Loading...
1/*
2 * Local APIC handling, local APIC timers
3 *
4 * (c) 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
5 *
6 * Fixes
7 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
8 * thanks to Eric Gilmore
9 * and Rolf G. Tews
10 * for testing these extensively.
11 * Maciej W. Rozycki : Various updates and fixes.
12 * Mikael Pettersson : Power Management for UP-APIC.
13 * Pavel Machek and
14 * Mikael Pettersson : PM converted to driver model.
15 */
16
17#include <linux/perf_event.h>
18#include <linux/kernel_stat.h>
19#include <linux/mc146818rtc.h>
20#include <linux/acpi_pmtmr.h>
21#include <linux/clockchips.h>
22#include <linux/interrupt.h>
23#include <linux/bootmem.h>
24#include <linux/ftrace.h>
25#include <linux/ioport.h>
26#include <linux/module.h>
27#include <linux/syscore_ops.h>
28#include <linux/delay.h>
29#include <linux/timex.h>
30#include <linux/i8253.h>
31#include <linux/dmar.h>
32#include <linux/init.h>
33#include <linux/cpu.h>
34#include <linux/dmi.h>
35#include <linux/smp.h>
36#include <linux/mm.h>
37
38#include <asm/perf_event.h>
39#include <asm/x86_init.h>
40#include <asm/pgalloc.h>
41#include <linux/atomic.h>
42#include <asm/mpspec.h>
43#include <asm/i8259.h>
44#include <asm/proto.h>
45#include <asm/apic.h>
46#include <asm/io_apic.h>
47#include <asm/desc.h>
48#include <asm/hpet.h>
49#include <asm/idle.h>
50#include <asm/mtrr.h>
51#include <asm/time.h>
52#include <asm/smp.h>
53#include <asm/mce.h>
54#include <asm/tsc.h>
55#include <asm/hypervisor.h>
56
57unsigned int num_processors;
58
59unsigned disabled_cpus __cpuinitdata;
60
61/* Processor that is doing the boot up */
62unsigned int boot_cpu_physical_apicid = -1U;
63
64/*
65 * The highest APIC ID seen during enumeration.
66 */
67unsigned int max_physical_apicid;
68
69/*
70 * Bitmask of physically existing CPUs:
71 */
72physid_mask_t phys_cpu_present_map;
73
74/*
75 * Map cpu index to physical APIC ID
76 */
77DEFINE_EARLY_PER_CPU(u16, x86_cpu_to_apicid, BAD_APICID);
78DEFINE_EARLY_PER_CPU(u16, x86_bios_cpu_apicid, BAD_APICID);
79EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
80EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
81
82#ifdef CONFIG_X86_32
83
84/*
85 * On x86_32, the mapping between cpu and logical apicid may vary
86 * depending on apic in use. The following early percpu variable is
87 * used for the mapping. This is where the behaviors of x86_64 and 32
88 * actually diverge. Let's keep it ugly for now.
89 */
90DEFINE_EARLY_PER_CPU(int, x86_cpu_to_logical_apicid, BAD_APICID);
91
92/*
93 * Knob to control our willingness to enable the local APIC.
94 *
95 * +1=force-enable
96 */
97static int force_enable_local_apic __initdata;
98/*
99 * APIC command line parameters
100 */
101static int __init parse_lapic(char *arg)
102{
103 force_enable_local_apic = 1;
104 return 0;
105}
106early_param("lapic", parse_lapic);
107/* Local APIC was disabled by the BIOS and enabled by the kernel */
108static int enabled_via_apicbase;
109
110/*
111 * Handle interrupt mode configuration register (IMCR).
112 * This register controls whether the interrupt signals
113 * that reach the BSP come from the master PIC or from the
114 * local APIC. Before entering Symmetric I/O Mode, either
115 * the BIOS or the operating system must switch out of
116 * PIC Mode by changing the IMCR.
117 */
118static inline void imcr_pic_to_apic(void)
119{
120 /* select IMCR register */
121 outb(0x70, 0x22);
122 /* NMI and 8259 INTR go through APIC */
123 outb(0x01, 0x23);
124}
125
126static inline void imcr_apic_to_pic(void)
127{
128 /* select IMCR register */
129 outb(0x70, 0x22);
130 /* NMI and 8259 INTR go directly to BSP */
131 outb(0x00, 0x23);
132}
133#endif
134
135#ifdef CONFIG_X86_64
136static int apic_calibrate_pmtmr __initdata;
137static __init int setup_apicpmtimer(char *s)
138{
139 apic_calibrate_pmtmr = 1;
140 notsc_setup(NULL);
141 return 0;
142}
143__setup("apicpmtimer", setup_apicpmtimer);
144#endif
145
146int x2apic_mode;
147#ifdef CONFIG_X86_X2APIC
148/* x2apic enabled before OS handover */
149static int x2apic_preenabled;
150static __init int setup_nox2apic(char *str)
151{
152 if (x2apic_enabled()) {
153 pr_warning("Bios already enabled x2apic, "
154 "can't enforce nox2apic");
155 return 0;
156 }
157
158 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
159 return 0;
160}
161early_param("nox2apic", setup_nox2apic);
162#endif
163
164unsigned long mp_lapic_addr;
165int disable_apic;
166/* Disable local APIC timer from the kernel commandline or via dmi quirk */
167static int disable_apic_timer __initdata;
168/* Local APIC timer works in C2 */
169int local_apic_timer_c2_ok;
170EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
171
172int first_system_vector = 0xfe;
173
174/*
175 * Debug level, exported for io_apic.c
176 */
177unsigned int apic_verbosity;
178
179int pic_mode;
180
181/* Have we found an MP table */
182int smp_found_config;
183
184static struct resource lapic_resource = {
185 .name = "Local APIC",
186 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
187};
188
189static unsigned int calibration_result;
190
191static void apic_pm_activate(void);
192
193static unsigned long apic_phys;
194
195/*
196 * Get the LAPIC version
197 */
198static inline int lapic_get_version(void)
199{
200 return GET_APIC_VERSION(apic_read(APIC_LVR));
201}
202
203/*
204 * Check, if the APIC is integrated or a separate chip
205 */
206static inline int lapic_is_integrated(void)
207{
208#ifdef CONFIG_X86_64
209 return 1;
210#else
211 return APIC_INTEGRATED(lapic_get_version());
212#endif
213}
214
215/*
216 * Check, whether this is a modern or a first generation APIC
217 */
218static int modern_apic(void)
219{
220 /* AMD systems use old APIC versions, so check the CPU */
221 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
222 boot_cpu_data.x86 >= 0xf)
223 return 1;
224 return lapic_get_version() >= 0x14;
225}
226
227/*
228 * right after this call apic become NOOP driven
229 * so apic->write/read doesn't do anything
230 */
231static void __init apic_disable(void)
232{
233 pr_info("APIC: switched to apic NOOP\n");
234 apic = &apic_noop;
235}
236
237void native_apic_wait_icr_idle(void)
238{
239 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
240 cpu_relax();
241}
242
243u32 native_safe_apic_wait_icr_idle(void)
244{
245 u32 send_status;
246 int timeout;
247
248 timeout = 0;
249 do {
250 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
251 if (!send_status)
252 break;
253 udelay(100);
254 } while (timeout++ < 1000);
255
256 return send_status;
257}
258
259void native_apic_icr_write(u32 low, u32 id)
260{
261 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
262 apic_write(APIC_ICR, low);
263}
264
265u64 native_apic_icr_read(void)
266{
267 u32 icr1, icr2;
268
269 icr2 = apic_read(APIC_ICR2);
270 icr1 = apic_read(APIC_ICR);
271
272 return icr1 | ((u64)icr2 << 32);
273}
274
275#ifdef CONFIG_X86_32
276/**
277 * get_physical_broadcast - Get number of physical broadcast IDs
278 */
279int get_physical_broadcast(void)
280{
281 return modern_apic() ? 0xff : 0xf;
282}
283#endif
284
285/**
286 * lapic_get_maxlvt - get the maximum number of local vector table entries
287 */
288int lapic_get_maxlvt(void)
289{
290 unsigned int v;
291
292 v = apic_read(APIC_LVR);
293 /*
294 * - we always have APIC integrated on 64bit mode
295 * - 82489DXs do not report # of LVT entries
296 */
297 return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
298}
299
300/*
301 * Local APIC timer
302 */
303
304/* Clock divisor */
305#define APIC_DIVISOR 16
306
307/*
308 * This function sets up the local APIC timer, with a timeout of
309 * 'clocks' APIC bus clock. During calibration we actually call
310 * this function twice on the boot CPU, once with a bogus timeout
311 * value, second time for real. The other (noncalibrating) CPUs
312 * call this function only once, with the real, calibrated value.
313 *
314 * We do reads before writes even if unnecessary, to get around the
315 * P5 APIC double write bug.
316 */
317static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
318{
319 unsigned int lvtt_value, tmp_value;
320
321 lvtt_value = LOCAL_TIMER_VECTOR;
322 if (!oneshot)
323 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
324 if (!lapic_is_integrated())
325 lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
326
327 if (!irqen)
328 lvtt_value |= APIC_LVT_MASKED;
329
330 apic_write(APIC_LVTT, lvtt_value);
331
332 /*
333 * Divide PICLK by 16
334 */
335 tmp_value = apic_read(APIC_TDCR);
336 apic_write(APIC_TDCR,
337 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
338 APIC_TDR_DIV_16);
339
340 if (!oneshot)
341 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
342}
343
344/*
345 * Setup extended LVT, AMD specific
346 *
347 * Software should use the LVT offsets the BIOS provides. The offsets
348 * are determined by the subsystems using it like those for MCE
349 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts
350 * are supported. Beginning with family 10h at least 4 offsets are
351 * available.
352 *
353 * Since the offsets must be consistent for all cores, we keep track
354 * of the LVT offsets in software and reserve the offset for the same
355 * vector also to be used on other cores. An offset is freed by
356 * setting the entry to APIC_EILVT_MASKED.
357 *
358 * If the BIOS is right, there should be no conflicts. Otherwise a
359 * "[Firmware Bug]: ..." error message is generated. However, if
360 * software does not properly determines the offsets, it is not
361 * necessarily a BIOS bug.
362 */
363
364static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
365
366static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
367{
368 return (old & APIC_EILVT_MASKED)
369 || (new == APIC_EILVT_MASKED)
370 || ((new & ~APIC_EILVT_MASKED) == old);
371}
372
373static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
374{
375 unsigned int rsvd; /* 0: uninitialized */
376
377 if (offset >= APIC_EILVT_NR_MAX)
378 return ~0;
379
380 rsvd = atomic_read(&eilvt_offsets[offset]) & ~APIC_EILVT_MASKED;
381 do {
382 if (rsvd &&
383 !eilvt_entry_is_changeable(rsvd, new))
384 /* may not change if vectors are different */
385 return rsvd;
386 rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
387 } while (rsvd != new);
388
389 return new;
390}
391
392/*
393 * If mask=1, the LVT entry does not generate interrupts while mask=0
394 * enables the vector. See also the BKDGs. Must be called with
395 * preemption disabled.
396 */
397
398int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
399{
400 unsigned long reg = APIC_EILVTn(offset);
401 unsigned int new, old, reserved;
402
403 new = (mask << 16) | (msg_type << 8) | vector;
404 old = apic_read(reg);
405 reserved = reserve_eilvt_offset(offset, new);
406
407 if (reserved != new) {
408 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
409 "vector 0x%x, but the register is already in use for "
410 "vector 0x%x on another cpu\n",
411 smp_processor_id(), reg, offset, new, reserved);
412 return -EINVAL;
413 }
414
415 if (!eilvt_entry_is_changeable(old, new)) {
416 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
417 "vector 0x%x, but the register is already in use for "
418 "vector 0x%x on this cpu\n",
419 smp_processor_id(), reg, offset, new, old);
420 return -EBUSY;
421 }
422
423 apic_write(reg, new);
424
425 return 0;
426}
427EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
428
429/*
430 * Program the next event, relative to now
431 */
432static int lapic_next_event(unsigned long delta,
433 struct clock_event_device *evt)
434{
435 apic_write(APIC_TMICT, delta);
436 return 0;
437}
438
439/*
440 * Setup the lapic timer in periodic or oneshot mode
441 */
442static void lapic_timer_setup(enum clock_event_mode mode,
443 struct clock_event_device *evt)
444{
445 unsigned long flags;
446 unsigned int v;
447
448 /* Lapic used as dummy for broadcast ? */
449 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
450 return;
451
452 local_irq_save(flags);
453
454 switch (mode) {
455 case CLOCK_EVT_MODE_PERIODIC:
456 case CLOCK_EVT_MODE_ONESHOT:
457 __setup_APIC_LVTT(calibration_result,
458 mode != CLOCK_EVT_MODE_PERIODIC, 1);
459 break;
460 case CLOCK_EVT_MODE_UNUSED:
461 case CLOCK_EVT_MODE_SHUTDOWN:
462 v = apic_read(APIC_LVTT);
463 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
464 apic_write(APIC_LVTT, v);
465 apic_write(APIC_TMICT, 0);
466 break;
467 case CLOCK_EVT_MODE_RESUME:
468 /* Nothing to do here */
469 break;
470 }
471
472 local_irq_restore(flags);
473}
474
475/*
476 * Local APIC timer broadcast function
477 */
478static void lapic_timer_broadcast(const struct cpumask *mask)
479{
480#ifdef CONFIG_SMP
481 apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
482#endif
483}
484
485
486/*
487 * The local apic timer can be used for any function which is CPU local.
488 */
489static struct clock_event_device lapic_clockevent = {
490 .name = "lapic",
491 .features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
492 | CLOCK_EVT_FEAT_C3STOP | CLOCK_EVT_FEAT_DUMMY,
493 .shift = 32,
494 .set_mode = lapic_timer_setup,
495 .set_next_event = lapic_next_event,
496 .broadcast = lapic_timer_broadcast,
497 .rating = 100,
498 .irq = -1,
499};
500static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
501
502/*
503 * Setup the local APIC timer for this CPU. Copy the initialized values
504 * of the boot CPU and register the clock event in the framework.
505 */
506static void __cpuinit setup_APIC_timer(void)
507{
508 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
509
510 if (this_cpu_has(X86_FEATURE_ARAT)) {
511 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
512 /* Make LAPIC timer preferrable over percpu HPET */
513 lapic_clockevent.rating = 150;
514 }
515
516 memcpy(levt, &lapic_clockevent, sizeof(*levt));
517 levt->cpumask = cpumask_of(smp_processor_id());
518
519 clockevents_register_device(levt);
520}
521
522/*
523 * In this functions we calibrate APIC bus clocks to the external timer.
524 *
525 * We want to do the calibration only once since we want to have local timer
526 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
527 * frequency.
528 *
529 * This was previously done by reading the PIT/HPET and waiting for a wrap
530 * around to find out, that a tick has elapsed. I have a box, where the PIT
531 * readout is broken, so it never gets out of the wait loop again. This was
532 * also reported by others.
533 *
534 * Monitoring the jiffies value is inaccurate and the clockevents
535 * infrastructure allows us to do a simple substitution of the interrupt
536 * handler.
537 *
538 * The calibration routine also uses the pm_timer when possible, as the PIT
539 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
540 * back to normal later in the boot process).
541 */
542
543#define LAPIC_CAL_LOOPS (HZ/10)
544
545static __initdata int lapic_cal_loops = -1;
546static __initdata long lapic_cal_t1, lapic_cal_t2;
547static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
548static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
549static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
550
551/*
552 * Temporary interrupt handler.
553 */
554static void __init lapic_cal_handler(struct clock_event_device *dev)
555{
556 unsigned long long tsc = 0;
557 long tapic = apic_read(APIC_TMCCT);
558 unsigned long pm = acpi_pm_read_early();
559
560 if (cpu_has_tsc)
561 rdtscll(tsc);
562
563 switch (lapic_cal_loops++) {
564 case 0:
565 lapic_cal_t1 = tapic;
566 lapic_cal_tsc1 = tsc;
567 lapic_cal_pm1 = pm;
568 lapic_cal_j1 = jiffies;
569 break;
570
571 case LAPIC_CAL_LOOPS:
572 lapic_cal_t2 = tapic;
573 lapic_cal_tsc2 = tsc;
574 if (pm < lapic_cal_pm1)
575 pm += ACPI_PM_OVRRUN;
576 lapic_cal_pm2 = pm;
577 lapic_cal_j2 = jiffies;
578 break;
579 }
580}
581
582static int __init
583calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
584{
585 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
586 const long pm_thresh = pm_100ms / 100;
587 unsigned long mult;
588 u64 res;
589
590#ifndef CONFIG_X86_PM_TIMER
591 return -1;
592#endif
593
594 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
595
596 /* Check, if the PM timer is available */
597 if (!deltapm)
598 return -1;
599
600 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
601
602 if (deltapm > (pm_100ms - pm_thresh) &&
603 deltapm < (pm_100ms + pm_thresh)) {
604 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
605 return 0;
606 }
607
608 res = (((u64)deltapm) * mult) >> 22;
609 do_div(res, 1000000);
610 pr_warning("APIC calibration not consistent "
611 "with PM-Timer: %ldms instead of 100ms\n",(long)res);
612
613 /* Correct the lapic counter value */
614 res = (((u64)(*delta)) * pm_100ms);
615 do_div(res, deltapm);
616 pr_info("APIC delta adjusted to PM-Timer: "
617 "%lu (%ld)\n", (unsigned long)res, *delta);
618 *delta = (long)res;
619
620 /* Correct the tsc counter value */
621 if (cpu_has_tsc) {
622 res = (((u64)(*deltatsc)) * pm_100ms);
623 do_div(res, deltapm);
624 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
625 "PM-Timer: %lu (%ld)\n",
626 (unsigned long)res, *deltatsc);
627 *deltatsc = (long)res;
628 }
629
630 return 0;
631}
632
633static int __init calibrate_APIC_clock(void)
634{
635 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
636 void (*real_handler)(struct clock_event_device *dev);
637 unsigned long deltaj;
638 long delta, deltatsc;
639 int pm_referenced = 0;
640
641 local_irq_disable();
642
643 /* Replace the global interrupt handler */
644 real_handler = global_clock_event->event_handler;
645 global_clock_event->event_handler = lapic_cal_handler;
646
647 /*
648 * Setup the APIC counter to maximum. There is no way the lapic
649 * can underflow in the 100ms detection time frame
650 */
651 __setup_APIC_LVTT(0xffffffff, 0, 0);
652
653 /* Let the interrupts run */
654 local_irq_enable();
655
656 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
657 cpu_relax();
658
659 local_irq_disable();
660
661 /* Restore the real event handler */
662 global_clock_event->event_handler = real_handler;
663
664 /* Build delta t1-t2 as apic timer counts down */
665 delta = lapic_cal_t1 - lapic_cal_t2;
666 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
667
668 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
669
670 /* we trust the PM based calibration if possible */
671 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
672 &delta, &deltatsc);
673
674 /* Calculate the scaled math multiplication factor */
675 lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
676 lapic_clockevent.shift);
677 lapic_clockevent.max_delta_ns =
678 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
679 lapic_clockevent.min_delta_ns =
680 clockevent_delta2ns(0xF, &lapic_clockevent);
681
682 calibration_result = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
683
684 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
685 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
686 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
687 calibration_result);
688
689 if (cpu_has_tsc) {
690 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
691 "%ld.%04ld MHz.\n",
692 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
693 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
694 }
695
696 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
697 "%u.%04u MHz.\n",
698 calibration_result / (1000000 / HZ),
699 calibration_result % (1000000 / HZ));
700
701 /*
702 * Do a sanity check on the APIC calibration result
703 */
704 if (calibration_result < (1000000 / HZ)) {
705 local_irq_enable();
706 pr_warning("APIC frequency too slow, disabling apic timer\n");
707 return -1;
708 }
709
710 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
711
712 /*
713 * PM timer calibration failed or not turned on
714 * so lets try APIC timer based calibration
715 */
716 if (!pm_referenced) {
717 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
718
719 /*
720 * Setup the apic timer manually
721 */
722 levt->event_handler = lapic_cal_handler;
723 lapic_timer_setup(CLOCK_EVT_MODE_PERIODIC, levt);
724 lapic_cal_loops = -1;
725
726 /* Let the interrupts run */
727 local_irq_enable();
728
729 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
730 cpu_relax();
731
732 /* Stop the lapic timer */
733 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt);
734
735 /* Jiffies delta */
736 deltaj = lapic_cal_j2 - lapic_cal_j1;
737 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
738
739 /* Check, if the jiffies result is consistent */
740 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
741 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
742 else
743 levt->features |= CLOCK_EVT_FEAT_DUMMY;
744 } else
745 local_irq_enable();
746
747 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
748 pr_warning("APIC timer disabled due to verification failure\n");
749 return -1;
750 }
751
752 return 0;
753}
754
755/*
756 * Setup the boot APIC
757 *
758 * Calibrate and verify the result.
759 */
760void __init setup_boot_APIC_clock(void)
761{
762 /*
763 * The local apic timer can be disabled via the kernel
764 * commandline or from the CPU detection code. Register the lapic
765 * timer as a dummy clock event source on SMP systems, so the
766 * broadcast mechanism is used. On UP systems simply ignore it.
767 */
768 if (disable_apic_timer) {
769 pr_info("Disabling APIC timer\n");
770 /* No broadcast on UP ! */
771 if (num_possible_cpus() > 1) {
772 lapic_clockevent.mult = 1;
773 setup_APIC_timer();
774 }
775 return;
776 }
777
778 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
779 "calibrating APIC timer ...\n");
780
781 if (calibrate_APIC_clock()) {
782 /* No broadcast on UP ! */
783 if (num_possible_cpus() > 1)
784 setup_APIC_timer();
785 return;
786 }
787
788 /*
789 * If nmi_watchdog is set to IO_APIC, we need the
790 * PIT/HPET going. Otherwise register lapic as a dummy
791 * device.
792 */
793 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
794
795 /* Setup the lapic or request the broadcast */
796 setup_APIC_timer();
797}
798
799void __cpuinit setup_secondary_APIC_clock(void)
800{
801 setup_APIC_timer();
802}
803
804/*
805 * The guts of the apic timer interrupt
806 */
807static void local_apic_timer_interrupt(void)
808{
809 int cpu = smp_processor_id();
810 struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
811
812 /*
813 * Normally we should not be here till LAPIC has been initialized but
814 * in some cases like kdump, its possible that there is a pending LAPIC
815 * timer interrupt from previous kernel's context and is delivered in
816 * new kernel the moment interrupts are enabled.
817 *
818 * Interrupts are enabled early and LAPIC is setup much later, hence
819 * its possible that when we get here evt->event_handler is NULL.
820 * Check for event_handler being NULL and discard the interrupt as
821 * spurious.
822 */
823 if (!evt->event_handler) {
824 pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
825 /* Switch it off */
826 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
827 return;
828 }
829
830 /*
831 * the NMI deadlock-detector uses this.
832 */
833 inc_irq_stat(apic_timer_irqs);
834
835 evt->event_handler(evt);
836}
837
838/*
839 * Local APIC timer interrupt. This is the most natural way for doing
840 * local interrupts, but local timer interrupts can be emulated by
841 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
842 *
843 * [ if a single-CPU system runs an SMP kernel then we call the local
844 * interrupt as well. Thus we cannot inline the local irq ... ]
845 */
846void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
847{
848 struct pt_regs *old_regs = set_irq_regs(regs);
849
850 /*
851 * NOTE! We'd better ACK the irq immediately,
852 * because timer handling can be slow.
853 */
854 ack_APIC_irq();
855 /*
856 * update_process_times() expects us to have done irq_enter().
857 * Besides, if we don't timer interrupts ignore the global
858 * interrupt lock, which is the WrongThing (tm) to do.
859 */
860 exit_idle();
861 irq_enter();
862 local_apic_timer_interrupt();
863 irq_exit();
864
865 set_irq_regs(old_regs);
866}
867
868int setup_profiling_timer(unsigned int multiplier)
869{
870 return -EINVAL;
871}
872
873/*
874 * Local APIC start and shutdown
875 */
876
877/**
878 * clear_local_APIC - shutdown the local APIC
879 *
880 * This is called, when a CPU is disabled and before rebooting, so the state of
881 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
882 * leftovers during boot.
883 */
884void clear_local_APIC(void)
885{
886 int maxlvt;
887 u32 v;
888
889 /* APIC hasn't been mapped yet */
890 if (!x2apic_mode && !apic_phys)
891 return;
892
893 maxlvt = lapic_get_maxlvt();
894 /*
895 * Masking an LVT entry can trigger a local APIC error
896 * if the vector is zero. Mask LVTERR first to prevent this.
897 */
898 if (maxlvt >= 3) {
899 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
900 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
901 }
902 /*
903 * Careful: we have to set masks only first to deassert
904 * any level-triggered sources.
905 */
906 v = apic_read(APIC_LVTT);
907 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
908 v = apic_read(APIC_LVT0);
909 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
910 v = apic_read(APIC_LVT1);
911 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
912 if (maxlvt >= 4) {
913 v = apic_read(APIC_LVTPC);
914 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
915 }
916
917 /* lets not touch this if we didn't frob it */
918#ifdef CONFIG_X86_THERMAL_VECTOR
919 if (maxlvt >= 5) {
920 v = apic_read(APIC_LVTTHMR);
921 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
922 }
923#endif
924#ifdef CONFIG_X86_MCE_INTEL
925 if (maxlvt >= 6) {
926 v = apic_read(APIC_LVTCMCI);
927 if (!(v & APIC_LVT_MASKED))
928 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
929 }
930#endif
931
932 /*
933 * Clean APIC state for other OSs:
934 */
935 apic_write(APIC_LVTT, APIC_LVT_MASKED);
936 apic_write(APIC_LVT0, APIC_LVT_MASKED);
937 apic_write(APIC_LVT1, APIC_LVT_MASKED);
938 if (maxlvt >= 3)
939 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
940 if (maxlvt >= 4)
941 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
942
943 /* Integrated APIC (!82489DX) ? */
944 if (lapic_is_integrated()) {
945 if (maxlvt > 3)
946 /* Clear ESR due to Pentium errata 3AP and 11AP */
947 apic_write(APIC_ESR, 0);
948 apic_read(APIC_ESR);
949 }
950}
951
952/**
953 * disable_local_APIC - clear and disable the local APIC
954 */
955void disable_local_APIC(void)
956{
957 unsigned int value;
958
959 /* APIC hasn't been mapped yet */
960 if (!x2apic_mode && !apic_phys)
961 return;
962
963 clear_local_APIC();
964
965 /*
966 * Disable APIC (implies clearing of registers
967 * for 82489DX!).
968 */
969 value = apic_read(APIC_SPIV);
970 value &= ~APIC_SPIV_APIC_ENABLED;
971 apic_write(APIC_SPIV, value);
972
973#ifdef CONFIG_X86_32
974 /*
975 * When LAPIC was disabled by the BIOS and enabled by the kernel,
976 * restore the disabled state.
977 */
978 if (enabled_via_apicbase) {
979 unsigned int l, h;
980
981 rdmsr(MSR_IA32_APICBASE, l, h);
982 l &= ~MSR_IA32_APICBASE_ENABLE;
983 wrmsr(MSR_IA32_APICBASE, l, h);
984 }
985#endif
986}
987
988/*
989 * If Linux enabled the LAPIC against the BIOS default disable it down before
990 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
991 * not power-off. Additionally clear all LVT entries before disable_local_APIC
992 * for the case where Linux didn't enable the LAPIC.
993 */
994void lapic_shutdown(void)
995{
996 unsigned long flags;
997
998 if (!cpu_has_apic && !apic_from_smp_config())
999 return;
1000
1001 local_irq_save(flags);
1002
1003#ifdef CONFIG_X86_32
1004 if (!enabled_via_apicbase)
1005 clear_local_APIC();
1006 else
1007#endif
1008 disable_local_APIC();
1009
1010
1011 local_irq_restore(flags);
1012}
1013
1014/*
1015 * This is to verify that we're looking at a real local APIC.
1016 * Check these against your board if the CPUs aren't getting
1017 * started for no apparent reason.
1018 */
1019int __init verify_local_APIC(void)
1020{
1021 unsigned int reg0, reg1;
1022
1023 /*
1024 * The version register is read-only in a real APIC.
1025 */
1026 reg0 = apic_read(APIC_LVR);
1027 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
1028 apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
1029 reg1 = apic_read(APIC_LVR);
1030 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
1031
1032 /*
1033 * The two version reads above should print the same
1034 * numbers. If the second one is different, then we
1035 * poke at a non-APIC.
1036 */
1037 if (reg1 != reg0)
1038 return 0;
1039
1040 /*
1041 * Check if the version looks reasonably.
1042 */
1043 reg1 = GET_APIC_VERSION(reg0);
1044 if (reg1 == 0x00 || reg1 == 0xff)
1045 return 0;
1046 reg1 = lapic_get_maxlvt();
1047 if (reg1 < 0x02 || reg1 == 0xff)
1048 return 0;
1049
1050 /*
1051 * The ID register is read/write in a real APIC.
1052 */
1053 reg0 = apic_read(APIC_ID);
1054 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
1055 apic_write(APIC_ID, reg0 ^ apic->apic_id_mask);
1056 reg1 = apic_read(APIC_ID);
1057 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
1058 apic_write(APIC_ID, reg0);
1059 if (reg1 != (reg0 ^ apic->apic_id_mask))
1060 return 0;
1061
1062 /*
1063 * The next two are just to see if we have sane values.
1064 * They're only really relevant if we're in Virtual Wire
1065 * compatibility mode, but most boxes are anymore.
1066 */
1067 reg0 = apic_read(APIC_LVT0);
1068 apic_printk(APIC_DEBUG, "Getting LVT0: %x\n", reg0);
1069 reg1 = apic_read(APIC_LVT1);
1070 apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
1071
1072 return 1;
1073}
1074
1075/**
1076 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1077 */
1078void __init sync_Arb_IDs(void)
1079{
1080 /*
1081 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1082 * needed on AMD.
1083 */
1084 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1085 return;
1086
1087 /*
1088 * Wait for idle.
1089 */
1090 apic_wait_icr_idle();
1091
1092 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1093 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1094 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1095}
1096
1097/*
1098 * An initial setup of the virtual wire mode.
1099 */
1100void __init init_bsp_APIC(void)
1101{
1102 unsigned int value;
1103
1104 /*
1105 * Don't do the setup now if we have a SMP BIOS as the
1106 * through-I/O-APIC virtual wire mode might be active.
1107 */
1108 if (smp_found_config || !cpu_has_apic)
1109 return;
1110
1111 /*
1112 * Do not trust the local APIC being empty at bootup.
1113 */
1114 clear_local_APIC();
1115
1116 /*
1117 * Enable APIC.
1118 */
1119 value = apic_read(APIC_SPIV);
1120 value &= ~APIC_VECTOR_MASK;
1121 value |= APIC_SPIV_APIC_ENABLED;
1122
1123#ifdef CONFIG_X86_32
1124 /* This bit is reserved on P4/Xeon and should be cleared */
1125 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1126 (boot_cpu_data.x86 == 15))
1127 value &= ~APIC_SPIV_FOCUS_DISABLED;
1128 else
1129#endif
1130 value |= APIC_SPIV_FOCUS_DISABLED;
1131 value |= SPURIOUS_APIC_VECTOR;
1132 apic_write(APIC_SPIV, value);
1133
1134 /*
1135 * Set up the virtual wire mode.
1136 */
1137 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1138 value = APIC_DM_NMI;
1139 if (!lapic_is_integrated()) /* 82489DX */
1140 value |= APIC_LVT_LEVEL_TRIGGER;
1141 apic_write(APIC_LVT1, value);
1142}
1143
1144static void __cpuinit lapic_setup_esr(void)
1145{
1146 unsigned int oldvalue, value, maxlvt;
1147
1148 if (!lapic_is_integrated()) {
1149 pr_info("No ESR for 82489DX.\n");
1150 return;
1151 }
1152
1153 if (apic->disable_esr) {
1154 /*
1155 * Something untraceable is creating bad interrupts on
1156 * secondary quads ... for the moment, just leave the
1157 * ESR disabled - we can't do anything useful with the
1158 * errors anyway - mbligh
1159 */
1160 pr_info("Leaving ESR disabled.\n");
1161 return;
1162 }
1163
1164 maxlvt = lapic_get_maxlvt();
1165 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1166 apic_write(APIC_ESR, 0);
1167 oldvalue = apic_read(APIC_ESR);
1168
1169 /* enables sending errors */
1170 value = ERROR_APIC_VECTOR;
1171 apic_write(APIC_LVTERR, value);
1172
1173 /*
1174 * spec says clear errors after enabling vector.
1175 */
1176 if (maxlvt > 3)
1177 apic_write(APIC_ESR, 0);
1178 value = apic_read(APIC_ESR);
1179 if (value != oldvalue)
1180 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1181 "vector: 0x%08x after: 0x%08x\n",
1182 oldvalue, value);
1183}
1184
1185/**
1186 * setup_local_APIC - setup the local APIC
1187 *
1188 * Used to setup local APIC while initializing BSP or bringin up APs.
1189 * Always called with preemption disabled.
1190 */
1191void __cpuinit setup_local_APIC(void)
1192{
1193 int cpu = smp_processor_id();
1194 unsigned int value, queued;
1195 int i, j, acked = 0;
1196 unsigned long long tsc = 0, ntsc;
1197 long long max_loops = cpu_khz;
1198
1199 if (cpu_has_tsc)
1200 rdtscll(tsc);
1201
1202 if (disable_apic) {
1203 disable_ioapic_support();
1204 return;
1205 }
1206
1207#ifdef CONFIG_X86_32
1208 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1209 if (lapic_is_integrated() && apic->disable_esr) {
1210 apic_write(APIC_ESR, 0);
1211 apic_write(APIC_ESR, 0);
1212 apic_write(APIC_ESR, 0);
1213 apic_write(APIC_ESR, 0);
1214 }
1215#endif
1216 perf_events_lapic_init();
1217
1218 /*
1219 * Double-check whether this APIC is really registered.
1220 * This is meaningless in clustered apic mode, so we skip it.
1221 */
1222 BUG_ON(!apic->apic_id_registered());
1223
1224 /*
1225 * Intel recommends to set DFR, LDR and TPR before enabling
1226 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1227 * document number 292116). So here it goes...
1228 */
1229 apic->init_apic_ldr();
1230
1231#ifdef CONFIG_X86_32
1232 /*
1233 * APIC LDR is initialized. If logical_apicid mapping was
1234 * initialized during get_smp_config(), make sure it matches the
1235 * actual value.
1236 */
1237 i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1238 WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1239 /* always use the value from LDR */
1240 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1241 logical_smp_processor_id();
1242
1243 /*
1244 * Some NUMA implementations (NUMAQ) don't initialize apicid to
1245 * node mapping during NUMA init. Now that logical apicid is
1246 * guaranteed to be known, give it another chance. This is already
1247 * a bit too late - percpu allocation has already happened without
1248 * proper NUMA affinity.
1249 */
1250 if (apic->x86_32_numa_cpu_node)
1251 set_apicid_to_node(early_per_cpu(x86_cpu_to_apicid, cpu),
1252 apic->x86_32_numa_cpu_node(cpu));
1253#endif
1254
1255 /*
1256 * Set Task Priority to 'accept all'. We never change this
1257 * later on.
1258 */
1259 value = apic_read(APIC_TASKPRI);
1260 value &= ~APIC_TPRI_MASK;
1261 apic_write(APIC_TASKPRI, value);
1262
1263 /*
1264 * After a crash, we no longer service the interrupts and a pending
1265 * interrupt from previous kernel might still have ISR bit set.
1266 *
1267 * Most probably by now CPU has serviced that pending interrupt and
1268 * it might not have done the ack_APIC_irq() because it thought,
1269 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1270 * does not clear the ISR bit and cpu thinks it has already serivced
1271 * the interrupt. Hence a vector might get locked. It was noticed
1272 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1273 */
1274 do {
1275 queued = 0;
1276 for (i = APIC_ISR_NR - 1; i >= 0; i--)
1277 queued |= apic_read(APIC_IRR + i*0x10);
1278
1279 for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1280 value = apic_read(APIC_ISR + i*0x10);
1281 for (j = 31; j >= 0; j--) {
1282 if (value & (1<<j)) {
1283 ack_APIC_irq();
1284 acked++;
1285 }
1286 }
1287 }
1288 if (acked > 256) {
1289 printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1290 acked);
1291 break;
1292 }
1293 if (cpu_has_tsc) {
1294 rdtscll(ntsc);
1295 max_loops = (cpu_khz << 10) - (ntsc - tsc);
1296 } else
1297 max_loops--;
1298 } while (queued && max_loops > 0);
1299 WARN_ON(max_loops <= 0);
1300
1301 /*
1302 * Now that we are all set up, enable the APIC
1303 */
1304 value = apic_read(APIC_SPIV);
1305 value &= ~APIC_VECTOR_MASK;
1306 /*
1307 * Enable APIC
1308 */
1309 value |= APIC_SPIV_APIC_ENABLED;
1310
1311#ifdef CONFIG_X86_32
1312 /*
1313 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1314 * certain networking cards. If high frequency interrupts are
1315 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1316 * entry is masked/unmasked at a high rate as well then sooner or
1317 * later IOAPIC line gets 'stuck', no more interrupts are received
1318 * from the device. If focus CPU is disabled then the hang goes
1319 * away, oh well :-(
1320 *
1321 * [ This bug can be reproduced easily with a level-triggered
1322 * PCI Ne2000 networking cards and PII/PIII processors, dual
1323 * BX chipset. ]
1324 */
1325 /*
1326 * Actually disabling the focus CPU check just makes the hang less
1327 * frequent as it makes the interrupt distributon model be more
1328 * like LRU than MRU (the short-term load is more even across CPUs).
1329 * See also the comment in end_level_ioapic_irq(). --macro
1330 */
1331
1332 /*
1333 * - enable focus processor (bit==0)
1334 * - 64bit mode always use processor focus
1335 * so no need to set it
1336 */
1337 value &= ~APIC_SPIV_FOCUS_DISABLED;
1338#endif
1339
1340 /*
1341 * Set spurious IRQ vector
1342 */
1343 value |= SPURIOUS_APIC_VECTOR;
1344 apic_write(APIC_SPIV, value);
1345
1346 /*
1347 * Set up LVT0, LVT1:
1348 *
1349 * set up through-local-APIC on the BP's LINT0. This is not
1350 * strictly necessary in pure symmetric-IO mode, but sometimes
1351 * we delegate interrupts to the 8259A.
1352 */
1353 /*
1354 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1355 */
1356 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1357 if (!cpu && (pic_mode || !value)) {
1358 value = APIC_DM_EXTINT;
1359 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1360 } else {
1361 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1362 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1363 }
1364 apic_write(APIC_LVT0, value);
1365
1366 /*
1367 * only the BP should see the LINT1 NMI signal, obviously.
1368 */
1369 if (!cpu)
1370 value = APIC_DM_NMI;
1371 else
1372 value = APIC_DM_NMI | APIC_LVT_MASKED;
1373 if (!lapic_is_integrated()) /* 82489DX */
1374 value |= APIC_LVT_LEVEL_TRIGGER;
1375 apic_write(APIC_LVT1, value);
1376
1377#ifdef CONFIG_X86_MCE_INTEL
1378 /* Recheck CMCI information after local APIC is up on CPU #0 */
1379 if (!cpu)
1380 cmci_recheck();
1381#endif
1382}
1383
1384void __cpuinit end_local_APIC_setup(void)
1385{
1386 lapic_setup_esr();
1387
1388#ifdef CONFIG_X86_32
1389 {
1390 unsigned int value;
1391 /* Disable the local apic timer */
1392 value = apic_read(APIC_LVTT);
1393 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1394 apic_write(APIC_LVTT, value);
1395 }
1396#endif
1397
1398 apic_pm_activate();
1399}
1400
1401void __init bsp_end_local_APIC_setup(void)
1402{
1403 end_local_APIC_setup();
1404
1405 /*
1406 * Now that local APIC setup is completed for BP, configure the fault
1407 * handling for interrupt remapping.
1408 */
1409 if (intr_remapping_enabled)
1410 enable_drhd_fault_handling();
1411
1412}
1413
1414#ifdef CONFIG_X86_X2APIC
1415void check_x2apic(void)
1416{
1417 if (x2apic_enabled()) {
1418 pr_info("x2apic enabled by BIOS, switching to x2apic ops\n");
1419 x2apic_preenabled = x2apic_mode = 1;
1420 }
1421}
1422
1423void enable_x2apic(void)
1424{
1425 int msr, msr2;
1426
1427 if (!x2apic_mode)
1428 return;
1429
1430 rdmsr(MSR_IA32_APICBASE, msr, msr2);
1431 if (!(msr & X2APIC_ENABLE)) {
1432 printk_once(KERN_INFO "Enabling x2apic\n");
1433 wrmsr(MSR_IA32_APICBASE, msr | X2APIC_ENABLE, msr2);
1434 }
1435}
1436#endif /* CONFIG_X86_X2APIC */
1437
1438int __init enable_IR(void)
1439{
1440#ifdef CONFIG_INTR_REMAP
1441 if (!intr_remapping_supported()) {
1442 pr_debug("intr-remapping not supported\n");
1443 return 0;
1444 }
1445
1446 if (!x2apic_preenabled && skip_ioapic_setup) {
1447 pr_info("Skipped enabling intr-remap because of skipping "
1448 "io-apic setup\n");
1449 return 0;
1450 }
1451
1452 if (enable_intr_remapping(x2apic_supported()))
1453 return 0;
1454
1455 pr_info("Enabled Interrupt-remapping\n");
1456
1457 return 1;
1458
1459#endif
1460 return 0;
1461}
1462
1463void __init enable_IR_x2apic(void)
1464{
1465 unsigned long flags;
1466 int ret, x2apic_enabled = 0;
1467 int dmar_table_init_ret;
1468
1469 dmar_table_init_ret = dmar_table_init();
1470 if (dmar_table_init_ret && !x2apic_supported())
1471 return;
1472
1473 ret = save_ioapic_entries();
1474 if (ret) {
1475 pr_info("Saving IO-APIC state failed: %d\n", ret);
1476 goto out;
1477 }
1478
1479 local_irq_save(flags);
1480 legacy_pic->mask_all();
1481 mask_ioapic_entries();
1482
1483 if (dmar_table_init_ret)
1484 ret = 0;
1485 else
1486 ret = enable_IR();
1487
1488 if (!ret) {
1489 /* IR is required if there is APIC ID > 255 even when running
1490 * under KVM
1491 */
1492 if (max_physical_apicid > 255 ||
1493 !hypervisor_x2apic_available())
1494 goto nox2apic;
1495 /*
1496 * without IR all CPUs can be addressed by IOAPIC/MSI
1497 * only in physical mode
1498 */
1499 x2apic_force_phys();
1500 }
1501
1502 x2apic_enabled = 1;
1503
1504 if (x2apic_supported() && !x2apic_mode) {
1505 x2apic_mode = 1;
1506 enable_x2apic();
1507 pr_info("Enabled x2apic\n");
1508 }
1509
1510nox2apic:
1511 if (!ret) /* IR enabling failed */
1512 restore_ioapic_entries();
1513 legacy_pic->restore_mask();
1514 local_irq_restore(flags);
1515
1516out:
1517 if (x2apic_enabled)
1518 return;
1519
1520 if (x2apic_preenabled)
1521 panic("x2apic: enabled by BIOS but kernel init failed.");
1522 else if (cpu_has_x2apic)
1523 pr_info("Not enabling x2apic, Intr-remapping init failed.\n");
1524}
1525
1526#ifdef CONFIG_X86_64
1527/*
1528 * Detect and enable local APICs on non-SMP boards.
1529 * Original code written by Keir Fraser.
1530 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1531 * not correctly set up (usually the APIC timer won't work etc.)
1532 */
1533static int __init detect_init_APIC(void)
1534{
1535 if (!cpu_has_apic) {
1536 pr_info("No local APIC present\n");
1537 return -1;
1538 }
1539
1540 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1541 return 0;
1542}
1543#else
1544
1545static int __init apic_verify(void)
1546{
1547 u32 features, h, l;
1548
1549 /*
1550 * The APIC feature bit should now be enabled
1551 * in `cpuid'
1552 */
1553 features = cpuid_edx(1);
1554 if (!(features & (1 << X86_FEATURE_APIC))) {
1555 pr_warning("Could not enable APIC!\n");
1556 return -1;
1557 }
1558 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1559 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1560
1561 /* The BIOS may have set up the APIC at some other address */
1562 rdmsr(MSR_IA32_APICBASE, l, h);
1563 if (l & MSR_IA32_APICBASE_ENABLE)
1564 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1565
1566 pr_info("Found and enabled local APIC!\n");
1567 return 0;
1568}
1569
1570int __init apic_force_enable(unsigned long addr)
1571{
1572 u32 h, l;
1573
1574 if (disable_apic)
1575 return -1;
1576
1577 /*
1578 * Some BIOSes disable the local APIC in the APIC_BASE
1579 * MSR. This can only be done in software for Intel P6 or later
1580 * and AMD K7 (Model > 1) or later.
1581 */
1582 rdmsr(MSR_IA32_APICBASE, l, h);
1583 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1584 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1585 l &= ~MSR_IA32_APICBASE_BASE;
1586 l |= MSR_IA32_APICBASE_ENABLE | addr;
1587 wrmsr(MSR_IA32_APICBASE, l, h);
1588 enabled_via_apicbase = 1;
1589 }
1590 return apic_verify();
1591}
1592
1593/*
1594 * Detect and initialize APIC
1595 */
1596static int __init detect_init_APIC(void)
1597{
1598 /* Disabled by kernel option? */
1599 if (disable_apic)
1600 return -1;
1601
1602 switch (boot_cpu_data.x86_vendor) {
1603 case X86_VENDOR_AMD:
1604 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1605 (boot_cpu_data.x86 >= 15))
1606 break;
1607 goto no_apic;
1608 case X86_VENDOR_INTEL:
1609 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1610 (boot_cpu_data.x86 == 5 && cpu_has_apic))
1611 break;
1612 goto no_apic;
1613 default:
1614 goto no_apic;
1615 }
1616
1617 if (!cpu_has_apic) {
1618 /*
1619 * Over-ride BIOS and try to enable the local APIC only if
1620 * "lapic" specified.
1621 */
1622 if (!force_enable_local_apic) {
1623 pr_info("Local APIC disabled by BIOS -- "
1624 "you can enable it with \"lapic\"\n");
1625 return -1;
1626 }
1627 if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1628 return -1;
1629 } else {
1630 if (apic_verify())
1631 return -1;
1632 }
1633
1634 apic_pm_activate();
1635
1636 return 0;
1637
1638no_apic:
1639 pr_info("No local APIC present or hardware disabled\n");
1640 return -1;
1641}
1642#endif
1643
1644/**
1645 * init_apic_mappings - initialize APIC mappings
1646 */
1647void __init init_apic_mappings(void)
1648{
1649 unsigned int new_apicid;
1650
1651 if (x2apic_mode) {
1652 boot_cpu_physical_apicid = read_apic_id();
1653 return;
1654 }
1655
1656 /* If no local APIC can be found return early */
1657 if (!smp_found_config && detect_init_APIC()) {
1658 /* lets NOP'ify apic operations */
1659 pr_info("APIC: disable apic facility\n");
1660 apic_disable();
1661 } else {
1662 apic_phys = mp_lapic_addr;
1663
1664 /*
1665 * acpi lapic path already maps that address in
1666 * acpi_register_lapic_address()
1667 */
1668 if (!acpi_lapic && !smp_found_config)
1669 register_lapic_address(apic_phys);
1670 }
1671
1672 /*
1673 * Fetch the APIC ID of the BSP in case we have a
1674 * default configuration (or the MP table is broken).
1675 */
1676 new_apicid = read_apic_id();
1677 if (boot_cpu_physical_apicid != new_apicid) {
1678 boot_cpu_physical_apicid = new_apicid;
1679 /*
1680 * yeah -- we lie about apic_version
1681 * in case if apic was disabled via boot option
1682 * but it's not a problem for SMP compiled kernel
1683 * since smp_sanity_check is prepared for such a case
1684 * and disable smp mode
1685 */
1686 apic_version[new_apicid] =
1687 GET_APIC_VERSION(apic_read(APIC_LVR));
1688 }
1689}
1690
1691void __init register_lapic_address(unsigned long address)
1692{
1693 mp_lapic_addr = address;
1694
1695 if (!x2apic_mode) {
1696 set_fixmap_nocache(FIX_APIC_BASE, address);
1697 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1698 APIC_BASE, mp_lapic_addr);
1699 }
1700 if (boot_cpu_physical_apicid == -1U) {
1701 boot_cpu_physical_apicid = read_apic_id();
1702 apic_version[boot_cpu_physical_apicid] =
1703 GET_APIC_VERSION(apic_read(APIC_LVR));
1704 }
1705}
1706
1707/*
1708 * This initializes the IO-APIC and APIC hardware if this is
1709 * a UP kernel.
1710 */
1711int apic_version[MAX_LOCAL_APIC];
1712
1713int __init APIC_init_uniprocessor(void)
1714{
1715 if (disable_apic) {
1716 pr_info("Apic disabled\n");
1717 return -1;
1718 }
1719#ifdef CONFIG_X86_64
1720 if (!cpu_has_apic) {
1721 disable_apic = 1;
1722 pr_info("Apic disabled by BIOS\n");
1723 return -1;
1724 }
1725#else
1726 if (!smp_found_config && !cpu_has_apic)
1727 return -1;
1728
1729 /*
1730 * Complain if the BIOS pretends there is one.
1731 */
1732 if (!cpu_has_apic &&
1733 APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
1734 pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
1735 boot_cpu_physical_apicid);
1736 return -1;
1737 }
1738#endif
1739
1740 default_setup_apic_routing();
1741
1742 verify_local_APIC();
1743 connect_bsp_APIC();
1744
1745#ifdef CONFIG_X86_64
1746 apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
1747#else
1748 /*
1749 * Hack: In case of kdump, after a crash, kernel might be booting
1750 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
1751 * might be zero if read from MP tables. Get it from LAPIC.
1752 */
1753# ifdef CONFIG_CRASH_DUMP
1754 boot_cpu_physical_apicid = read_apic_id();
1755# endif
1756#endif
1757 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1758 setup_local_APIC();
1759
1760#ifdef CONFIG_X86_IO_APIC
1761 /*
1762 * Now enable IO-APICs, actually call clear_IO_APIC
1763 * We need clear_IO_APIC before enabling error vector
1764 */
1765 if (!skip_ioapic_setup && nr_ioapics)
1766 enable_IO_APIC();
1767#endif
1768
1769 bsp_end_local_APIC_setup();
1770
1771#ifdef CONFIG_X86_IO_APIC
1772 if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
1773 setup_IO_APIC();
1774 else {
1775 nr_ioapics = 0;
1776 }
1777#endif
1778
1779 x86_init.timers.setup_percpu_clockev();
1780 return 0;
1781}
1782
1783/*
1784 * Local APIC interrupts
1785 */
1786
1787/*
1788 * This interrupt should _never_ happen with our APIC/SMP architecture
1789 */
1790void smp_spurious_interrupt(struct pt_regs *regs)
1791{
1792 u32 v;
1793
1794 exit_idle();
1795 irq_enter();
1796 /*
1797 * Check if this really is a spurious interrupt and ACK it
1798 * if it is a vectored one. Just in case...
1799 * Spurious interrupts should not be ACKed.
1800 */
1801 v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
1802 if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
1803 ack_APIC_irq();
1804
1805 inc_irq_stat(irq_spurious_count);
1806
1807 /* see sw-dev-man vol 3, chapter 7.4.13.5 */
1808 pr_info("spurious APIC interrupt on CPU#%d, "
1809 "should never happen.\n", smp_processor_id());
1810 irq_exit();
1811}
1812
1813/*
1814 * This interrupt should never happen with our APIC/SMP architecture
1815 */
1816void smp_error_interrupt(struct pt_regs *regs)
1817{
1818 u32 v0, v1;
1819 u32 i = 0;
1820 static const char * const error_interrupt_reason[] = {
1821 "Send CS error", /* APIC Error Bit 0 */
1822 "Receive CS error", /* APIC Error Bit 1 */
1823 "Send accept error", /* APIC Error Bit 2 */
1824 "Receive accept error", /* APIC Error Bit 3 */
1825 "Redirectable IPI", /* APIC Error Bit 4 */
1826 "Send illegal vector", /* APIC Error Bit 5 */
1827 "Received illegal vector", /* APIC Error Bit 6 */
1828 "Illegal register address", /* APIC Error Bit 7 */
1829 };
1830
1831 exit_idle();
1832 irq_enter();
1833 /* First tickle the hardware, only then report what went on. -- REW */
1834 v0 = apic_read(APIC_ESR);
1835 apic_write(APIC_ESR, 0);
1836 v1 = apic_read(APIC_ESR);
1837 ack_APIC_irq();
1838 atomic_inc(&irq_err_count);
1839
1840 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x(%02x)",
1841 smp_processor_id(), v0 , v1);
1842
1843 v1 = v1 & 0xff;
1844 while (v1) {
1845 if (v1 & 0x1)
1846 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
1847 i++;
1848 v1 >>= 1;
1849 };
1850
1851 apic_printk(APIC_DEBUG, KERN_CONT "\n");
1852
1853 irq_exit();
1854}
1855
1856/**
1857 * connect_bsp_APIC - attach the APIC to the interrupt system
1858 */
1859void __init connect_bsp_APIC(void)
1860{
1861#ifdef CONFIG_X86_32
1862 if (pic_mode) {
1863 /*
1864 * Do not trust the local APIC being empty at bootup.
1865 */
1866 clear_local_APIC();
1867 /*
1868 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
1869 * local APIC to INT and NMI lines.
1870 */
1871 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
1872 "enabling APIC mode.\n");
1873 imcr_pic_to_apic();
1874 }
1875#endif
1876 if (apic->enable_apic_mode)
1877 apic->enable_apic_mode();
1878}
1879
1880/**
1881 * disconnect_bsp_APIC - detach the APIC from the interrupt system
1882 * @virt_wire_setup: indicates, whether virtual wire mode is selected
1883 *
1884 * Virtual wire mode is necessary to deliver legacy interrupts even when the
1885 * APIC is disabled.
1886 */
1887void disconnect_bsp_APIC(int virt_wire_setup)
1888{
1889 unsigned int value;
1890
1891#ifdef CONFIG_X86_32
1892 if (pic_mode) {
1893 /*
1894 * Put the board back into PIC mode (has an effect only on
1895 * certain older boards). Note that APIC interrupts, including
1896 * IPIs, won't work beyond this point! The only exception are
1897 * INIT IPIs.
1898 */
1899 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
1900 "entering PIC mode.\n");
1901 imcr_apic_to_pic();
1902 return;
1903 }
1904#endif
1905
1906 /* Go back to Virtual Wire compatibility mode */
1907
1908 /* For the spurious interrupt use vector F, and enable it */
1909 value = apic_read(APIC_SPIV);
1910 value &= ~APIC_VECTOR_MASK;
1911 value |= APIC_SPIV_APIC_ENABLED;
1912 value |= 0xf;
1913 apic_write(APIC_SPIV, value);
1914
1915 if (!virt_wire_setup) {
1916 /*
1917 * For LVT0 make it edge triggered, active high,
1918 * external and enabled
1919 */
1920 value = apic_read(APIC_LVT0);
1921 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1922 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1923 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1924 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1925 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
1926 apic_write(APIC_LVT0, value);
1927 } else {
1928 /* Disable LVT0 */
1929 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1930 }
1931
1932 /*
1933 * For LVT1 make it edge triggered, active high,
1934 * nmi and enabled
1935 */
1936 value = apic_read(APIC_LVT1);
1937 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1938 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1939 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1940 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1941 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
1942 apic_write(APIC_LVT1, value);
1943}
1944
1945void __cpuinit generic_processor_info(int apicid, int version)
1946{
1947 int cpu, max = nr_cpu_ids;
1948 bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
1949 phys_cpu_present_map);
1950
1951 /*
1952 * If boot cpu has not been detected yet, then only allow upto
1953 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
1954 */
1955 if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
1956 apicid != boot_cpu_physical_apicid) {
1957 int thiscpu = max + disabled_cpus - 1;
1958
1959 pr_warning(
1960 "ACPI: NR_CPUS/possible_cpus limit of %i almost"
1961 " reached. Keeping one slot for boot cpu."
1962 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
1963
1964 disabled_cpus++;
1965 return;
1966 }
1967
1968 if (num_processors >= nr_cpu_ids) {
1969 int thiscpu = max + disabled_cpus;
1970
1971 pr_warning(
1972 "ACPI: NR_CPUS/possible_cpus limit of %i reached."
1973 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
1974
1975 disabled_cpus++;
1976 return;
1977 }
1978
1979 num_processors++;
1980 if (apicid == boot_cpu_physical_apicid) {
1981 /*
1982 * x86_bios_cpu_apicid is required to have processors listed
1983 * in same order as logical cpu numbers. Hence the first
1984 * entry is BSP, and so on.
1985 * boot_cpu_init() already hold bit 0 in cpu_present_mask
1986 * for BSP.
1987 */
1988 cpu = 0;
1989 } else
1990 cpu = cpumask_next_zero(-1, cpu_present_mask);
1991
1992 /*
1993 * Validate version
1994 */
1995 if (version == 0x0) {
1996 pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
1997 cpu, apicid);
1998 version = 0x10;
1999 }
2000 apic_version[apicid] = version;
2001
2002 if (version != apic_version[boot_cpu_physical_apicid]) {
2003 pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2004 apic_version[boot_cpu_physical_apicid], cpu, version);
2005 }
2006
2007 physid_set(apicid, phys_cpu_present_map);
2008 if (apicid > max_physical_apicid)
2009 max_physical_apicid = apicid;
2010
2011#if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2012 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2013 early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2014#endif
2015#ifdef CONFIG_X86_32
2016 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2017 apic->x86_32_early_logical_apicid(cpu);
2018#endif
2019 set_cpu_possible(cpu, true);
2020 set_cpu_present(cpu, true);
2021}
2022
2023int hard_smp_processor_id(void)
2024{
2025 return read_apic_id();
2026}
2027
2028void default_init_apic_ldr(void)
2029{
2030 unsigned long val;
2031
2032 apic_write(APIC_DFR, APIC_DFR_VALUE);
2033 val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
2034 val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
2035 apic_write(APIC_LDR, val);
2036}
2037
2038/*
2039 * Power management
2040 */
2041#ifdef CONFIG_PM
2042
2043static struct {
2044 /*
2045 * 'active' is true if the local APIC was enabled by us and
2046 * not the BIOS; this signifies that we are also responsible
2047 * for disabling it before entering apm/acpi suspend
2048 */
2049 int active;
2050 /* r/w apic fields */
2051 unsigned int apic_id;
2052 unsigned int apic_taskpri;
2053 unsigned int apic_ldr;
2054 unsigned int apic_dfr;
2055 unsigned int apic_spiv;
2056 unsigned int apic_lvtt;
2057 unsigned int apic_lvtpc;
2058 unsigned int apic_lvt0;
2059 unsigned int apic_lvt1;
2060 unsigned int apic_lvterr;
2061 unsigned int apic_tmict;
2062 unsigned int apic_tdcr;
2063 unsigned int apic_thmr;
2064} apic_pm_state;
2065
2066static int lapic_suspend(void)
2067{
2068 unsigned long flags;
2069 int maxlvt;
2070
2071 if (!apic_pm_state.active)
2072 return 0;
2073
2074 maxlvt = lapic_get_maxlvt();
2075
2076 apic_pm_state.apic_id = apic_read(APIC_ID);
2077 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2078 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2079 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2080 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2081 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2082 if (maxlvt >= 4)
2083 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2084 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2085 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2086 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2087 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2088 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2089#ifdef CONFIG_X86_THERMAL_VECTOR
2090 if (maxlvt >= 5)
2091 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2092#endif
2093
2094 local_irq_save(flags);
2095 disable_local_APIC();
2096
2097 if (intr_remapping_enabled)
2098 disable_intr_remapping();
2099
2100 local_irq_restore(flags);
2101 return 0;
2102}
2103
2104static void lapic_resume(void)
2105{
2106 unsigned int l, h;
2107 unsigned long flags;
2108 int maxlvt;
2109
2110 if (!apic_pm_state.active)
2111 return;
2112
2113 local_irq_save(flags);
2114 if (intr_remapping_enabled) {
2115 /*
2116 * IO-APIC and PIC have their own resume routines.
2117 * We just mask them here to make sure the interrupt
2118 * subsystem is completely quiet while we enable x2apic
2119 * and interrupt-remapping.
2120 */
2121 mask_ioapic_entries();
2122 legacy_pic->mask_all();
2123 }
2124
2125 if (x2apic_mode)
2126 enable_x2apic();
2127 else {
2128 /*
2129 * Make sure the APICBASE points to the right address
2130 *
2131 * FIXME! This will be wrong if we ever support suspend on
2132 * SMP! We'll need to do this as part of the CPU restore!
2133 */
2134 rdmsr(MSR_IA32_APICBASE, l, h);
2135 l &= ~MSR_IA32_APICBASE_BASE;
2136 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2137 wrmsr(MSR_IA32_APICBASE, l, h);
2138 }
2139
2140 maxlvt = lapic_get_maxlvt();
2141 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2142 apic_write(APIC_ID, apic_pm_state.apic_id);
2143 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2144 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2145 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2146 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2147 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2148 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2149#if defined(CONFIG_X86_MCE_P4THERMAL) || defined(CONFIG_X86_MCE_INTEL)
2150 if (maxlvt >= 5)
2151 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2152#endif
2153 if (maxlvt >= 4)
2154 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2155 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2156 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2157 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2158 apic_write(APIC_ESR, 0);
2159 apic_read(APIC_ESR);
2160 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2161 apic_write(APIC_ESR, 0);
2162 apic_read(APIC_ESR);
2163
2164 if (intr_remapping_enabled)
2165 reenable_intr_remapping(x2apic_mode);
2166
2167 local_irq_restore(flags);
2168}
2169
2170/*
2171 * This device has no shutdown method - fully functioning local APICs
2172 * are needed on every CPU up until machine_halt/restart/poweroff.
2173 */
2174
2175static struct syscore_ops lapic_syscore_ops = {
2176 .resume = lapic_resume,
2177 .suspend = lapic_suspend,
2178};
2179
2180static void __cpuinit apic_pm_activate(void)
2181{
2182 apic_pm_state.active = 1;
2183}
2184
2185static int __init init_lapic_sysfs(void)
2186{
2187 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2188 if (cpu_has_apic)
2189 register_syscore_ops(&lapic_syscore_ops);
2190
2191 return 0;
2192}
2193
2194/* local apic needs to resume before other devices access its registers. */
2195core_initcall(init_lapic_sysfs);
2196
2197#else /* CONFIG_PM */
2198
2199static void apic_pm_activate(void) { }
2200
2201#endif /* CONFIG_PM */
2202
2203#ifdef CONFIG_X86_64
2204
2205static int __cpuinit apic_cluster_num(void)
2206{
2207 int i, clusters, zeros;
2208 unsigned id;
2209 u16 *bios_cpu_apicid;
2210 DECLARE_BITMAP(clustermap, NUM_APIC_CLUSTERS);
2211
2212 bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
2213 bitmap_zero(clustermap, NUM_APIC_CLUSTERS);
2214
2215 for (i = 0; i < nr_cpu_ids; i++) {
2216 /* are we being called early in kernel startup? */
2217 if (bios_cpu_apicid) {
2218 id = bios_cpu_apicid[i];
2219 } else if (i < nr_cpu_ids) {
2220 if (cpu_present(i))
2221 id = per_cpu(x86_bios_cpu_apicid, i);
2222 else
2223 continue;
2224 } else
2225 break;
2226
2227 if (id != BAD_APICID)
2228 __set_bit(APIC_CLUSTERID(id), clustermap);
2229 }
2230
2231 /* Problem: Partially populated chassis may not have CPUs in some of
2232 * the APIC clusters they have been allocated. Only present CPUs have
2233 * x86_bios_cpu_apicid entries, thus causing zeroes in the bitmap.
2234 * Since clusters are allocated sequentially, count zeros only if
2235 * they are bounded by ones.
2236 */
2237 clusters = 0;
2238 zeros = 0;
2239 for (i = 0; i < NUM_APIC_CLUSTERS; i++) {
2240 if (test_bit(i, clustermap)) {
2241 clusters += 1 + zeros;
2242 zeros = 0;
2243 } else
2244 ++zeros;
2245 }
2246
2247 return clusters;
2248}
2249
2250static int __cpuinitdata multi_checked;
2251static int __cpuinitdata multi;
2252
2253static int __cpuinit set_multi(const struct dmi_system_id *d)
2254{
2255 if (multi)
2256 return 0;
2257 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2258 multi = 1;
2259 return 0;
2260}
2261
2262static const __cpuinitconst struct dmi_system_id multi_dmi_table[] = {
2263 {
2264 .callback = set_multi,
2265 .ident = "IBM System Summit2",
2266 .matches = {
2267 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2268 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2269 },
2270 },
2271 {}
2272};
2273
2274static void __cpuinit dmi_check_multi(void)
2275{
2276 if (multi_checked)
2277 return;
2278
2279 dmi_check_system(multi_dmi_table);
2280 multi_checked = 1;
2281}
2282
2283/*
2284 * apic_is_clustered_box() -- Check if we can expect good TSC
2285 *
2286 * Thus far, the major user of this is IBM's Summit2 series:
2287 * Clustered boxes may have unsynced TSC problems if they are
2288 * multi-chassis.
2289 * Use DMI to check them
2290 */
2291__cpuinit int apic_is_clustered_box(void)
2292{
2293 dmi_check_multi();
2294 if (multi)
2295 return 1;
2296
2297 if (!is_vsmp_box())
2298 return 0;
2299
2300 /*
2301 * ScaleMP vSMPowered boxes have one cluster per board and TSCs are
2302 * not guaranteed to be synced between boards
2303 */
2304 if (apic_cluster_num() > 1)
2305 return 1;
2306
2307 return 0;
2308}
2309#endif
2310
2311/*
2312 * APIC command line parameters
2313 */
2314static int __init setup_disableapic(char *arg)
2315{
2316 disable_apic = 1;
2317 setup_clear_cpu_cap(X86_FEATURE_APIC);
2318 return 0;
2319}
2320early_param("disableapic", setup_disableapic);
2321
2322/* same as disableapic, for compatibility */
2323static int __init setup_nolapic(char *arg)
2324{
2325 return setup_disableapic(arg);
2326}
2327early_param("nolapic", setup_nolapic);
2328
2329static int __init parse_lapic_timer_c2_ok(char *arg)
2330{
2331 local_apic_timer_c2_ok = 1;
2332 return 0;
2333}
2334early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2335
2336static int __init parse_disable_apic_timer(char *arg)
2337{
2338 disable_apic_timer = 1;
2339 return 0;
2340}
2341early_param("noapictimer", parse_disable_apic_timer);
2342
2343static int __init parse_nolapic_timer(char *arg)
2344{
2345 disable_apic_timer = 1;
2346 return 0;
2347}
2348early_param("nolapic_timer", parse_nolapic_timer);
2349
2350static int __init apic_set_verbosity(char *arg)
2351{
2352 if (!arg) {
2353#ifdef CONFIG_X86_64
2354 skip_ioapic_setup = 0;
2355 return 0;
2356#endif
2357 return -EINVAL;
2358 }
2359
2360 if (strcmp("debug", arg) == 0)
2361 apic_verbosity = APIC_DEBUG;
2362 else if (strcmp("verbose", arg) == 0)
2363 apic_verbosity = APIC_VERBOSE;
2364 else {
2365 pr_warning("APIC Verbosity level %s not recognised"
2366 " use apic=verbose or apic=debug\n", arg);
2367 return -EINVAL;
2368 }
2369
2370 return 0;
2371}
2372early_param("apic", apic_set_verbosity);
2373
2374static int __init lapic_insert_resource(void)
2375{
2376 if (!apic_phys)
2377 return -1;
2378
2379 /* Put local APIC into the resource map. */
2380 lapic_resource.start = apic_phys;
2381 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2382 insert_resource(&iomem_resource, &lapic_resource);
2383
2384 return 0;
2385}
2386
2387/*
2388 * need call insert after e820_reserve_resources()
2389 * that is using request_resource
2390 */
2391late_initcall(lapic_insert_resource);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Local APIC handling, local APIC timers
4 *
5 * (c) 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
6 *
7 * Fixes
8 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
9 * thanks to Eric Gilmore
10 * and Rolf G. Tews
11 * for testing these extensively.
12 * Maciej W. Rozycki : Various updates and fixes.
13 * Mikael Pettersson : Power Management for UP-APIC.
14 * Pavel Machek and
15 * Mikael Pettersson : PM converted to driver model.
16 */
17
18#include <linux/perf_event.h>
19#include <linux/kernel_stat.h>
20#include <linux/mc146818rtc.h>
21#include <linux/acpi_pmtmr.h>
22#include <linux/clockchips.h>
23#include <linux/interrupt.h>
24#include <linux/memblock.h>
25#include <linux/ftrace.h>
26#include <linux/ioport.h>
27#include <linux/export.h>
28#include <linux/syscore_ops.h>
29#include <linux/delay.h>
30#include <linux/timex.h>
31#include <linux/i8253.h>
32#include <linux/dmar.h>
33#include <linux/init.h>
34#include <linux/cpu.h>
35#include <linux/dmi.h>
36#include <linux/smp.h>
37#include <linux/mm.h>
38
39#include <asm/trace/irq_vectors.h>
40#include <asm/irq_remapping.h>
41#include <asm/perf_event.h>
42#include <asm/x86_init.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>
48#include <asm/apic.h>
49#include <asm/acpi.h>
50#include <asm/io_apic.h>
51#include <asm/desc.h>
52#include <asm/hpet.h>
53#include <asm/mtrr.h>
54#include <asm/time.h>
55#include <asm/smp.h>
56#include <asm/mce.h>
57#include <asm/tsc.h>
58#include <asm/hypervisor.h>
59#include <asm/cpu_device_id.h>
60#include <asm/intel-family.h>
61#include <asm/irq_regs.h>
62
63unsigned int num_processors;
64
65unsigned disabled_cpus;
66
67/* Processor that is doing the boot up */
68unsigned int boot_cpu_physical_apicid __ro_after_init = -1U;
69EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
70
71u8 boot_cpu_apic_version __ro_after_init;
72
73/*
74 * The highest APIC ID seen during enumeration.
75 */
76static unsigned int max_physical_apicid;
77
78/*
79 * Bitmask of physically existing CPUs:
80 */
81physid_mask_t phys_cpu_present_map;
82
83/*
84 * Processor to be disabled specified by kernel parameter
85 * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
86 * avoid undefined behaviour caused by sending INIT from AP to BSP.
87 */
88static unsigned int disabled_cpu_apicid __ro_after_init = BAD_APICID;
89
90/*
91 * This variable controls which CPUs receive external NMIs. By default,
92 * external NMIs are delivered only to the BSP.
93 */
94static int apic_extnmi __ro_after_init = APIC_EXTNMI_BSP;
95
96/*
97 * Map cpu index to physical APIC ID
98 */
99DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
100DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
101DEFINE_EARLY_PER_CPU_READ_MOSTLY(u32, x86_cpu_to_acpiid, U32_MAX);
102EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
103EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
104EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_acpiid);
105
106#ifdef CONFIG_X86_32
107
108/*
109 * On x86_32, the mapping between cpu and logical apicid may vary
110 * depending on apic in use. The following early percpu variable is
111 * used for the mapping. This is where the behaviors of x86_64 and 32
112 * actually diverge. Let's keep it ugly for now.
113 */
114DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
115
116/* Local APIC was disabled by the BIOS and enabled by the kernel */
117static int enabled_via_apicbase __ro_after_init;
118
119/*
120 * Handle interrupt mode configuration register (IMCR).
121 * This register controls whether the interrupt signals
122 * that reach the BSP come from the master PIC or from the
123 * local APIC. Before entering Symmetric I/O Mode, either
124 * the BIOS or the operating system must switch out of
125 * PIC Mode by changing the IMCR.
126 */
127static inline void imcr_pic_to_apic(void)
128{
129 /* select IMCR register */
130 outb(0x70, 0x22);
131 /* NMI and 8259 INTR go through APIC */
132 outb(0x01, 0x23);
133}
134
135static inline void imcr_apic_to_pic(void)
136{
137 /* select IMCR register */
138 outb(0x70, 0x22);
139 /* NMI and 8259 INTR go directly to BSP */
140 outb(0x00, 0x23);
141}
142#endif
143
144/*
145 * Knob to control our willingness to enable the local APIC.
146 *
147 * +1=force-enable
148 */
149static int force_enable_local_apic __initdata;
150
151/*
152 * APIC command line parameters
153 */
154static int __init parse_lapic(char *arg)
155{
156 if (IS_ENABLED(CONFIG_X86_32) && !arg)
157 force_enable_local_apic = 1;
158 else if (arg && !strncmp(arg, "notscdeadline", 13))
159 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
160 return 0;
161}
162early_param("lapic", parse_lapic);
163
164#ifdef CONFIG_X86_64
165static int apic_calibrate_pmtmr __initdata;
166static __init int setup_apicpmtimer(char *s)
167{
168 apic_calibrate_pmtmr = 1;
169 notsc_setup(NULL);
170 return 0;
171}
172__setup("apicpmtimer", setup_apicpmtimer);
173#endif
174
175unsigned long mp_lapic_addr __ro_after_init;
176int disable_apic __ro_after_init;
177/* Disable local APIC timer from the kernel commandline or via dmi quirk */
178static int disable_apic_timer __initdata;
179/* Local APIC timer works in C2 */
180int local_apic_timer_c2_ok __ro_after_init;
181EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
182
183/*
184 * Debug level, exported for io_apic.c
185 */
186int apic_verbosity __ro_after_init;
187
188int pic_mode __ro_after_init;
189
190/* Have we found an MP table */
191int smp_found_config __ro_after_init;
192
193static struct resource lapic_resource = {
194 .name = "Local APIC",
195 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
196};
197
198unsigned int lapic_timer_period = 0;
199
200static void apic_pm_activate(void);
201
202static unsigned long apic_phys __ro_after_init;
203
204/*
205 * Get the LAPIC version
206 */
207static inline int lapic_get_version(void)
208{
209 return GET_APIC_VERSION(apic_read(APIC_LVR));
210}
211
212/*
213 * Check, if the APIC is integrated or a separate chip
214 */
215static inline int lapic_is_integrated(void)
216{
217 return APIC_INTEGRATED(lapic_get_version());
218}
219
220/*
221 * Check, whether this is a modern or a first generation APIC
222 */
223static int modern_apic(void)
224{
225 /* AMD systems use old APIC versions, so check the CPU */
226 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
227 boot_cpu_data.x86 >= 0xf)
228 return 1;
229
230 /* Hygon systems use modern APIC */
231 if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
232 return 1;
233
234 return lapic_get_version() >= 0x14;
235}
236
237/*
238 * right after this call apic become NOOP driven
239 * so apic->write/read doesn't do anything
240 */
241static void __init apic_disable(void)
242{
243 pr_info("APIC: switched to apic NOOP\n");
244 apic = &apic_noop;
245}
246
247void native_apic_wait_icr_idle(void)
248{
249 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
250 cpu_relax();
251}
252
253u32 native_safe_apic_wait_icr_idle(void)
254{
255 u32 send_status;
256 int timeout;
257
258 timeout = 0;
259 do {
260 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
261 if (!send_status)
262 break;
263 inc_irq_stat(icr_read_retry_count);
264 udelay(100);
265 } while (timeout++ < 1000);
266
267 return send_status;
268}
269
270void native_apic_icr_write(u32 low, u32 id)
271{
272 unsigned long flags;
273
274 local_irq_save(flags);
275 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
276 apic_write(APIC_ICR, low);
277 local_irq_restore(flags);
278}
279
280u64 native_apic_icr_read(void)
281{
282 u32 icr1, icr2;
283
284 icr2 = apic_read(APIC_ICR2);
285 icr1 = apic_read(APIC_ICR);
286
287 return icr1 | ((u64)icr2 << 32);
288}
289
290#ifdef CONFIG_X86_32
291/**
292 * get_physical_broadcast - Get number of physical broadcast IDs
293 */
294int get_physical_broadcast(void)
295{
296 return modern_apic() ? 0xff : 0xf;
297}
298#endif
299
300/**
301 * lapic_get_maxlvt - get the maximum number of local vector table entries
302 */
303int lapic_get_maxlvt(void)
304{
305 /*
306 * - we always have APIC integrated on 64bit mode
307 * - 82489DXs do not report # of LVT entries
308 */
309 return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
310}
311
312/*
313 * Local APIC timer
314 */
315
316/* Clock divisor */
317#define APIC_DIVISOR 16
318#define TSC_DIVISOR 8
319
320/*
321 * This function sets up the local APIC timer, with a timeout of
322 * 'clocks' APIC bus clock. During calibration we actually call
323 * this function twice on the boot CPU, once with a bogus timeout
324 * value, second time for real. The other (noncalibrating) CPUs
325 * call this function only once, with the real, calibrated value.
326 *
327 * We do reads before writes even if unnecessary, to get around the
328 * P5 APIC double write bug.
329 */
330static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
331{
332 unsigned int lvtt_value, tmp_value;
333
334 lvtt_value = LOCAL_TIMER_VECTOR;
335 if (!oneshot)
336 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
337 else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
338 lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
339
340 if (!lapic_is_integrated())
341 lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
342
343 if (!irqen)
344 lvtt_value |= APIC_LVT_MASKED;
345
346 apic_write(APIC_LVTT, lvtt_value);
347
348 if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
349 /*
350 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
351 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
352 * According to Intel, MFENCE can do the serialization here.
353 */
354 asm volatile("mfence" : : : "memory");
355 return;
356 }
357
358 /*
359 * Divide PICLK by 16
360 */
361 tmp_value = apic_read(APIC_TDCR);
362 apic_write(APIC_TDCR,
363 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
364 APIC_TDR_DIV_16);
365
366 if (!oneshot)
367 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
368}
369
370/*
371 * Setup extended LVT, AMD specific
372 *
373 * Software should use the LVT offsets the BIOS provides. The offsets
374 * are determined by the subsystems using it like those for MCE
375 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts
376 * are supported. Beginning with family 10h at least 4 offsets are
377 * available.
378 *
379 * Since the offsets must be consistent for all cores, we keep track
380 * of the LVT offsets in software and reserve the offset for the same
381 * vector also to be used on other cores. An offset is freed by
382 * setting the entry to APIC_EILVT_MASKED.
383 *
384 * If the BIOS is right, there should be no conflicts. Otherwise a
385 * "[Firmware Bug]: ..." error message is generated. However, if
386 * software does not properly determines the offsets, it is not
387 * necessarily a BIOS bug.
388 */
389
390static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
391
392static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
393{
394 return (old & APIC_EILVT_MASKED)
395 || (new == APIC_EILVT_MASKED)
396 || ((new & ~APIC_EILVT_MASKED) == old);
397}
398
399static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
400{
401 unsigned int rsvd, vector;
402
403 if (offset >= APIC_EILVT_NR_MAX)
404 return ~0;
405
406 rsvd = atomic_read(&eilvt_offsets[offset]);
407 do {
408 vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */
409 if (vector && !eilvt_entry_is_changeable(vector, new))
410 /* may not change if vectors are different */
411 return rsvd;
412 rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
413 } while (rsvd != new);
414
415 rsvd &= ~APIC_EILVT_MASKED;
416 if (rsvd && rsvd != vector)
417 pr_info("LVT offset %d assigned for vector 0x%02x\n",
418 offset, rsvd);
419
420 return new;
421}
422
423/*
424 * If mask=1, the LVT entry does not generate interrupts while mask=0
425 * enables the vector. See also the BKDGs. Must be called with
426 * preemption disabled.
427 */
428
429int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
430{
431 unsigned long reg = APIC_EILVTn(offset);
432 unsigned int new, old, reserved;
433
434 new = (mask << 16) | (msg_type << 8) | vector;
435 old = apic_read(reg);
436 reserved = reserve_eilvt_offset(offset, new);
437
438 if (reserved != new) {
439 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
440 "vector 0x%x, but the register is already in use for "
441 "vector 0x%x on another cpu\n",
442 smp_processor_id(), reg, offset, new, reserved);
443 return -EINVAL;
444 }
445
446 if (!eilvt_entry_is_changeable(old, new)) {
447 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
448 "vector 0x%x, but the register is already in use for "
449 "vector 0x%x on this cpu\n",
450 smp_processor_id(), reg, offset, new, old);
451 return -EBUSY;
452 }
453
454 apic_write(reg, new);
455
456 return 0;
457}
458EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
459
460/*
461 * Program the next event, relative to now
462 */
463static int lapic_next_event(unsigned long delta,
464 struct clock_event_device *evt)
465{
466 apic_write(APIC_TMICT, delta);
467 return 0;
468}
469
470static int lapic_next_deadline(unsigned long delta,
471 struct clock_event_device *evt)
472{
473 u64 tsc;
474
475 tsc = rdtsc();
476 wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
477 return 0;
478}
479
480static int lapic_timer_shutdown(struct clock_event_device *evt)
481{
482 unsigned int v;
483
484 /* Lapic used as dummy for broadcast ? */
485 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
486 return 0;
487
488 v = apic_read(APIC_LVTT);
489 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
490 apic_write(APIC_LVTT, v);
491 apic_write(APIC_TMICT, 0);
492 return 0;
493}
494
495static inline int
496lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
497{
498 /* Lapic used as dummy for broadcast ? */
499 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
500 return 0;
501
502 __setup_APIC_LVTT(lapic_timer_period, oneshot, 1);
503 return 0;
504}
505
506static int lapic_timer_set_periodic(struct clock_event_device *evt)
507{
508 return lapic_timer_set_periodic_oneshot(evt, false);
509}
510
511static int lapic_timer_set_oneshot(struct clock_event_device *evt)
512{
513 return lapic_timer_set_periodic_oneshot(evt, true);
514}
515
516/*
517 * Local APIC timer broadcast function
518 */
519static void lapic_timer_broadcast(const struct cpumask *mask)
520{
521#ifdef CONFIG_SMP
522 apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
523#endif
524}
525
526
527/*
528 * The local apic timer can be used for any function which is CPU local.
529 */
530static struct clock_event_device lapic_clockevent = {
531 .name = "lapic",
532 .features = CLOCK_EVT_FEAT_PERIODIC |
533 CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
534 | CLOCK_EVT_FEAT_DUMMY,
535 .shift = 32,
536 .set_state_shutdown = lapic_timer_shutdown,
537 .set_state_periodic = lapic_timer_set_periodic,
538 .set_state_oneshot = lapic_timer_set_oneshot,
539 .set_state_oneshot_stopped = lapic_timer_shutdown,
540 .set_next_event = lapic_next_event,
541 .broadcast = lapic_timer_broadcast,
542 .rating = 100,
543 .irq = -1,
544};
545static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
546
547static const struct x86_cpu_id deadline_match[] __initconst = {
548 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x2, 0x2), 0x3a), /* EP */
549 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x4, 0x4), 0x0f), /* EX */
550
551 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_X, 0x0b000020),
552
553 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x2, 0x2), 0x00000011),
554 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x3, 0x3), 0x0700000e),
555 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x4, 0x4), 0x0f00000c),
556 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x5, 0x5), 0x0e000003),
557
558 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x3, 0x3), 0x01000136),
559 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x4, 0x4), 0x02000014),
560 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x5, 0xf), 0),
561
562 X86_MATCH_INTEL_FAM6_MODEL( HASWELL, 0x22),
563 X86_MATCH_INTEL_FAM6_MODEL( HASWELL_L, 0x20),
564 X86_MATCH_INTEL_FAM6_MODEL( HASWELL_G, 0x17),
565
566 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL, 0x25),
567 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_G, 0x17),
568
569 X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE_L, 0xb2),
570 X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE, 0xb2),
571
572 X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE_L, 0x52),
573 X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE, 0x52),
574
575 {},
576};
577
578static __init bool apic_validate_deadline_timer(void)
579{
580 const struct x86_cpu_id *m;
581 u32 rev;
582
583 if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
584 return false;
585 if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
586 return true;
587
588 m = x86_match_cpu(deadline_match);
589 if (!m)
590 return true;
591
592 rev = (u32)m->driver_data;
593
594 if (boot_cpu_data.microcode >= rev)
595 return true;
596
597 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
598 pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
599 "please update microcode to version: 0x%x (or later)\n", rev);
600 return false;
601}
602
603/*
604 * Setup the local APIC timer for this CPU. Copy the initialized values
605 * of the boot CPU and register the clock event in the framework.
606 */
607static void setup_APIC_timer(void)
608{
609 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
610
611 if (this_cpu_has(X86_FEATURE_ARAT)) {
612 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
613 /* Make LAPIC timer preferrable over percpu HPET */
614 lapic_clockevent.rating = 150;
615 }
616
617 memcpy(levt, &lapic_clockevent, sizeof(*levt));
618 levt->cpumask = cpumask_of(smp_processor_id());
619
620 if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
621 levt->name = "lapic-deadline";
622 levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
623 CLOCK_EVT_FEAT_DUMMY);
624 levt->set_next_event = lapic_next_deadline;
625 clockevents_config_and_register(levt,
626 tsc_khz * (1000 / TSC_DIVISOR),
627 0xF, ~0UL);
628 } else
629 clockevents_register_device(levt);
630}
631
632/*
633 * Install the updated TSC frequency from recalibration at the TSC
634 * deadline clockevent devices.
635 */
636static void __lapic_update_tsc_freq(void *info)
637{
638 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
639
640 if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
641 return;
642
643 clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
644}
645
646void lapic_update_tsc_freq(void)
647{
648 /*
649 * The clockevent device's ->mult and ->shift can both be
650 * changed. In order to avoid races, schedule the frequency
651 * update code on each CPU.
652 */
653 on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
654}
655
656/*
657 * In this functions we calibrate APIC bus clocks to the external timer.
658 *
659 * We want to do the calibration only once since we want to have local timer
660 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
661 * frequency.
662 *
663 * This was previously done by reading the PIT/HPET and waiting for a wrap
664 * around to find out, that a tick has elapsed. I have a box, where the PIT
665 * readout is broken, so it never gets out of the wait loop again. This was
666 * also reported by others.
667 *
668 * Monitoring the jiffies value is inaccurate and the clockevents
669 * infrastructure allows us to do a simple substitution of the interrupt
670 * handler.
671 *
672 * The calibration routine also uses the pm_timer when possible, as the PIT
673 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
674 * back to normal later in the boot process).
675 */
676
677#define LAPIC_CAL_LOOPS (HZ/10)
678
679static __initdata int lapic_cal_loops = -1;
680static __initdata long lapic_cal_t1, lapic_cal_t2;
681static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
682static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
683static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
684
685/*
686 * Temporary interrupt handler and polled calibration function.
687 */
688static void __init lapic_cal_handler(struct clock_event_device *dev)
689{
690 unsigned long long tsc = 0;
691 long tapic = apic_read(APIC_TMCCT);
692 unsigned long pm = acpi_pm_read_early();
693
694 if (boot_cpu_has(X86_FEATURE_TSC))
695 tsc = rdtsc();
696
697 switch (lapic_cal_loops++) {
698 case 0:
699 lapic_cal_t1 = tapic;
700 lapic_cal_tsc1 = tsc;
701 lapic_cal_pm1 = pm;
702 lapic_cal_j1 = jiffies;
703 break;
704
705 case LAPIC_CAL_LOOPS:
706 lapic_cal_t2 = tapic;
707 lapic_cal_tsc2 = tsc;
708 if (pm < lapic_cal_pm1)
709 pm += ACPI_PM_OVRRUN;
710 lapic_cal_pm2 = pm;
711 lapic_cal_j2 = jiffies;
712 break;
713 }
714}
715
716static int __init
717calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
718{
719 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
720 const long pm_thresh = pm_100ms / 100;
721 unsigned long mult;
722 u64 res;
723
724#ifndef CONFIG_X86_PM_TIMER
725 return -1;
726#endif
727
728 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
729
730 /* Check, if the PM timer is available */
731 if (!deltapm)
732 return -1;
733
734 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
735
736 if (deltapm > (pm_100ms - pm_thresh) &&
737 deltapm < (pm_100ms + pm_thresh)) {
738 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
739 return 0;
740 }
741
742 res = (((u64)deltapm) * mult) >> 22;
743 do_div(res, 1000000);
744 pr_warn("APIC calibration not consistent "
745 "with PM-Timer: %ldms instead of 100ms\n", (long)res);
746
747 /* Correct the lapic counter value */
748 res = (((u64)(*delta)) * pm_100ms);
749 do_div(res, deltapm);
750 pr_info("APIC delta adjusted to PM-Timer: "
751 "%lu (%ld)\n", (unsigned long)res, *delta);
752 *delta = (long)res;
753
754 /* Correct the tsc counter value */
755 if (boot_cpu_has(X86_FEATURE_TSC)) {
756 res = (((u64)(*deltatsc)) * pm_100ms);
757 do_div(res, deltapm);
758 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
759 "PM-Timer: %lu (%ld)\n",
760 (unsigned long)res, *deltatsc);
761 *deltatsc = (long)res;
762 }
763
764 return 0;
765}
766
767static int __init lapic_init_clockevent(void)
768{
769 if (!lapic_timer_period)
770 return -1;
771
772 /* Calculate the scaled math multiplication factor */
773 lapic_clockevent.mult = div_sc(lapic_timer_period/APIC_DIVISOR,
774 TICK_NSEC, lapic_clockevent.shift);
775 lapic_clockevent.max_delta_ns =
776 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
777 lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
778 lapic_clockevent.min_delta_ns =
779 clockevent_delta2ns(0xF, &lapic_clockevent);
780 lapic_clockevent.min_delta_ticks = 0xF;
781
782 return 0;
783}
784
785bool __init apic_needs_pit(void)
786{
787 /*
788 * If the frequencies are not known, PIT is required for both TSC
789 * and apic timer calibration.
790 */
791 if (!tsc_khz || !cpu_khz)
792 return true;
793
794 /* Is there an APIC at all or is it disabled? */
795 if (!boot_cpu_has(X86_FEATURE_APIC) || disable_apic)
796 return true;
797
798 /*
799 * If interrupt delivery mode is legacy PIC or virtual wire without
800 * configuration, the local APIC timer wont be set up. Make sure
801 * that the PIT is initialized.
802 */
803 if (apic_intr_mode == APIC_PIC ||
804 apic_intr_mode == APIC_VIRTUAL_WIRE_NO_CONFIG)
805 return true;
806
807 /* Virt guests may lack ARAT, but still have DEADLINE */
808 if (!boot_cpu_has(X86_FEATURE_ARAT))
809 return true;
810
811 /* Deadline timer is based on TSC so no further PIT action required */
812 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
813 return false;
814
815 /* APIC timer disabled? */
816 if (disable_apic_timer)
817 return true;
818 /*
819 * The APIC timer frequency is known already, no PIT calibration
820 * required. If unknown, let the PIT be initialized.
821 */
822 return lapic_timer_period == 0;
823}
824
825static int __init calibrate_APIC_clock(void)
826{
827 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
828 u64 tsc_perj = 0, tsc_start = 0;
829 unsigned long jif_start;
830 unsigned long deltaj;
831 long delta, deltatsc;
832 int pm_referenced = 0;
833
834 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
835 return 0;
836
837 /*
838 * Check if lapic timer has already been calibrated by platform
839 * specific routine, such as tsc calibration code. If so just fill
840 * in the clockevent structure and return.
841 */
842 if (!lapic_init_clockevent()) {
843 apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
844 lapic_timer_period);
845 /*
846 * Direct calibration methods must have an always running
847 * local APIC timer, no need for broadcast timer.
848 */
849 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
850 return 0;
851 }
852
853 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
854 "calibrating APIC timer ...\n");
855
856 /*
857 * There are platforms w/o global clockevent devices. Instead of
858 * making the calibration conditional on that, use a polling based
859 * approach everywhere.
860 */
861 local_irq_disable();
862
863 /*
864 * Setup the APIC counter to maximum. There is no way the lapic
865 * can underflow in the 100ms detection time frame
866 */
867 __setup_APIC_LVTT(0xffffffff, 0, 0);
868
869 /*
870 * Methods to terminate the calibration loop:
871 * 1) Global clockevent if available (jiffies)
872 * 2) TSC if available and frequency is known
873 */
874 jif_start = READ_ONCE(jiffies);
875
876 if (tsc_khz) {
877 tsc_start = rdtsc();
878 tsc_perj = div_u64((u64)tsc_khz * 1000, HZ);
879 }
880
881 /*
882 * Enable interrupts so the tick can fire, if a global
883 * clockevent device is available
884 */
885 local_irq_enable();
886
887 while (lapic_cal_loops <= LAPIC_CAL_LOOPS) {
888 /* Wait for a tick to elapse */
889 while (1) {
890 if (tsc_khz) {
891 u64 tsc_now = rdtsc();
892 if ((tsc_now - tsc_start) >= tsc_perj) {
893 tsc_start += tsc_perj;
894 break;
895 }
896 } else {
897 unsigned long jif_now = READ_ONCE(jiffies);
898
899 if (time_after(jif_now, jif_start)) {
900 jif_start = jif_now;
901 break;
902 }
903 }
904 cpu_relax();
905 }
906
907 /* Invoke the calibration routine */
908 local_irq_disable();
909 lapic_cal_handler(NULL);
910 local_irq_enable();
911 }
912
913 local_irq_disable();
914
915 /* Build delta t1-t2 as apic timer counts down */
916 delta = lapic_cal_t1 - lapic_cal_t2;
917 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
918
919 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
920
921 /* we trust the PM based calibration if possible */
922 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
923 &delta, &deltatsc);
924
925 lapic_timer_period = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
926 lapic_init_clockevent();
927
928 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
929 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
930 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
931 lapic_timer_period);
932
933 if (boot_cpu_has(X86_FEATURE_TSC)) {
934 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
935 "%ld.%04ld MHz.\n",
936 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
937 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
938 }
939
940 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
941 "%u.%04u MHz.\n",
942 lapic_timer_period / (1000000 / HZ),
943 lapic_timer_period % (1000000 / HZ));
944
945 /*
946 * Do a sanity check on the APIC calibration result
947 */
948 if (lapic_timer_period < (1000000 / HZ)) {
949 local_irq_enable();
950 pr_warn("APIC frequency too slow, disabling apic timer\n");
951 return -1;
952 }
953
954 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
955
956 /*
957 * PM timer calibration failed or not turned on so lets try APIC
958 * timer based calibration, if a global clockevent device is
959 * available.
960 */
961 if (!pm_referenced && global_clock_event) {
962 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
963
964 /*
965 * Setup the apic timer manually
966 */
967 levt->event_handler = lapic_cal_handler;
968 lapic_timer_set_periodic(levt);
969 lapic_cal_loops = -1;
970
971 /* Let the interrupts run */
972 local_irq_enable();
973
974 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
975 cpu_relax();
976
977 /* Stop the lapic timer */
978 local_irq_disable();
979 lapic_timer_shutdown(levt);
980
981 /* Jiffies delta */
982 deltaj = lapic_cal_j2 - lapic_cal_j1;
983 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
984
985 /* Check, if the jiffies result is consistent */
986 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
987 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
988 else
989 levt->features |= CLOCK_EVT_FEAT_DUMMY;
990 }
991 local_irq_enable();
992
993 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
994 pr_warn("APIC timer disabled due to verification failure\n");
995 return -1;
996 }
997
998 return 0;
999}
1000
1001/*
1002 * Setup the boot APIC
1003 *
1004 * Calibrate and verify the result.
1005 */
1006void __init setup_boot_APIC_clock(void)
1007{
1008 /*
1009 * The local apic timer can be disabled via the kernel
1010 * commandline or from the CPU detection code. Register the lapic
1011 * timer as a dummy clock event source on SMP systems, so the
1012 * broadcast mechanism is used. On UP systems simply ignore it.
1013 */
1014 if (disable_apic_timer) {
1015 pr_info("Disabling APIC timer\n");
1016 /* No broadcast on UP ! */
1017 if (num_possible_cpus() > 1) {
1018 lapic_clockevent.mult = 1;
1019 setup_APIC_timer();
1020 }
1021 return;
1022 }
1023
1024 if (calibrate_APIC_clock()) {
1025 /* No broadcast on UP ! */
1026 if (num_possible_cpus() > 1)
1027 setup_APIC_timer();
1028 return;
1029 }
1030
1031 /*
1032 * If nmi_watchdog is set to IO_APIC, we need the
1033 * PIT/HPET going. Otherwise register lapic as a dummy
1034 * device.
1035 */
1036 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
1037
1038 /* Setup the lapic or request the broadcast */
1039 setup_APIC_timer();
1040 amd_e400_c1e_apic_setup();
1041}
1042
1043void setup_secondary_APIC_clock(void)
1044{
1045 setup_APIC_timer();
1046 amd_e400_c1e_apic_setup();
1047}
1048
1049/*
1050 * The guts of the apic timer interrupt
1051 */
1052static void local_apic_timer_interrupt(void)
1053{
1054 struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1055
1056 /*
1057 * Normally we should not be here till LAPIC has been initialized but
1058 * in some cases like kdump, its possible that there is a pending LAPIC
1059 * timer interrupt from previous kernel's context and is delivered in
1060 * new kernel the moment interrupts are enabled.
1061 *
1062 * Interrupts are enabled early and LAPIC is setup much later, hence
1063 * its possible that when we get here evt->event_handler is NULL.
1064 * Check for event_handler being NULL and discard the interrupt as
1065 * spurious.
1066 */
1067 if (!evt->event_handler) {
1068 pr_warn("Spurious LAPIC timer interrupt on cpu %d\n",
1069 smp_processor_id());
1070 /* Switch it off */
1071 lapic_timer_shutdown(evt);
1072 return;
1073 }
1074
1075 /*
1076 * the NMI deadlock-detector uses this.
1077 */
1078 inc_irq_stat(apic_timer_irqs);
1079
1080 evt->event_handler(evt);
1081}
1082
1083/*
1084 * Local APIC timer interrupt. This is the most natural way for doing
1085 * local interrupts, but local timer interrupts can be emulated by
1086 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1087 *
1088 * [ if a single-CPU system runs an SMP kernel then we call the local
1089 * interrupt as well. Thus we cannot inline the local irq ... ]
1090 */
1091DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt)
1092{
1093 struct pt_regs *old_regs = set_irq_regs(regs);
1094
1095 ack_APIC_irq();
1096 trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1097 local_apic_timer_interrupt();
1098 trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1099
1100 set_irq_regs(old_regs);
1101}
1102
1103int setup_profiling_timer(unsigned int multiplier)
1104{
1105 return -EINVAL;
1106}
1107
1108/*
1109 * Local APIC start and shutdown
1110 */
1111
1112/**
1113 * clear_local_APIC - shutdown the local APIC
1114 *
1115 * This is called, when a CPU is disabled and before rebooting, so the state of
1116 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1117 * leftovers during boot.
1118 */
1119void clear_local_APIC(void)
1120{
1121 int maxlvt;
1122 u32 v;
1123
1124 /* APIC hasn't been mapped yet */
1125 if (!x2apic_mode && !apic_phys)
1126 return;
1127
1128 maxlvt = lapic_get_maxlvt();
1129 /*
1130 * Masking an LVT entry can trigger a local APIC error
1131 * if the vector is zero. Mask LVTERR first to prevent this.
1132 */
1133 if (maxlvt >= 3) {
1134 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1135 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1136 }
1137 /*
1138 * Careful: we have to set masks only first to deassert
1139 * any level-triggered sources.
1140 */
1141 v = apic_read(APIC_LVTT);
1142 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1143 v = apic_read(APIC_LVT0);
1144 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1145 v = apic_read(APIC_LVT1);
1146 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1147 if (maxlvt >= 4) {
1148 v = apic_read(APIC_LVTPC);
1149 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1150 }
1151
1152 /* lets not touch this if we didn't frob it */
1153#ifdef CONFIG_X86_THERMAL_VECTOR
1154 if (maxlvt >= 5) {
1155 v = apic_read(APIC_LVTTHMR);
1156 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1157 }
1158#endif
1159#ifdef CONFIG_X86_MCE_INTEL
1160 if (maxlvt >= 6) {
1161 v = apic_read(APIC_LVTCMCI);
1162 if (!(v & APIC_LVT_MASKED))
1163 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1164 }
1165#endif
1166
1167 /*
1168 * Clean APIC state for other OSs:
1169 */
1170 apic_write(APIC_LVTT, APIC_LVT_MASKED);
1171 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1172 apic_write(APIC_LVT1, APIC_LVT_MASKED);
1173 if (maxlvt >= 3)
1174 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1175 if (maxlvt >= 4)
1176 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1177
1178 /* Integrated APIC (!82489DX) ? */
1179 if (lapic_is_integrated()) {
1180 if (maxlvt > 3)
1181 /* Clear ESR due to Pentium errata 3AP and 11AP */
1182 apic_write(APIC_ESR, 0);
1183 apic_read(APIC_ESR);
1184 }
1185}
1186
1187/**
1188 * apic_soft_disable - Clears and software disables the local APIC on hotplug
1189 *
1190 * Contrary to disable_local_APIC() this does not touch the enable bit in
1191 * MSR_IA32_APICBASE. Clearing that bit on systems based on the 3 wire APIC
1192 * bus would require a hardware reset as the APIC would lose track of bus
1193 * arbitration. On systems with FSB delivery APICBASE could be disabled,
1194 * but it has to be guaranteed that no interrupt is sent to the APIC while
1195 * in that state and it's not clear from the SDM whether it still responds
1196 * to INIT/SIPI messages. Stay on the safe side and use software disable.
1197 */
1198void apic_soft_disable(void)
1199{
1200 u32 value;
1201
1202 clear_local_APIC();
1203
1204 /* Soft disable APIC (implies clearing of registers for 82489DX!). */
1205 value = apic_read(APIC_SPIV);
1206 value &= ~APIC_SPIV_APIC_ENABLED;
1207 apic_write(APIC_SPIV, value);
1208}
1209
1210/**
1211 * disable_local_APIC - clear and disable the local APIC
1212 */
1213void disable_local_APIC(void)
1214{
1215 /* APIC hasn't been mapped yet */
1216 if (!x2apic_mode && !apic_phys)
1217 return;
1218
1219 apic_soft_disable();
1220
1221#ifdef CONFIG_X86_32
1222 /*
1223 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1224 * restore the disabled state.
1225 */
1226 if (enabled_via_apicbase) {
1227 unsigned int l, h;
1228
1229 rdmsr(MSR_IA32_APICBASE, l, h);
1230 l &= ~MSR_IA32_APICBASE_ENABLE;
1231 wrmsr(MSR_IA32_APICBASE, l, h);
1232 }
1233#endif
1234}
1235
1236/*
1237 * If Linux enabled the LAPIC against the BIOS default disable it down before
1238 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1239 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1240 * for the case where Linux didn't enable the LAPIC.
1241 */
1242void lapic_shutdown(void)
1243{
1244 unsigned long flags;
1245
1246 if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1247 return;
1248
1249 local_irq_save(flags);
1250
1251#ifdef CONFIG_X86_32
1252 if (!enabled_via_apicbase)
1253 clear_local_APIC();
1254 else
1255#endif
1256 disable_local_APIC();
1257
1258
1259 local_irq_restore(flags);
1260}
1261
1262/**
1263 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1264 */
1265void __init sync_Arb_IDs(void)
1266{
1267 /*
1268 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1269 * needed on AMD.
1270 */
1271 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1272 return;
1273
1274 /*
1275 * Wait for idle.
1276 */
1277 apic_wait_icr_idle();
1278
1279 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1280 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1281 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1282}
1283
1284enum apic_intr_mode_id apic_intr_mode __ro_after_init;
1285
1286static int __init __apic_intr_mode_select(void)
1287{
1288 /* Check kernel option */
1289 if (disable_apic) {
1290 pr_info("APIC disabled via kernel command line\n");
1291 return APIC_PIC;
1292 }
1293
1294 /* Check BIOS */
1295#ifdef CONFIG_X86_64
1296 /* On 64-bit, the APIC must be integrated, Check local APIC only */
1297 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1298 disable_apic = 1;
1299 pr_info("APIC disabled by BIOS\n");
1300 return APIC_PIC;
1301 }
1302#else
1303 /* On 32-bit, the APIC may be integrated APIC or 82489DX */
1304
1305 /* Neither 82489DX nor integrated APIC ? */
1306 if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1307 disable_apic = 1;
1308 return APIC_PIC;
1309 }
1310
1311 /* If the BIOS pretends there is an integrated APIC ? */
1312 if (!boot_cpu_has(X86_FEATURE_APIC) &&
1313 APIC_INTEGRATED(boot_cpu_apic_version)) {
1314 disable_apic = 1;
1315 pr_err(FW_BUG "Local APIC %d not detected, force emulation\n",
1316 boot_cpu_physical_apicid);
1317 return APIC_PIC;
1318 }
1319#endif
1320
1321 /* Check MP table or ACPI MADT configuration */
1322 if (!smp_found_config) {
1323 disable_ioapic_support();
1324 if (!acpi_lapic) {
1325 pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1326 return APIC_VIRTUAL_WIRE_NO_CONFIG;
1327 }
1328 return APIC_VIRTUAL_WIRE;
1329 }
1330
1331#ifdef CONFIG_SMP
1332 /* If SMP should be disabled, then really disable it! */
1333 if (!setup_max_cpus) {
1334 pr_info("APIC: SMP mode deactivated\n");
1335 return APIC_SYMMETRIC_IO_NO_ROUTING;
1336 }
1337
1338 if (read_apic_id() != boot_cpu_physical_apicid) {
1339 panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1340 read_apic_id(), boot_cpu_physical_apicid);
1341 /* Or can we switch back to PIC here? */
1342 }
1343#endif
1344
1345 return APIC_SYMMETRIC_IO;
1346}
1347
1348/* Select the interrupt delivery mode for the BSP */
1349void __init apic_intr_mode_select(void)
1350{
1351 apic_intr_mode = __apic_intr_mode_select();
1352}
1353
1354/*
1355 * An initial setup of the virtual wire mode.
1356 */
1357void __init init_bsp_APIC(void)
1358{
1359 unsigned int value;
1360
1361 /*
1362 * Don't do the setup now if we have a SMP BIOS as the
1363 * through-I/O-APIC virtual wire mode might be active.
1364 */
1365 if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1366 return;
1367
1368 /*
1369 * Do not trust the local APIC being empty at bootup.
1370 */
1371 clear_local_APIC();
1372
1373 /*
1374 * Enable APIC.
1375 */
1376 value = apic_read(APIC_SPIV);
1377 value &= ~APIC_VECTOR_MASK;
1378 value |= APIC_SPIV_APIC_ENABLED;
1379
1380#ifdef CONFIG_X86_32
1381 /* This bit is reserved on P4/Xeon and should be cleared */
1382 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1383 (boot_cpu_data.x86 == 15))
1384 value &= ~APIC_SPIV_FOCUS_DISABLED;
1385 else
1386#endif
1387 value |= APIC_SPIV_FOCUS_DISABLED;
1388 value |= SPURIOUS_APIC_VECTOR;
1389 apic_write(APIC_SPIV, value);
1390
1391 /*
1392 * Set up the virtual wire mode.
1393 */
1394 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1395 value = APIC_DM_NMI;
1396 if (!lapic_is_integrated()) /* 82489DX */
1397 value |= APIC_LVT_LEVEL_TRIGGER;
1398 if (apic_extnmi == APIC_EXTNMI_NONE)
1399 value |= APIC_LVT_MASKED;
1400 apic_write(APIC_LVT1, value);
1401}
1402
1403static void __init apic_bsp_setup(bool upmode);
1404
1405/* Init the interrupt delivery mode for the BSP */
1406void __init apic_intr_mode_init(void)
1407{
1408 bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1409
1410 switch (apic_intr_mode) {
1411 case APIC_PIC:
1412 pr_info("APIC: Keep in PIC mode(8259)\n");
1413 return;
1414 case APIC_VIRTUAL_WIRE:
1415 pr_info("APIC: Switch to virtual wire mode setup\n");
1416 default_setup_apic_routing();
1417 break;
1418 case APIC_VIRTUAL_WIRE_NO_CONFIG:
1419 pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1420 upmode = true;
1421 default_setup_apic_routing();
1422 break;
1423 case APIC_SYMMETRIC_IO:
1424 pr_info("APIC: Switch to symmetric I/O mode setup\n");
1425 default_setup_apic_routing();
1426 break;
1427 case APIC_SYMMETRIC_IO_NO_ROUTING:
1428 pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1429 break;
1430 }
1431
1432 apic_bsp_setup(upmode);
1433}
1434
1435static void lapic_setup_esr(void)
1436{
1437 unsigned int oldvalue, value, maxlvt;
1438
1439 if (!lapic_is_integrated()) {
1440 pr_info("No ESR for 82489DX.\n");
1441 return;
1442 }
1443
1444 if (apic->disable_esr) {
1445 /*
1446 * Something untraceable is creating bad interrupts on
1447 * secondary quads ... for the moment, just leave the
1448 * ESR disabled - we can't do anything useful with the
1449 * errors anyway - mbligh
1450 */
1451 pr_info("Leaving ESR disabled.\n");
1452 return;
1453 }
1454
1455 maxlvt = lapic_get_maxlvt();
1456 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1457 apic_write(APIC_ESR, 0);
1458 oldvalue = apic_read(APIC_ESR);
1459
1460 /* enables sending errors */
1461 value = ERROR_APIC_VECTOR;
1462 apic_write(APIC_LVTERR, value);
1463
1464 /*
1465 * spec says clear errors after enabling vector.
1466 */
1467 if (maxlvt > 3)
1468 apic_write(APIC_ESR, 0);
1469 value = apic_read(APIC_ESR);
1470 if (value != oldvalue)
1471 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1472 "vector: 0x%08x after: 0x%08x\n",
1473 oldvalue, value);
1474}
1475
1476#define APIC_IR_REGS APIC_ISR_NR
1477#define APIC_IR_BITS (APIC_IR_REGS * 32)
1478#define APIC_IR_MAPSIZE (APIC_IR_BITS / BITS_PER_LONG)
1479
1480union apic_ir {
1481 unsigned long map[APIC_IR_MAPSIZE];
1482 u32 regs[APIC_IR_REGS];
1483};
1484
1485static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr)
1486{
1487 int i, bit;
1488
1489 /* Read the IRRs */
1490 for (i = 0; i < APIC_IR_REGS; i++)
1491 irr->regs[i] = apic_read(APIC_IRR + i * 0x10);
1492
1493 /* Read the ISRs */
1494 for (i = 0; i < APIC_IR_REGS; i++)
1495 isr->regs[i] = apic_read(APIC_ISR + i * 0x10);
1496
1497 /*
1498 * If the ISR map is not empty. ACK the APIC and run another round
1499 * to verify whether a pending IRR has been unblocked and turned
1500 * into a ISR.
1501 */
1502 if (!bitmap_empty(isr->map, APIC_IR_BITS)) {
1503 /*
1504 * There can be multiple ISR bits set when a high priority
1505 * interrupt preempted a lower priority one. Issue an ACK
1506 * per set bit.
1507 */
1508 for_each_set_bit(bit, isr->map, APIC_IR_BITS)
1509 ack_APIC_irq();
1510 return true;
1511 }
1512
1513 return !bitmap_empty(irr->map, APIC_IR_BITS);
1514}
1515
1516/*
1517 * After a crash, we no longer service the interrupts and a pending
1518 * interrupt from previous kernel might still have ISR bit set.
1519 *
1520 * Most probably by now the CPU has serviced that pending interrupt and it
1521 * might not have done the ack_APIC_irq() because it thought, interrupt
1522 * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear
1523 * the ISR bit and cpu thinks it has already serivced the interrupt. Hence
1524 * a vector might get locked. It was noticed for timer irq (vector
1525 * 0x31). Issue an extra EOI to clear ISR.
1526 *
1527 * If there are pending IRR bits they turn into ISR bits after a higher
1528 * priority ISR bit has been acked.
1529 */
1530static void apic_pending_intr_clear(void)
1531{
1532 union apic_ir irr, isr;
1533 unsigned int i;
1534
1535 /* 512 loops are way oversized and give the APIC a chance to obey. */
1536 for (i = 0; i < 512; i++) {
1537 if (!apic_check_and_ack(&irr, &isr))
1538 return;
1539 }
1540 /* Dump the IRR/ISR content if that failed */
1541 pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map);
1542}
1543
1544/**
1545 * setup_local_APIC - setup the local APIC
1546 *
1547 * Used to setup local APIC while initializing BSP or bringing up APs.
1548 * Always called with preemption disabled.
1549 */
1550static void setup_local_APIC(void)
1551{
1552 int cpu = smp_processor_id();
1553 unsigned int value;
1554
1555 if (disable_apic) {
1556 disable_ioapic_support();
1557 return;
1558 }
1559
1560 /*
1561 * If this comes from kexec/kcrash the APIC might be enabled in
1562 * SPIV. Soft disable it before doing further initialization.
1563 */
1564 value = apic_read(APIC_SPIV);
1565 value &= ~APIC_SPIV_APIC_ENABLED;
1566 apic_write(APIC_SPIV, value);
1567
1568#ifdef CONFIG_X86_32
1569 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1570 if (lapic_is_integrated() && apic->disable_esr) {
1571 apic_write(APIC_ESR, 0);
1572 apic_write(APIC_ESR, 0);
1573 apic_write(APIC_ESR, 0);
1574 apic_write(APIC_ESR, 0);
1575 }
1576#endif
1577 /*
1578 * Double-check whether this APIC is really registered.
1579 * This is meaningless in clustered apic mode, so we skip it.
1580 */
1581 BUG_ON(!apic->apic_id_registered());
1582
1583 /*
1584 * Intel recommends to set DFR, LDR and TPR before enabling
1585 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1586 * document number 292116). So here it goes...
1587 */
1588 apic->init_apic_ldr();
1589
1590#ifdef CONFIG_X86_32
1591 if (apic->dest_logical) {
1592 int logical_apicid, ldr_apicid;
1593
1594 /*
1595 * APIC LDR is initialized. If logical_apicid mapping was
1596 * initialized during get_smp_config(), make sure it matches
1597 * the actual value.
1598 */
1599 logical_apicid = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1600 ldr_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1601 if (logical_apicid != BAD_APICID)
1602 WARN_ON(logical_apicid != ldr_apicid);
1603 /* Always use the value from LDR. */
1604 early_per_cpu(x86_cpu_to_logical_apicid, cpu) = ldr_apicid;
1605 }
1606#endif
1607
1608 /*
1609 * Set Task Priority to 'accept all except vectors 0-31'. An APIC
1610 * vector in the 16-31 range could be delivered if TPR == 0, but we
1611 * would think it's an exception and terrible things will happen. We
1612 * never change this later on.
1613 */
1614 value = apic_read(APIC_TASKPRI);
1615 value &= ~APIC_TPRI_MASK;
1616 value |= 0x10;
1617 apic_write(APIC_TASKPRI, value);
1618
1619 /* Clear eventually stale ISR/IRR bits */
1620 apic_pending_intr_clear();
1621
1622 /*
1623 * Now that we are all set up, enable the APIC
1624 */
1625 value = apic_read(APIC_SPIV);
1626 value &= ~APIC_VECTOR_MASK;
1627 /*
1628 * Enable APIC
1629 */
1630 value |= APIC_SPIV_APIC_ENABLED;
1631
1632#ifdef CONFIG_X86_32
1633 /*
1634 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1635 * certain networking cards. If high frequency interrupts are
1636 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1637 * entry is masked/unmasked at a high rate as well then sooner or
1638 * later IOAPIC line gets 'stuck', no more interrupts are received
1639 * from the device. If focus CPU is disabled then the hang goes
1640 * away, oh well :-(
1641 *
1642 * [ This bug can be reproduced easily with a level-triggered
1643 * PCI Ne2000 networking cards and PII/PIII processors, dual
1644 * BX chipset. ]
1645 */
1646 /*
1647 * Actually disabling the focus CPU check just makes the hang less
1648 * frequent as it makes the interrupt distributon model be more
1649 * like LRU than MRU (the short-term load is more even across CPUs).
1650 */
1651
1652 /*
1653 * - enable focus processor (bit==0)
1654 * - 64bit mode always use processor focus
1655 * so no need to set it
1656 */
1657 value &= ~APIC_SPIV_FOCUS_DISABLED;
1658#endif
1659
1660 /*
1661 * Set spurious IRQ vector
1662 */
1663 value |= SPURIOUS_APIC_VECTOR;
1664 apic_write(APIC_SPIV, value);
1665
1666 perf_events_lapic_init();
1667
1668 /*
1669 * Set up LVT0, LVT1:
1670 *
1671 * set up through-local-APIC on the boot CPU's LINT0. This is not
1672 * strictly necessary in pure symmetric-IO mode, but sometimes
1673 * we delegate interrupts to the 8259A.
1674 */
1675 /*
1676 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1677 */
1678 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1679 if (!cpu && (pic_mode || !value || skip_ioapic_setup)) {
1680 value = APIC_DM_EXTINT;
1681 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1682 } else {
1683 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1684 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1685 }
1686 apic_write(APIC_LVT0, value);
1687
1688 /*
1689 * Only the BSP sees the LINT1 NMI signal by default. This can be
1690 * modified by apic_extnmi= boot option.
1691 */
1692 if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1693 apic_extnmi == APIC_EXTNMI_ALL)
1694 value = APIC_DM_NMI;
1695 else
1696 value = APIC_DM_NMI | APIC_LVT_MASKED;
1697
1698 /* Is 82489DX ? */
1699 if (!lapic_is_integrated())
1700 value |= APIC_LVT_LEVEL_TRIGGER;
1701 apic_write(APIC_LVT1, value);
1702
1703#ifdef CONFIG_X86_MCE_INTEL
1704 /* Recheck CMCI information after local APIC is up on CPU #0 */
1705 if (!cpu)
1706 cmci_recheck();
1707#endif
1708}
1709
1710static void end_local_APIC_setup(void)
1711{
1712 lapic_setup_esr();
1713
1714#ifdef CONFIG_X86_32
1715 {
1716 unsigned int value;
1717 /* Disable the local apic timer */
1718 value = apic_read(APIC_LVTT);
1719 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1720 apic_write(APIC_LVTT, value);
1721 }
1722#endif
1723
1724 apic_pm_activate();
1725}
1726
1727/*
1728 * APIC setup function for application processors. Called from smpboot.c
1729 */
1730void apic_ap_setup(void)
1731{
1732 setup_local_APIC();
1733 end_local_APIC_setup();
1734}
1735
1736#ifdef CONFIG_X86_X2APIC
1737int x2apic_mode;
1738
1739enum {
1740 X2APIC_OFF,
1741 X2APIC_ON,
1742 X2APIC_DISABLED,
1743};
1744static int x2apic_state;
1745
1746static void __x2apic_disable(void)
1747{
1748 u64 msr;
1749
1750 if (!boot_cpu_has(X86_FEATURE_APIC))
1751 return;
1752
1753 rdmsrl(MSR_IA32_APICBASE, msr);
1754 if (!(msr & X2APIC_ENABLE))
1755 return;
1756 /* Disable xapic and x2apic first and then reenable xapic mode */
1757 wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1758 wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1759 printk_once(KERN_INFO "x2apic disabled\n");
1760}
1761
1762static void __x2apic_enable(void)
1763{
1764 u64 msr;
1765
1766 rdmsrl(MSR_IA32_APICBASE, msr);
1767 if (msr & X2APIC_ENABLE)
1768 return;
1769 wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1770 printk_once(KERN_INFO "x2apic enabled\n");
1771}
1772
1773static int __init setup_nox2apic(char *str)
1774{
1775 if (x2apic_enabled()) {
1776 int apicid = native_apic_msr_read(APIC_ID);
1777
1778 if (apicid >= 255) {
1779 pr_warn("Apicid: %08x, cannot enforce nox2apic\n",
1780 apicid);
1781 return 0;
1782 }
1783 pr_warn("x2apic already enabled.\n");
1784 __x2apic_disable();
1785 }
1786 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1787 x2apic_state = X2APIC_DISABLED;
1788 x2apic_mode = 0;
1789 return 0;
1790}
1791early_param("nox2apic", setup_nox2apic);
1792
1793/* Called from cpu_init() to enable x2apic on (secondary) cpus */
1794void x2apic_setup(void)
1795{
1796 /*
1797 * If x2apic is not in ON state, disable it if already enabled
1798 * from BIOS.
1799 */
1800 if (x2apic_state != X2APIC_ON) {
1801 __x2apic_disable();
1802 return;
1803 }
1804 __x2apic_enable();
1805}
1806
1807static __init void x2apic_disable(void)
1808{
1809 u32 x2apic_id, state = x2apic_state;
1810
1811 x2apic_mode = 0;
1812 x2apic_state = X2APIC_DISABLED;
1813
1814 if (state != X2APIC_ON)
1815 return;
1816
1817 x2apic_id = read_apic_id();
1818 if (x2apic_id >= 255)
1819 panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1820
1821 __x2apic_disable();
1822 register_lapic_address(mp_lapic_addr);
1823}
1824
1825static __init void x2apic_enable(void)
1826{
1827 if (x2apic_state != X2APIC_OFF)
1828 return;
1829
1830 x2apic_mode = 1;
1831 x2apic_state = X2APIC_ON;
1832 __x2apic_enable();
1833}
1834
1835static __init void try_to_enable_x2apic(int remap_mode)
1836{
1837 if (x2apic_state == X2APIC_DISABLED)
1838 return;
1839
1840 if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1841 /* IR is required if there is APIC ID > 255 even when running
1842 * under KVM
1843 */
1844 if (max_physical_apicid > 255 ||
1845 !x86_init.hyper.x2apic_available()) {
1846 pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1847 x2apic_disable();
1848 return;
1849 }
1850
1851 /*
1852 * without IR all CPUs can be addressed by IOAPIC/MSI
1853 * only in physical mode
1854 */
1855 x2apic_phys = 1;
1856 }
1857 x2apic_enable();
1858}
1859
1860void __init check_x2apic(void)
1861{
1862 if (x2apic_enabled()) {
1863 pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1864 x2apic_mode = 1;
1865 x2apic_state = X2APIC_ON;
1866 } else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1867 x2apic_state = X2APIC_DISABLED;
1868 }
1869}
1870#else /* CONFIG_X86_X2APIC */
1871static int __init validate_x2apic(void)
1872{
1873 if (!apic_is_x2apic_enabled())
1874 return 0;
1875 /*
1876 * Checkme: Can we simply turn off x2apic here instead of panic?
1877 */
1878 panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n");
1879}
1880early_initcall(validate_x2apic);
1881
1882static inline void try_to_enable_x2apic(int remap_mode) { }
1883static inline void __x2apic_enable(void) { }
1884#endif /* !CONFIG_X86_X2APIC */
1885
1886void __init enable_IR_x2apic(void)
1887{
1888 unsigned long flags;
1889 int ret, ir_stat;
1890
1891 if (skip_ioapic_setup) {
1892 pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1893 return;
1894 }
1895
1896 ir_stat = irq_remapping_prepare();
1897 if (ir_stat < 0 && !x2apic_supported())
1898 return;
1899
1900 ret = save_ioapic_entries();
1901 if (ret) {
1902 pr_info("Saving IO-APIC state failed: %d\n", ret);
1903 return;
1904 }
1905
1906 local_irq_save(flags);
1907 legacy_pic->mask_all();
1908 mask_ioapic_entries();
1909
1910 /* If irq_remapping_prepare() succeeded, try to enable it */
1911 if (ir_stat >= 0)
1912 ir_stat = irq_remapping_enable();
1913 /* ir_stat contains the remap mode or an error code */
1914 try_to_enable_x2apic(ir_stat);
1915
1916 if (ir_stat < 0)
1917 restore_ioapic_entries();
1918 legacy_pic->restore_mask();
1919 local_irq_restore(flags);
1920}
1921
1922#ifdef CONFIG_X86_64
1923/*
1924 * Detect and enable local APICs on non-SMP boards.
1925 * Original code written by Keir Fraser.
1926 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1927 * not correctly set up (usually the APIC timer won't work etc.)
1928 */
1929static int __init detect_init_APIC(void)
1930{
1931 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1932 pr_info("No local APIC present\n");
1933 return -1;
1934 }
1935
1936 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1937 return 0;
1938}
1939#else
1940
1941static int __init apic_verify(void)
1942{
1943 u32 features, h, l;
1944
1945 /*
1946 * The APIC feature bit should now be enabled
1947 * in `cpuid'
1948 */
1949 features = cpuid_edx(1);
1950 if (!(features & (1 << X86_FEATURE_APIC))) {
1951 pr_warn("Could not enable APIC!\n");
1952 return -1;
1953 }
1954 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1955 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1956
1957 /* The BIOS may have set up the APIC at some other address */
1958 if (boot_cpu_data.x86 >= 6) {
1959 rdmsr(MSR_IA32_APICBASE, l, h);
1960 if (l & MSR_IA32_APICBASE_ENABLE)
1961 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1962 }
1963
1964 pr_info("Found and enabled local APIC!\n");
1965 return 0;
1966}
1967
1968int __init apic_force_enable(unsigned long addr)
1969{
1970 u32 h, l;
1971
1972 if (disable_apic)
1973 return -1;
1974
1975 /*
1976 * Some BIOSes disable the local APIC in the APIC_BASE
1977 * MSR. This can only be done in software for Intel P6 or later
1978 * and AMD K7 (Model > 1) or later.
1979 */
1980 if (boot_cpu_data.x86 >= 6) {
1981 rdmsr(MSR_IA32_APICBASE, l, h);
1982 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1983 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1984 l &= ~MSR_IA32_APICBASE_BASE;
1985 l |= MSR_IA32_APICBASE_ENABLE | addr;
1986 wrmsr(MSR_IA32_APICBASE, l, h);
1987 enabled_via_apicbase = 1;
1988 }
1989 }
1990 return apic_verify();
1991}
1992
1993/*
1994 * Detect and initialize APIC
1995 */
1996static int __init detect_init_APIC(void)
1997{
1998 /* Disabled by kernel option? */
1999 if (disable_apic)
2000 return -1;
2001
2002 switch (boot_cpu_data.x86_vendor) {
2003 case X86_VENDOR_AMD:
2004 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
2005 (boot_cpu_data.x86 >= 15))
2006 break;
2007 goto no_apic;
2008 case X86_VENDOR_HYGON:
2009 break;
2010 case X86_VENDOR_INTEL:
2011 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
2012 (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
2013 break;
2014 goto no_apic;
2015 default:
2016 goto no_apic;
2017 }
2018
2019 if (!boot_cpu_has(X86_FEATURE_APIC)) {
2020 /*
2021 * Over-ride BIOS and try to enable the local APIC only if
2022 * "lapic" specified.
2023 */
2024 if (!force_enable_local_apic) {
2025 pr_info("Local APIC disabled by BIOS -- "
2026 "you can enable it with \"lapic\"\n");
2027 return -1;
2028 }
2029 if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
2030 return -1;
2031 } else {
2032 if (apic_verify())
2033 return -1;
2034 }
2035
2036 apic_pm_activate();
2037
2038 return 0;
2039
2040no_apic:
2041 pr_info("No local APIC present or hardware disabled\n");
2042 return -1;
2043}
2044#endif
2045
2046/**
2047 * init_apic_mappings - initialize APIC mappings
2048 */
2049void __init init_apic_mappings(void)
2050{
2051 unsigned int new_apicid;
2052
2053 if (apic_validate_deadline_timer())
2054 pr_info("TSC deadline timer available\n");
2055
2056 if (x2apic_mode) {
2057 boot_cpu_physical_apicid = read_apic_id();
2058 return;
2059 }
2060
2061 /* If no local APIC can be found return early */
2062 if (!smp_found_config && detect_init_APIC()) {
2063 /* lets NOP'ify apic operations */
2064 pr_info("APIC: disable apic facility\n");
2065 apic_disable();
2066 } else {
2067 apic_phys = mp_lapic_addr;
2068
2069 /*
2070 * If the system has ACPI MADT tables or MP info, the LAPIC
2071 * address is already registered.
2072 */
2073 if (!acpi_lapic && !smp_found_config)
2074 register_lapic_address(apic_phys);
2075 }
2076
2077 /*
2078 * Fetch the APIC ID of the BSP in case we have a
2079 * default configuration (or the MP table is broken).
2080 */
2081 new_apicid = read_apic_id();
2082 if (boot_cpu_physical_apicid != new_apicid) {
2083 boot_cpu_physical_apicid = new_apicid;
2084 /*
2085 * yeah -- we lie about apic_version
2086 * in case if apic was disabled via boot option
2087 * but it's not a problem for SMP compiled kernel
2088 * since apic_intr_mode_select is prepared for such
2089 * a case and disable smp mode
2090 */
2091 boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2092 }
2093}
2094
2095void __init register_lapic_address(unsigned long address)
2096{
2097 mp_lapic_addr = address;
2098
2099 if (!x2apic_mode) {
2100 set_fixmap_nocache(FIX_APIC_BASE, address);
2101 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
2102 APIC_BASE, address);
2103 }
2104 if (boot_cpu_physical_apicid == -1U) {
2105 boot_cpu_physical_apicid = read_apic_id();
2106 boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2107 }
2108}
2109
2110/*
2111 * Local APIC interrupts
2112 */
2113
2114/**
2115 * spurious_interrupt - Catch all for interrupts raised on unused vectors
2116 * @regs: Pointer to pt_regs on stack
2117 * @vector: The vector number
2118 *
2119 * This is invoked from ASM entry code to catch all interrupts which
2120 * trigger on an entry which is routed to the common_spurious idtentry
2121 * point.
2122 *
2123 * Also called from sysvec_spurious_apic_interrupt().
2124 */
2125DEFINE_IDTENTRY_IRQ(spurious_interrupt)
2126{
2127 u32 v;
2128
2129 trace_spurious_apic_entry(vector);
2130
2131 inc_irq_stat(irq_spurious_count);
2132
2133 /*
2134 * If this is a spurious interrupt then do not acknowledge
2135 */
2136 if (vector == SPURIOUS_APIC_VECTOR) {
2137 /* See SDM vol 3 */
2138 pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n",
2139 smp_processor_id());
2140 goto out;
2141 }
2142
2143 /*
2144 * If it is a vectored one, verify it's set in the ISR. If set,
2145 * acknowledge it.
2146 */
2147 v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2148 if (v & (1 << (vector & 0x1f))) {
2149 pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n",
2150 vector, smp_processor_id());
2151 ack_APIC_irq();
2152 } else {
2153 pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n",
2154 vector, smp_processor_id());
2155 }
2156out:
2157 trace_spurious_apic_exit(vector);
2158}
2159
2160DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt)
2161{
2162 __spurious_interrupt(regs, SPURIOUS_APIC_VECTOR);
2163}
2164
2165/*
2166 * This interrupt should never happen with our APIC/SMP architecture
2167 */
2168DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt)
2169{
2170 static const char * const error_interrupt_reason[] = {
2171 "Send CS error", /* APIC Error Bit 0 */
2172 "Receive CS error", /* APIC Error Bit 1 */
2173 "Send accept error", /* APIC Error Bit 2 */
2174 "Receive accept error", /* APIC Error Bit 3 */
2175 "Redirectable IPI", /* APIC Error Bit 4 */
2176 "Send illegal vector", /* APIC Error Bit 5 */
2177 "Received illegal vector", /* APIC Error Bit 6 */
2178 "Illegal register address", /* APIC Error Bit 7 */
2179 };
2180 u32 v, i = 0;
2181
2182 trace_error_apic_entry(ERROR_APIC_VECTOR);
2183
2184 /* First tickle the hardware, only then report what went on. -- REW */
2185 if (lapic_get_maxlvt() > 3) /* Due to the Pentium erratum 3AP. */
2186 apic_write(APIC_ESR, 0);
2187 v = apic_read(APIC_ESR);
2188 ack_APIC_irq();
2189 atomic_inc(&irq_err_count);
2190
2191 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2192 smp_processor_id(), v);
2193
2194 v &= 0xff;
2195 while (v) {
2196 if (v & 0x1)
2197 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2198 i++;
2199 v >>= 1;
2200 }
2201
2202 apic_printk(APIC_DEBUG, KERN_CONT "\n");
2203
2204 trace_error_apic_exit(ERROR_APIC_VECTOR);
2205}
2206
2207/**
2208 * connect_bsp_APIC - attach the APIC to the interrupt system
2209 */
2210static void __init connect_bsp_APIC(void)
2211{
2212#ifdef CONFIG_X86_32
2213 if (pic_mode) {
2214 /*
2215 * Do not trust the local APIC being empty at bootup.
2216 */
2217 clear_local_APIC();
2218 /*
2219 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
2220 * local APIC to INT and NMI lines.
2221 */
2222 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2223 "enabling APIC mode.\n");
2224 imcr_pic_to_apic();
2225 }
2226#endif
2227}
2228
2229/**
2230 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2231 * @virt_wire_setup: indicates, whether virtual wire mode is selected
2232 *
2233 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2234 * APIC is disabled.
2235 */
2236void disconnect_bsp_APIC(int virt_wire_setup)
2237{
2238 unsigned int value;
2239
2240#ifdef CONFIG_X86_32
2241 if (pic_mode) {
2242 /*
2243 * Put the board back into PIC mode (has an effect only on
2244 * certain older boards). Note that APIC interrupts, including
2245 * IPIs, won't work beyond this point! The only exception are
2246 * INIT IPIs.
2247 */
2248 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2249 "entering PIC mode.\n");
2250 imcr_apic_to_pic();
2251 return;
2252 }
2253#endif
2254
2255 /* Go back to Virtual Wire compatibility mode */
2256
2257 /* For the spurious interrupt use vector F, and enable it */
2258 value = apic_read(APIC_SPIV);
2259 value &= ~APIC_VECTOR_MASK;
2260 value |= APIC_SPIV_APIC_ENABLED;
2261 value |= 0xf;
2262 apic_write(APIC_SPIV, value);
2263
2264 if (!virt_wire_setup) {
2265 /*
2266 * For LVT0 make it edge triggered, active high,
2267 * external and enabled
2268 */
2269 value = apic_read(APIC_LVT0);
2270 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2271 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2272 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2273 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2274 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2275 apic_write(APIC_LVT0, value);
2276 } else {
2277 /* Disable LVT0 */
2278 apic_write(APIC_LVT0, APIC_LVT_MASKED);
2279 }
2280
2281 /*
2282 * For LVT1 make it edge triggered, active high,
2283 * nmi and enabled
2284 */
2285 value = apic_read(APIC_LVT1);
2286 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2287 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2288 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2289 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2290 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2291 apic_write(APIC_LVT1, value);
2292}
2293
2294/*
2295 * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2296 * contiguously, it equals to current allocated max logical CPU ID plus 1.
2297 * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2298 * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2299 *
2300 * NOTE: Reserve 0 for BSP.
2301 */
2302static int nr_logical_cpuids = 1;
2303
2304/*
2305 * Used to store mapping between logical CPU IDs and APIC IDs.
2306 */
2307static int cpuid_to_apicid[] = {
2308 [0 ... NR_CPUS - 1] = -1,
2309};
2310
2311#ifdef CONFIG_SMP
2312/**
2313 * apic_id_is_primary_thread - Check whether APIC ID belongs to a primary thread
2314 * @apicid: APIC ID to check
2315 */
2316bool apic_id_is_primary_thread(unsigned int apicid)
2317{
2318 u32 mask;
2319
2320 if (smp_num_siblings == 1)
2321 return true;
2322 /* Isolate the SMT bit(s) in the APICID and check for 0 */
2323 mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
2324 return !(apicid & mask);
2325}
2326#endif
2327
2328/*
2329 * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2330 * and cpuid_to_apicid[] synchronized.
2331 */
2332static int allocate_logical_cpuid(int apicid)
2333{
2334 int i;
2335
2336 /*
2337 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2338 * check if the kernel has allocated a cpuid for it.
2339 */
2340 for (i = 0; i < nr_logical_cpuids; i++) {
2341 if (cpuid_to_apicid[i] == apicid)
2342 return i;
2343 }
2344
2345 /* Allocate a new cpuid. */
2346 if (nr_logical_cpuids >= nr_cpu_ids) {
2347 WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %u reached. "
2348 "Processor %d/0x%x and the rest are ignored.\n",
2349 nr_cpu_ids, nr_logical_cpuids, apicid);
2350 return -EINVAL;
2351 }
2352
2353 cpuid_to_apicid[nr_logical_cpuids] = apicid;
2354 return nr_logical_cpuids++;
2355}
2356
2357int generic_processor_info(int apicid, int version)
2358{
2359 int cpu, max = nr_cpu_ids;
2360 bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2361 phys_cpu_present_map);
2362
2363 /*
2364 * boot_cpu_physical_apicid is designed to have the apicid
2365 * returned by read_apic_id(), i.e, the apicid of the
2366 * currently booting-up processor. However, on some platforms,
2367 * it is temporarily modified by the apicid reported as BSP
2368 * through MP table. Concretely:
2369 *
2370 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2371 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2372 *
2373 * This function is executed with the modified
2374 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2375 * parameter doesn't work to disable APs on kdump 2nd kernel.
2376 *
2377 * Since fixing handling of boot_cpu_physical_apicid requires
2378 * another discussion and tests on each platform, we leave it
2379 * for now and here we use read_apic_id() directly in this
2380 * function, generic_processor_info().
2381 */
2382 if (disabled_cpu_apicid != BAD_APICID &&
2383 disabled_cpu_apicid != read_apic_id() &&
2384 disabled_cpu_apicid == apicid) {
2385 int thiscpu = num_processors + disabled_cpus;
2386
2387 pr_warn("APIC: Disabling requested cpu."
2388 " Processor %d/0x%x ignored.\n", thiscpu, apicid);
2389
2390 disabled_cpus++;
2391 return -ENODEV;
2392 }
2393
2394 /*
2395 * If boot cpu has not been detected yet, then only allow upto
2396 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2397 */
2398 if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2399 apicid != boot_cpu_physical_apicid) {
2400 int thiscpu = max + disabled_cpus - 1;
2401
2402 pr_warn("APIC: NR_CPUS/possible_cpus limit of %i almost"
2403 " reached. Keeping one slot for boot cpu."
2404 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2405
2406 disabled_cpus++;
2407 return -ENODEV;
2408 }
2409
2410 if (num_processors >= nr_cpu_ids) {
2411 int thiscpu = max + disabled_cpus;
2412
2413 pr_warn("APIC: NR_CPUS/possible_cpus limit of %i reached. "
2414 "Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2415
2416 disabled_cpus++;
2417 return -EINVAL;
2418 }
2419
2420 if (apicid == boot_cpu_physical_apicid) {
2421 /*
2422 * x86_bios_cpu_apicid is required to have processors listed
2423 * in same order as logical cpu numbers. Hence the first
2424 * entry is BSP, and so on.
2425 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2426 * for BSP.
2427 */
2428 cpu = 0;
2429
2430 /* Logical cpuid 0 is reserved for BSP. */
2431 cpuid_to_apicid[0] = apicid;
2432 } else {
2433 cpu = allocate_logical_cpuid(apicid);
2434 if (cpu < 0) {
2435 disabled_cpus++;
2436 return -EINVAL;
2437 }
2438 }
2439
2440 /*
2441 * Validate version
2442 */
2443 if (version == 0x0) {
2444 pr_warn("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2445 cpu, apicid);
2446 version = 0x10;
2447 }
2448
2449 if (version != boot_cpu_apic_version) {
2450 pr_warn("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2451 boot_cpu_apic_version, cpu, version);
2452 }
2453
2454 if (apicid > max_physical_apicid)
2455 max_physical_apicid = apicid;
2456
2457#if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2458 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2459 early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2460#endif
2461#ifdef CONFIG_X86_32
2462 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2463 apic->x86_32_early_logical_apicid(cpu);
2464#endif
2465 set_cpu_possible(cpu, true);
2466 physid_set(apicid, phys_cpu_present_map);
2467 set_cpu_present(cpu, true);
2468 num_processors++;
2469
2470 return cpu;
2471}
2472
2473int hard_smp_processor_id(void)
2474{
2475 return read_apic_id();
2476}
2477
2478/*
2479 * Override the generic EOI implementation with an optimized version.
2480 * Only called during early boot when only one CPU is active and with
2481 * interrupts disabled, so we know this does not race with actual APIC driver
2482 * use.
2483 */
2484void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2485{
2486 struct apic **drv;
2487
2488 for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2489 /* Should happen once for each apic */
2490 WARN_ON((*drv)->eoi_write == eoi_write);
2491 (*drv)->native_eoi_write = (*drv)->eoi_write;
2492 (*drv)->eoi_write = eoi_write;
2493 }
2494}
2495
2496static void __init apic_bsp_up_setup(void)
2497{
2498#ifdef CONFIG_X86_64
2499 apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2500#else
2501 /*
2502 * Hack: In case of kdump, after a crash, kernel might be booting
2503 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2504 * might be zero if read from MP tables. Get it from LAPIC.
2505 */
2506# ifdef CONFIG_CRASH_DUMP
2507 boot_cpu_physical_apicid = read_apic_id();
2508# endif
2509#endif
2510 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2511}
2512
2513/**
2514 * apic_bsp_setup - Setup function for local apic and io-apic
2515 * @upmode: Force UP mode (for APIC_init_uniprocessor)
2516 */
2517static void __init apic_bsp_setup(bool upmode)
2518{
2519 connect_bsp_APIC();
2520 if (upmode)
2521 apic_bsp_up_setup();
2522 setup_local_APIC();
2523
2524 enable_IO_APIC();
2525 end_local_APIC_setup();
2526 irq_remap_enable_fault_handling();
2527 setup_IO_APIC();
2528}
2529
2530#ifdef CONFIG_UP_LATE_INIT
2531void __init up_late_init(void)
2532{
2533 if (apic_intr_mode == APIC_PIC)
2534 return;
2535
2536 /* Setup local timer */
2537 x86_init.timers.setup_percpu_clockev();
2538}
2539#endif
2540
2541/*
2542 * Power management
2543 */
2544#ifdef CONFIG_PM
2545
2546static struct {
2547 /*
2548 * 'active' is true if the local APIC was enabled by us and
2549 * not the BIOS; this signifies that we are also responsible
2550 * for disabling it before entering apm/acpi suspend
2551 */
2552 int active;
2553 /* r/w apic fields */
2554 unsigned int apic_id;
2555 unsigned int apic_taskpri;
2556 unsigned int apic_ldr;
2557 unsigned int apic_dfr;
2558 unsigned int apic_spiv;
2559 unsigned int apic_lvtt;
2560 unsigned int apic_lvtpc;
2561 unsigned int apic_lvt0;
2562 unsigned int apic_lvt1;
2563 unsigned int apic_lvterr;
2564 unsigned int apic_tmict;
2565 unsigned int apic_tdcr;
2566 unsigned int apic_thmr;
2567 unsigned int apic_cmci;
2568} apic_pm_state;
2569
2570static int lapic_suspend(void)
2571{
2572 unsigned long flags;
2573 int maxlvt;
2574
2575 if (!apic_pm_state.active)
2576 return 0;
2577
2578 maxlvt = lapic_get_maxlvt();
2579
2580 apic_pm_state.apic_id = apic_read(APIC_ID);
2581 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2582 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2583 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2584 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2585 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2586 if (maxlvt >= 4)
2587 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2588 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2589 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2590 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2591 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2592 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2593#ifdef CONFIG_X86_THERMAL_VECTOR
2594 if (maxlvt >= 5)
2595 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2596#endif
2597#ifdef CONFIG_X86_MCE_INTEL
2598 if (maxlvt >= 6)
2599 apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2600#endif
2601
2602 local_irq_save(flags);
2603
2604 /*
2605 * Mask IOAPIC before disabling the local APIC to prevent stale IRR
2606 * entries on some implementations.
2607 */
2608 mask_ioapic_entries();
2609
2610 disable_local_APIC();
2611
2612 irq_remapping_disable();
2613
2614 local_irq_restore(flags);
2615 return 0;
2616}
2617
2618static void lapic_resume(void)
2619{
2620 unsigned int l, h;
2621 unsigned long flags;
2622 int maxlvt;
2623
2624 if (!apic_pm_state.active)
2625 return;
2626
2627 local_irq_save(flags);
2628
2629 /*
2630 * IO-APIC and PIC have their own resume routines.
2631 * We just mask them here to make sure the interrupt
2632 * subsystem is completely quiet while we enable x2apic
2633 * and interrupt-remapping.
2634 */
2635 mask_ioapic_entries();
2636 legacy_pic->mask_all();
2637
2638 if (x2apic_mode) {
2639 __x2apic_enable();
2640 } else {
2641 /*
2642 * Make sure the APICBASE points to the right address
2643 *
2644 * FIXME! This will be wrong if we ever support suspend on
2645 * SMP! We'll need to do this as part of the CPU restore!
2646 */
2647 if (boot_cpu_data.x86 >= 6) {
2648 rdmsr(MSR_IA32_APICBASE, l, h);
2649 l &= ~MSR_IA32_APICBASE_BASE;
2650 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2651 wrmsr(MSR_IA32_APICBASE, l, h);
2652 }
2653 }
2654
2655 maxlvt = lapic_get_maxlvt();
2656 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2657 apic_write(APIC_ID, apic_pm_state.apic_id);
2658 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2659 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2660 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2661 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2662 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2663 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2664#ifdef CONFIG_X86_THERMAL_VECTOR
2665 if (maxlvt >= 5)
2666 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2667#endif
2668#ifdef CONFIG_X86_MCE_INTEL
2669 if (maxlvt >= 6)
2670 apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2671#endif
2672 if (maxlvt >= 4)
2673 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2674 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2675 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2676 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2677 apic_write(APIC_ESR, 0);
2678 apic_read(APIC_ESR);
2679 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2680 apic_write(APIC_ESR, 0);
2681 apic_read(APIC_ESR);
2682
2683 irq_remapping_reenable(x2apic_mode);
2684
2685 local_irq_restore(flags);
2686}
2687
2688/*
2689 * This device has no shutdown method - fully functioning local APICs
2690 * are needed on every CPU up until machine_halt/restart/poweroff.
2691 */
2692
2693static struct syscore_ops lapic_syscore_ops = {
2694 .resume = lapic_resume,
2695 .suspend = lapic_suspend,
2696};
2697
2698static void apic_pm_activate(void)
2699{
2700 apic_pm_state.active = 1;
2701}
2702
2703static int __init init_lapic_sysfs(void)
2704{
2705 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2706 if (boot_cpu_has(X86_FEATURE_APIC))
2707 register_syscore_ops(&lapic_syscore_ops);
2708
2709 return 0;
2710}
2711
2712/* local apic needs to resume before other devices access its registers. */
2713core_initcall(init_lapic_sysfs);
2714
2715#else /* CONFIG_PM */
2716
2717static void apic_pm_activate(void) { }
2718
2719#endif /* CONFIG_PM */
2720
2721#ifdef CONFIG_X86_64
2722
2723static int multi_checked;
2724static int multi;
2725
2726static int set_multi(const struct dmi_system_id *d)
2727{
2728 if (multi)
2729 return 0;
2730 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2731 multi = 1;
2732 return 0;
2733}
2734
2735static const struct dmi_system_id multi_dmi_table[] = {
2736 {
2737 .callback = set_multi,
2738 .ident = "IBM System Summit2",
2739 .matches = {
2740 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2741 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2742 },
2743 },
2744 {}
2745};
2746
2747static void dmi_check_multi(void)
2748{
2749 if (multi_checked)
2750 return;
2751
2752 dmi_check_system(multi_dmi_table);
2753 multi_checked = 1;
2754}
2755
2756/*
2757 * apic_is_clustered_box() -- Check if we can expect good TSC
2758 *
2759 * Thus far, the major user of this is IBM's Summit2 series:
2760 * Clustered boxes may have unsynced TSC problems if they are
2761 * multi-chassis.
2762 * Use DMI to check them
2763 */
2764int apic_is_clustered_box(void)
2765{
2766 dmi_check_multi();
2767 return multi;
2768}
2769#endif
2770
2771/*
2772 * APIC command line parameters
2773 */
2774static int __init setup_disableapic(char *arg)
2775{
2776 disable_apic = 1;
2777 setup_clear_cpu_cap(X86_FEATURE_APIC);
2778 return 0;
2779}
2780early_param("disableapic", setup_disableapic);
2781
2782/* same as disableapic, for compatibility */
2783static int __init setup_nolapic(char *arg)
2784{
2785 return setup_disableapic(arg);
2786}
2787early_param("nolapic", setup_nolapic);
2788
2789static int __init parse_lapic_timer_c2_ok(char *arg)
2790{
2791 local_apic_timer_c2_ok = 1;
2792 return 0;
2793}
2794early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2795
2796static int __init parse_disable_apic_timer(char *arg)
2797{
2798 disable_apic_timer = 1;
2799 return 0;
2800}
2801early_param("noapictimer", parse_disable_apic_timer);
2802
2803static int __init parse_nolapic_timer(char *arg)
2804{
2805 disable_apic_timer = 1;
2806 return 0;
2807}
2808early_param("nolapic_timer", parse_nolapic_timer);
2809
2810static int __init apic_set_verbosity(char *arg)
2811{
2812 if (!arg) {
2813#ifdef CONFIG_X86_64
2814 skip_ioapic_setup = 0;
2815 return 0;
2816#endif
2817 return -EINVAL;
2818 }
2819
2820 if (strcmp("debug", arg) == 0)
2821 apic_verbosity = APIC_DEBUG;
2822 else if (strcmp("verbose", arg) == 0)
2823 apic_verbosity = APIC_VERBOSE;
2824#ifdef CONFIG_X86_64
2825 else {
2826 pr_warn("APIC Verbosity level %s not recognised"
2827 " use apic=verbose or apic=debug\n", arg);
2828 return -EINVAL;
2829 }
2830#endif
2831
2832 return 0;
2833}
2834early_param("apic", apic_set_verbosity);
2835
2836static int __init lapic_insert_resource(void)
2837{
2838 if (!apic_phys)
2839 return -1;
2840
2841 /* Put local APIC into the resource map. */
2842 lapic_resource.start = apic_phys;
2843 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2844 insert_resource(&iomem_resource, &lapic_resource);
2845
2846 return 0;
2847}
2848
2849/*
2850 * need call insert after e820__reserve_resources()
2851 * that is using request_resource
2852 */
2853late_initcall(lapic_insert_resource);
2854
2855static int __init apic_set_disabled_cpu_apicid(char *arg)
2856{
2857 if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2858 return -EINVAL;
2859
2860 return 0;
2861}
2862early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2863
2864static int __init apic_set_extnmi(char *arg)
2865{
2866 if (!arg)
2867 return -EINVAL;
2868
2869 if (!strncmp("all", arg, 3))
2870 apic_extnmi = APIC_EXTNMI_ALL;
2871 else if (!strncmp("none", arg, 4))
2872 apic_extnmi = APIC_EXTNMI_NONE;
2873 else if (!strncmp("bsp", arg, 3))
2874 apic_extnmi = APIC_EXTNMI_BSP;
2875 else {
2876 pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2877 return -EINVAL;
2878 }
2879
2880 return 0;
2881}
2882early_param("apic_extnmi", apic_set_extnmi);