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/memblock.h>
24 #include <linux/ftrace.h>
25 #include <linux/ioport.h>
26 #include <linux/export.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/trace/irq_vectors.h>
39 #include <asm/irq_remapping.h>
40 #include <asm/perf_event.h>
41 #include <asm/x86_init.h>
42 #include <asm/pgalloc.h>
43 #include <linux/atomic.h>
44 #include <asm/mpspec.h>
45 #include <asm/i8259.h>
46 #include <asm/proto.h>
47 #include <asm/traps.h>
49 #include <asm/io_apic.h>
57 #include <asm/hypervisor.h>
58 #include <asm/cpu_device_id.h>
59 #include <asm/intel-family.h>
60 #include <asm/irq_regs.h>
62 unsigned int num_processors;
64 unsigned disabled_cpus;
66 /* Processor that is doing the boot up */
67 unsigned int boot_cpu_physical_apicid = -1U;
68 EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
70 u8 boot_cpu_apic_version;
73 * The highest APIC ID seen during enumeration.
75 static unsigned int max_physical_apicid;
78 * Bitmask of physically existing CPUs:
80 physid_mask_t phys_cpu_present_map;
83 * Processor to be disabled specified by kernel parameter
84 * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
85 * avoid undefined behaviour caused by sending INIT from AP to BSP.
87 static unsigned int disabled_cpu_apicid __read_mostly = BAD_APICID;
90 * This variable controls which CPUs receive external NMIs. By default,
91 * external NMIs are delivered only to the BSP.
93 static int apic_extnmi = APIC_EXTNMI_BSP;
96 * Map cpu index to physical APIC ID
98 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
99 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
100 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u32, x86_cpu_to_acpiid, U32_MAX);
101 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
102 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
103 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_acpiid);
108 * On x86_32, the mapping between cpu and logical apicid may vary
109 * depending on apic in use. The following early percpu variable is
110 * used for the mapping. This is where the behaviors of x86_64 and 32
111 * actually diverge. Let's keep it ugly for now.
113 DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
115 /* Local APIC was disabled by the BIOS and enabled by the kernel */
116 static int enabled_via_apicbase;
119 * Handle interrupt mode configuration register (IMCR).
120 * This register controls whether the interrupt signals
121 * that reach the BSP come from the master PIC or from the
122 * local APIC. Before entering Symmetric I/O Mode, either
123 * the BIOS or the operating system must switch out of
124 * PIC Mode by changing the IMCR.
126 static inline void imcr_pic_to_apic(void)
128 /* select IMCR register */
130 /* NMI and 8259 INTR go through APIC */
134 static inline void imcr_apic_to_pic(void)
136 /* select IMCR register */
138 /* NMI and 8259 INTR go directly to BSP */
144 * Knob to control our willingness to enable the local APIC.
148 static int force_enable_local_apic __initdata;
151 * APIC command line parameters
153 static int __init parse_lapic(char *arg)
155 if (IS_ENABLED(CONFIG_X86_32) && !arg)
156 force_enable_local_apic = 1;
157 else if (arg && !strncmp(arg, "notscdeadline", 13))
158 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
161 early_param("lapic", parse_lapic);
164 static int apic_calibrate_pmtmr __initdata;
165 static __init int setup_apicpmtimer(char *s)
167 apic_calibrate_pmtmr = 1;
171 __setup("apicpmtimer", setup_apicpmtimer);
174 unsigned long mp_lapic_addr;
176 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
177 static int disable_apic_timer __initdata;
178 /* Local APIC timer works in C2 */
179 int local_apic_timer_c2_ok;
180 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
183 * Debug level, exported for io_apic.c
185 unsigned int apic_verbosity;
189 /* Have we found an MP table */
190 int smp_found_config;
192 static struct resource lapic_resource = {
193 .name = "Local APIC",
194 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
197 unsigned int lapic_timer_frequency = 0;
199 static void apic_pm_activate(void);
201 static unsigned long apic_phys;
204 * Get the LAPIC version
206 static inline int lapic_get_version(void)
208 return GET_APIC_VERSION(apic_read(APIC_LVR));
212 * Check, if the APIC is integrated or a separate chip
214 static inline int lapic_is_integrated(void)
216 return APIC_INTEGRATED(lapic_get_version());
220 * Check, whether this is a modern or a first generation APIC
222 static int modern_apic(void)
224 /* AMD systems use old APIC versions, so check the CPU */
225 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
226 boot_cpu_data.x86 >= 0xf)
229 /* Hygon systems use modern APIC */
230 if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
233 return lapic_get_version() >= 0x14;
237 * right after this call apic become NOOP driven
238 * so apic->write/read doesn't do anything
240 static void __init apic_disable(void)
242 pr_info("APIC: switched to apic NOOP\n");
246 void native_apic_wait_icr_idle(void)
248 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
252 u32 native_safe_apic_wait_icr_idle(void)
259 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
262 inc_irq_stat(icr_read_retry_count);
264 } while (timeout++ < 1000);
269 void native_apic_icr_write(u32 low, u32 id)
273 local_irq_save(flags);
274 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
275 apic_write(APIC_ICR, low);
276 local_irq_restore(flags);
279 u64 native_apic_icr_read(void)
283 icr2 = apic_read(APIC_ICR2);
284 icr1 = apic_read(APIC_ICR);
286 return icr1 | ((u64)icr2 << 32);
291 * get_physical_broadcast - Get number of physical broadcast IDs
293 int get_physical_broadcast(void)
295 return modern_apic() ? 0xff : 0xf;
300 * lapic_get_maxlvt - get the maximum number of local vector table entries
302 int lapic_get_maxlvt(void)
305 * - we always have APIC integrated on 64bit mode
306 * - 82489DXs do not report # of LVT entries
308 return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
316 #define APIC_DIVISOR 16
317 #define TSC_DIVISOR 8
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 else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
337 lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
339 if (!lapic_is_integrated())
340 lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
343 lvtt_value |= APIC_LVT_MASKED;
345 apic_write(APIC_LVTT, lvtt_value);
347 if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
349 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
350 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
351 * According to Intel, MFENCE can do the serialization here.
353 asm volatile("mfence" : : : "memory");
355 printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
362 tmp_value = apic_read(APIC_TDCR);
363 apic_write(APIC_TDCR,
364 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
368 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
372 * Setup extended LVT, AMD specific
374 * Software should use the LVT offsets the BIOS provides. The offsets
375 * are determined by the subsystems using it like those for MCE
376 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts
377 * are supported. Beginning with family 10h at least 4 offsets are
380 * Since the offsets must be consistent for all cores, we keep track
381 * of the LVT offsets in software and reserve the offset for the same
382 * vector also to be used on other cores. An offset is freed by
383 * setting the entry to APIC_EILVT_MASKED.
385 * If the BIOS is right, there should be no conflicts. Otherwise a
386 * "[Firmware Bug]: ..." error message is generated. However, if
387 * software does not properly determines the offsets, it is not
388 * necessarily a BIOS bug.
391 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
393 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
395 return (old & APIC_EILVT_MASKED)
396 || (new == APIC_EILVT_MASKED)
397 || ((new & ~APIC_EILVT_MASKED) == old);
400 static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
402 unsigned int rsvd, vector;
404 if (offset >= APIC_EILVT_NR_MAX)
407 rsvd = atomic_read(&eilvt_offsets[offset]);
409 vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */
410 if (vector && !eilvt_entry_is_changeable(vector, new))
411 /* may not change if vectors are different */
413 rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
414 } while (rsvd != new);
416 rsvd &= ~APIC_EILVT_MASKED;
417 if (rsvd && rsvd != vector)
418 pr_info("LVT offset %d assigned for vector 0x%02x\n",
425 * If mask=1, the LVT entry does not generate interrupts while mask=0
426 * enables the vector. See also the BKDGs. Must be called with
427 * preemption disabled.
430 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
432 unsigned long reg = APIC_EILVTn(offset);
433 unsigned int new, old, reserved;
435 new = (mask << 16) | (msg_type << 8) | vector;
436 old = apic_read(reg);
437 reserved = reserve_eilvt_offset(offset, new);
439 if (reserved != new) {
440 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
441 "vector 0x%x, but the register is already in use for "
442 "vector 0x%x on another cpu\n",
443 smp_processor_id(), reg, offset, new, reserved);
447 if (!eilvt_entry_is_changeable(old, new)) {
448 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
449 "vector 0x%x, but the register is already in use for "
450 "vector 0x%x on this cpu\n",
451 smp_processor_id(), reg, offset, new, old);
455 apic_write(reg, new);
459 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
462 * Program the next event, relative to now
464 static int lapic_next_event(unsigned long delta,
465 struct clock_event_device *evt)
467 apic_write(APIC_TMICT, delta);
471 static int lapic_next_deadline(unsigned long delta,
472 struct clock_event_device *evt)
477 wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
481 static int lapic_timer_shutdown(struct clock_event_device *evt)
485 /* Lapic used as dummy for broadcast ? */
486 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
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);
497 lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
499 /* Lapic used as dummy for broadcast ? */
500 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
503 __setup_APIC_LVTT(lapic_timer_frequency, oneshot, 1);
507 static int lapic_timer_set_periodic(struct clock_event_device *evt)
509 return lapic_timer_set_periodic_oneshot(evt, false);
512 static int lapic_timer_set_oneshot(struct clock_event_device *evt)
514 return lapic_timer_set_periodic_oneshot(evt, true);
518 * Local APIC timer broadcast function
520 static void lapic_timer_broadcast(const struct cpumask *mask)
523 apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
529 * The local apic timer can be used for any function which is CPU local.
531 static struct clock_event_device lapic_clockevent = {
533 .features = CLOCK_EVT_FEAT_PERIODIC |
534 CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
535 | CLOCK_EVT_FEAT_DUMMY,
537 .set_state_shutdown = lapic_timer_shutdown,
538 .set_state_periodic = lapic_timer_set_periodic,
539 .set_state_oneshot = lapic_timer_set_oneshot,
540 .set_state_oneshot_stopped = lapic_timer_shutdown,
541 .set_next_event = lapic_next_event,
542 .broadcast = lapic_timer_broadcast,
546 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
548 #define DEADLINE_MODEL_MATCH_FUNC(model, func) \
549 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&func }
551 #define DEADLINE_MODEL_MATCH_REV(model, rev) \
552 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)rev }
554 static u32 hsx_deadline_rev(void)
556 switch (boot_cpu_data.x86_stepping) {
557 case 0x02: return 0x3a; /* EP */
558 case 0x04: return 0x0f; /* EX */
564 static u32 bdx_deadline_rev(void)
566 switch (boot_cpu_data.x86_stepping) {
567 case 0x02: return 0x00000011;
568 case 0x03: return 0x0700000e;
569 case 0x04: return 0x0f00000c;
570 case 0x05: return 0x0e000003;
576 static u32 skx_deadline_rev(void)
578 switch (boot_cpu_data.x86_stepping) {
579 case 0x03: return 0x01000136;
580 case 0x04: return 0x02000014;
583 if (boot_cpu_data.x86_stepping > 4)
589 static const struct x86_cpu_id deadline_match[] = {
590 DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_HASWELL_X, hsx_deadline_rev),
591 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_X, 0x0b000020),
592 DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_BROADWELL_XEON_D, bdx_deadline_rev),
593 DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_SKYLAKE_X, skx_deadline_rev),
595 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_CORE, 0x22),
596 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_ULT, 0x20),
597 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_GT3E, 0x17),
599 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_CORE, 0x25),
600 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_GT3E, 0x17),
602 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_MOBILE, 0xb2),
603 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_DESKTOP, 0xb2),
605 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_MOBILE, 0x52),
606 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_DESKTOP, 0x52),
611 static void apic_check_deadline_errata(void)
613 const struct x86_cpu_id *m;
616 if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER) ||
617 boot_cpu_has(X86_FEATURE_HYPERVISOR))
620 m = x86_match_cpu(deadline_match);
625 * Function pointers will have the MSB set due to address layout,
626 * immediate revisions will not.
628 if ((long)m->driver_data < 0)
629 rev = ((u32 (*)(void))(m->driver_data))();
631 rev = (u32)m->driver_data;
633 if (boot_cpu_data.microcode >= rev)
636 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
637 pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
638 "please update microcode to version: 0x%x (or later)\n", rev);
642 * Setup the local APIC timer for this CPU. Copy the initialized values
643 * of the boot CPU and register the clock event in the framework.
645 static void setup_APIC_timer(void)
647 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
649 if (this_cpu_has(X86_FEATURE_ARAT)) {
650 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
651 /* Make LAPIC timer preferrable over percpu HPET */
652 lapic_clockevent.rating = 150;
655 memcpy(levt, &lapic_clockevent, sizeof(*levt));
656 levt->cpumask = cpumask_of(smp_processor_id());
658 if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
659 levt->name = "lapic-deadline";
660 levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
661 CLOCK_EVT_FEAT_DUMMY);
662 levt->set_next_event = lapic_next_deadline;
663 clockevents_config_and_register(levt,
664 tsc_khz * (1000 / TSC_DIVISOR),
667 clockevents_register_device(levt);
671 * Install the updated TSC frequency from recalibration at the TSC
672 * deadline clockevent devices.
674 static void __lapic_update_tsc_freq(void *info)
676 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
678 if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
681 clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
684 void lapic_update_tsc_freq(void)
687 * The clockevent device's ->mult and ->shift can both be
688 * changed. In order to avoid races, schedule the frequency
689 * update code on each CPU.
691 on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
695 * In this functions we calibrate APIC bus clocks to the external timer.
697 * We want to do the calibration only once since we want to have local timer
698 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
701 * This was previously done by reading the PIT/HPET and waiting for a wrap
702 * around to find out, that a tick has elapsed. I have a box, where the PIT
703 * readout is broken, so it never gets out of the wait loop again. This was
704 * also reported by others.
706 * Monitoring the jiffies value is inaccurate and the clockevents
707 * infrastructure allows us to do a simple substitution of the interrupt
710 * The calibration routine also uses the pm_timer when possible, as the PIT
711 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
712 * back to normal later in the boot process).
715 #define LAPIC_CAL_LOOPS (HZ/10)
717 static __initdata int lapic_cal_loops = -1;
718 static __initdata long lapic_cal_t1, lapic_cal_t2;
719 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
720 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
721 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
724 * Temporary interrupt handler.
726 static void __init lapic_cal_handler(struct clock_event_device *dev)
728 unsigned long long tsc = 0;
729 long tapic = apic_read(APIC_TMCCT);
730 unsigned long pm = acpi_pm_read_early();
732 if (boot_cpu_has(X86_FEATURE_TSC))
735 switch (lapic_cal_loops++) {
737 lapic_cal_t1 = tapic;
738 lapic_cal_tsc1 = tsc;
740 lapic_cal_j1 = jiffies;
743 case LAPIC_CAL_LOOPS:
744 lapic_cal_t2 = tapic;
745 lapic_cal_tsc2 = tsc;
746 if (pm < lapic_cal_pm1)
747 pm += ACPI_PM_OVRRUN;
749 lapic_cal_j2 = jiffies;
755 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
757 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
758 const long pm_thresh = pm_100ms / 100;
762 #ifndef CONFIG_X86_PM_TIMER
766 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
768 /* Check, if the PM timer is available */
772 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
774 if (deltapm > (pm_100ms - pm_thresh) &&
775 deltapm < (pm_100ms + pm_thresh)) {
776 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
780 res = (((u64)deltapm) * mult) >> 22;
781 do_div(res, 1000000);
782 pr_warning("APIC calibration not consistent "
783 "with PM-Timer: %ldms instead of 100ms\n",(long)res);
785 /* Correct the lapic counter value */
786 res = (((u64)(*delta)) * pm_100ms);
787 do_div(res, deltapm);
788 pr_info("APIC delta adjusted to PM-Timer: "
789 "%lu (%ld)\n", (unsigned long)res, *delta);
792 /* Correct the tsc counter value */
793 if (boot_cpu_has(X86_FEATURE_TSC)) {
794 res = (((u64)(*deltatsc)) * pm_100ms);
795 do_div(res, deltapm);
796 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
797 "PM-Timer: %lu (%ld)\n",
798 (unsigned long)res, *deltatsc);
799 *deltatsc = (long)res;
805 static int __init lapic_init_clockevent(void)
807 if (!lapic_timer_frequency)
810 /* Calculate the scaled math multiplication factor */
811 lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
812 TICK_NSEC, lapic_clockevent.shift);
813 lapic_clockevent.max_delta_ns =
814 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
815 lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
816 lapic_clockevent.min_delta_ns =
817 clockevent_delta2ns(0xF, &lapic_clockevent);
818 lapic_clockevent.min_delta_ticks = 0xF;
823 static int __init calibrate_APIC_clock(void)
825 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
826 void (*real_handler)(struct clock_event_device *dev);
827 unsigned long deltaj;
828 long delta, deltatsc;
829 int pm_referenced = 0;
831 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
835 * Check if lapic timer has already been calibrated by platform
836 * specific routine, such as tsc calibration code. If so just fill
837 * in the clockevent structure and return.
839 if (!lapic_init_clockevent()) {
840 apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
841 lapic_timer_frequency);
843 * Direct calibration methods must have an always running
844 * local APIC timer, no need for broadcast timer.
846 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
850 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
851 "calibrating APIC timer ...\n");
855 /* Replace the global interrupt handler */
856 real_handler = global_clock_event->event_handler;
857 global_clock_event->event_handler = lapic_cal_handler;
860 * Setup the APIC counter to maximum. There is no way the lapic
861 * can underflow in the 100ms detection time frame
863 __setup_APIC_LVTT(0xffffffff, 0, 0);
865 /* Let the interrupts run */
868 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
873 /* Restore the real event handler */
874 global_clock_event->event_handler = real_handler;
876 /* Build delta t1-t2 as apic timer counts down */
877 delta = lapic_cal_t1 - lapic_cal_t2;
878 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
880 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
882 /* we trust the PM based calibration if possible */
883 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
886 lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
887 lapic_init_clockevent();
889 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
890 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
891 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
892 lapic_timer_frequency);
894 if (boot_cpu_has(X86_FEATURE_TSC)) {
895 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
897 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
898 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
901 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
903 lapic_timer_frequency / (1000000 / HZ),
904 lapic_timer_frequency % (1000000 / HZ));
907 * Do a sanity check on the APIC calibration result
909 if (lapic_timer_frequency < (1000000 / HZ)) {
911 pr_warning("APIC frequency too slow, disabling apic timer\n");
915 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
918 * PM timer calibration failed or not turned on
919 * so lets try APIC timer based calibration
921 if (!pm_referenced) {
922 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
925 * Setup the apic timer manually
927 levt->event_handler = lapic_cal_handler;
928 lapic_timer_set_periodic(levt);
929 lapic_cal_loops = -1;
931 /* Let the interrupts run */
934 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
937 /* Stop the lapic timer */
939 lapic_timer_shutdown(levt);
942 deltaj = lapic_cal_j2 - lapic_cal_j1;
943 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
945 /* Check, if the jiffies result is consistent */
946 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
947 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
949 levt->features |= CLOCK_EVT_FEAT_DUMMY;
953 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
954 pr_warning("APIC timer disabled due to verification failure\n");
962 * Setup the boot APIC
964 * Calibrate and verify the result.
966 void __init setup_boot_APIC_clock(void)
969 * The local apic timer can be disabled via the kernel
970 * commandline or from the CPU detection code. Register the lapic
971 * timer as a dummy clock event source on SMP systems, so the
972 * broadcast mechanism is used. On UP systems simply ignore it.
974 if (disable_apic_timer) {
975 pr_info("Disabling APIC timer\n");
976 /* No broadcast on UP ! */
977 if (num_possible_cpus() > 1) {
978 lapic_clockevent.mult = 1;
984 if (calibrate_APIC_clock()) {
985 /* No broadcast on UP ! */
986 if (num_possible_cpus() > 1)
992 * If nmi_watchdog is set to IO_APIC, we need the
993 * PIT/HPET going. Otherwise register lapic as a dummy
996 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
998 /* Setup the lapic or request the broadcast */
1000 amd_e400_c1e_apic_setup();
1003 void setup_secondary_APIC_clock(void)
1006 amd_e400_c1e_apic_setup();
1010 * The guts of the apic timer interrupt
1012 static void local_apic_timer_interrupt(void)
1014 struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1017 * Normally we should not be here till LAPIC has been initialized but
1018 * in some cases like kdump, its possible that there is a pending LAPIC
1019 * timer interrupt from previous kernel's context and is delivered in
1020 * new kernel the moment interrupts are enabled.
1022 * Interrupts are enabled early and LAPIC is setup much later, hence
1023 * its possible that when we get here evt->event_handler is NULL.
1024 * Check for event_handler being NULL and discard the interrupt as
1027 if (!evt->event_handler) {
1028 pr_warning("Spurious LAPIC timer interrupt on cpu %d\n",
1029 smp_processor_id());
1031 lapic_timer_shutdown(evt);
1036 * the NMI deadlock-detector uses this.
1038 inc_irq_stat(apic_timer_irqs);
1040 evt->event_handler(evt);
1044 * Local APIC timer interrupt. This is the most natural way for doing
1045 * local interrupts, but local timer interrupts can be emulated by
1046 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1048 * [ if a single-CPU system runs an SMP kernel then we call the local
1049 * interrupt as well. Thus we cannot inline the local irq ... ]
1051 __visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
1053 struct pt_regs *old_regs = set_irq_regs(regs);
1056 * NOTE! We'd better ACK the irq immediately,
1057 * because timer handling can be slow.
1059 * update_process_times() expects us to have done irq_enter().
1060 * Besides, if we don't timer interrupts ignore the global
1061 * interrupt lock, which is the WrongThing (tm) to do.
1064 trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1065 local_apic_timer_interrupt();
1066 trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1069 set_irq_regs(old_regs);
1072 int setup_profiling_timer(unsigned int multiplier)
1078 * Local APIC start and shutdown
1082 * clear_local_APIC - shutdown the local APIC
1084 * This is called, when a CPU is disabled and before rebooting, so the state of
1085 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1086 * leftovers during boot.
1088 void clear_local_APIC(void)
1093 /* APIC hasn't been mapped yet */
1094 if (!x2apic_mode && !apic_phys)
1097 maxlvt = lapic_get_maxlvt();
1099 * Masking an LVT entry can trigger a local APIC error
1100 * if the vector is zero. Mask LVTERR first to prevent this.
1103 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1104 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1107 * Careful: we have to set masks only first to deassert
1108 * any level-triggered sources.
1110 v = apic_read(APIC_LVTT);
1111 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1112 v = apic_read(APIC_LVT0);
1113 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1114 v = apic_read(APIC_LVT1);
1115 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1117 v = apic_read(APIC_LVTPC);
1118 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1121 /* lets not touch this if we didn't frob it */
1122 #ifdef CONFIG_X86_THERMAL_VECTOR
1124 v = apic_read(APIC_LVTTHMR);
1125 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1128 #ifdef CONFIG_X86_MCE_INTEL
1130 v = apic_read(APIC_LVTCMCI);
1131 if (!(v & APIC_LVT_MASKED))
1132 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1137 * Clean APIC state for other OSs:
1139 apic_write(APIC_LVTT, APIC_LVT_MASKED);
1140 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1141 apic_write(APIC_LVT1, APIC_LVT_MASKED);
1143 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1145 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1147 /* Integrated APIC (!82489DX) ? */
1148 if (lapic_is_integrated()) {
1150 /* Clear ESR due to Pentium errata 3AP and 11AP */
1151 apic_write(APIC_ESR, 0);
1152 apic_read(APIC_ESR);
1157 * disable_local_APIC - clear and disable the local APIC
1159 void disable_local_APIC(void)
1163 /* APIC hasn't been mapped yet */
1164 if (!x2apic_mode && !apic_phys)
1170 * Disable APIC (implies clearing of registers
1173 value = apic_read(APIC_SPIV);
1174 value &= ~APIC_SPIV_APIC_ENABLED;
1175 apic_write(APIC_SPIV, value);
1177 #ifdef CONFIG_X86_32
1179 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1180 * restore the disabled state.
1182 if (enabled_via_apicbase) {
1185 rdmsr(MSR_IA32_APICBASE, l, h);
1186 l &= ~MSR_IA32_APICBASE_ENABLE;
1187 wrmsr(MSR_IA32_APICBASE, l, h);
1193 * If Linux enabled the LAPIC against the BIOS default disable it down before
1194 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1195 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1196 * for the case where Linux didn't enable the LAPIC.
1198 void lapic_shutdown(void)
1200 unsigned long flags;
1202 if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1205 local_irq_save(flags);
1207 #ifdef CONFIG_X86_32
1208 if (!enabled_via_apicbase)
1212 disable_local_APIC();
1215 local_irq_restore(flags);
1219 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1221 void __init sync_Arb_IDs(void)
1224 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1227 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1233 apic_wait_icr_idle();
1235 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1236 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1237 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1240 enum apic_intr_mode_id apic_intr_mode;
1242 static int __init apic_intr_mode_select(void)
1244 /* Check kernel option */
1246 pr_info("APIC disabled via kernel command line\n");
1251 #ifdef CONFIG_X86_64
1252 /* On 64-bit, the APIC must be integrated, Check local APIC only */
1253 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1255 pr_info("APIC disabled by BIOS\n");
1259 /* On 32-bit, the APIC may be integrated APIC or 82489DX */
1261 /* Neither 82489DX nor integrated APIC ? */
1262 if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1267 /* If the BIOS pretends there is an integrated APIC ? */
1268 if (!boot_cpu_has(X86_FEATURE_APIC) &&
1269 APIC_INTEGRATED(boot_cpu_apic_version)) {
1271 pr_err(FW_BUG "Local APIC %d not detected, force emulation\n",
1272 boot_cpu_physical_apicid);
1277 /* Check MP table or ACPI MADT configuration */
1278 if (!smp_found_config) {
1279 disable_ioapic_support();
1281 pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1282 return APIC_VIRTUAL_WIRE_NO_CONFIG;
1284 return APIC_VIRTUAL_WIRE;
1288 /* If SMP should be disabled, then really disable it! */
1289 if (!setup_max_cpus) {
1290 pr_info("APIC: SMP mode deactivated\n");
1291 return APIC_SYMMETRIC_IO_NO_ROUTING;
1294 if (read_apic_id() != boot_cpu_physical_apicid) {
1295 panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1296 read_apic_id(), boot_cpu_physical_apicid);
1297 /* Or can we switch back to PIC here? */
1301 return APIC_SYMMETRIC_IO;
1305 * An initial setup of the virtual wire mode.
1307 void __init init_bsp_APIC(void)
1312 * Don't do the setup now if we have a SMP BIOS as the
1313 * through-I/O-APIC virtual wire mode might be active.
1315 if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1319 * Do not trust the local APIC being empty at bootup.
1326 value = apic_read(APIC_SPIV);
1327 value &= ~APIC_VECTOR_MASK;
1328 value |= APIC_SPIV_APIC_ENABLED;
1330 #ifdef CONFIG_X86_32
1331 /* This bit is reserved on P4/Xeon and should be cleared */
1332 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1333 (boot_cpu_data.x86 == 15))
1334 value &= ~APIC_SPIV_FOCUS_DISABLED;
1337 value |= APIC_SPIV_FOCUS_DISABLED;
1338 value |= SPURIOUS_APIC_VECTOR;
1339 apic_write(APIC_SPIV, value);
1342 * Set up the virtual wire mode.
1344 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1345 value = APIC_DM_NMI;
1346 if (!lapic_is_integrated()) /* 82489DX */
1347 value |= APIC_LVT_LEVEL_TRIGGER;
1348 if (apic_extnmi == APIC_EXTNMI_NONE)
1349 value |= APIC_LVT_MASKED;
1350 apic_write(APIC_LVT1, value);
1353 /* Init the interrupt delivery mode for the BSP */
1354 void __init apic_intr_mode_init(void)
1356 bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1358 apic_intr_mode = apic_intr_mode_select();
1360 switch (apic_intr_mode) {
1362 pr_info("APIC: Keep in PIC mode(8259)\n");
1364 case APIC_VIRTUAL_WIRE:
1365 pr_info("APIC: Switch to virtual wire mode setup\n");
1366 default_setup_apic_routing();
1368 case APIC_VIRTUAL_WIRE_NO_CONFIG:
1369 pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1371 default_setup_apic_routing();
1373 case APIC_SYMMETRIC_IO:
1374 pr_info("APIC: Switch to symmetric I/O mode setup\n");
1375 default_setup_apic_routing();
1377 case APIC_SYMMETRIC_IO_NO_ROUTING:
1378 pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1382 apic_bsp_setup(upmode);
1385 static void lapic_setup_esr(void)
1387 unsigned int oldvalue, value, maxlvt;
1389 if (!lapic_is_integrated()) {
1390 pr_info("No ESR for 82489DX.\n");
1394 if (apic->disable_esr) {
1396 * Something untraceable is creating bad interrupts on
1397 * secondary quads ... for the moment, just leave the
1398 * ESR disabled - we can't do anything useful with the
1399 * errors anyway - mbligh
1401 pr_info("Leaving ESR disabled.\n");
1405 maxlvt = lapic_get_maxlvt();
1406 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1407 apic_write(APIC_ESR, 0);
1408 oldvalue = apic_read(APIC_ESR);
1410 /* enables sending errors */
1411 value = ERROR_APIC_VECTOR;
1412 apic_write(APIC_LVTERR, value);
1415 * spec says clear errors after enabling vector.
1418 apic_write(APIC_ESR, 0);
1419 value = apic_read(APIC_ESR);
1420 if (value != oldvalue)
1421 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1422 "vector: 0x%08x after: 0x%08x\n",
1426 static void apic_pending_intr_clear(void)
1428 long long max_loops = cpu_khz ? cpu_khz : 1000000;
1429 unsigned long long tsc = 0, ntsc;
1430 unsigned int queued;
1431 unsigned long value;
1432 int i, j, acked = 0;
1434 if (boot_cpu_has(X86_FEATURE_TSC))
1437 * After a crash, we no longer service the interrupts and a pending
1438 * interrupt from previous kernel might still have ISR bit set.
1440 * Most probably by now CPU has serviced that pending interrupt and
1441 * it might not have done the ack_APIC_irq() because it thought,
1442 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1443 * does not clear the ISR bit and cpu thinks it has already serivced
1444 * the interrupt. Hence a vector might get locked. It was noticed
1445 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1449 for (i = APIC_ISR_NR - 1; i >= 0; i--)
1450 queued |= apic_read(APIC_IRR + i*0x10);
1452 for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1453 value = apic_read(APIC_ISR + i*0x10);
1454 for_each_set_bit(j, &value, 32) {
1460 pr_err("LAPIC pending interrupts after %d EOI\n", acked);
1464 if (boot_cpu_has(X86_FEATURE_TSC) && cpu_khz) {
1466 max_loops = (cpu_khz << 10) - (ntsc - tsc);
1471 } while (queued && max_loops > 0);
1472 WARN_ON(max_loops <= 0);
1476 * setup_local_APIC - setup the local APIC
1478 * Used to setup local APIC while initializing BSP or bringing up APs.
1479 * Always called with preemption disabled.
1481 static void setup_local_APIC(void)
1483 int cpu = smp_processor_id();
1485 #ifdef CONFIG_X86_32
1486 int logical_apicid, ldr_apicid;
1491 disable_ioapic_support();
1495 #ifdef CONFIG_X86_32
1496 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1497 if (lapic_is_integrated() && apic->disable_esr) {
1498 apic_write(APIC_ESR, 0);
1499 apic_write(APIC_ESR, 0);
1500 apic_write(APIC_ESR, 0);
1501 apic_write(APIC_ESR, 0);
1504 perf_events_lapic_init();
1507 * Double-check whether this APIC is really registered.
1508 * This is meaningless in clustered apic mode, so we skip it.
1510 BUG_ON(!apic->apic_id_registered());
1513 * Intel recommends to set DFR, LDR and TPR before enabling
1514 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1515 * document number 292116). So here it goes...
1517 apic->init_apic_ldr();
1519 #ifdef CONFIG_X86_32
1521 * APIC LDR is initialized. If logical_apicid mapping was
1522 * initialized during get_smp_config(), make sure it matches the
1525 logical_apicid = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1526 ldr_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1527 WARN_ON(logical_apicid != BAD_APICID && logical_apicid != ldr_apicid);
1528 /* always use the value from LDR */
1529 early_per_cpu(x86_cpu_to_logical_apicid, cpu) = ldr_apicid;
1533 * Set Task Priority to 'accept all'. We never change this
1536 value = apic_read(APIC_TASKPRI);
1537 value &= ~APIC_TPRI_MASK;
1538 apic_write(APIC_TASKPRI, value);
1540 apic_pending_intr_clear();
1543 * Now that we are all set up, enable the APIC
1545 value = apic_read(APIC_SPIV);
1546 value &= ~APIC_VECTOR_MASK;
1550 value |= APIC_SPIV_APIC_ENABLED;
1552 #ifdef CONFIG_X86_32
1554 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1555 * certain networking cards. If high frequency interrupts are
1556 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1557 * entry is masked/unmasked at a high rate as well then sooner or
1558 * later IOAPIC line gets 'stuck', no more interrupts are received
1559 * from the device. If focus CPU is disabled then the hang goes
1562 * [ This bug can be reproduced easily with a level-triggered
1563 * PCI Ne2000 networking cards and PII/PIII processors, dual
1567 * Actually disabling the focus CPU check just makes the hang less
1568 * frequent as it makes the interrupt distributon model be more
1569 * like LRU than MRU (the short-term load is more even across CPUs).
1573 * - enable focus processor (bit==0)
1574 * - 64bit mode always use processor focus
1575 * so no need to set it
1577 value &= ~APIC_SPIV_FOCUS_DISABLED;
1581 * Set spurious IRQ vector
1583 value |= SPURIOUS_APIC_VECTOR;
1584 apic_write(APIC_SPIV, value);
1587 * Set up LVT0, LVT1:
1589 * set up through-local-APIC on the boot CPU's LINT0. This is not
1590 * strictly necessary in pure symmetric-IO mode, but sometimes
1591 * we delegate interrupts to the 8259A.
1594 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1596 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1597 if (!cpu && (pic_mode || !value || skip_ioapic_setup)) {
1598 value = APIC_DM_EXTINT;
1599 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1601 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1602 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1604 apic_write(APIC_LVT0, value);
1607 * Only the BSP sees the LINT1 NMI signal by default. This can be
1608 * modified by apic_extnmi= boot option.
1610 if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1611 apic_extnmi == APIC_EXTNMI_ALL)
1612 value = APIC_DM_NMI;
1614 value = APIC_DM_NMI | APIC_LVT_MASKED;
1617 if (!lapic_is_integrated())
1618 value |= APIC_LVT_LEVEL_TRIGGER;
1619 apic_write(APIC_LVT1, value);
1621 #ifdef CONFIG_X86_MCE_INTEL
1622 /* Recheck CMCI information after local APIC is up on CPU #0 */
1628 static void end_local_APIC_setup(void)
1632 #ifdef CONFIG_X86_32
1635 /* Disable the local apic timer */
1636 value = apic_read(APIC_LVTT);
1637 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1638 apic_write(APIC_LVTT, value);
1646 * APIC setup function for application processors. Called from smpboot.c
1648 void apic_ap_setup(void)
1651 end_local_APIC_setup();
1654 #ifdef CONFIG_X86_X2APIC
1662 static int x2apic_state;
1664 static void __x2apic_disable(void)
1668 if (!boot_cpu_has(X86_FEATURE_APIC))
1671 rdmsrl(MSR_IA32_APICBASE, msr);
1672 if (!(msr & X2APIC_ENABLE))
1674 /* Disable xapic and x2apic first and then reenable xapic mode */
1675 wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1676 wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1677 printk_once(KERN_INFO "x2apic disabled\n");
1680 static void __x2apic_enable(void)
1684 rdmsrl(MSR_IA32_APICBASE, msr);
1685 if (msr & X2APIC_ENABLE)
1687 wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1688 printk_once(KERN_INFO "x2apic enabled\n");
1691 static int __init setup_nox2apic(char *str)
1693 if (x2apic_enabled()) {
1694 int apicid = native_apic_msr_read(APIC_ID);
1696 if (apicid >= 255) {
1697 pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
1701 pr_warning("x2apic already enabled.\n");
1704 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1705 x2apic_state = X2APIC_DISABLED;
1709 early_param("nox2apic", setup_nox2apic);
1711 /* Called from cpu_init() to enable x2apic on (secondary) cpus */
1712 void x2apic_setup(void)
1715 * If x2apic is not in ON state, disable it if already enabled
1718 if (x2apic_state != X2APIC_ON) {
1725 static __init void x2apic_disable(void)
1727 u32 x2apic_id, state = x2apic_state;
1730 x2apic_state = X2APIC_DISABLED;
1732 if (state != X2APIC_ON)
1735 x2apic_id = read_apic_id();
1736 if (x2apic_id >= 255)
1737 panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1740 register_lapic_address(mp_lapic_addr);
1743 static __init void x2apic_enable(void)
1745 if (x2apic_state != X2APIC_OFF)
1749 x2apic_state = X2APIC_ON;
1753 static __init void try_to_enable_x2apic(int remap_mode)
1755 if (x2apic_state == X2APIC_DISABLED)
1758 if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1759 /* IR is required if there is APIC ID > 255 even when running
1762 if (max_physical_apicid > 255 ||
1763 !x86_init.hyper.x2apic_available()) {
1764 pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1770 * without IR all CPUs can be addressed by IOAPIC/MSI
1771 * only in physical mode
1778 void __init check_x2apic(void)
1780 if (x2apic_enabled()) {
1781 pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1783 x2apic_state = X2APIC_ON;
1784 } else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1785 x2apic_state = X2APIC_DISABLED;
1788 #else /* CONFIG_X86_X2APIC */
1789 static int __init validate_x2apic(void)
1791 if (!apic_is_x2apic_enabled())
1794 * Checkme: Can we simply turn off x2apic here instead of panic?
1796 panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n");
1798 early_initcall(validate_x2apic);
1800 static inline void try_to_enable_x2apic(int remap_mode) { }
1801 static inline void __x2apic_enable(void) { }
1802 #endif /* !CONFIG_X86_X2APIC */
1804 void __init enable_IR_x2apic(void)
1806 unsigned long flags;
1809 if (skip_ioapic_setup) {
1810 pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1814 ir_stat = irq_remapping_prepare();
1815 if (ir_stat < 0 && !x2apic_supported())
1818 ret = save_ioapic_entries();
1820 pr_info("Saving IO-APIC state failed: %d\n", ret);
1824 local_irq_save(flags);
1825 legacy_pic->mask_all();
1826 mask_ioapic_entries();
1828 /* If irq_remapping_prepare() succeeded, try to enable it */
1830 ir_stat = irq_remapping_enable();
1831 /* ir_stat contains the remap mode or an error code */
1832 try_to_enable_x2apic(ir_stat);
1835 restore_ioapic_entries();
1836 legacy_pic->restore_mask();
1837 local_irq_restore(flags);
1840 #ifdef CONFIG_X86_64
1842 * Detect and enable local APICs on non-SMP boards.
1843 * Original code written by Keir Fraser.
1844 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1845 * not correctly set up (usually the APIC timer won't work etc.)
1847 static int __init detect_init_APIC(void)
1849 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1850 pr_info("No local APIC present\n");
1854 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1859 static int __init apic_verify(void)
1864 * The APIC feature bit should now be enabled
1867 features = cpuid_edx(1);
1868 if (!(features & (1 << X86_FEATURE_APIC))) {
1869 pr_warning("Could not enable APIC!\n");
1872 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1873 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1875 /* The BIOS may have set up the APIC at some other address */
1876 if (boot_cpu_data.x86 >= 6) {
1877 rdmsr(MSR_IA32_APICBASE, l, h);
1878 if (l & MSR_IA32_APICBASE_ENABLE)
1879 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1882 pr_info("Found and enabled local APIC!\n");
1886 int __init apic_force_enable(unsigned long addr)
1894 * Some BIOSes disable the local APIC in the APIC_BASE
1895 * MSR. This can only be done in software for Intel P6 or later
1896 * and AMD K7 (Model > 1) or later.
1898 if (boot_cpu_data.x86 >= 6) {
1899 rdmsr(MSR_IA32_APICBASE, l, h);
1900 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1901 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1902 l &= ~MSR_IA32_APICBASE_BASE;
1903 l |= MSR_IA32_APICBASE_ENABLE | addr;
1904 wrmsr(MSR_IA32_APICBASE, l, h);
1905 enabled_via_apicbase = 1;
1908 return apic_verify();
1912 * Detect and initialize APIC
1914 static int __init detect_init_APIC(void)
1916 /* Disabled by kernel option? */
1920 switch (boot_cpu_data.x86_vendor) {
1921 case X86_VENDOR_AMD:
1922 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1923 (boot_cpu_data.x86 >= 15))
1926 case X86_VENDOR_HYGON:
1928 case X86_VENDOR_INTEL:
1929 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1930 (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
1937 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1939 * Over-ride BIOS and try to enable the local APIC only if
1940 * "lapic" specified.
1942 if (!force_enable_local_apic) {
1943 pr_info("Local APIC disabled by BIOS -- "
1944 "you can enable it with \"lapic\"\n");
1947 if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1959 pr_info("No local APIC present or hardware disabled\n");
1965 * init_apic_mappings - initialize APIC mappings
1967 void __init init_apic_mappings(void)
1969 unsigned int new_apicid;
1971 apic_check_deadline_errata();
1974 boot_cpu_physical_apicid = read_apic_id();
1978 /* If no local APIC can be found return early */
1979 if (!smp_found_config && detect_init_APIC()) {
1980 /* lets NOP'ify apic operations */
1981 pr_info("APIC: disable apic facility\n");
1984 apic_phys = mp_lapic_addr;
1987 * If the system has ACPI MADT tables or MP info, the LAPIC
1988 * address is already registered.
1990 if (!acpi_lapic && !smp_found_config)
1991 register_lapic_address(apic_phys);
1995 * Fetch the APIC ID of the BSP in case we have a
1996 * default configuration (or the MP table is broken).
1998 new_apicid = read_apic_id();
1999 if (boot_cpu_physical_apicid != new_apicid) {
2000 boot_cpu_physical_apicid = new_apicid;
2002 * yeah -- we lie about apic_version
2003 * in case if apic was disabled via boot option
2004 * but it's not a problem for SMP compiled kernel
2005 * since apic_intr_mode_select is prepared for such
2006 * a case and disable smp mode
2008 boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2012 void __init register_lapic_address(unsigned long address)
2014 mp_lapic_addr = address;
2017 set_fixmap_nocache(FIX_APIC_BASE, address);
2018 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
2019 APIC_BASE, address);
2021 if (boot_cpu_physical_apicid == -1U) {
2022 boot_cpu_physical_apicid = read_apic_id();
2023 boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2028 * Local APIC interrupts
2032 * This interrupt should _never_ happen with our APIC/SMP architecture
2034 __visible void __irq_entry smp_spurious_interrupt(struct pt_regs *regs)
2036 u8 vector = ~regs->orig_ax;
2040 trace_spurious_apic_entry(vector);
2043 * Check if this really is a spurious interrupt and ACK it
2044 * if it is a vectored one. Just in case...
2045 * Spurious interrupts should not be ACKed.
2047 v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2048 if (v & (1 << (vector & 0x1f)))
2051 inc_irq_stat(irq_spurious_count);
2053 /* see sw-dev-man vol 3, chapter 7.4.13.5 */
2054 pr_info("spurious APIC interrupt through vector %02x on CPU#%d, "
2055 "should never happen.\n", vector, smp_processor_id());
2057 trace_spurious_apic_exit(vector);
2062 * This interrupt should never happen with our APIC/SMP architecture
2064 __visible void __irq_entry smp_error_interrupt(struct pt_regs *regs)
2066 static const char * const error_interrupt_reason[] = {
2067 "Send CS error", /* APIC Error Bit 0 */
2068 "Receive CS error", /* APIC Error Bit 1 */
2069 "Send accept error", /* APIC Error Bit 2 */
2070 "Receive accept error", /* APIC Error Bit 3 */
2071 "Redirectable IPI", /* APIC Error Bit 4 */
2072 "Send illegal vector", /* APIC Error Bit 5 */
2073 "Received illegal vector", /* APIC Error Bit 6 */
2074 "Illegal register address", /* APIC Error Bit 7 */
2079 trace_error_apic_entry(ERROR_APIC_VECTOR);
2081 /* First tickle the hardware, only then report what went on. -- REW */
2082 if (lapic_get_maxlvt() > 3) /* Due to the Pentium erratum 3AP. */
2083 apic_write(APIC_ESR, 0);
2084 v = apic_read(APIC_ESR);
2086 atomic_inc(&irq_err_count);
2088 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2089 smp_processor_id(), v);
2094 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2099 apic_printk(APIC_DEBUG, KERN_CONT "\n");
2101 trace_error_apic_exit(ERROR_APIC_VECTOR);
2106 * connect_bsp_APIC - attach the APIC to the interrupt system
2108 static void __init connect_bsp_APIC(void)
2110 #ifdef CONFIG_X86_32
2113 * Do not trust the local APIC being empty at bootup.
2117 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
2118 * local APIC to INT and NMI lines.
2120 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2121 "enabling APIC mode.\n");
2128 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2129 * @virt_wire_setup: indicates, whether virtual wire mode is selected
2131 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2134 void disconnect_bsp_APIC(int virt_wire_setup)
2138 #ifdef CONFIG_X86_32
2141 * Put the board back into PIC mode (has an effect only on
2142 * certain older boards). Note that APIC interrupts, including
2143 * IPIs, won't work beyond this point! The only exception are
2146 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2147 "entering PIC mode.\n");
2153 /* Go back to Virtual Wire compatibility mode */
2155 /* For the spurious interrupt use vector F, and enable it */
2156 value = apic_read(APIC_SPIV);
2157 value &= ~APIC_VECTOR_MASK;
2158 value |= APIC_SPIV_APIC_ENABLED;
2160 apic_write(APIC_SPIV, value);
2162 if (!virt_wire_setup) {
2164 * For LVT0 make it edge triggered, active high,
2165 * external and enabled
2167 value = apic_read(APIC_LVT0);
2168 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2169 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2170 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2171 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2172 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2173 apic_write(APIC_LVT0, value);
2176 apic_write(APIC_LVT0, APIC_LVT_MASKED);
2180 * For LVT1 make it edge triggered, active high,
2183 value = apic_read(APIC_LVT1);
2184 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2185 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2186 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2187 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2188 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2189 apic_write(APIC_LVT1, value);
2193 * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2194 * contiguously, it equals to current allocated max logical CPU ID plus 1.
2195 * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2196 * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2198 * NOTE: Reserve 0 for BSP.
2200 static int nr_logical_cpuids = 1;
2203 * Used to store mapping between logical CPU IDs and APIC IDs.
2205 static int cpuid_to_apicid[] = {
2206 [0 ... NR_CPUS - 1] = -1,
2211 * apic_id_is_primary_thread - Check whether APIC ID belongs to a primary thread
2212 * @id: APIC ID to check
2214 bool apic_id_is_primary_thread(unsigned int apicid)
2218 if (smp_num_siblings == 1)
2220 /* Isolate the SMT bit(s) in the APICID and check for 0 */
2221 mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
2222 return !(apicid & mask);
2227 * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2228 * and cpuid_to_apicid[] synchronized.
2230 static int allocate_logical_cpuid(int apicid)
2235 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2236 * check if the kernel has allocated a cpuid for it.
2238 for (i = 0; i < nr_logical_cpuids; i++) {
2239 if (cpuid_to_apicid[i] == apicid)
2243 /* Allocate a new cpuid. */
2244 if (nr_logical_cpuids >= nr_cpu_ids) {
2245 WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %u reached. "
2246 "Processor %d/0x%x and the rest are ignored.\n",
2247 nr_cpu_ids, nr_logical_cpuids, apicid);
2251 cpuid_to_apicid[nr_logical_cpuids] = apicid;
2252 return nr_logical_cpuids++;
2255 int generic_processor_info(int apicid, int version)
2257 int cpu, max = nr_cpu_ids;
2258 bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2259 phys_cpu_present_map);
2262 * boot_cpu_physical_apicid is designed to have the apicid
2263 * returned by read_apic_id(), i.e, the apicid of the
2264 * currently booting-up processor. However, on some platforms,
2265 * it is temporarily modified by the apicid reported as BSP
2266 * through MP table. Concretely:
2268 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2269 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2271 * This function is executed with the modified
2272 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2273 * parameter doesn't work to disable APs on kdump 2nd kernel.
2275 * Since fixing handling of boot_cpu_physical_apicid requires
2276 * another discussion and tests on each platform, we leave it
2277 * for now and here we use read_apic_id() directly in this
2278 * function, generic_processor_info().
2280 if (disabled_cpu_apicid != BAD_APICID &&
2281 disabled_cpu_apicid != read_apic_id() &&
2282 disabled_cpu_apicid == apicid) {
2283 int thiscpu = num_processors + disabled_cpus;
2285 pr_warning("APIC: Disabling requested cpu."
2286 " Processor %d/0x%x ignored.\n",
2294 * If boot cpu has not been detected yet, then only allow upto
2295 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2297 if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2298 apicid != boot_cpu_physical_apicid) {
2299 int thiscpu = max + disabled_cpus - 1;
2302 "APIC: NR_CPUS/possible_cpus limit of %i almost"
2303 " reached. Keeping one slot for boot cpu."
2304 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2310 if (num_processors >= nr_cpu_ids) {
2311 int thiscpu = max + disabled_cpus;
2313 pr_warning("APIC: NR_CPUS/possible_cpus limit of %i "
2314 "reached. Processor %d/0x%x ignored.\n",
2315 max, thiscpu, apicid);
2321 if (apicid == boot_cpu_physical_apicid) {
2323 * x86_bios_cpu_apicid is required to have processors listed
2324 * in same order as logical cpu numbers. Hence the first
2325 * entry is BSP, and so on.
2326 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2331 /* Logical cpuid 0 is reserved for BSP. */
2332 cpuid_to_apicid[0] = apicid;
2334 cpu = allocate_logical_cpuid(apicid);
2344 if (version == 0x0) {
2345 pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2350 if (version != boot_cpu_apic_version) {
2351 pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2352 boot_cpu_apic_version, cpu, version);
2355 if (apicid > max_physical_apicid)
2356 max_physical_apicid = apicid;
2358 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2359 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2360 early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2362 #ifdef CONFIG_X86_32
2363 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2364 apic->x86_32_early_logical_apicid(cpu);
2366 set_cpu_possible(cpu, true);
2367 physid_set(apicid, phys_cpu_present_map);
2368 set_cpu_present(cpu, true);
2374 int hard_smp_processor_id(void)
2376 return read_apic_id();
2380 * Override the generic EOI implementation with an optimized version.
2381 * Only called during early boot when only one CPU is active and with
2382 * interrupts disabled, so we know this does not race with actual APIC driver
2385 void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2389 for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2390 /* Should happen once for each apic */
2391 WARN_ON((*drv)->eoi_write == eoi_write);
2392 (*drv)->native_eoi_write = (*drv)->eoi_write;
2393 (*drv)->eoi_write = eoi_write;
2397 static void __init apic_bsp_up_setup(void)
2399 #ifdef CONFIG_X86_64
2400 apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2403 * Hack: In case of kdump, after a crash, kernel might be booting
2404 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2405 * might be zero if read from MP tables. Get it from LAPIC.
2407 # ifdef CONFIG_CRASH_DUMP
2408 boot_cpu_physical_apicid = read_apic_id();
2411 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2415 * apic_bsp_setup - Setup function for local apic and io-apic
2416 * @upmode: Force UP mode (for APIC_init_uniprocessor)
2419 * apic_id of BSP APIC
2421 void __init apic_bsp_setup(bool upmode)
2425 apic_bsp_up_setup();
2429 end_local_APIC_setup();
2430 irq_remap_enable_fault_handling();
2434 #ifdef CONFIG_UP_LATE_INIT
2435 void __init up_late_init(void)
2437 if (apic_intr_mode == APIC_PIC)
2440 /* Setup local timer */
2441 x86_init.timers.setup_percpu_clockev();
2452 * 'active' is true if the local APIC was enabled by us and
2453 * not the BIOS; this signifies that we are also responsible
2454 * for disabling it before entering apm/acpi suspend
2457 /* r/w apic fields */
2458 unsigned int apic_id;
2459 unsigned int apic_taskpri;
2460 unsigned int apic_ldr;
2461 unsigned int apic_dfr;
2462 unsigned int apic_spiv;
2463 unsigned int apic_lvtt;
2464 unsigned int apic_lvtpc;
2465 unsigned int apic_lvt0;
2466 unsigned int apic_lvt1;
2467 unsigned int apic_lvterr;
2468 unsigned int apic_tmict;
2469 unsigned int apic_tdcr;
2470 unsigned int apic_thmr;
2471 unsigned int apic_cmci;
2474 static int lapic_suspend(void)
2476 unsigned long flags;
2479 if (!apic_pm_state.active)
2482 maxlvt = lapic_get_maxlvt();
2484 apic_pm_state.apic_id = apic_read(APIC_ID);
2485 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2486 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2487 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2488 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2489 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2491 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2492 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2493 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2494 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2495 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2496 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2497 #ifdef CONFIG_X86_THERMAL_VECTOR
2499 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2501 #ifdef CONFIG_X86_MCE_INTEL
2503 apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2506 local_irq_save(flags);
2507 disable_local_APIC();
2509 irq_remapping_disable();
2511 local_irq_restore(flags);
2515 static void lapic_resume(void)
2518 unsigned long flags;
2521 if (!apic_pm_state.active)
2524 local_irq_save(flags);
2527 * IO-APIC and PIC have their own resume routines.
2528 * We just mask them here to make sure the interrupt
2529 * subsystem is completely quiet while we enable x2apic
2530 * and interrupt-remapping.
2532 mask_ioapic_entries();
2533 legacy_pic->mask_all();
2539 * Make sure the APICBASE points to the right address
2541 * FIXME! This will be wrong if we ever support suspend on
2542 * SMP! We'll need to do this as part of the CPU restore!
2544 if (boot_cpu_data.x86 >= 6) {
2545 rdmsr(MSR_IA32_APICBASE, l, h);
2546 l &= ~MSR_IA32_APICBASE_BASE;
2547 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2548 wrmsr(MSR_IA32_APICBASE, l, h);
2552 maxlvt = lapic_get_maxlvt();
2553 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2554 apic_write(APIC_ID, apic_pm_state.apic_id);
2555 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2556 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2557 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2558 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2559 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2560 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2561 #ifdef CONFIG_X86_THERMAL_VECTOR
2563 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2565 #ifdef CONFIG_X86_MCE_INTEL
2567 apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2570 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2571 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2572 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2573 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2574 apic_write(APIC_ESR, 0);
2575 apic_read(APIC_ESR);
2576 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2577 apic_write(APIC_ESR, 0);
2578 apic_read(APIC_ESR);
2580 irq_remapping_reenable(x2apic_mode);
2582 local_irq_restore(flags);
2586 * This device has no shutdown method - fully functioning local APICs
2587 * are needed on every CPU up until machine_halt/restart/poweroff.
2590 static struct syscore_ops lapic_syscore_ops = {
2591 .resume = lapic_resume,
2592 .suspend = lapic_suspend,
2595 static void apic_pm_activate(void)
2597 apic_pm_state.active = 1;
2600 static int __init init_lapic_sysfs(void)
2602 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2603 if (boot_cpu_has(X86_FEATURE_APIC))
2604 register_syscore_ops(&lapic_syscore_ops);
2609 /* local apic needs to resume before other devices access its registers. */
2610 core_initcall(init_lapic_sysfs);
2612 #else /* CONFIG_PM */
2614 static void apic_pm_activate(void) { }
2616 #endif /* CONFIG_PM */
2618 #ifdef CONFIG_X86_64
2620 static int multi_checked;
2623 static int set_multi(const struct dmi_system_id *d)
2627 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2632 static const struct dmi_system_id multi_dmi_table[] = {
2634 .callback = set_multi,
2635 .ident = "IBM System Summit2",
2637 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2638 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2644 static void dmi_check_multi(void)
2649 dmi_check_system(multi_dmi_table);
2654 * apic_is_clustered_box() -- Check if we can expect good TSC
2656 * Thus far, the major user of this is IBM's Summit2 series:
2657 * Clustered boxes may have unsynced TSC problems if they are
2659 * Use DMI to check them
2661 int apic_is_clustered_box(void)
2669 * APIC command line parameters
2671 static int __init setup_disableapic(char *arg)
2674 setup_clear_cpu_cap(X86_FEATURE_APIC);
2677 early_param("disableapic", setup_disableapic);
2679 /* same as disableapic, for compatibility */
2680 static int __init setup_nolapic(char *arg)
2682 return setup_disableapic(arg);
2684 early_param("nolapic", setup_nolapic);
2686 static int __init parse_lapic_timer_c2_ok(char *arg)
2688 local_apic_timer_c2_ok = 1;
2691 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2693 static int __init parse_disable_apic_timer(char *arg)
2695 disable_apic_timer = 1;
2698 early_param("noapictimer", parse_disable_apic_timer);
2700 static int __init parse_nolapic_timer(char *arg)
2702 disable_apic_timer = 1;
2705 early_param("nolapic_timer", parse_nolapic_timer);
2707 static int __init apic_set_verbosity(char *arg)
2710 #ifdef CONFIG_X86_64
2711 skip_ioapic_setup = 0;
2717 if (strcmp("debug", arg) == 0)
2718 apic_verbosity = APIC_DEBUG;
2719 else if (strcmp("verbose", arg) == 0)
2720 apic_verbosity = APIC_VERBOSE;
2721 #ifdef CONFIG_X86_64
2723 pr_warning("APIC Verbosity level %s not recognised"
2724 " use apic=verbose or apic=debug\n", arg);
2731 early_param("apic", apic_set_verbosity);
2733 static int __init lapic_insert_resource(void)
2738 /* Put local APIC into the resource map. */
2739 lapic_resource.start = apic_phys;
2740 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2741 insert_resource(&iomem_resource, &lapic_resource);
2747 * need call insert after e820__reserve_resources()
2748 * that is using request_resource
2750 late_initcall(lapic_insert_resource);
2752 static int __init apic_set_disabled_cpu_apicid(char *arg)
2754 if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2759 early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2761 static int __init apic_set_extnmi(char *arg)
2766 if (!strncmp("all", arg, 3))
2767 apic_extnmi = APIC_EXTNMI_ALL;
2768 else if (!strncmp("none", arg, 4))
2769 apic_extnmi = APIC_EXTNMI_NONE;
2770 else if (!strncmp("bsp", arg, 3))
2771 apic_extnmi = APIC_EXTNMI_BSP;
2773 pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2779 early_param("apic_extnmi", apic_set_extnmi);