuinit lapic_setup_esr(void)
1182 unsigned int oldvalue, value, maxlvt;
1184 if (!lapic_is_integrated()) {
1185 pr_info("No ESR for 82489DX.\n");
1189 if (apic->disable_esr) {
1191 * Something untraceable is creating bad interrupts on
1192 * secondary quads ... for the moment, just leave the
1193 * ESR disabled - we can't do anything useful with the
1194 * errors anyway - mbligh
1196 pr_info("Leaving ESR disabled.\n");
1200 maxlvt = lapic_get_maxlvt();
1201 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1202 apic_write(APIC_ESR, 0);
1203 oldvalue = apic_read(APIC_ESR);
1205 /* enables sending errors */
1206 value = ERROR_APIC_VECTOR;
1207 apic_write(APIC_LVTERR, value);
1210 * spec says clear errors after enabling vector.
1213 apic_write(APIC_ESR, 0);
1214 value = apic_read(APIC_ESR);
1215 if (value != oldvalue)
1216 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1217 "vector: 0x%08x after: 0x%08x\n",
1222 * setup_local_APIC - setup the local APIC
1224 * Used to setup local APIC while initializing BSP or bringin up APs.
1225 * Always called with preemption disabled.
1227 void __cpuinit setup_local_APIC(void)
1229 int cpu = smp_processor_id();
1230 unsigned int value, queued;
1231 int i, j, acked = 0;
1232 unsigned long long tsc = 0, ntsc;
1233 long long max_loops = cpu_khz;
1239 disable_ioapic_support();
1243 #ifdef CONFIG_X86_32
1244 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1245 if (lapic_is_integrated() && apic->disable_esr) {
1246 apic_write(APIC_ESR, 0);
1247 apic_write(APIC_ESR, 0);
1248 apic_write(APIC_ESR, 0);
1249 apic_write(APIC_ESR, 0);
1252 perf_events_lapic_init();
1255 * Double-check whether this APIC is really registered.
1256 * This is meaningless in clustered apic mode, so we skip it.
1258 BUG_ON(!apic->apic_id_registered());
1261 * Intel recommends to set DFR, LDR and TPR before enabling
1262 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1263 * document number 292116). So here it goes...
1265 apic->init_apic_ldr();
1267 #ifdef CONFIG_X86_32
1269 * APIC LDR is initialized. If logical_apicid mapping was
1270 * initialized during get_smp_config(), make sure it matches the
1273 i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1274 WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1275 /* always use the value from LDR */
1276 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1277 logical_smp_processor_id();
1280 * Some NUMA implementations (NUMAQ) don't initialize apicid to
1281 * node mapping during NUMA init. Now that logical apicid is
1282 * guaranteed to be known, give it another chance. This is already
1283 * a bit too late - percpu allocation has already happened without
1284 * proper NUMA affinity.
1286 if (apic->x86_32_numa_cpu_node)
1287 set_apicid_to_node(early_per_cpu(x86_cpu_to_apicid, cpu),
1288 apic->x86_32_numa_cpu_node(cpu));
1292 * Set Task Priority to 'accept all'. We never change this
1295 value = apic_read(APIC_TASKPRI);
1296 value &= ~APIC_TPRI_MASK;
1297 apic_write(APIC_TASKPRI, value);
1300 * After a crash, we no longer service the interrupts and a pending
1301 * interrupt from previous kernel might still have ISR bit set.
1303 * Most probably by now CPU has serviced that pending interrupt and
1304 * it might not have done the ack_APIC_irq() because it thought,
1305 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1306 * does not clear the ISR bit and cpu thinks it has already serivced
1307 * the interrupt. Hence a vector might get locked. It was noticed
1308 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1312 for (i = APIC_ISR_NR - 1; i >= 0; i--)
1313 queued |= apic_read(APIC_IRR + i*0x10);
1315 for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1316 value = apic_read(APIC_ISR + i*0x10);
1317 for (j = 31; j >= 0; j--) {
1318 if (value & (1<<j)) {
1325 printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1332 max_loops = (cpu_khz << 10) - (ntsc - tsc);
1336 } while (queued && max_loops > 0);
1337 WARN_ON(max_loops <= 0);
1340 * Now that we are all set up, enable the APIC
1342 value = apic_read(APIC_SPIV);
1343 value &= ~APIC_VECTOR_MASK;
1347 value |= APIC_SPIV_APIC_ENABLED;
1349 #ifdef CONFIG_X86_32
1351 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1352 * certain networking cards. If high frequency interrupts are
1353 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1354 * entry is masked/unmasked at a high rate as well then sooner or
1355 * later IOAPIC line gets 'stuck', no more interrupts are received
1356 * from the device. If focus CPU is disabled then the hang goes
1359 * [ This bug can be reproduced easily with a level-triggered
1360 * PCI Ne2000 networking cards and PII/PIII processors, dual
1364 * Actually disabling the focus CPU check just makes the hang less
1365 * frequent as it makes the interrupt distributon model be more
1366 * like LRU than MRU (the short-term load is more even across CPUs).
1367 * See also the comment in end_level_ioapic_irq(). --macro
1371 * - enable focus processor (bit==0)
1372 * - 64bit mode always use processor focus
1373 * so no need to set it
1375 value &= ~APIC_SPIV_FOCUS_DISABLED;
1379 * Set spurious IRQ vector
1381 value |= SPURIOUS_APIC_VECTOR;
1382 apic_write(APIC_SPIV, value);
1385 * Set up LVT0, LVT1:
1387 * set up through-local-APIC on the BP's LINT0. This is not
1388 * strictly necessary in pure symmetric-IO mode, but sometimes
1389 * we delegate interrupts to the 8259A.
1392 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1394 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1395 if (!cpu && (pic_mode || !value)) {
1396 value = APIC_DM_EXTINT;
1397 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1399 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1400 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1402 apic_write(APIC_LVT0, value);
1405 * only the BP should see the LINT1 NMI signal, obviously.
1408 value = APIC_DM_NMI;
1410 value = APIC_DM_NMI | APIC_LVT_MASKED;
1411 if (!lapic_is_integrated()) /* 82489DX */
1412 value |= APIC_LVT_LEVEL_TRIGGER;
1413 apic_write(APIC_LVT1, value);
1415 #ifdef CONFIG_X86_MCE_INTEL
1416 /* Recheck CMCI information after local APIC is up on CPU #0 */
1422 void __cpuinit end_local_APIC_setup(void)
1426 #ifdef CONFIG_X86_32
1429 /* Disable the local apic timer */
1430 value = apic_read(APIC_LVTT);
1431 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1432 apic_write(APIC_LVTT, value);
1439 void __init bsp_end_local_APIC_setup(void)
1441 end_local_APIC_setup();
1444 * Now that local APIC setup is completed for BP, configure the fault
1445 * handling for interrupt remapping.
1447 if (irq_remapping_enabled)
1448 irq_remap_enable_fault_handling();
1452 #ifdef CONFIG_X86_X2APIC
1454 * Need to disable xapic and x2apic at the same time and then enable xapic mode
1456 static inline void __disable_x2apic(u64 msr)
1458 wrmsrl(MSR_IA32_APICBASE,
1459 msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1460 wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1463 static __init void disable_x2apic(void)
1467 if (!cpu_has_x2apic)
1470 rdmsrl(MSR_IA32_APICBASE, msr);
1471 if (msr & X2APIC_ENABLE) {
1472 u32 x2apic_id = read_apic_id();
1474 if (x2apic_id >= 255)
1475 panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1477 pr_info("Disabling x2apic\n");
1478 __disable_x2apic(msr);
1481 clear_cpu_cap(&cpu_data(0), X86_FEATURE_X2APIC);
1482 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1485 x2apic_disabled = 1;
1488 register_lapic_address(mp_lapic_addr);
1492 void check_x2apic(void)
1494 if (x2apic_enabled()) {
1495 pr_info("x2apic enabled by BIOS, switching to x2apic ops\n");
1496 x2apic_preenabled = x2apic_mode = 1;
1500 void enable_x2apic(void)
1504 rdmsrl(MSR_IA32_APICBASE, msr);
1505 if (x2apic_disabled) {
1506 __disable_x2apic(msr);
1513 if (!(msr & X2APIC_ENABLE)) {
1514 printk_once(KERN_INFO "Enabling x2apic\n");
1515 wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1518 #endif /* CONFIG_X86_X2APIC */
1520 int __init enable_IR(void)
1522 #ifdef CONFIG_IRQ_REMAP
1523 if (!irq_remapping_supported()) {
1524 pr_debug("intr-remapping not supported\n");
1528 if (!x2apic_preenabled && skip_ioapic_setup) {
1529 pr_info("Skipped enabling intr-remap because of skipping "
1534 return irq_remapping_enable();
1539 void __init enable_IR_x2apic(void)
1541 unsigned long flags;
1542 int ret, x2apic_enabled = 0;
1543 int hardware_init_ret;
1545 /* Make sure irq_remap_ops are initialized */
1546 setup_irq_remapping_ops();
1548 hardware_init_ret = irq_remapping_prepare();
1549 if (hardware_init_ret && !x2apic_supported())
1552 ret = save_ioapic_entries();
1554 pr_info("Saving IO-APIC state failed: %d\n", ret);
1558 local_irq_save(flags);
1559 legacy_pic->mask_all();
1560 mask_ioapic_entries();
1562 if (x2apic_preenabled && nox2apic)
1565 if (hardware_init_ret)
1570 if (!x2apic_supported())
1574 /* IR is required if there is APIC ID > 255 even when running
1577 if (max_physical_apicid > 255 ||
1578 !hypervisor_x2apic_available()) {
1579 if (x2apic_preenabled)
1584 * without IR all CPUs can be addressed by IOAPIC/MSI
1585 * only in physical mode
1587 x2apic_force_phys();
1590 if (ret == IRQ_REMAP_XAPIC_MODE) {
1591 pr_info("x2apic not enabled, IRQ remapping is in xapic mode\n");
1597 if (x2apic_supported() && !x2apic_mode) {
1600 pr_info("Enabled x2apic\n");
1604 if (ret < 0) /* IR enabling failed */
1605 restore_ioapic_entries();
1606 legacy_pic->restore_mask();
1607 local_irq_restore(flags);
1610 #ifdef CONFIG_X86_64
1612 * Detect and enable local APICs on non-SMP boards.
1613 * Original code written by Keir Fraser.
1614 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1615 * not correctly set up (usually the APIC timer won't work etc.)
1617 static int __init detect_init_APIC(void)
1619 if (!cpu_has_apic) {
1620 pr_info("No local APIC present\n");
1624 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1629 static int __init apic_verify(void)
1634 * The APIC feature bit should now be enabled
1637 features = cpuid_edx(1);
1638 if (!(features & (1 << X86_FEATURE_APIC))) {
1639 pr_warning("Could not enable APIC!\n");
1642 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1643 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1645 /* The BIOS may have set up the APIC at some other address */
1646 if (boot_cpu_data.x86 >= 6) {
1647 rdmsr(MSR_IA32_APICBASE, l, h);
1648 if (l & MSR_IA32_APICBASE_ENABLE)
1649 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1652 pr_info("Found and enabled local APIC!\n");
1656 int __init apic_force_enable(unsigned long addr)
1664 * Some BIOSes disable the local APIC in the APIC_BASE
1665 * MSR. This can only be done in software for Intel P6 or later
1666 * and AMD K7 (Model > 1) or later.
1668 if (boot_cpu_data.x86 >= 6) {
1669 rdmsr(MSR_IA32_APICBASE, l, h);
1670 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1671 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1672 l &= ~MSR_IA32_APICBASE_BASE;
1673 l |= MSR_IA32_APICBASE_ENABLE | addr;
1674 wrmsr(MSR_IA32_APICBASE, l, h);
1675 enabled_via_apicbase = 1;
1678 return apic_verify();
1682 * Detect and initialize APIC
1684 static int __init detect_init_APIC(void)
1686 /* Disabled by kernel option? */
1690 switch (boot_cpu_data.x86_vendor) {
1691 case X86_VENDOR_AMD:
1692 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1693 (boot_cpu_data.x86 >= 15))
1696 case X86_VENDOR_INTEL:
1697 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1698 (boot_cpu_data.x86 == 5 && cpu_has_apic))
1705 if (!cpu_has_apic) {
1707 * Over-ride BIOS and try to enable the local APIC only if
1708 * "lapic" specified.
1710 if (!force_enable_local_apic) {
1711 pr_info("Local APIC disabled by BIOS -- "
1712 "you can enable it with \"lapic\"\n");
1715 if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1727 pr_info("No local APIC present or hardware disabled\n");
1733 * init_apic_mappings - initialize APIC mappings
1735 void __init init_apic_mappings(void)
1737 unsigned int new_apicid;
1740 boot_cpu_physical_apicid = read_apic_id();
1744 /* If no local APIC can be found return early */
1745 if (!smp_found_config && detect_init_APIC()) {
1746 /* lets NOP'ify apic operations */
1747 pr_info("APIC: disable apic facility\n");
1750 apic_phys = mp_lapic_addr;
1753 * acpi lapic path already maps that address in
1754 * acpi_register_lapic_address()
1756 if (!acpi_lapic && !smp_found_config)
1757 register_lapic_address(apic_phys);
1761 * Fetch the APIC ID of the BSP in case we have a
1762 * default configuration (or the MP table is broken).
1764 new_apicid = read_apic_id();
1765 if (boot_cpu_physical_apicid != new_apicid) {
1766 boot_cpu_physical_apicid = new_apicid;
1768 * yeah -- we lie about apic_version
1769 * in case if apic was disabled via boot option
1770 * but it's not a problem for SMP compiled kernel
1771 * since smp_sanity_check is prepared for such a case
1772 * and disable smp mode
1774 apic_version[new_apicid] =
1775 GET_APIC_VERSION(apic_read(APIC_LVR));
1779 void __init register_lapic_address(unsigned long address)
1781 mp_lapic_addr = address;
1784 set_fixmap_nocache(FIX_APIC_BASE, address);
1785 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1786 APIC_BASE, mp_lapic_addr);
1788 if (boot_cpu_physical_apicid == -1U) {
1789 boot_cpu_physical_apicid = read_apic_id();
1790 apic_version[boot_cpu_physical_apicid] =
1791 GET_APIC_VERSION(apic_read(APIC_LVR));
1796 * This initializes the IO-APIC and APIC hardware if this is
1799 int apic_version[MAX_LOCAL_APIC];
1801 int __init APIC_init_uniprocessor(void)
1804 pr_info("Apic disabled\n");
1807 #ifdef CONFIG_X86_64
1808 if (!cpu_has_apic) {
1810 pr_info("Apic disabled by BIOS\n");
1814 if (!smp_found_config && !cpu_has_apic)
1818 * Complain if the BIOS pretends there is one.
1820 if (!cpu_has_apic &&
1821 APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
1822 pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
1823 boot_cpu_physical_apicid);
1828 default_setup_apic_routing();
1830 verify_local_APIC();
1833 #ifdef CONFIG_X86_64
1834 apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
1837 * Hack: In case of kdump, after a crash, kernel might be booting
1838 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
1839 * might be zero if read from MP tables. Get it from LAPIC.
1841 # ifdef CONFIG_CRASH_DUMP
1842 boot_cpu_physical_apicid = read_apic_id();
1845 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1848 #ifdef CONFIG_X86_IO_APIC
1850 * Now enable IO-APICs, actually call clear_IO_APIC
1851 * We need clear_IO_APIC before enabling error vector
1853 if (!skip_ioapic_setup && nr_ioapics)
1857 bsp_end_local_APIC_setup();
1859 #ifdef CONFIG_X86_IO_APIC
1860 if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
1867 x86_init.timers.setup_percpu_clockev();
1872 * Local APIC interrupts
1876 * This interrupt should _never_ happen with our APIC/SMP architecture
1878 void smp_spurious_interrupt(struct pt_regs *regs)
1885 * Check if this really is a spurious interrupt and ACK it
1886 * if it is a vectored one. Just in case...
1887 * Spurious interrupts should not be ACKed.
1889 v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
1890 if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
1893 inc_irq_stat(irq_spurious_count);
1895 /* see sw-dev-man vol 3, chapter 7.4.13.5 */
1896 pr_info("spurious APIC interrupt on CPU#%d, "
1897 "should never happen.\n", smp_processor_id());
1902 * This interrupt should never happen with our APIC/SMP architecture
1904 void smp_error_interrupt(struct pt_regs *regs)
1908 static const char * const error_interrupt_reason[] = {
1909 "Send CS error", /* APIC Error Bit 0 */
1910 "Receive CS error", /* APIC Error Bit 1 */
1911 "Send accept error", /* APIC Error Bit 2 */
1912 "Receive accept error", /* APIC Error Bit 3 */
1913 "Redirectable IPI", /* APIC Error Bit 4 */
1914 "Send illegal vector", /* APIC Error Bit 5 */
1915 "Received illegal vector", /* APIC Error Bit 6 */
1916 "Illegal register address", /* APIC Error Bit 7 */
1921 /* First tickle the hardware, only then report what went on. -- REW */
1922 v0 = apic_read(APIC_ESR);
1923 apic_write(APIC_ESR, 0);
1924 v1 = apic_read(APIC_ESR);
1926 atomic_inc(&irq_err_count);
1928 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x(%02x)",
1929 smp_processor_id(), v0 , v1);
1934 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
1939 apic_printk(APIC_DEBUG, KERN_CONT "\n");
1945 * connect_bsp_APIC - attach the APIC to the interrupt system
1947 void __init connect_bsp_APIC(void)
1949 #ifdef CONFIG_X86_32
1952 * Do not trust the local APIC being empty at bootup.
1956 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
1957 * local APIC to INT and NMI lines.
1959 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
1960 "enabling APIC mode.\n");
1964 if (apic->enable_apic_mode)
1965 apic->enable_apic_mode();
1969 * disconnect_bsp_APIC - detach the APIC from the interrupt system
1970 * @virt_wire_setup: indicates, whether virtual wire mode is selected
1972 * Virtual wire mode is necessary to deliver legacy interrupts even when the
1975 void disconnect_bsp_APIC(int virt_wire_setup)
1979 #ifdef CONFIG_X86_32
1982 * Put the board back into PIC mode (has an effect only on
1983 * certain older boards). Note that APIC interrupts, including
1984 * IPIs, won't work beyond this point! The only exception are
1987 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
1988 "entering PIC mode.\n");
1994 /* Go back to Virtual Wire compatibility mode */
1996 /* For the spurious interrupt use vector F, and enable it */
1997 value = apic_read(APIC_SPIV);
1998 value &= ~APIC_VECTOR_MASK;
1999 value |= APIC_SPIV_APIC_ENABLED;
2001 apic_write(APIC_SPIV, value);
2003 if (!virt_wire_setup) {
2005 * For LVT0 make it edge triggered, active high,
2006 * external and enabled
2008 value = apic_read(APIC_LVT0);
2009 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2010 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2011 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2012 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2013 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2014 apic_write(APIC_LVT0, value);
2017 apic_write(APIC_LVT0, APIC_LVT_MASKED);
2021 * For LVT1 make it edge triggered, active high,
2024 value = apic_read(APIC_LVT1);
2025 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2026 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2027 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2028 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2029 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2030 apic_write(APIC_LVT1, value);
2033 void __cpuinit generic_processor_info(int apicid, int version)
2035 int cpu, max = nr_cpu_ids;
2036 bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2037 phys_cpu_present_map);
2040 * If boot cpu has not been detected yet, then only allow upto
2041 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2043 if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2044 apicid != boot_cpu_physical_apicid) {
2045 int thiscpu = max + disabled_cpus - 1;
2048 "ACPI: NR_CPUS/possible_cpus limit of %i almost"
2049 " reached. Keeping one slot for boot cpu."
2050 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2056 if (num_processors >= nr_cpu_ids) {
2057 int thiscpu = max + disabled_cpus;
2060 "ACPI: NR_CPUS/possible_cpus limit of %i reached."
2061 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2068 if (apicid == boot_cpu_physical_apicid) {
2070 * x86_bios_cpu_apicid is required to have processors listed
2071 * in same order as logical cpu numbers. Hence the first
2072 * entry is BSP, and so on.
2073 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2078 cpu = cpumask_next_zero(-1, cpu_present_mask);
2083 if (version == 0x0) {
2084 pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2088 apic_version[apicid] = version;
2090 if (version != apic_version[boot_cpu_physical_apicid]) {
2091 pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2092 apic_version[boot_cpu_physical_apicid], cpu, version);
2095 physid_set(apicid, phys_cpu_present_map);
2096 if (apicid > max_physical_apicid)
2097 max_physical_apicid = apicid;
2099 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2100 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2101 early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2103 #ifdef CONFIG_X86_32
2104 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2105 apic->x86_32_early_logical_apicid(cpu);
2107 set_cpu_possible(cpu, true);
2108 set_cpu_present(cpu, true);
2111 int hard_smp_processor_id(void)
2113 return read_apic_id();
2116 void default_init_apic_ldr(void)
2120 apic_write(APIC_DFR, APIC_DFR_VALUE);
2121 val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
2122 val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
2123 apic_write(APIC_LDR, val);
2133 * 'active' is true if the local APIC was enabled by us and
2134 * not the BIOS; this signifies that we are also responsible
2135 * for disabling it before entering apm/acpi suspend
2138 /* r/w apic fields */
2139 unsigned int apic_id;
2140 unsigned int apic_taskpri;
2141 unsigned int apic_ldr;
2142 unsigned int apic_dfr;
2143 unsigned int apic_spiv;
2144 unsigned int apic_lvtt;
2145 unsigned int apic_lvtpc;
2146 unsigned int apic_lvt0;
2147 unsigned int apic_lvt1;
2148 unsigned int apic_lvterr;
2149 unsigned int apic_tmict;
2150 unsigned int apic_tdcr;
2151 unsigned int apic_thmr;
2154 static int lapic_suspend(void)
2156 unsigned long flags;
2159 if (!apic_pm_state.active)
2162 maxlvt = lapic_get_maxlvt();
2164 apic_pm_state.apic_id = apic_read(APIC_ID);
2165 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2166 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2167 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2168 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2169 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2171 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2172 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2173 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2174 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2175 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2176 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2177 #ifdef CONFIG_X86_THERMAL_VECTOR
2179 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2182 local_irq_save(flags);
2183 disable_local_APIC();
2185 if (irq_remapping_enabled)
2186 irq_remapping_disable();
2188 local_irq_restore(flags);
2192 static void lapic_resume(void)
2195 unsigned long flags;
2198 if (!apic_pm_state.active)
2201 local_irq_save(flags);
2202 if (irq_remapping_enabled) {
2204 * IO-APIC and PIC have their own resume routines.
2205 * We just mask them here to make sure the interrupt
2206 * subsystem is completely quiet while we enable x2apic
2207 * and interrupt-remapping.
2209 mask_ioapic_entries();
2210 legacy_pic->mask_all();
2217 * Make sure the APICBASE points to the right address
2219 * FIXME! This will be wrong if we ever support suspend on
2220 * SMP! We'll need to do this as part of the CPU restore!
2222 if (boot_cpu_data.x86 >= 6) {
2223 rdmsr(MSR_IA32_APICBASE, l, h);
2224 l &= ~MSR_IA32_APICBASE_BASE;
2225 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2226 wrmsr(MSR_IA32_APICBASE, l, h);
2230 maxlvt = lapic_get_maxlvt();
2231 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2232 apic_write(APIC_ID, apic_pm_state.apic_id);
2233 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2234 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2235 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2236 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2237 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2238 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2239 #if defined(CONFIG_X86_MCE_P4THERMAL) || defined(CONFIG_X86_MCE_INTEL)
2241 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2244 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2245 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2246 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2247 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2248 apic_write(APIC_ESR, 0);
2249 apic_read(APIC_ESR);
2250 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2251 apic_write(APIC_ESR, 0);
2252 apic_read(APIC_ESR);
2254 if (irq_remapping_enabled)
2255 irq_remapping_reenable(x2apic_mode);
2257 local_irq_restore(flags);
2261 * This device has no shutdown method - fully functioning local APICs
2262 * are needed on every CPU up until machine_halt/restart/poweroff.
2265 static struct syscore_ops lapic_syscore_ops = {
2266 .resume = lapic_resume,
2267 .suspend = lapic_suspend,
2270 static void __cpuinit apic_pm_activate(void)
2272 apic_pm_state.active = 1;
2275 static int __init init_lapic_sysfs(void)
2277 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2279 register_syscore_ops(&lapic_syscore_ops);
2284 /* local apic needs to resume before other devices access its registers. */
2285 core_initcall(init_lapic_sysfs);
2287 #else /* CONFIG_PM */
2289 static void apic_pm_activate(void) { }
2291 #endif /* CONFIG_PM */
2293 #ifdef CONFIG_X86_64
2295 static int __cpuinit apic_cluster_num(void)
2297 int i, clusters, zeros;
2299 u16 *bios_cpu_apicid;
2300 DECLARE_BITMAP(clustermap, NUM_APIC_CLUSTERS);
2302 bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
2303 bitmap_zero(clustermap, NUM_APIC_CLUSTERS);
2305 for (i = 0; i < nr_cpu_ids; i++) {
2306 /* are we being called early in kernel startup? */
2307 if (bios_cpu_apicid) {
2308 id = bios_cpu_apicid[i];
2309 } else if (i < nr_cpu_ids) {
2311 id = per_cpu(x86_bios_cpu_apicid, i);
2317 if (id != BAD_APICID)
2318 __set_bit(APIC_CLUSTERID(id), clustermap);
2321 /* Problem: Partially populated chassis may not have CPUs in some of
2322 * the APIC clusters they have been allocated. Only present CPUs have
2323 * x86_bios_cpu_apicid entries, thus causing zeroes in the bitmap.
2324 * Since clusters are allocated sequentially, count zeros only if
2325 * they are bounded by ones.
2329 for (i = 0; i < NUM_APIC_CLUSTERS; i++) {
2330 if (test_bit(i, clustermap)) {
2331 clusters += 1 + zeros;
2340 static int __cpuinitdata multi_checked;
2341 static int __cpuinitdata multi;
2343 static int __cpuinit set_multi(const struct dmi_system_id *d)
2347 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2352 static const __cpuinitconst struct dmi_system_id multi_dmi_table[] = {
2354 .callback = set_multi,
2355 .ident = "IBM System Summit2",
2357 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2358 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2364 static void __cpuinit dmi_check_multi(void)
2369 dmi_check_system(multi_dmi_table);
2374 * apic_is_clustered_box() -- Check if we can expect good TSC
2376 * Thus far, the major user of this is IBM's Summit2 series:
2377 * Clustered boxes may have unsynced TSC problems if they are
2379 * Use DMI to check them
2381 __cpuinit int apic_is_clustered_box(void)
2391 * ScaleMP vSMPowered boxes have one cluster per board and TSCs are
2392 * not guaranteed to be synced between boards
2394 if (apic_cluster_num() > 1)
2402 * APIC command line parameters
2404 static int __init setup_disableapic(char *arg)
2407 setup_clear_cpu_cap(X86_FEATURE_APIC);
2410 early_param("disableapic", setup_disableapic);
2412 /* same as disableapic, for compatibility */
2413 static int __init setup_nolapic(char *arg)
2415 return setup_disableapic(arg);
2417 early_param("nolapic", setup_nolapic);
2419 static int __init parse_lapic_timer_c2_ok(char *arg)
2421 local_apic_timer_c2_ok = 1;
2424 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2426 static int __init parse_disable_apic_timer(char *arg)
2428 disable_apic_timer = 1;
2431 early_param("noapictimer", parse_disable_apic_timer);
2433 static int __init parse_nolapic_timer(char *arg)
2435 disable_apic_timer = 1;
2438 early_param("nolapic_timer", parse_nolapic_timer);
2440 static int __init apic_set_verbosity(char *arg)
2443 #ifdef CONFIG_X86_64
2444 skip_ioapic_setup = 0;
2450 if (strcmp("debug", arg) == 0)
2451 apic_verbosity = APIC_DEBUG;
2452 else if (strcmp("verbose", arg) == 0)
2453 apic_verbosity = APIC_VERBOSE;
2455 pr_warning("APIC Verbosity level %s not recognised"
2456 " use apic=verbose or apic=debug\n", arg);
2462 early_param("apic", apic_set_verbosity);
2464 static int __init lapic_insert_resource(void)
2469 /* Put local APIC into the resource map. */
2470 lapic_resource.start = apic_phys;
2471 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2472 insert_resource(&iomem_resource, &lapic_resource);
2478 * need call insert after e820_reserve_resources()
2479 * that is using request_resource
2481 late_initcall(lapic_insert_resource);