2 * Local APIC handling, local APIC timers
4 * (c) 1999, 2000 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/init.h>
20 #include <linux/delay.h>
21 #include <linux/bootmem.h>
22 #include <linux/interrupt.h>
23 #include <linux/mc146818rtc.h>
24 #include <linux/kernel_stat.h>
25 #include <linux/sysdev.h>
26 #include <linux/cpu.h>
27 #include <linux/clockchips.h>
28 #include <linux/acpi_pmtmr.h>
29 #include <linux/module.h>
30 #include <linux/dmi.h>
32 #include <asm/atomic.h>
35 #include <asm/mpspec.h>
37 #include <asm/arch_hooks.h>
39 #include <asm/i8253.h>
42 #include <mach_apic.h>
43 #include <mach_apicdef.h>
49 #if ((SPURIOUS_APIC_VECTOR & 0x0F) != 0x0F)
50 # error SPURIOUS_APIC_VECTOR definition error
54 * Knob to control our willingness to enable the local APIC.
56 * -1=force-disable, +1=force-enable
58 static int enable_local_apic __initdata;
60 /* Local APIC timer verification ok */
61 static int local_apic_timer_verify_ok;
62 /* Disable local APIC timer from the kernel commandline or via dmi quirk
63 or using CPU MSR check */
64 int local_apic_timer_disabled;
65 /* Local APIC timer works in C2 */
66 int local_apic_timer_c2_ok;
67 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
70 * Debug level, exported for io_apic.c
74 static unsigned int calibration_result;
76 static int lapic_next_event(unsigned long delta,
77 struct clock_event_device *evt);
78 static void lapic_timer_setup(enum clock_event_mode mode,
79 struct clock_event_device *evt);
80 static void lapic_timer_broadcast(cpumask_t mask);
81 static void apic_pm_activate(void);
84 * The local apic timer can be used for any function which is CPU local.
86 static struct clock_event_device lapic_clockevent = {
88 .features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
89 | CLOCK_EVT_FEAT_C3STOP | CLOCK_EVT_FEAT_DUMMY,
91 .set_mode = lapic_timer_setup,
92 .set_next_event = lapic_next_event,
93 .broadcast = lapic_timer_broadcast,
97 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
99 /* Local APIC was disabled by the BIOS and enabled by the kernel */
100 static int enabled_via_apicbase;
103 * Get the LAPIC version
105 static inline int lapic_get_version(void)
107 return GET_APIC_VERSION(apic_read(APIC_LVR));
111 * Check, if the APIC is integrated or a separate chip
113 static inline int lapic_is_integrated(void)
115 return APIC_INTEGRATED(lapic_get_version());
119 * Check, whether this is a modern or a first generation APIC
121 static int modern_apic(void)
123 /* AMD systems use old APIC versions, so check the CPU */
124 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
125 boot_cpu_data.x86 >= 0xf)
127 return lapic_get_version() >= 0x14;
130 void apic_wait_icr_idle(void)
132 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
136 u32 safe_apic_wait_icr_idle(void)
143 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
147 } while (timeout++ < 1000);
153 * enable_NMI_through_LVT0 - enable NMI through local vector table 0
155 void __cpuinit enable_NMI_through_LVT0(void)
157 unsigned int v = APIC_DM_NMI;
159 /* Level triggered for 82489DX */
160 if (!lapic_is_integrated())
161 v |= APIC_LVT_LEVEL_TRIGGER;
162 apic_write_around(APIC_LVT0, v);
166 * get_physical_broadcast - Get number of physical broadcast IDs
168 int get_physical_broadcast(void)
170 return modern_apic() ? 0xff : 0xf;
174 * lapic_get_maxlvt - get the maximum number of local vector table entries
176 int lapic_get_maxlvt(void)
178 unsigned int v = apic_read(APIC_LVR);
180 /* 82489DXs do not report # of LVT entries. */
181 return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
188 /* Clock divisor is set to 16 */
189 #define APIC_DIVISOR 16
192 * This function sets up the local APIC timer, with a timeout of
193 * 'clocks' APIC bus clock. During calibration we actually call
194 * this function twice on the boot CPU, once with a bogus timeout
195 * value, second time for real. The other (noncalibrating) CPUs
196 * call this function only once, with the real, calibrated value.
198 * We do reads before writes even if unnecessary, to get around the
199 * P5 APIC double write bug.
201 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
203 unsigned int lvtt_value, tmp_value;
205 lvtt_value = LOCAL_TIMER_VECTOR;
207 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
208 if (!lapic_is_integrated())
209 lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
212 lvtt_value |= APIC_LVT_MASKED;
214 apic_write_around(APIC_LVTT, lvtt_value);
219 tmp_value = apic_read(APIC_TDCR);
220 apic_write_around(APIC_TDCR, (tmp_value
221 & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE))
225 apic_write_around(APIC_TMICT, clocks/APIC_DIVISOR);
229 * Program the next event, relative to now
231 static int lapic_next_event(unsigned long delta,
232 struct clock_event_device *evt)
234 apic_write_around(APIC_TMICT, delta);
239 * Setup the lapic timer in periodic or oneshot mode
241 static void lapic_timer_setup(enum clock_event_mode mode,
242 struct clock_event_device *evt)
247 /* Lapic used for broadcast ? */
248 if (!local_apic_timer_verify_ok)
251 local_irq_save(flags);
254 case CLOCK_EVT_MODE_PERIODIC:
255 case CLOCK_EVT_MODE_ONESHOT:
256 __setup_APIC_LVTT(calibration_result,
257 mode != CLOCK_EVT_MODE_PERIODIC, 1);
259 case CLOCK_EVT_MODE_UNUSED:
260 case CLOCK_EVT_MODE_SHUTDOWN:
261 v = apic_read(APIC_LVTT);
262 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
263 apic_write_around(APIC_LVTT, v);
265 case CLOCK_EVT_MODE_RESUME:
266 /* Nothing to do here */
270 local_irq_restore(flags);
274 * Local APIC timer broadcast function
276 static void lapic_timer_broadcast(cpumask_t mask)
279 send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
284 * Setup the local APIC timer for this CPU. Copy the initilized values
285 * of the boot CPU and register the clock event in the framework.
287 static void __devinit setup_APIC_timer(void)
289 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
291 memcpy(levt, &lapic_clockevent, sizeof(*levt));
292 levt->cpumask = cpumask_of_cpu(smp_processor_id());
294 clockevents_register_device(levt);
298 * In this functions we calibrate APIC bus clocks to the external timer.
300 * We want to do the calibration only once since we want to have local timer
301 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
304 * This was previously done by reading the PIT/HPET and waiting for a wrap
305 * around to find out, that a tick has elapsed. I have a box, where the PIT
306 * readout is broken, so it never gets out of the wait loop again. This was
307 * also reported by others.
309 * Monitoring the jiffies value is inaccurate and the clockevents
310 * infrastructure allows us to do a simple substitution of the interrupt
313 * The calibration routine also uses the pm_timer when possible, as the PIT
314 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
315 * back to normal later in the boot process).
318 #define LAPIC_CAL_LOOPS (HZ/10)
320 static __initdata int lapic_cal_loops = -1;
321 static __initdata long lapic_cal_t1, lapic_cal_t2;
322 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
323 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
324 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
327 * Temporary interrupt handler.
329 static void __init lapic_cal_handler(struct clock_event_device *dev)
331 unsigned long long tsc = 0;
332 long tapic = apic_read(APIC_TMCCT);
333 unsigned long pm = acpi_pm_read_early();
338 switch (lapic_cal_loops++) {
340 lapic_cal_t1 = tapic;
341 lapic_cal_tsc1 = tsc;
343 lapic_cal_j1 = jiffies;
346 case LAPIC_CAL_LOOPS:
347 lapic_cal_t2 = tapic;
348 lapic_cal_tsc2 = tsc;
349 if (pm < lapic_cal_pm1)
350 pm += ACPI_PM_OVRRUN;
352 lapic_cal_j2 = jiffies;
358 * Setup the boot APIC
360 * Calibrate and verify the result.
362 void __init setup_boot_APIC_clock(void)
364 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
365 const long pm_100ms = PMTMR_TICKS_PER_SEC/10;
366 const long pm_thresh = pm_100ms/100;
367 void (*real_handler)(struct clock_event_device *dev);
368 unsigned long deltaj;
370 int pm_referenced = 0;
373 * The local apic timer can be disabled via the kernel
374 * commandline or from the CPU detection code. Register the lapic
375 * timer as a dummy clock event source on SMP systems, so the
376 * broadcast mechanism is used. On UP systems simply ignore it.
378 if (local_apic_timer_disabled) {
379 /* No broadcast on UP ! */
380 if (num_possible_cpus() > 1) {
381 lapic_clockevent.mult = 1;
387 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
388 "calibrating APIC timer ...\n");
392 /* Replace the global interrupt handler */
393 real_handler = global_clock_event->event_handler;
394 global_clock_event->event_handler = lapic_cal_handler;
397 * Setup the APIC counter to 1e9. There is no way the lapic
398 * can underflow in the 100ms detection time frame
400 __setup_APIC_LVTT(1000000000, 0, 0);
402 /* Let the interrupts run */
405 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
410 /* Restore the real event handler */
411 global_clock_event->event_handler = real_handler;
413 /* Build delta t1-t2 as apic timer counts down */
414 delta = lapic_cal_t1 - lapic_cal_t2;
415 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
417 /* Check, if the PM timer is available */
418 deltapm = lapic_cal_pm2 - lapic_cal_pm1;
419 apic_printk(APIC_VERBOSE, "... PM timer delta = %ld\n", deltapm);
425 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
427 if (deltapm > (pm_100ms - pm_thresh) &&
428 deltapm < (pm_100ms + pm_thresh)) {
429 apic_printk(APIC_VERBOSE, "... PM timer result ok\n");
431 res = (((u64) deltapm) * mult) >> 22;
432 do_div(res, 1000000);
433 printk(KERN_WARNING "APIC calibration not consistent "
434 "with PM Timer: %ldms instead of 100ms\n",
436 /* Correct the lapic counter value */
437 res = (((u64) delta) * pm_100ms);
438 do_div(res, deltapm);
439 printk(KERN_INFO "APIC delta adjusted to PM-Timer: "
440 "%lu (%ld)\n", (unsigned long) res, delta);
446 /* Calculate the scaled math multiplication factor */
447 lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS, 32);
448 lapic_clockevent.max_delta_ns =
449 clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
450 lapic_clockevent.min_delta_ns =
451 clockevent_delta2ns(0xF, &lapic_clockevent);
453 calibration_result = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
455 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
456 apic_printk(APIC_VERBOSE, "..... mult: %ld\n", lapic_clockevent.mult);
457 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
461 delta = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
462 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
464 (delta / LAPIC_CAL_LOOPS) / (1000000 / HZ),
465 (delta / LAPIC_CAL_LOOPS) % (1000000 / HZ));
468 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
470 calibration_result / (1000000 / HZ),
471 calibration_result % (1000000 / HZ));
473 local_apic_timer_verify_ok = 1;
476 * Do a sanity check on the APIC calibration result
478 if (calibration_result < (1000000 / HZ)) {
481 "APIC frequency too slow, disabling apic timer\n");
482 /* No broadcast on UP ! */
483 if (num_possible_cpus() > 1)
488 /* We trust the pm timer based calibration */
489 if (!pm_referenced) {
490 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
493 * Setup the apic timer manually
495 levt->event_handler = lapic_cal_handler;
496 lapic_timer_setup(CLOCK_EVT_MODE_PERIODIC, levt);
497 lapic_cal_loops = -1;
499 /* Let the interrupts run */
502 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
507 /* Stop the lapic timer */
508 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt);
513 deltaj = lapic_cal_j2 - lapic_cal_j1;
514 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
516 /* Check, if the jiffies result is consistent */
517 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
518 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
520 local_apic_timer_verify_ok = 0;
524 if (!local_apic_timer_verify_ok) {
526 "APIC timer disabled due to verification failure.\n");
527 /* No broadcast on UP ! */
528 if (num_possible_cpus() == 1)
532 * If nmi_watchdog is set to IO_APIC, we need the
533 * PIT/HPET going. Otherwise register lapic as a dummy
536 if (nmi_watchdog != NMI_IO_APIC)
537 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
539 printk(KERN_WARNING "APIC timer registered as dummy,"
540 " due to nmi_watchdog=1!\n");
543 /* Setup the lapic or request the broadcast */
547 void __devinit setup_secondary_APIC_clock(void)
553 * The guts of the apic timer interrupt
555 static void local_apic_timer_interrupt(void)
557 int cpu = smp_processor_id();
558 struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
561 * Normally we should not be here till LAPIC has been initialized but
562 * in some cases like kdump, its possible that there is a pending LAPIC
563 * timer interrupt from previous kernel's context and is delivered in
564 * new kernel the moment interrupts are enabled.
566 * Interrupts are enabled early and LAPIC is setup much later, hence
567 * its possible that when we get here evt->event_handler is NULL.
568 * Check for event_handler being NULL and discard the interrupt as
571 if (!evt->event_handler) {
573 "Spurious LAPIC timer interrupt on cpu %d\n", cpu);
575 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
580 * the NMI deadlock-detector uses this.
582 per_cpu(irq_stat, cpu).apic_timer_irqs++;
584 evt->event_handler(evt);
588 * Local APIC timer interrupt. This is the most natural way for doing
589 * local interrupts, but local timer interrupts can be emulated by
590 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
592 * [ if a single-CPU system runs an SMP kernel then we call the local
593 * interrupt as well. Thus we cannot inline the local irq ... ]
595 void smp_apic_timer_interrupt(struct pt_regs *regs)
597 struct pt_regs *old_regs = set_irq_regs(regs);
600 * NOTE! We'd better ACK the irq immediately,
601 * because timer handling can be slow.
605 * update_process_times() expects us to have done irq_enter().
606 * Besides, if we don't timer interrupts ignore the global
607 * interrupt lock, which is the WrongThing (tm) to do.
610 local_apic_timer_interrupt();
613 set_irq_regs(old_regs);
616 int setup_profiling_timer(unsigned int multiplier)
622 * Local APIC start and shutdown
626 * clear_local_APIC - shutdown the local APIC
628 * This is called, when a CPU is disabled and before rebooting, so the state of
629 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
630 * leftovers during boot.
632 void clear_local_APIC(void)
634 int maxlvt = lapic_get_maxlvt();
638 * Masking an LVT entry can trigger a local APIC error
639 * if the vector is zero. Mask LVTERR first to prevent this.
642 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
643 apic_write_around(APIC_LVTERR, v | APIC_LVT_MASKED);
646 * Careful: we have to set masks only first to deassert
647 * any level-triggered sources.
649 v = apic_read(APIC_LVTT);
650 apic_write_around(APIC_LVTT, v | APIC_LVT_MASKED);
651 v = apic_read(APIC_LVT0);
652 apic_write_around(APIC_LVT0, v | APIC_LVT_MASKED);
653 v = apic_read(APIC_LVT1);
654 apic_write_around(APIC_LVT1, v | APIC_LVT_MASKED);
656 v = apic_read(APIC_LVTPC);
657 apic_write_around(APIC_LVTPC, v | APIC_LVT_MASKED);
660 /* lets not touch this if we didn't frob it */
661 #ifdef CONFIG_X86_MCE_P4THERMAL
663 v = apic_read(APIC_LVTTHMR);
664 apic_write_around(APIC_LVTTHMR, v | APIC_LVT_MASKED);
668 * Clean APIC state for other OSs:
670 apic_write_around(APIC_LVTT, APIC_LVT_MASKED);
671 apic_write_around(APIC_LVT0, APIC_LVT_MASKED);
672 apic_write_around(APIC_LVT1, APIC_LVT_MASKED);
674 apic_write_around(APIC_LVTERR, APIC_LVT_MASKED);
676 apic_write_around(APIC_LVTPC, APIC_LVT_MASKED);
678 #ifdef CONFIG_X86_MCE_P4THERMAL
680 apic_write_around(APIC_LVTTHMR, APIC_LVT_MASKED);
682 /* Integrated APIC (!82489DX) ? */
683 if (lapic_is_integrated()) {
685 /* Clear ESR due to Pentium errata 3AP and 11AP */
686 apic_write(APIC_ESR, 0);
692 * disable_local_APIC - clear and disable the local APIC
694 void disable_local_APIC(void)
701 * Disable APIC (implies clearing of registers
704 value = apic_read(APIC_SPIV);
705 value &= ~APIC_SPIV_APIC_ENABLED;
706 apic_write_around(APIC_SPIV, value);
709 * When LAPIC was disabled by the BIOS and enabled by the kernel,
710 * restore the disabled state.
712 if (enabled_via_apicbase) {
715 rdmsr(MSR_IA32_APICBASE, l, h);
716 l &= ~MSR_IA32_APICBASE_ENABLE;
717 wrmsr(MSR_IA32_APICBASE, l, h);
722 * If Linux enabled the LAPIC against the BIOS default disable it down before
723 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
724 * not power-off. Additionally clear all LVT entries before disable_local_APIC
725 * for the case where Linux didn't enable the LAPIC.
727 void lapic_shutdown(void)
734 local_irq_save(flags);
737 if (enabled_via_apicbase)
738 disable_local_APIC();
740 local_irq_restore(flags);
744 * This is to verify that we're looking at a real local APIC.
745 * Check these against your board if the CPUs aren't getting
746 * started for no apparent reason.
748 int __init verify_local_APIC(void)
750 unsigned int reg0, reg1;
753 * The version register is read-only in a real APIC.
755 reg0 = apic_read(APIC_LVR);
756 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
757 apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
758 reg1 = apic_read(APIC_LVR);
759 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
762 * The two version reads above should print the same
763 * numbers. If the second one is different, then we
764 * poke at a non-APIC.
770 * Check if the version looks reasonably.
772 reg1 = GET_APIC_VERSION(reg0);
773 if (reg1 == 0x00 || reg1 == 0xff)
775 reg1 = lapic_get_maxlvt();
776 if (reg1 < 0x02 || reg1 == 0xff)
780 * The ID register is read/write in a real APIC.
782 reg0 = apic_read(APIC_ID);
783 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
786 * The next two are just to see if we have sane values.
787 * They're only really relevant if we're in Virtual Wire
788 * compatibility mode, but most boxes are anymore.
790 reg0 = apic_read(APIC_LVT0);
791 apic_printk(APIC_DEBUG, "Getting LVT0: %x\n", reg0);
792 reg1 = apic_read(APIC_LVT1);
793 apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
799 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
801 void __init sync_Arb_IDs(void)
804 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
807 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
812 apic_wait_icr_idle();
814 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
815 apic_write_around(APIC_ICR, APIC_DEST_ALLINC | APIC_INT_LEVELTRIG
820 * An initial setup of the virtual wire mode.
822 void __init init_bsp_APIC(void)
827 * Don't do the setup now if we have a SMP BIOS as the
828 * through-I/O-APIC virtual wire mode might be active.
830 if (smp_found_config || !cpu_has_apic)
834 * Do not trust the local APIC being empty at bootup.
841 value = apic_read(APIC_SPIV);
842 value &= ~APIC_VECTOR_MASK;
843 value |= APIC_SPIV_APIC_ENABLED;
845 /* This bit is reserved on P4/Xeon and should be cleared */
846 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
847 (boot_cpu_data.x86 == 15))
848 value &= ~APIC_SPIV_FOCUS_DISABLED;
850 value |= APIC_SPIV_FOCUS_DISABLED;
851 value |= SPURIOUS_APIC_VECTOR;
852 apic_write_around(APIC_SPIV, value);
855 * Set up the virtual wire mode.
857 apic_write_around(APIC_LVT0, APIC_DM_EXTINT);
859 if (!lapic_is_integrated()) /* 82489DX */
860 value |= APIC_LVT_LEVEL_TRIGGER;
861 apic_write_around(APIC_LVT1, value);
865 * setup_local_APIC - setup the local APIC
867 void __cpuinit setup_local_APIC(void)
869 unsigned long oldvalue, value, maxlvt, integrated;
872 /* Pound the ESR really hard over the head with a big hammer - mbligh */
874 apic_write(APIC_ESR, 0);
875 apic_write(APIC_ESR, 0);
876 apic_write(APIC_ESR, 0);
877 apic_write(APIC_ESR, 0);
880 integrated = lapic_is_integrated();
883 * Double-check whether this APIC is really registered.
885 if (!apic_id_registered())
889 * Intel recommends to set DFR, LDR and TPR before enabling
890 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
891 * document number 292116). So here it goes...
896 * Set Task Priority to 'accept all'. We never change this
899 value = apic_read(APIC_TASKPRI);
900 value &= ~APIC_TPRI_MASK;
901 apic_write_around(APIC_TASKPRI, value);
904 * After a crash, we no longer service the interrupts and a pending
905 * interrupt from previous kernel might still have ISR bit set.
907 * Most probably by now CPU has serviced that pending interrupt and
908 * it might not have done the ack_APIC_irq() because it thought,
909 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
910 * does not clear the ISR bit and cpu thinks it has already serivced
911 * the interrupt. Hence a vector might get locked. It was noticed
912 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
914 for (i = APIC_ISR_NR - 1; i >= 0; i--) {
915 value = apic_read(APIC_ISR + i*0x10);
916 for (j = 31; j >= 0; j--) {
923 * Now that we are all set up, enable the APIC
925 value = apic_read(APIC_SPIV);
926 value &= ~APIC_VECTOR_MASK;
930 value |= APIC_SPIV_APIC_ENABLED;
933 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
934 * certain networking cards. If high frequency interrupts are
935 * happening on a particular IOAPIC pin, plus the IOAPIC routing
936 * entry is masked/unmasked at a high rate as well then sooner or
937 * later IOAPIC line gets 'stuck', no more interrupts are received
938 * from the device. If focus CPU is disabled then the hang goes
941 * [ This bug can be reproduced easily with a level-triggered
942 * PCI Ne2000 networking cards and PII/PIII processors, dual
946 * Actually disabling the focus CPU check just makes the hang less
947 * frequent as it makes the interrupt distributon model be more
948 * like LRU than MRU (the short-term load is more even across CPUs).
949 * See also the comment in end_level_ioapic_irq(). --macro
952 /* Enable focus processor (bit==0) */
953 value &= ~APIC_SPIV_FOCUS_DISABLED;
956 * Set spurious IRQ vector
958 value |= SPURIOUS_APIC_VECTOR;
959 apic_write_around(APIC_SPIV, value);
964 * set up through-local-APIC on the BP's LINT0. This is not
965 * strictly necessary in pure symmetric-IO mode, but sometimes
966 * we delegate interrupts to the 8259A.
969 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
971 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
972 if (!smp_processor_id() && (pic_mode || !value)) {
973 value = APIC_DM_EXTINT;
974 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n",
977 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
978 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n",
981 apic_write_around(APIC_LVT0, value);
984 * only the BP should see the LINT1 NMI signal, obviously.
986 if (!smp_processor_id())
989 value = APIC_DM_NMI | APIC_LVT_MASKED;
990 if (!integrated) /* 82489DX */
991 value |= APIC_LVT_LEVEL_TRIGGER;
992 apic_write_around(APIC_LVT1, value);
994 if (integrated && !esr_disable) {
996 maxlvt = lapic_get_maxlvt();
997 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
998 apic_write(APIC_ESR, 0);
999 oldvalue = apic_read(APIC_ESR);
1001 /* enables sending errors */
1002 value = ERROR_APIC_VECTOR;
1003 apic_write_around(APIC_LVTERR, value);
1005 * spec says clear errors after enabling vector.
1008 apic_write(APIC_ESR, 0);
1009 value = apic_read(APIC_ESR);
1010 if (value != oldvalue)
1011 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1012 "vector: 0x%08lx after: 0x%08lx\n",
1017 * Something untraceable is creating bad interrupts on
1018 * secondary quads ... for the moment, just leave the
1019 * ESR disabled - we can't do anything useful with the
1020 * errors anyway - mbligh
1022 printk(KERN_INFO "Leaving ESR disabled.\n");
1024 printk(KERN_INFO "No ESR for 82489DX.\n");
1027 /* Disable the local apic timer */
1028 value = apic_read(APIC_LVTT);
1029 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1030 apic_write_around(APIC_LVTT, value);
1032 setup_apic_nmi_watchdog(NULL);
1037 * Detect and initialize APIC
1039 static int __init detect_init_APIC(void)
1043 /* Disabled by kernel option? */
1044 if (enable_local_apic < 0)
1047 switch (boot_cpu_data.x86_vendor) {
1048 case X86_VENDOR_AMD:
1049 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1050 (boot_cpu_data.x86 == 15))
1053 case X86_VENDOR_INTEL:
1054 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1055 (boot_cpu_data.x86 == 5 && cpu_has_apic))
1062 if (!cpu_has_apic) {
1064 * Over-ride BIOS and try to enable the local APIC only if
1065 * "lapic" specified.
1067 if (enable_local_apic <= 0) {
1068 printk(KERN_INFO "Local APIC disabled by BIOS -- "
1069 "you can enable it with \"lapic\"\n");
1073 * Some BIOSes disable the local APIC in the APIC_BASE
1074 * MSR. This can only be done in software for Intel P6 or later
1075 * and AMD K7 (Model > 1) or later.
1077 rdmsr(MSR_IA32_APICBASE, l, h);
1078 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1080 "Local APIC disabled by BIOS -- reenabling.\n");
1081 l &= ~MSR_IA32_APICBASE_BASE;
1082 l |= MSR_IA32_APICBASE_ENABLE | APIC_DEFAULT_PHYS_BASE;
1083 wrmsr(MSR_IA32_APICBASE, l, h);
1084 enabled_via_apicbase = 1;
1088 * The APIC feature bit should now be enabled
1091 features = cpuid_edx(1);
1092 if (!(features & (1 << X86_FEATURE_APIC))) {
1093 printk(KERN_WARNING "Could not enable APIC!\n");
1096 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1097 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1099 /* The BIOS may have set up the APIC at some other address */
1100 rdmsr(MSR_IA32_APICBASE, l, h);
1101 if (l & MSR_IA32_APICBASE_ENABLE)
1102 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1104 if (nmi_watchdog != NMI_NONE && nmi_watchdog != NMI_DISABLED)
1105 nmi_watchdog = NMI_LOCAL_APIC;
1107 printk(KERN_INFO "Found and enabled local APIC!\n");
1114 printk(KERN_INFO "No local APIC present or hardware disabled\n");
1119 * init_apic_mappings - initialize APIC mappings
1121 void __init init_apic_mappings(void)
1123 unsigned long apic_phys;
1126 * If no local APIC can be found then set up a fake all
1127 * zeroes page to simulate the local APIC and another
1128 * one for the IO-APIC.
1130 if (!smp_found_config && detect_init_APIC()) {
1131 apic_phys = (unsigned long) alloc_bootmem_pages(PAGE_SIZE);
1132 apic_phys = __pa(apic_phys);
1134 apic_phys = mp_lapic_addr;
1136 set_fixmap_nocache(FIX_APIC_BASE, apic_phys);
1137 printk(KERN_DEBUG "mapped APIC to %08lx (%08lx)\n", APIC_BASE,
1141 * Fetch the APIC ID of the BSP in case we have a
1142 * default configuration (or the MP table is broken).
1144 if (boot_cpu_physical_apicid == -1U)
1145 boot_cpu_physical_apicid = GET_APIC_ID(apic_read(APIC_ID));
1147 #ifdef CONFIG_X86_IO_APIC
1149 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
1152 for (i = 0; i < nr_ioapics; i++) {
1153 if (smp_found_config) {
1154 ioapic_phys = mp_ioapics[i].mpc_apicaddr;
1157 "WARNING: bogus zero IO-APIC "
1158 "address found in MPTABLE, "
1159 "disabling IO/APIC support!\n");
1160 smp_found_config = 0;
1161 skip_ioapic_setup = 1;
1162 goto fake_ioapic_page;
1166 ioapic_phys = (unsigned long)
1167 alloc_bootmem_pages(PAGE_SIZE);
1168 ioapic_phys = __pa(ioapic_phys);
1170 set_fixmap_nocache(idx, ioapic_phys);
1171 printk(KERN_DEBUG "mapped IOAPIC to %08lx (%08lx)\n",
1172 __fix_to_virt(idx), ioapic_phys);
1180 * This initializes the IO-APIC and APIC hardware if this is
1183 int __init APIC_init_uniprocessor(void)
1185 if (enable_local_apic < 0)
1186 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1188 if (!smp_found_config && !cpu_has_apic)
1192 * Complain if the BIOS pretends there is one.
1194 if (!cpu_has_apic &&
1195 APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
1196 printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
1197 boot_cpu_physical_apicid);
1198 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1202 verify_local_APIC();
1207 * Hack: In case of kdump, after a crash, kernel might be booting
1208 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
1209 * might be zero if read from MP tables. Get it from LAPIC.
1211 #ifdef CONFIG_CRASH_DUMP
1212 boot_cpu_physical_apicid = GET_APIC_ID(apic_read(APIC_ID));
1214 phys_cpu_present_map = physid_mask_of_physid(boot_cpu_physical_apicid);
1218 #ifdef CONFIG_X86_IO_APIC
1219 if (smp_found_config)
1220 if (!skip_ioapic_setup && nr_ioapics)
1229 * Local APIC interrupts
1233 * This interrupt should _never_ happen with our APIC/SMP architecture
1235 void smp_spurious_interrupt(struct pt_regs *regs)
1241 * Check if this really is a spurious interrupt and ACK it
1242 * if it is a vectored one. Just in case...
1243 * Spurious interrupts should not be ACKed.
1245 v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
1246 if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
1249 /* see sw-dev-man vol 3, chapter 7.4.13.5 */
1250 printk(KERN_INFO "spurious APIC interrupt on CPU#%d, "
1251 "should never happen.\n", smp_processor_id());
1252 __get_cpu_var(irq_stat).irq_spurious_count++;
1257 * This interrupt should never happen with our APIC/SMP architecture
1259 void smp_error_interrupt(struct pt_regs *regs)
1261 unsigned long v, v1;
1264 /* First tickle the hardware, only then report what went on. -- REW */
1265 v = apic_read(APIC_ESR);
1266 apic_write(APIC_ESR, 0);
1267 v1 = apic_read(APIC_ESR);
1269 atomic_inc(&irq_err_count);
1271 /* Here is what the APIC error bits mean:
1274 2: Send accept error
1275 3: Receive accept error
1277 5: Send illegal vector
1278 6: Received illegal vector
1279 7: Illegal register address
1281 printk(KERN_DEBUG "APIC error on CPU%d: %02lx(%02lx)\n",
1282 smp_processor_id(), v , v1);
1287 * Initialize APIC interrupts
1289 void __init apic_intr_init(void)
1294 /* self generated IPI for local APIC timer */
1295 set_intr_gate(LOCAL_TIMER_VECTOR, apic_timer_interrupt);
1297 /* IPI vectors for APIC spurious and error interrupts */
1298 set_intr_gate(SPURIOUS_APIC_VECTOR, spurious_interrupt);
1299 set_intr_gate(ERROR_APIC_VECTOR, error_interrupt);
1301 /* thermal monitor LVT interrupt */
1302 #ifdef CONFIG_X86_MCE_P4THERMAL
1303 set_intr_gate(THERMAL_APIC_VECTOR, thermal_interrupt);
1308 * connect_bsp_APIC - attach the APIC to the interrupt system
1310 void __init connect_bsp_APIC(void)
1314 * Do not trust the local APIC being empty at bootup.
1318 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
1319 * local APIC to INT and NMI lines.
1321 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
1322 "enabling APIC mode.\n");
1330 * disconnect_bsp_APIC - detach the APIC from the interrupt system
1331 * @virt_wire_setup: indicates, whether virtual wire mode is selected
1333 * Virtual wire mode is necessary to deliver legacy interrupts even when the
1336 void disconnect_bsp_APIC(int virt_wire_setup)
1340 * Put the board back into PIC mode (has an effect only on
1341 * certain older boards). Note that APIC interrupts, including
1342 * IPIs, won't work beyond this point! The only exception are
1345 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
1346 "entering PIC mode.\n");
1350 /* Go back to Virtual Wire compatibility mode */
1351 unsigned long value;
1353 /* For the spurious interrupt use vector F, and enable it */
1354 value = apic_read(APIC_SPIV);
1355 value &= ~APIC_VECTOR_MASK;
1356 value |= APIC_SPIV_APIC_ENABLED;
1358 apic_write_around(APIC_SPIV, value);
1360 if (!virt_wire_setup) {
1362 * For LVT0 make it edge triggered, active high,
1363 * external and enabled
1365 value = apic_read(APIC_LVT0);
1366 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1367 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1368 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1369 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1370 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
1371 apic_write_around(APIC_LVT0, value);
1374 apic_write_around(APIC_LVT0, APIC_LVT_MASKED);
1378 * For LVT1 make it edge triggered, active high, nmi and
1381 value = apic_read(APIC_LVT1);
1383 APIC_MODE_MASK | APIC_SEND_PENDING |
1384 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1385 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1386 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1387 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
1388 apic_write_around(APIC_LVT1, value);
1399 /* r/w apic fields */
1400 unsigned int apic_id;
1401 unsigned int apic_taskpri;
1402 unsigned int apic_ldr;
1403 unsigned int apic_dfr;
1404 unsigned int apic_spiv;
1405 unsigned int apic_lvtt;
1406 unsigned int apic_lvtpc;
1407 unsigned int apic_lvt0;
1408 unsigned int apic_lvt1;
1409 unsigned int apic_lvterr;
1410 unsigned int apic_tmict;
1411 unsigned int apic_tdcr;
1412 unsigned int apic_thmr;
1415 static int lapic_suspend(struct sys_device *dev, pm_message_t state)
1417 unsigned long flags;
1420 if (!apic_pm_state.active)
1423 maxlvt = lapic_get_maxlvt();
1425 apic_pm_state.apic_id = apic_read(APIC_ID);
1426 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
1427 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
1428 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
1429 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
1430 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
1432 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
1433 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
1434 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
1435 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
1436 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
1437 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
1438 #ifdef CONFIG_X86_MCE_P4THERMAL
1440 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
1443 local_irq_save(flags);
1444 disable_local_APIC();
1445 local_irq_restore(flags);
1449 static int lapic_resume(struct sys_device *dev)
1452 unsigned long flags;
1455 if (!apic_pm_state.active)
1458 maxlvt = lapic_get_maxlvt();
1460 local_irq_save(flags);
1463 * Make sure the APICBASE points to the right address
1465 * FIXME! This will be wrong if we ever support suspend on
1466 * SMP! We'll need to do this as part of the CPU restore!
1468 rdmsr(MSR_IA32_APICBASE, l, h);
1469 l &= ~MSR_IA32_APICBASE_BASE;
1470 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
1471 wrmsr(MSR_IA32_APICBASE, l, h);
1473 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
1474 apic_write(APIC_ID, apic_pm_state.apic_id);
1475 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
1476 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
1477 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
1478 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
1479 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
1480 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
1481 #ifdef CONFIG_X86_MCE_P4THERMAL
1483 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
1486 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
1487 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
1488 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
1489 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
1490 apic_write(APIC_ESR, 0);
1491 apic_read(APIC_ESR);
1492 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
1493 apic_write(APIC_ESR, 0);
1494 apic_read(APIC_ESR);
1495 local_irq_restore(flags);
1500 * This device has no shutdown method - fully functioning local APICs
1501 * are needed on every CPU up until machine_halt/restart/poweroff.
1504 static struct sysdev_class lapic_sysclass = {
1506 .resume = lapic_resume,
1507 .suspend = lapic_suspend,
1510 static struct sys_device device_lapic = {
1512 .cls = &lapic_sysclass,
1515 static void __devinit apic_pm_activate(void)
1517 apic_pm_state.active = 1;
1520 static int __init init_lapic_sysfs(void)
1526 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
1528 error = sysdev_class_register(&lapic_sysclass);
1530 error = sysdev_register(&device_lapic);
1533 device_initcall(init_lapic_sysfs);
1535 #else /* CONFIG_PM */
1537 static void apic_pm_activate(void) { }
1539 #endif /* CONFIG_PM */
1542 * APIC command line parameters
1544 static int __init parse_lapic(char *arg)
1546 enable_local_apic = 1;
1549 early_param("lapic", parse_lapic);
1551 static int __init parse_nolapic(char *arg)
1553 enable_local_apic = -1;
1554 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1557 early_param("nolapic", parse_nolapic);
1559 static int __init parse_disable_lapic_timer(char *arg)
1561 local_apic_timer_disabled = 1;
1564 early_param("nolapic_timer", parse_disable_lapic_timer);
1566 static int __init parse_lapic_timer_c2_ok(char *arg)
1568 local_apic_timer_c2_ok = 1;
1571 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
1573 static int __init apic_set_verbosity(char *str)
1575 if (strcmp("debug", str) == 0)
1576 apic_verbosity = APIC_DEBUG;
1577 else if (strcmp("verbose", str) == 0)
1578 apic_verbosity = APIC_VERBOSE;
1581 __setup("apic=", apic_set_verbosity);