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/irq_remapping.h>
39#include <asm/perf_event.h>
40#include <asm/x86_init.h>
41#include <asm/pgalloc.h>
42#include <linux/atomic.h>
43#include <asm/mpspec.h>
44#include <asm/i8259.h>
45#include <asm/proto.h>
46#include <asm/apic.h>
47#include <asm/io_apic.h>
48#include <asm/desc.h>
49#include <asm/hpet.h>
50#include <asm/idle.h>
51#include <asm/mtrr.h>
52#include <asm/time.h>
53#include <asm/smp.h>
54#include <asm/mce.h>
55#include <asm/tsc.h>
56#include <asm/hypervisor.h>
57
58unsigned int num_processors;
59
60unsigned disabled_cpus __cpuinitdata;
61
62/* Processor that is doing the boot up */
63unsigned int boot_cpu_physical_apicid = -1U;
64
65/*
66 * The highest APIC ID seen during enumeration.
67 */
68unsigned int max_physical_apicid;
69
70/*
71 * Bitmask of physically existing CPUs:
72 */
73physid_mask_t phys_cpu_present_map;
74
75/*
76 * Map cpu index to physical APIC ID
77 */
78DEFINE_EARLY_PER_CPU(u16, x86_cpu_to_apicid, BAD_APICID);
79DEFINE_EARLY_PER_CPU(u16, x86_bios_cpu_apicid, BAD_APICID);
80EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
81EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
82
83#ifdef CONFIG_X86_32
84
85/*
86 * On x86_32, the mapping between cpu and logical apicid may vary
87 * depending on apic in use. The following early percpu variable is
88 * used for the mapping. This is where the behaviors of x86_64 and 32
89 * actually diverge. Let's keep it ugly for now.
90 */
91DEFINE_EARLY_PER_CPU(int, x86_cpu_to_logical_apicid, BAD_APICID);
92
93/*
94 * Knob to control our willingness to enable the local APIC.
95 *
96 * +1=force-enable
97 */
98static int force_enable_local_apic __initdata;
99/*
100 * APIC command line parameters
101 */
102static int __init parse_lapic(char *arg)
103{
104 force_enable_local_apic = 1;
105 return 0;
106}
107early_param("lapic", parse_lapic);
108/* Local APIC was disabled by the BIOS and enabled by the kernel */
109static int enabled_via_apicbase;
110
111/*
112 * Handle interrupt mode configuration register (IMCR).
113 * This register controls whether the interrupt signals
114 * that reach the BSP come from the master PIC or from the
115 * local APIC. Before entering Symmetric I/O Mode, either
116 * the BIOS or the operating system must switch out of
117 * PIC Mode by changing the IMCR.
118 */
119static inline void imcr_pic_to_apic(void)
120{
121 /* select IMCR register */
122 outb(0x70, 0x22);
123 /* NMI and 8259 INTR go through APIC */
124 outb(0x01, 0x23);
125}
126
127static inline void imcr_apic_to_pic(void)
128{
129 /* select IMCR register */
130 outb(0x70, 0x22);
131 /* NMI and 8259 INTR go directly to BSP */
132 outb(0x00, 0x23);
133}
134#endif
135
136#ifdef CONFIG_X86_64
137static int apic_calibrate_pmtmr __initdata;
138static __init int setup_apicpmtimer(char *s)
139{
140 apic_calibrate_pmtmr = 1;
141 notsc_setup(NULL);
142 return 0;
143}
144__setup("apicpmtimer", setup_apicpmtimer);
145#endif
146
147int x2apic_mode;
148#ifdef CONFIG_X86_X2APIC
149/* x2apic enabled before OS handover */
150int x2apic_preenabled;
151static int x2apic_disabled;
152static int nox2apic;
153static __init int setup_nox2apic(char *str)
154{
155 if (x2apic_enabled()) {
156 int apicid = native_apic_msr_read(APIC_ID);
157
158 if (apicid >= 255) {
159 pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
160 apicid);
161 return 0;
162 }
163
164 pr_warning("x2apic already enabled. will disable it\n");
165 } else
166 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
167
168 nox2apic = 1;
169
170 return 0;
171}
172early_param("nox2apic", setup_nox2apic);
173#endif
174
175unsigned long mp_lapic_addr;
176int disable_apic;
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;
181EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
182
183int first_system_vector = 0xfe;
184
185/*
186 * Debug level, exported for io_apic.c
187 */
188unsigned int apic_verbosity;
189
190int pic_mode;
191
192/* Have we found an MP table */
193int smp_found_config;
194
195static struct resource lapic_resource = {
196 .name = "Local APIC",
197 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
198};
199
200unsigned int lapic_timer_frequency = 0;
201
202static void apic_pm_activate(void);
203
204static unsigned long apic_phys;
205
206/*
207 * Get the LAPIC version
208 */
209static inline int lapic_get_version(void)
210{
211 return GET_APIC_VERSION(apic_read(APIC_LVR));
212}
213
214/*
215 * Check, if the APIC is integrated or a separate chip
216 */
217static inline int lapic_is_integrated(void)
218{
219#ifdef CONFIG_X86_64
220 return 1;
221#else
222 return APIC_INTEGRATED(lapic_get_version());
223#endif
224}
225
226/*
227 * Check, whether this is a modern or a first generation APIC
228 */
229static int modern_apic(void)
230{
231 /* AMD systems use old APIC versions, so check the CPU */
232 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
233 boot_cpu_data.x86 >= 0xf)
234 return 1;
235 return lapic_get_version() >= 0x14;
236}
237
238/*
239 * right after this call apic become NOOP driven
240 * so apic->write/read doesn't do anything
241 */
242static void __init apic_disable(void)
243{
244 pr_info("APIC: switched to apic NOOP\n");
245 apic = &apic_noop;
246}
247
248void native_apic_wait_icr_idle(void)
249{
250 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
251 cpu_relax();
252}
253
254u32 native_safe_apic_wait_icr_idle(void)
255{
256 u32 send_status;
257 int timeout;
258
259 timeout = 0;
260 do {
261 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
262 if (!send_status)
263 break;
264 inc_irq_stat(icr_read_retry_count);
265 udelay(100);
266 } while (timeout++ < 1000);
267
268 return send_status;
269}
270
271void native_apic_icr_write(u32 low, u32 id)
272{
273 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
274 apic_write(APIC_ICR, low);
275}
276
277u64 native_apic_icr_read(void)
278{
279 u32 icr1, icr2;
280
281 icr2 = apic_read(APIC_ICR2);
282 icr1 = apic_read(APIC_ICR);
283
284 return icr1 | ((u64)icr2 << 32);
285}
286
287#ifdef CONFIG_X86_32
288/**
289 * get_physical_broadcast - Get number of physical broadcast IDs
290 */
291int get_physical_broadcast(void)
292{
293 return modern_apic() ? 0xff : 0xf;
294}
295#endif
296
297/**
298 * lapic_get_maxlvt - get the maximum number of local vector table entries
299 */
300int lapic_get_maxlvt(void)
301{
302 unsigned int v;
303
304 v = apic_read(APIC_LVR);
305 /*
306 * - we always have APIC integrated on 64bit mode
307 * - 82489DXs do not report # of LVT entries
308 */
309 return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
310}
311
312/*
313 * Local APIC timer
314 */
315
316/* Clock divisor */
317#define APIC_DIVISOR 16
318
319/*
320 * This function sets up the local APIC timer, with a timeout of
321 * 'clocks' APIC bus clock. During calibration we actually call
322 * this function twice on the boot CPU, once with a bogus timeout
323 * value, second time for real. The other (noncalibrating) CPUs
324 * call this function only once, with the real, calibrated value.
325 *
326 * We do reads before writes even if unnecessary, to get around the
327 * P5 APIC double write bug.
328 */
329static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
330{
331 unsigned int lvtt_value, tmp_value;
332
333 lvtt_value = LOCAL_TIMER_VECTOR;
334 if (!oneshot)
335 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
336 if (!lapic_is_integrated())
337 lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
338
339 if (!irqen)
340 lvtt_value |= APIC_LVT_MASKED;
341
342 apic_write(APIC_LVTT, lvtt_value);
343
344 /*
345 * Divide PICLK by 16
346 */
347 tmp_value = apic_read(APIC_TDCR);
348 apic_write(APIC_TDCR,
349 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
350 APIC_TDR_DIV_16);
351
352 if (!oneshot)
353 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
354}
355
356/*
357 * Setup extended LVT, AMD specific
358 *
359 * Software should use the LVT offsets the BIOS provides. The offsets
360 * are determined by the subsystems using it like those for MCE
361 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts
362 * are supported. Beginning with family 10h at least 4 offsets are
363 * available.
364 *
365 * Since the offsets must be consistent for all cores, we keep track
366 * of the LVT offsets in software and reserve the offset for the same
367 * vector also to be used on other cores. An offset is freed by
368 * setting the entry to APIC_EILVT_MASKED.
369 *
370 * If the BIOS is right, there should be no conflicts. Otherwise a
371 * "[Firmware Bug]: ..." error message is generated. However, if
372 * software does not properly determines the offsets, it is not
373 * necessarily a BIOS bug.
374 */
375
376static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
377
378static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
379{
380 return (old & APIC_EILVT_MASKED)
381 || (new == APIC_EILVT_MASKED)
382 || ((new & ~APIC_EILVT_MASKED) == old);
383}
384
385static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
386{
387 unsigned int rsvd, vector;
388
389 if (offset >= APIC_EILVT_NR_MAX)
390 return ~0;
391
392 rsvd = atomic_read(&eilvt_offsets[offset]);
393 do {
394 vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */
395 if (vector && !eilvt_entry_is_changeable(vector, new))
396 /* may not change if vectors are different */
397 return rsvd;
398 rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
399 } while (rsvd != new);
400
401 rsvd &= ~APIC_EILVT_MASKED;
402 if (rsvd && rsvd != vector)
403 pr_info("LVT offset %d assigned for vector 0x%02x\n",
404 offset, rsvd);
405
406 return new;
407}
408
409/*
410 * If mask=1, the LVT entry does not generate interrupts while mask=0
411 * enables the vector. See also the BKDGs. Must be called with
412 * preemption disabled.
413 */
414
415int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
416{
417 unsigned long reg = APIC_EILVTn(offset);
418 unsigned int new, old, reserved;
419
420 new = (mask << 16) | (msg_type << 8) | vector;
421 old = apic_read(reg);
422 reserved = reserve_eilvt_offset(offset, new);
423
424 if (reserved != new) {
425 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
426 "vector 0x%x, but the register is already in use for "
427 "vector 0x%x on another cpu\n",
428 smp_processor_id(), reg, offset, new, reserved);
429 return -EINVAL;
430 }
431
432 if (!eilvt_entry_is_changeable(old, new)) {
433 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
434 "vector 0x%x, but the register is already in use for "
435 "vector 0x%x on this cpu\n",
436 smp_processor_id(), reg, offset, new, old);
437 return -EBUSY;
438 }
439
440 apic_write(reg, new);
441
442 return 0;
443}
444EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
445
446/*
447 * Program the next event, relative to now
448 */
449static int lapic_next_event(unsigned long delta,
450 struct clock_event_device *evt)
451{
452 apic_write(APIC_TMICT, delta);
453 return 0;
454}
455
456/*
457 * Setup the lapic timer in periodic or oneshot mode
458 */
459static void lapic_timer_setup(enum clock_event_mode mode,
460 struct clock_event_device *evt)
461{
462 unsigned long flags;
463 unsigned int v;
464
465 /* Lapic used as dummy for broadcast ? */
466 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
467 return;
468
469 local_irq_save(flags);
470
471 switch (mode) {
472 case CLOCK_EVT_MODE_PERIODIC:
473 case CLOCK_EVT_MODE_ONESHOT:
474 __setup_APIC_LVTT(lapic_timer_frequency,
475 mode != CLOCK_EVT_MODE_PERIODIC, 1);
476 break;
477 case CLOCK_EVT_MODE_UNUSED:
478 case CLOCK_EVT_MODE_SHUTDOWN:
479 v = apic_read(APIC_LVTT);
480 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
481 apic_write(APIC_LVTT, v);
482 apic_write(APIC_TMICT, 0);
483 break;
484 case CLOCK_EVT_MODE_RESUME:
485 /* Nothing to do here */
486 break;
487 }
488
489 local_irq_restore(flags);
490}
491
492/*
493 * Local APIC timer broadcast function
494 */
495static void lapic_timer_broadcast(const struct cpumask *mask)
496{
497#ifdef CONFIG_SMP
498 apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
499#endif
500}
501
502
503/*
504 * The local apic timer can be used for any function which is CPU local.
505 */
506static struct clock_event_device lapic_clockevent = {
507 .name = "lapic",
508 .features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
509 | CLOCK_EVT_FEAT_C3STOP | CLOCK_EVT_FEAT_DUMMY,
510 .shift = 32,
511 .set_mode = lapic_timer_setup,
512 .set_next_event = lapic_next_event,
513 .broadcast = lapic_timer_broadcast,
514 .rating = 100,
515 .irq = -1,
516};
517static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
518
519/*
520 * Setup the local APIC timer for this CPU. Copy the initialized values
521 * of the boot CPU and register the clock event in the framework.
522 */
523static void __cpuinit setup_APIC_timer(void)
524{
525 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
526
527 if (this_cpu_has(X86_FEATURE_ARAT)) {
528 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
529 /* Make LAPIC timer preferrable over percpu HPET */
530 lapic_clockevent.rating = 150;
531 }
532
533 memcpy(levt, &lapic_clockevent, sizeof(*levt));
534 levt->cpumask = cpumask_of(smp_processor_id());
535
536 clockevents_register_device(levt);
537}
538
539/*
540 * In this functions we calibrate APIC bus clocks to the external timer.
541 *
542 * We want to do the calibration only once since we want to have local timer
543 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
544 * frequency.
545 *
546 * This was previously done by reading the PIT/HPET and waiting for a wrap
547 * around to find out, that a tick has elapsed. I have a box, where the PIT
548 * readout is broken, so it never gets out of the wait loop again. This was
549 * also reported by others.
550 *
551 * Monitoring the jiffies value is inaccurate and the clockevents
552 * infrastructure allows us to do a simple substitution of the interrupt
553 * handler.
554 *
555 * The calibration routine also uses the pm_timer when possible, as the PIT
556 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
557 * back to normal later in the boot process).
558 */
559
560#define LAPIC_CAL_LOOPS (HZ/10)
561
562static __initdata int lapic_cal_loops = -1;
563static __initdata long lapic_cal_t1, lapic_cal_t2;
564static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
565static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
566static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
567
568/*
569 * Temporary interrupt handler.
570 */
571static void __init lapic_cal_handler(struct clock_event_device *dev)
572{
573 unsigned long long tsc = 0;
574 long tapic = apic_read(APIC_TMCCT);
575 unsigned long pm = acpi_pm_read_early();
576
577 if (cpu_has_tsc)
578 rdtscll(tsc);
579
580 switch (lapic_cal_loops++) {
581 case 0:
582 lapic_cal_t1 = tapic;
583 lapic_cal_tsc1 = tsc;
584 lapic_cal_pm1 = pm;
585 lapic_cal_j1 = jiffies;
586 break;
587
588 case LAPIC_CAL_LOOPS:
589 lapic_cal_t2 = tapic;
590 lapic_cal_tsc2 = tsc;
591 if (pm < lapic_cal_pm1)
592 pm += ACPI_PM_OVRRUN;
593 lapic_cal_pm2 = pm;
594 lapic_cal_j2 = jiffies;
595 break;
596 }
597}
598
599static int __init
600calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
601{
602 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
603 const long pm_thresh = pm_100ms / 100;
604 unsigned long mult;
605 u64 res;
606
607#ifndef CONFIG_X86_PM_TIMER
608 return -1;
609#endif
610
611 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
612
613 /* Check, if the PM timer is available */
614 if (!deltapm)
615 return -1;
616
617 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
618
619 if (deltapm > (pm_100ms - pm_thresh) &&
620 deltapm < (pm_100ms + pm_thresh)) {
621 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
622 return 0;
623 }
624
625 res = (((u64)deltapm) * mult) >> 22;
626 do_div(res, 1000000);
627 pr_warning("APIC calibration not consistent "
628 "with PM-Timer: %ldms instead of 100ms\n",(long)res);
629
630 /* Correct the lapic counter value */
631 res = (((u64)(*delta)) * pm_100ms);
632 do_div(res, deltapm);
633 pr_info("APIC delta adjusted to PM-Timer: "
634 "%lu (%ld)\n", (unsigned long)res, *delta);
635 *delta = (long)res;
636
637 /* Correct the tsc counter value */
638 if (cpu_has_tsc) {
639 res = (((u64)(*deltatsc)) * pm_100ms);
640 do_div(res, deltapm);
641 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
642 "PM-Timer: %lu (%ld)\n",
643 (unsigned long)res, *deltatsc);
644 *deltatsc = (long)res;
645 }
646
647 return 0;
648}
649
650static int __init calibrate_APIC_clock(void)
651{
652 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
653 void (*real_handler)(struct clock_event_device *dev);
654 unsigned long deltaj;
655 long delta, deltatsc;
656 int pm_referenced = 0;
657
658 /**
659 * check if lapic timer has already been calibrated by platform
660 * specific routine, such as tsc calibration code. if so, we just fill
661 * in the clockevent structure and return.
662 */
663
664 if (lapic_timer_frequency) {
665 apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
666 lapic_timer_frequency);
667 lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
668 TICK_NSEC, lapic_clockevent.shift);
669 lapic_clockevent.max_delta_ns =
670 clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
671 lapic_clockevent.min_delta_ns =
672 clockevent_delta2ns(0xF, &lapic_clockevent);
673 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
674 return 0;
675 }
676
677 local_irq_disable();
678
679 /* Replace the global interrupt handler */
680 real_handler = global_clock_event->event_handler;
681 global_clock_event->event_handler = lapic_cal_handler;
682
683 /*
684 * Setup the APIC counter to maximum. There is no way the lapic
685 * can underflow in the 100ms detection time frame
686 */
687 __setup_APIC_LVTT(0xffffffff, 0, 0);
688
689 /* Let the interrupts run */
690 local_irq_enable();
691
692 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
693 cpu_relax();
694
695 local_irq_disable();
696
697 /* Restore the real event handler */
698 global_clock_event->event_handler = real_handler;
699
700 /* Build delta t1-t2 as apic timer counts down */
701 delta = lapic_cal_t1 - lapic_cal_t2;
702 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
703
704 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
705
706 /* we trust the PM based calibration if possible */
707 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
708 &delta, &deltatsc);
709
710 /* Calculate the scaled math multiplication factor */
711 lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
712 lapic_clockevent.shift);
713 lapic_clockevent.max_delta_ns =
714 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
715 lapic_clockevent.min_delta_ns =
716 clockevent_delta2ns(0xF, &lapic_clockevent);
717
718 lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
719
720 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
721 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
722 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
723 lapic_timer_frequency);
724
725 if (cpu_has_tsc) {
726 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
727 "%ld.%04ld MHz.\n",
728 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
729 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
730 }
731
732 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
733 "%u.%04u MHz.\n",
734 lapic_timer_frequency / (1000000 / HZ),
735 lapic_timer_frequency % (1000000 / HZ));
736
737 /*
738 * Do a sanity check on the APIC calibration result
739 */
740 if (lapic_timer_frequency < (1000000 / HZ)) {
741 local_irq_enable();
742 pr_warning("APIC frequency too slow, disabling apic timer\n");
743 return -1;
744 }
745
746 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
747
748 /*
749 * PM timer calibration failed or not turned on
750 * so lets try APIC timer based calibration
751 */
752 if (!pm_referenced) {
753 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
754
755 /*
756 * Setup the apic timer manually
757 */
758 levt->event_handler = lapic_cal_handler;
759 lapic_timer_setup(CLOCK_EVT_MODE_PERIODIC, levt);
760 lapic_cal_loops = -1;
761
762 /* Let the interrupts run */
763 local_irq_enable();
764
765 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
766 cpu_relax();
767
768 /* Stop the lapic timer */
769 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt);
770
771 /* Jiffies delta */
772 deltaj = lapic_cal_j2 - lapic_cal_j1;
773 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
774
775 /* Check, if the jiffies result is consistent */
776 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
777 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
778 else
779 levt->features |= CLOCK_EVT_FEAT_DUMMY;
780 } else
781 local_irq_enable();
782
783 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
784 pr_warning("APIC timer disabled due to verification failure\n");
785 return -1;
786 }
787
788 return 0;
789}
790
791/*
792 * Setup the boot APIC
793 *
794 * Calibrate and verify the result.
795 */
796void __init setup_boot_APIC_clock(void)
797{
798 /*
799 * The local apic timer can be disabled via the kernel
800 * commandline or from the CPU detection code. Register the lapic
801 * timer as a dummy clock event source on SMP systems, so the
802 * broadcast mechanism is used. On UP systems simply ignore it.
803 */
804 if (disable_apic_timer) {
805 pr_info("Disabling APIC timer\n");
806 /* No broadcast on UP ! */
807 if (num_possible_cpus() > 1) {
808 lapic_clockevent.mult = 1;
809 setup_APIC_timer();
810 }
811 return;
812 }
813
814 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
815 "calibrating APIC timer ...\n");
816
817 if (calibrate_APIC_clock()) {
818 /* No broadcast on UP ! */
819 if (num_possible_cpus() > 1)
820 setup_APIC_timer();
821 return;
822 }
823
824 /*
825 * If nmi_watchdog is set to IO_APIC, we need the
826 * PIT/HPET going. Otherwise register lapic as a dummy
827 * device.
828 */
829 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
830
831 /* Setup the lapic or request the broadcast */
832 setup_APIC_timer();
833}
834
835void __cpuinit setup_secondary_APIC_clock(void)
836{
837 setup_APIC_timer();
838}
839
840/*
841 * The guts of the apic timer interrupt
842 */
843static void local_apic_timer_interrupt(void)
844{
845 int cpu = smp_processor_id();
846 struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
847
848 /*
849 * Normally we should not be here till LAPIC has been initialized but
850 * in some cases like kdump, its possible that there is a pending LAPIC
851 * timer interrupt from previous kernel's context and is delivered in
852 * new kernel the moment interrupts are enabled.
853 *
854 * Interrupts are enabled early and LAPIC is setup much later, hence
855 * its possible that when we get here evt->event_handler is NULL.
856 * Check for event_handler being NULL and discard the interrupt as
857 * spurious.
858 */
859 if (!evt->event_handler) {
860 pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
861 /* Switch it off */
862 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
863 return;
864 }
865
866 /*
867 * the NMI deadlock-detector uses this.
868 */
869 inc_irq_stat(apic_timer_irqs);
870
871 evt->event_handler(evt);
872}
873
874/*
875 * Local APIC timer interrupt. This is the most natural way for doing
876 * local interrupts, but local timer interrupts can be emulated by
877 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
878 *
879 * [ if a single-CPU system runs an SMP kernel then we call the local
880 * interrupt as well. Thus we cannot inline the local irq ... ]
881 */
882void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
883{
884 struct pt_regs *old_regs = set_irq_regs(regs);
885
886 /*
887 * NOTE! We'd better ACK the irq immediately,
888 * because timer handling can be slow.
889 */
890 ack_APIC_irq();
891 /*
892 * update_process_times() expects us to have done irq_enter().
893 * Besides, if we don't timer interrupts ignore the global
894 * interrupt lock, which is the WrongThing (tm) to do.
895 */
896 irq_enter();
897 exit_idle();
898 local_apic_timer_interrupt();
899 irq_exit();
900
901 set_irq_regs(old_regs);
902}
903
904int setup_profiling_timer(unsigned int multiplier)
905{
906 return -EINVAL;
907}
908
909/*
910 * Local APIC start and shutdown
911 */
912
913/**
914 * clear_local_APIC - shutdown the local APIC
915 *
916 * This is called, when a CPU is disabled and before rebooting, so the state of
917 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
918 * leftovers during boot.
919 */
920void clear_local_APIC(void)
921{
922 int maxlvt;
923 u32 v;
924
925 /* APIC hasn't been mapped yet */
926 if (!x2apic_mode && !apic_phys)
927 return;
928
929 maxlvt = lapic_get_maxlvt();
930 /*
931 * Masking an LVT entry can trigger a local APIC error
932 * if the vector is zero. Mask LVTERR first to prevent this.
933 */
934 if (maxlvt >= 3) {
935 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
936 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
937 }
938 /*
939 * Careful: we have to set masks only first to deassert
940 * any level-triggered sources.
941 */
942 v = apic_read(APIC_LVTT);
943 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
944 v = apic_read(APIC_LVT0);
945 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
946 v = apic_read(APIC_LVT1);
947 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
948 if (maxlvt >= 4) {
949 v = apic_read(APIC_LVTPC);
950 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
951 }
952
953 /* lets not touch this if we didn't frob it */
954#ifdef CONFIG_X86_THERMAL_VECTOR
955 if (maxlvt >= 5) {
956 v = apic_read(APIC_LVTTHMR);
957 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
958 }
959#endif
960#ifdef CONFIG_X86_MCE_INTEL
961 if (maxlvt >= 6) {
962 v = apic_read(APIC_LVTCMCI);
963 if (!(v & APIC_LVT_MASKED))
964 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
965 }
966#endif
967
968 /*
969 * Clean APIC state for other OSs:
970 */
971 apic_write(APIC_LVTT, APIC_LVT_MASKED);
972 apic_write(APIC_LVT0, APIC_LVT_MASKED);
973 apic_write(APIC_LVT1, APIC_LVT_MASKED);
974 if (maxlvt >= 3)
975 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
976 if (maxlvt >= 4)
977 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
978
979 /* Integrated APIC (!82489DX) ? */
980 if (lapic_is_integrated()) {
981 if (maxlvt > 3)
982 /* Clear ESR due to Pentium errata 3AP and 11AP */
983 apic_write(APIC_ESR, 0);
984 apic_read(APIC_ESR);
985 }
986}
987
988/**
989 * disable_local_APIC - clear and disable the local APIC
990 */
991void disable_local_APIC(void)
992{
993 unsigned int value;
994
995 /* APIC hasn't been mapped yet */
996 if (!x2apic_mode && !apic_phys)
997 return;
998
999 clear_local_APIC();
1000
1001 /*
1002 * Disable APIC (implies clearing of registers
1003 * for 82489DX!).
1004 */
1005 value = apic_read(APIC_SPIV);
1006 value &= ~APIC_SPIV_APIC_ENABLED;
1007 apic_write(APIC_SPIV, value);
1008
1009#ifdef CONFIG_X86_32
1010 /*
1011 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1012 * restore the disabled state.
1013 */
1014 if (enabled_via_apicbase) {
1015 unsigned int l, h;
1016
1017 rdmsr(MSR_IA32_APICBASE, l, h);
1018 l &= ~MSR_IA32_APICBASE_ENABLE;
1019 wrmsr(MSR_IA32_APICBASE, l, h);
1020 }
1021#endif
1022}
1023
1024/*
1025 * If Linux enabled the LAPIC against the BIOS default disable it down before
1026 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1027 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1028 * for the case where Linux didn't enable the LAPIC.
1029 */
1030void lapic_shutdown(void)
1031{
1032 unsigned long flags;
1033
1034 if (!cpu_has_apic && !apic_from_smp_config())
1035 return;
1036
1037 local_irq_save(flags);
1038
1039#ifdef CONFIG_X86_32
1040 if (!enabled_via_apicbase)
1041 clear_local_APIC();
1042 else
1043#endif
1044 disable_local_APIC();
1045
1046
1047 local_irq_restore(flags);
1048}
1049
1050/*
1051 * This is to verify that we're looking at a real local APIC.
1052 * Check these against your board if the CPUs aren't getting
1053 * started for no apparent reason.
1054 */
1055int __init verify_local_APIC(void)
1056{
1057 unsigned int reg0, reg1;
1058
1059 /*
1060 * The version register is read-only in a real APIC.
1061 */
1062 reg0 = apic_read(APIC_LVR);
1063 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
1064 apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
1065 reg1 = apic_read(APIC_LVR);
1066 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
1067
1068 /*
1069 * The two version reads above should print the same
1070 * numbers. If the second one is different, then we
1071 * poke at a non-APIC.
1072 */
1073 if (reg1 != reg0)
1074 return 0;
1075
1076 /*
1077 * Check if the version looks reasonably.
1078 */
1079 reg1 = GET_APIC_VERSION(reg0);
1080 if (reg1 == 0x00 || reg1 == 0xff)
1081 return 0;
1082 reg1 = lapic_get_maxlvt();
1083 if (reg1 < 0x02 || reg1 == 0xff)
1084 return 0;
1085
1086 /*
1087 * The ID register is read/write in a real APIC.
1088 */
1089 reg0 = apic_read(APIC_ID);
1090 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
1091 apic_write(APIC_ID, reg0 ^ apic->apic_id_mask);
1092 reg1 = apic_read(APIC_ID);
1093 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
1094 apic_write(APIC_ID, reg0);
1095 if (reg1 != (reg0 ^ apic->apic_id_mask))
1096 return 0;
1097
1098 /*
1099 * The next two are just to see if we have sane values.
1100 * They're only really relevant if we're in Virtual Wire
1101 * compatibility mode, but most boxes are anymore.
1102 */
1103 reg0 = apic_read(APIC_LVT0);
1104 apic_printk(APIC_DEBUG, "Getting LVT0: %x\n", reg0);
1105 reg1 = apic_read(APIC_LVT1);
1106 apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
1107
1108 return 1;
1109}
1110
1111/**
1112 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1113 */
1114void __init sync_Arb_IDs(void)
1115{
1116 /*
1117 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1118 * needed on AMD.
1119 */
1120 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1121 return;
1122
1123 /*
1124 * Wait for idle.
1125 */
1126 apic_wait_icr_idle();
1127
1128 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1129 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1130 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1131}
1132
1133/*
1134 * An initial setup of the virtual wire mode.
1135 */
1136void __init init_bsp_APIC(void)
1137{
1138 unsigned int value;
1139
1140 /*
1141 * Don't do the setup now if we have a SMP BIOS as the
1142 * through-I/O-APIC virtual wire mode might be active.
1143 */
1144 if (smp_found_config || !cpu_has_apic)
1145 return;
1146
1147 /*
1148 * Do not trust the local APIC being empty at bootup.
1149 */
1150 clear_local_APIC();
1151
1152 /*
1153 * Enable APIC.
1154 */
1155 value = apic_read(APIC_SPIV);
1156 value &= ~APIC_VECTOR_MASK;
1157 value |= APIC_SPIV_APIC_ENABLED;
1158
1159#ifdef CONFIG_X86_32
1160 /* This bit is reserved on P4/Xeon and should be cleared */
1161 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1162 (boot_cpu_data.x86 == 15))
1163 value &= ~APIC_SPIV_FOCUS_DISABLED;
1164 else
1165#endif
1166 value |= APIC_SPIV_FOCUS_DISABLED;
1167 value |= SPURIOUS_APIC_VECTOR;
1168 apic_write(APIC_SPIV, value);
1169
1170 /*
1171 * Set up the virtual wire mode.
1172 */
1173 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1174 value = APIC_DM_NMI;
1175 if (!lapic_is_integrated()) /* 82489DX */
1176 value |= APIC_LVT_LEVEL_TRIGGER;
1177 apic_write(APIC_LVT1, value);
1178}
1179
1180static void __cpuinit lapic_setup_esr(void)
1181{
1182 unsigned int oldvalue, value, maxlvt;
1183
1184 if (!lapic_is_integrated()) {
1185 pr_info("No ESR for 82489DX.\n");
1186 return;
1187 }
1188
1189 if (apic->disable_esr) {
1190 /*
1191 * Something untraceable is creating bad interrupts on
1192 * secondary quads ... for the moment, just leave the
1193 * ESR disabled - we can't do anything useful with the
1194 * errors anyway - mbligh
1195 */
1196 pr_info("Leaving ESR disabled.\n");
1197 return;
1198 }
1199
1200 maxlvt = lapic_get_maxlvt();
1201 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1202 apic_write(APIC_ESR, 0);
1203 oldvalue = apic_read(APIC_ESR);
1204
1205 /* enables sending errors */
1206 value = ERROR_APIC_VECTOR;
1207 apic_write(APIC_LVTERR, value);
1208
1209 /*
1210 * spec says clear errors after enabling vector.
1211 */
1212 if (maxlvt > 3)
1213 apic_write(APIC_ESR, 0);
1214 value = apic_read(APIC_ESR);
1215 if (value != oldvalue)
1216 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1217 "vector: 0x%08x after: 0x%08x\n",
1218 oldvalue, value);
1219}
1220
1221/**
1222 * setup_local_APIC - setup the local APIC
1223 *
1224 * Used to setup local APIC while initializing BSP or bringin up APs.
1225 * Always called with preemption disabled.
1226 */
1227void __cpuinit setup_local_APIC(void)
1228{
1229 int cpu = smp_processor_id();
1230 unsigned int value, queued;
1231 int i, j, acked = 0;
1232 unsigned long long tsc = 0, ntsc;
1233 long long max_loops = cpu_khz;
1234
1235 if (cpu_has_tsc)
1236 rdtscll(tsc);
1237
1238 if (disable_apic) {
1239 disable_ioapic_support();
1240 return;
1241 }
1242
1243#ifdef CONFIG_X86_32
1244 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1245 if (lapic_is_integrated() && apic->disable_esr) {
1246 apic_write(APIC_ESR, 0);
1247 apic_write(APIC_ESR, 0);
1248 apic_write(APIC_ESR, 0);
1249 apic_write(APIC_ESR, 0);
1250 }
1251#endif
1252 perf_events_lapic_init();
1253
1254 /*
1255 * Double-check whether this APIC is really registered.
1256 * This is meaningless in clustered apic mode, so we skip it.
1257 */
1258 BUG_ON(!apic->apic_id_registered());
1259
1260 /*
1261 * Intel recommends to set DFR, LDR and TPR before enabling
1262 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1263 * document number 292116). So here it goes...
1264 */
1265 apic->init_apic_ldr();
1266
1267#ifdef CONFIG_X86_32
1268 /*
1269 * APIC LDR is initialized. If logical_apicid mapping was
1270 * initialized during get_smp_config(), make sure it matches the
1271 * actual value.
1272 */
1273 i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1274 WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1275 /* always use the value from LDR */
1276 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1277 logical_smp_processor_id();
1278
1279 /*
1280 * Some NUMA implementations (NUMAQ) don't initialize apicid to
1281 * node mapping during NUMA init. Now that logical apicid is
1282 * guaranteed to be known, give it another chance. This is already
1283 * a bit too late - percpu allocation has already happened without
1284 * proper NUMA affinity.
1285 */
1286 if (apic->x86_32_numa_cpu_node)
1287 set_apicid_to_node(early_per_cpu(x86_cpu_to_apicid, cpu),
1288 apic->x86_32_numa_cpu_node(cpu));
1289#endif
1290
1291 /*
1292 * Set Task Priority to 'accept all'. We never change this
1293 * later on.
1294 */
1295 value = apic_read(APIC_TASKPRI);
1296 value &= ~APIC_TPRI_MASK;
1297 apic_write(APIC_TASKPRI, value);
1298
1299 /*
1300 * After a crash, we no longer service the interrupts and a pending
1301 * interrupt from previous kernel might still have ISR bit set.
1302 *
1303 * Most probably by now CPU has serviced that pending interrupt and
1304 * it might not have done the ack_APIC_irq() because it thought,
1305 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1306 * does not clear the ISR bit and cpu thinks it has already serivced
1307 * the interrupt. Hence a vector might get locked. It was noticed
1308 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1309 */
1310 do {
1311 queued = 0;
1312 for (i = APIC_ISR_NR - 1; i >= 0; i--)
1313 queued |= apic_read(APIC_IRR + i*0x10);
1314
1315 for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1316 value = apic_read(APIC_ISR + i*0x10);
1317 for (j = 31; j >= 0; j--) {
1318 if (value & (1<<j)) {
1319 ack_APIC_irq();
1320 acked++;
1321 }
1322 }
1323 }
1324 if (acked > 256) {
1325 printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1326 acked);
1327 break;
1328 }
1329 if (queued) {
1330 if (cpu_has_tsc) {
1331 rdtscll(ntsc);
1332 max_loops = (cpu_khz << 10) - (ntsc - tsc);
1333 } else
1334 max_loops--;
1335 }
1336 } while (queued && max_loops > 0);
1337 WARN_ON(max_loops <= 0);
1338
1339 /*
1340 * Now that we are all set up, enable the APIC
1341 */
1342 value = apic_read(APIC_SPIV);
1343 value &= ~APIC_VECTOR_MASK;
1344 /*
1345 * Enable APIC
1346 */
1347 value |= APIC_SPIV_APIC_ENABLED;
1348
1349#ifdef CONFIG_X86_32
1350 /*
1351 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1352 * certain networking cards. If high frequency interrupts are
1353 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1354 * entry is masked/unmasked at a high rate as well then sooner or
1355 * later IOAPIC line gets 'stuck', no more interrupts are received
1356 * from the device. If focus CPU is disabled then the hang goes
1357 * away, oh well :-(
1358 *
1359 * [ This bug can be reproduced easily with a level-triggered
1360 * PCI Ne2000 networking cards and PII/PIII processors, dual
1361 * BX chipset. ]
1362 */
1363 /*
1364 * Actually disabling the focus CPU check just makes the hang less
1365 * frequent as it makes the interrupt distributon model be more
1366 * like LRU than MRU (the short-term load is more even across CPUs).
1367 * See also the comment in end_level_ioapic_irq(). --macro
1368 */
1369
1370 /*
1371 * - enable focus processor (bit==0)
1372 * - 64bit mode always use processor focus
1373 * so no need to set it
1374 */
1375 value &= ~APIC_SPIV_FOCUS_DISABLED;
1376#endif
1377
1378 /*
1379 * Set spurious IRQ vector
1380 */
1381 value |= SPURIOUS_APIC_VECTOR;
1382 apic_write(APIC_SPIV, value);
1383
1384 /*
1385 * Set up LVT0, LVT1:
1386 *
1387 * set up through-local-APIC on the BP's LINT0. This is not
1388 * strictly necessary in pure symmetric-IO mode, but sometimes
1389 * we delegate interrupts to the 8259A.
1390 */
1391 /*
1392 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1393 */
1394 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1395 if (!cpu && (pic_mode || !value)) {
1396 value = APIC_DM_EXTINT;
1397 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1398 } else {
1399 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1400 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1401 }
1402 apic_write(APIC_LVT0, value);
1403
1404 /*
1405 * only the BP should see the LINT1 NMI signal, obviously.
1406 */
1407 if (!cpu)
1408 value = APIC_DM_NMI;
1409 else
1410 value = APIC_DM_NMI | APIC_LVT_MASKED;
1411 if (!lapic_is_integrated()) /* 82489DX */
1412 value |= APIC_LVT_LEVEL_TRIGGER;
1413 apic_write(APIC_LVT1, value);
1414
1415#ifdef CONFIG_X86_MCE_INTEL
1416 /* Recheck CMCI information after local APIC is up on CPU #0 */
1417 if (!cpu)
1418 cmci_recheck();
1419#endif
1420}
1421
1422void __cpuinit end_local_APIC_setup(void)
1423{
1424 lapic_setup_esr();
1425
1426#ifdef CONFIG_X86_32
1427 {
1428 unsigned int value;
1429 /* Disable the local apic timer */
1430 value = apic_read(APIC_LVTT);
1431 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1432 apic_write(APIC_LVTT, value);
1433 }
1434#endif
1435
1436 apic_pm_activate();
1437}
1438
1439void __init bsp_end_local_APIC_setup(void)
1440{
1441 end_local_APIC_setup();
1442
1443 /*
1444 * Now that local APIC setup is completed for BP, configure the fault
1445 * handling for interrupt remapping.
1446 */
1447 if (irq_remapping_enabled)
1448 irq_remap_enable_fault_handling();
1449
1450}
1451
1452#ifdef CONFIG_X86_X2APIC
1453/*
1454 * Need to disable xapic and x2apic at the same time and then enable xapic mode
1455 */
1456static inline void __disable_x2apic(u64 msr)
1457{
1458 wrmsrl(MSR_IA32_APICBASE,
1459 msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1460 wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1461}
1462
1463static __init void disable_x2apic(void)
1464{
1465 u64 msr;
1466
1467 if (!cpu_has_x2apic)
1468 return;
1469
1470 rdmsrl(MSR_IA32_APICBASE, msr);
1471 if (msr & X2APIC_ENABLE) {
1472 u32 x2apic_id = read_apic_id();
1473
1474 if (x2apic_id >= 255)
1475 panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1476
1477 pr_info("Disabling x2apic\n");
1478 __disable_x2apic(msr);
1479
1480 if (nox2apic) {
1481 clear_cpu_cap(&cpu_data(0), X86_FEATURE_X2APIC);
1482 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1483 }
1484
1485 x2apic_disabled = 1;
1486 x2apic_mode = 0;
1487
1488 register_lapic_address(mp_lapic_addr);
1489 }
1490}
1491
1492void check_x2apic(void)
1493{
1494 if (x2apic_enabled()) {
1495 pr_info("x2apic enabled by BIOS, switching to x2apic ops\n");
1496 x2apic_preenabled = x2apic_mode = 1;
1497 }
1498}
1499
1500void enable_x2apic(void)
1501{
1502 u64 msr;
1503
1504 rdmsrl(MSR_IA32_APICBASE, msr);
1505 if (x2apic_disabled) {
1506 __disable_x2apic(msr);
1507 return;
1508 }
1509
1510 if (!x2apic_mode)
1511 return;
1512
1513 if (!(msr & X2APIC_ENABLE)) {
1514 printk_once(KERN_INFO "Enabling x2apic\n");
1515 wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1516 }
1517}
1518#endif /* CONFIG_X86_X2APIC */
1519
1520int __init enable_IR(void)
1521{
1522#ifdef CONFIG_IRQ_REMAP
1523 if (!irq_remapping_supported()) {
1524 pr_debug("intr-remapping not supported\n");
1525 return -1;
1526 }
1527
1528 if (!x2apic_preenabled && skip_ioapic_setup) {
1529 pr_info("Skipped enabling intr-remap because of skipping "
1530 "io-apic setup\n");
1531 return -1;
1532 }
1533
1534 return irq_remapping_enable();
1535#endif
1536 return -1;
1537}
1538
1539void __init enable_IR_x2apic(void)
1540{
1541 unsigned long flags;
1542 int ret, x2apic_enabled = 0;
1543 int hardware_init_ret;
1544
1545 /* Make sure irq_remap_ops are initialized */
1546 setup_irq_remapping_ops();
1547
1548 hardware_init_ret = irq_remapping_prepare();
1549 if (hardware_init_ret && !x2apic_supported())
1550 return;
1551
1552 ret = save_ioapic_entries();
1553 if (ret) {
1554 pr_info("Saving IO-APIC state failed: %d\n", ret);
1555 return;
1556 }
1557
1558 local_irq_save(flags);
1559 legacy_pic->mask_all();
1560 mask_ioapic_entries();
1561
1562 if (x2apic_preenabled && nox2apic)
1563 disable_x2apic();
1564
1565 if (hardware_init_ret)
1566 ret = -1;
1567 else
1568 ret = enable_IR();
1569
1570 if (!x2apic_supported())
1571 goto skip_x2apic;
1572
1573 if (ret < 0) {
1574 /* IR is required if there is APIC ID > 255 even when running
1575 * under KVM
1576 */
1577 if (max_physical_apicid > 255 ||
1578 !hypervisor_x2apic_available()) {
1579 if (x2apic_preenabled)
1580 disable_x2apic();
1581 goto skip_x2apic;
1582 }
1583 /*
1584 * without IR all CPUs can be addressed by IOAPIC/MSI
1585 * only in physical mode
1586 */
1587 x2apic_force_phys();
1588 }
1589
1590 if (ret == IRQ_REMAP_XAPIC_MODE) {
1591 pr_info("x2apic not enabled, IRQ remapping is in xapic mode\n");
1592 goto skip_x2apic;
1593 }
1594
1595 x2apic_enabled = 1;
1596
1597 if (x2apic_supported() && !x2apic_mode) {
1598 x2apic_mode = 1;
1599 enable_x2apic();
1600 pr_info("Enabled x2apic\n");
1601 }
1602
1603skip_x2apic:
1604 if (ret < 0) /* IR enabling failed */
1605 restore_ioapic_entries();
1606 legacy_pic->restore_mask();
1607 local_irq_restore(flags);
1608}
1609
1610#ifdef CONFIG_X86_64
1611/*
1612 * Detect and enable local APICs on non-SMP boards.
1613 * Original code written by Keir Fraser.
1614 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1615 * not correctly set up (usually the APIC timer won't work etc.)
1616 */
1617static int __init detect_init_APIC(void)
1618{
1619 if (!cpu_has_apic) {
1620 pr_info("No local APIC present\n");
1621 return -1;
1622 }
1623
1624 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1625 return 0;
1626}
1627#else
1628
1629static int __init apic_verify(void)
1630{
1631 u32 features, h, l;
1632
1633 /*
1634 * The APIC feature bit should now be enabled
1635 * in `cpuid'
1636 */
1637 features = cpuid_edx(1);
1638 if (!(features & (1 << X86_FEATURE_APIC))) {
1639 pr_warning("Could not enable APIC!\n");
1640 return -1;
1641 }
1642 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1643 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1644
1645 /* The BIOS may have set up the APIC at some other address */
1646 if (boot_cpu_data.x86 >= 6) {
1647 rdmsr(MSR_IA32_APICBASE, l, h);
1648 if (l & MSR_IA32_APICBASE_ENABLE)
1649 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1650 }
1651
1652 pr_info("Found and enabled local APIC!\n");
1653 return 0;
1654}
1655
1656int __init apic_force_enable(unsigned long addr)
1657{
1658 u32 h, l;
1659
1660 if (disable_apic)
1661 return -1;
1662
1663 /*
1664 * Some BIOSes disable the local APIC in the APIC_BASE
1665 * MSR. This can only be done in software for Intel P6 or later
1666 * and AMD K7 (Model > 1) or later.
1667 */
1668 if (boot_cpu_data.x86 >= 6) {
1669 rdmsr(MSR_IA32_APICBASE, l, h);
1670 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1671 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1672 l &= ~MSR_IA32_APICBASE_BASE;
1673 l |= MSR_IA32_APICBASE_ENABLE | addr;
1674 wrmsr(MSR_IA32_APICBASE, l, h);
1675 enabled_via_apicbase = 1;
1676 }
1677 }
1678 return apic_verify();
1679}
1680
1681/*
1682 * Detect and initialize APIC
1683 */
1684static int __init detect_init_APIC(void)
1685{
1686 /* Disabled by kernel option? */
1687 if (disable_apic)
1688 return -1;
1689
1690 switch (boot_cpu_data.x86_vendor) {
1691 case X86_VENDOR_AMD:
1692 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1693 (boot_cpu_data.x86 >= 15))
1694 break;
1695 goto no_apic;
1696 case X86_VENDOR_INTEL:
1697 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1698 (boot_cpu_data.x86 == 5 && cpu_has_apic))
1699 break;
1700 goto no_apic;
1701 default:
1702 goto no_apic;
1703 }
1704
1705 if (!cpu_has_apic) {
1706 /*
1707 * Over-ride BIOS and try to enable the local APIC only if
1708 * "lapic" specified.
1709 */
1710 if (!force_enable_local_apic) {
1711 pr_info("Local APIC disabled by BIOS -- "
1712 "you can enable it with \"lapic\"\n");
1713 return -1;
1714 }
1715 if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1716 return -1;
1717 } else {
1718 if (apic_verify())
1719 return -1;
1720 }
1721
1722 apic_pm_activate();
1723
1724 return 0;
1725
1726no_apic:
1727 pr_info("No local APIC present or hardware disabled\n");
1728 return -1;
1729}
1730#endif
1731
1732/**
1733 * init_apic_mappings - initialize APIC mappings
1734 */
1735void __init init_apic_mappings(void)
1736{
1737 unsigned int new_apicid;
1738
1739 if (x2apic_mode) {
1740 boot_cpu_physical_apicid = read_apic_id();
1741 return;
1742 }
1743
1744 /* If no local APIC can be found return early */
1745 if (!smp_found_config && detect_init_APIC()) {
1746 /* lets NOP'ify apic operations */
1747 pr_info("APIC: disable apic facility\n");
1748 apic_disable();
1749 } else {
1750 apic_phys = mp_lapic_addr;
1751
1752 /*
1753 * acpi lapic path already maps that address in
1754 * acpi_register_lapic_address()
1755 */
1756 if (!acpi_lapic && !smp_found_config)
1757 register_lapic_address(apic_phys);
1758 }
1759
1760 /*
1761 * Fetch the APIC ID of the BSP in case we have a
1762 * default configuration (or the MP table is broken).
1763 */
1764 new_apicid = read_apic_id();
1765 if (boot_cpu_physical_apicid != new_apicid) {
1766 boot_cpu_physical_apicid = new_apicid;
1767 /*
1768 * yeah -- we lie about apic_version
1769 * in case if apic was disabled via boot option
1770 * but it's not a problem for SMP compiled kernel
1771 * since smp_sanity_check is prepared for such a case
1772 * and disable smp mode
1773 */
1774 apic_version[new_apicid] =
1775 GET_APIC_VERSION(apic_read(APIC_LVR));
1776 }
1777}
1778
1779void __init register_lapic_address(unsigned long address)
1780{
1781 mp_lapic_addr = address;
1782
1783 if (!x2apic_mode) {
1784 set_fixmap_nocache(FIX_APIC_BASE, address);
1785 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1786 APIC_BASE, mp_lapic_addr);
1787 }
1788 if (boot_cpu_physical_apicid == -1U) {
1789 boot_cpu_physical_apicid = read_apic_id();
1790 apic_version[boot_cpu_physical_apicid] =
1791 GET_APIC_VERSION(apic_read(APIC_LVR));
1792 }
1793}
1794
1795/*
1796 * This initializes the IO-APIC and APIC hardware if this is
1797 * a UP kernel.
1798 */
1799int apic_version[MAX_LOCAL_APIC];
1800
1801int __init APIC_init_uniprocessor(void)
1802{
1803 if (disable_apic) {
1804 pr_info("Apic disabled\n");
1805 return -1;
1806 }
1807#ifdef CONFIG_X86_64
1808 if (!cpu_has_apic) {
1809 disable_apic = 1;
1810 pr_info("Apic disabled by BIOS\n");
1811 return -1;
1812 }
1813#else
1814 if (!smp_found_config && !cpu_has_apic)
1815 return -1;
1816
1817 /*
1818 * Complain if the BIOS pretends there is one.
1819 */
1820 if (!cpu_has_apic &&
1821 APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
1822 pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
1823 boot_cpu_physical_apicid);
1824 return -1;
1825 }
1826#endif
1827
1828 default_setup_apic_routing();
1829
1830 verify_local_APIC();
1831 connect_bsp_APIC();
1832
1833#ifdef CONFIG_X86_64
1834 apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
1835#else
1836 /*
1837 * Hack: In case of kdump, after a crash, kernel might be booting
1838 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
1839 * might be zero if read from MP tables. Get it from LAPIC.
1840 */
1841# ifdef CONFIG_CRASH_DUMP
1842 boot_cpu_physical_apicid = read_apic_id();
1843# endif
1844#endif
1845 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1846 setup_local_APIC();
1847
1848#ifdef CONFIG_X86_IO_APIC
1849 /*
1850 * Now enable IO-APICs, actually call clear_IO_APIC
1851 * We need clear_IO_APIC before enabling error vector
1852 */
1853 if (!skip_ioapic_setup && nr_ioapics)
1854 enable_IO_APIC();
1855#endif
1856
1857 bsp_end_local_APIC_setup();
1858
1859#ifdef CONFIG_X86_IO_APIC
1860 if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
1861 setup_IO_APIC();
1862 else {
1863 nr_ioapics = 0;
1864 }
1865#endif
1866
1867 x86_init.timers.setup_percpu_clockev();
1868 return 0;
1869}
1870
1871/*
1872 * Local APIC interrupts
1873 */
1874
1875/*
1876 * This interrupt should _never_ happen with our APIC/SMP architecture
1877 */
1878void smp_spurious_interrupt(struct pt_regs *regs)
1879{
1880 u32 v;
1881
1882 irq_enter();
1883 exit_idle();
1884 /*
1885 * Check if this really is a spurious interrupt and ACK it
1886 * if it is a vectored one. Just in case...
1887 * Spurious interrupts should not be ACKed.
1888 */
1889 v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
1890 if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
1891 ack_APIC_irq();
1892
1893 inc_irq_stat(irq_spurious_count);
1894
1895 /* see sw-dev-man vol 3, chapter 7.4.13.5 */
1896 pr_info("spurious APIC interrupt on CPU#%d, "
1897 "should never happen.\n", smp_processor_id());
1898 irq_exit();
1899}
1900
1901/*
1902 * This interrupt should never happen with our APIC/SMP architecture
1903 */
1904void smp_error_interrupt(struct pt_regs *regs)
1905{
1906 u32 v0, v1;
1907 u32 i = 0;
1908 static const char * const error_interrupt_reason[] = {
1909 "Send CS error", /* APIC Error Bit 0 */
1910 "Receive CS error", /* APIC Error Bit 1 */
1911 "Send accept error", /* APIC Error Bit 2 */
1912 "Receive accept error", /* APIC Error Bit 3 */
1913 "Redirectable IPI", /* APIC Error Bit 4 */
1914 "Send illegal vector", /* APIC Error Bit 5 */
1915 "Received illegal vector", /* APIC Error Bit 6 */
1916 "Illegal register address", /* APIC Error Bit 7 */
1917 };
1918
1919 irq_enter();
1920 exit_idle();
1921 /* First tickle the hardware, only then report what went on. -- REW */
1922 v0 = apic_read(APIC_ESR);
1923 apic_write(APIC_ESR, 0);
1924 v1 = apic_read(APIC_ESR);
1925 ack_APIC_irq();
1926 atomic_inc(&irq_err_count);
1927
1928 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x(%02x)",
1929 smp_processor_id(), v0 , v1);
1930
1931 v1 = v1 & 0xff;
1932 while (v1) {
1933 if (v1 & 0x1)
1934 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
1935 i++;
1936 v1 >>= 1;
1937 };
1938
1939 apic_printk(APIC_DEBUG, KERN_CONT "\n");
1940
1941 irq_exit();
1942}
1943
1944/**
1945 * connect_bsp_APIC - attach the APIC to the interrupt system
1946 */
1947void __init connect_bsp_APIC(void)
1948{
1949#ifdef CONFIG_X86_32
1950 if (pic_mode) {
1951 /*
1952 * Do not trust the local APIC being empty at bootup.
1953 */
1954 clear_local_APIC();
1955 /*
1956 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
1957 * local APIC to INT and NMI lines.
1958 */
1959 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
1960 "enabling APIC mode.\n");
1961 imcr_pic_to_apic();
1962 }
1963#endif
1964 if (apic->enable_apic_mode)
1965 apic->enable_apic_mode();
1966}
1967
1968/**
1969 * disconnect_bsp_APIC - detach the APIC from the interrupt system
1970 * @virt_wire_setup: indicates, whether virtual wire mode is selected
1971 *
1972 * Virtual wire mode is necessary to deliver legacy interrupts even when the
1973 * APIC is disabled.
1974 */
1975void disconnect_bsp_APIC(int virt_wire_setup)
1976{
1977 unsigned int value;
1978
1979#ifdef CONFIG_X86_32
1980 if (pic_mode) {
1981 /*
1982 * Put the board back into PIC mode (has an effect only on
1983 * certain older boards). Note that APIC interrupts, including
1984 * IPIs, won't work beyond this point! The only exception are
1985 * INIT IPIs.
1986 */
1987 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
1988 "entering PIC mode.\n");
1989 imcr_apic_to_pic();
1990 return;
1991 }
1992#endif
1993
1994 /* Go back to Virtual Wire compatibility mode */
1995
1996 /* For the spurious interrupt use vector F, and enable it */
1997 value = apic_read(APIC_SPIV);
1998 value &= ~APIC_VECTOR_MASK;
1999 value |= APIC_SPIV_APIC_ENABLED;
2000 value |= 0xf;
2001 apic_write(APIC_SPIV, value);
2002
2003 if (!virt_wire_setup) {
2004 /*
2005 * For LVT0 make it edge triggered, active high,
2006 * external and enabled
2007 */
2008 value = apic_read(APIC_LVT0);
2009 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2010 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2011 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2012 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2013 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2014 apic_write(APIC_LVT0, value);
2015 } else {
2016 /* Disable LVT0 */
2017 apic_write(APIC_LVT0, APIC_LVT_MASKED);
2018 }
2019
2020 /*
2021 * For LVT1 make it edge triggered, active high,
2022 * nmi and enabled
2023 */
2024 value = apic_read(APIC_LVT1);
2025 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2026 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2027 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2028 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2029 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2030 apic_write(APIC_LVT1, value);
2031}
2032
2033void __cpuinit generic_processor_info(int apicid, int version)
2034{
2035 int cpu, max = nr_cpu_ids;
2036 bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2037 phys_cpu_present_map);
2038
2039 /*
2040 * If boot cpu has not been detected yet, then only allow upto
2041 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2042 */
2043 if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2044 apicid != boot_cpu_physical_apicid) {
2045 int thiscpu = max + disabled_cpus - 1;
2046
2047 pr_warning(
2048 "ACPI: NR_CPUS/possible_cpus limit of %i almost"
2049 " reached. Keeping one slot for boot cpu."
2050 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2051
2052 disabled_cpus++;
2053 return;
2054 }
2055
2056 if (num_processors >= nr_cpu_ids) {
2057 int thiscpu = max + disabled_cpus;
2058
2059 pr_warning(
2060 "ACPI: NR_CPUS/possible_cpus limit of %i reached."
2061 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2062
2063 disabled_cpus++;
2064 return;
2065 }
2066
2067 num_processors++;
2068 if (apicid == boot_cpu_physical_apicid) {
2069 /*
2070 * x86_bios_cpu_apicid is required to have processors listed
2071 * in same order as logical cpu numbers. Hence the first
2072 * entry is BSP, and so on.
2073 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2074 * for BSP.
2075 */
2076 cpu = 0;
2077 } else
2078 cpu = cpumask_next_zero(-1, cpu_present_mask);
2079
2080 /*
2081 * Validate version
2082 */
2083 if (version == 0x0) {
2084 pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2085 cpu, apicid);
2086 version = 0x10;
2087 }
2088 apic_version[apicid] = version;
2089
2090 if (version != apic_version[boot_cpu_physical_apicid]) {
2091 pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2092 apic_version[boot_cpu_physical_apicid], cpu, version);
2093 }
2094
2095 physid_set(apicid, phys_cpu_present_map);
2096 if (apicid > max_physical_apicid)
2097 max_physical_apicid = apicid;
2098
2099#if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2100 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2101 early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2102#endif
2103#ifdef CONFIG_X86_32
2104 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2105 apic->x86_32_early_logical_apicid(cpu);
2106#endif
2107 set_cpu_possible(cpu, true);
2108 set_cpu_present(cpu, true);
2109}
2110
2111int hard_smp_processor_id(void)
2112{
2113 return read_apic_id();
2114}
2115
2116void default_init_apic_ldr(void)
2117{
2118 unsigned long val;
2119
2120 apic_write(APIC_DFR, APIC_DFR_VALUE);
2121 val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
2122 val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
2123 apic_write(APIC_LDR, val);
2124}
2125
2126/*
2127 * Power management
2128 */
2129#ifdef CONFIG_PM
2130
2131static struct {
2132 /*
2133 * 'active' is true if the local APIC was enabled by us and
2134 * not the BIOS; this signifies that we are also responsible
2135 * for disabling it before entering apm/acpi suspend
2136 */
2137 int active;
2138 /* r/w apic fields */
2139 unsigned int apic_id;
2140 unsigned int apic_taskpri;
2141 unsigned int apic_ldr;
2142 unsigned int apic_dfr;
2143 unsigned int apic_spiv;
2144 unsigned int apic_lvtt;
2145 unsigned int apic_lvtpc;
2146 unsigned int apic_lvt0;
2147 unsigned int apic_lvt1;
2148 unsigned int apic_lvterr;
2149 unsigned int apic_tmict;
2150 unsigned int apic_tdcr;
2151 unsigned int apic_thmr;
2152} apic_pm_state;
2153
2154static int lapic_suspend(void)
2155{
2156 unsigned long flags;
2157 int maxlvt;
2158
2159 if (!apic_pm_state.active)
2160 return 0;
2161
2162 maxlvt = lapic_get_maxlvt();
2163
2164 apic_pm_state.apic_id = apic_read(APIC_ID);
2165 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2166 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2167 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2168 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2169 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2170 if (maxlvt >= 4)
2171 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2172 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2173 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2174 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2175 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2176 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2177#ifdef CONFIG_X86_THERMAL_VECTOR
2178 if (maxlvt >= 5)
2179 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2180#endif
2181
2182 local_irq_save(flags);
2183 disable_local_APIC();
2184
2185 if (irq_remapping_enabled)
2186 irq_remapping_disable();
2187
2188 local_irq_restore(flags);
2189 return 0;
2190}
2191
2192static void lapic_resume(void)
2193{
2194 unsigned int l, h;
2195 unsigned long flags;
2196 int maxlvt;
2197
2198 if (!apic_pm_state.active)
2199 return;
2200
2201 local_irq_save(flags);
2202 if (irq_remapping_enabled) {
2203 /*
2204 * IO-APIC and PIC have their own resume routines.
2205 * We just mask them here to make sure the interrupt
2206 * subsystem is completely quiet while we enable x2apic
2207 * and interrupt-remapping.
2208 */
2209 mask_ioapic_entries();
2210 legacy_pic->mask_all();
2211 }
2212
2213 if (x2apic_mode)
2214 enable_x2apic();
2215 else {
2216 /*
2217 * Make sure the APICBASE points to the right address
2218 *
2219 * FIXME! This will be wrong if we ever support suspend on
2220 * SMP! We'll need to do this as part of the CPU restore!
2221 */
2222 if (boot_cpu_data.x86 >= 6) {
2223 rdmsr(MSR_IA32_APICBASE, l, h);
2224 l &= ~MSR_IA32_APICBASE_BASE;
2225 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2226 wrmsr(MSR_IA32_APICBASE, l, h);
2227 }
2228 }
2229
2230 maxlvt = lapic_get_maxlvt();
2231 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2232 apic_write(APIC_ID, apic_pm_state.apic_id);
2233 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2234 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2235 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2236 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2237 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2238 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2239#if defined(CONFIG_X86_MCE_P4THERMAL) || defined(CONFIG_X86_MCE_INTEL)
2240 if (maxlvt >= 5)
2241 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2242#endif
2243 if (maxlvt >= 4)
2244 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2245 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2246 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2247 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2248 apic_write(APIC_ESR, 0);
2249 apic_read(APIC_ESR);
2250 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2251 apic_write(APIC_ESR, 0);
2252 apic_read(APIC_ESR);
2253
2254 if (irq_remapping_enabled)
2255 irq_remapping_reenable(x2apic_mode);
2256
2257 local_irq_restore(flags);
2258}
2259
2260/*
2261 * This device has no shutdown method - fully functioning local APICs
2262 * are needed on every CPU up until machine_halt/restart/poweroff.
2263 */
2264
2265static struct syscore_ops lapic_syscore_ops = {
2266 .resume = lapic_resume,
2267 .suspend = lapic_suspend,
2268};
2269
2270static void __cpuinit apic_pm_activate(void)
2271{
2272 apic_pm_state.active = 1;
2273}
2274
2275static int __init init_lapic_sysfs(void)
2276{
2277 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2278 if (cpu_has_apic)
2279 register_syscore_ops(&lapic_syscore_ops);
2280
2281 return 0;
2282}
2283
2284/* local apic needs to resume before other devices access its registers. */
2285core_initcall(init_lapic_sysfs);
2286
2287#else /* CONFIG_PM */
2288
2289static void apic_pm_activate(void) { }
2290
2291#endif /* CONFIG_PM */
2292
2293#ifdef CONFIG_X86_64
2294
2295static int __cpuinit apic_cluster_num(void)
2296{
2297 int i, clusters, zeros;
2298 unsigned id;
2299 u16 *bios_cpu_apicid;
2300 DECLARE_BITMAP(clustermap, NUM_APIC_CLUSTERS);
2301
2302 bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
2303 bitmap_zero(clustermap, NUM_APIC_CLUSTERS);
2304
2305 for (i = 0; i < nr_cpu_ids; i++) {
2306 /* are we being called early in kernel startup? */
2307 if (bios_cpu_apicid) {
2308 id = bios_cpu_apicid[i];
2309 } else if (i < nr_cpu_ids) {
2310 if (cpu_present(i))
2311 id = per_cpu(x86_bios_cpu_apicid, i);
2312 else
2313 continue;
2314 } else
2315 break;
2316
2317 if (id != BAD_APICID)
2318 __set_bit(APIC_CLUSTERID(id), clustermap);
2319 }
2320
2321 /* Problem: Partially populated chassis may not have CPUs in some of
2322 * the APIC clusters they have been allocated. Only present CPUs have
2323 * x86_bios_cpu_apicid entries, thus causing zeroes in the bitmap.
2324 * Since clusters are allocated sequentially, count zeros only if
2325 * they are bounded by ones.
2326 */
2327 clusters = 0;
2328 zeros = 0;
2329 for (i = 0; i < NUM_APIC_CLUSTERS; i++) {
2330 if (test_bit(i, clustermap)) {
2331 clusters += 1 + zeros;
2332 zeros = 0;
2333 } else
2334 ++zeros;
2335 }
2336
2337 return clusters;
2338}
2339
2340static int __cpuinitdata multi_checked;
2341static int __cpuinitdata multi;
2342
2343static int __cpuinit set_multi(const struct dmi_system_id *d)
2344{
2345 if (multi)
2346 return 0;
2347 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2348 multi = 1;
2349 return 0;
2350}
2351
2352static const __cpuinitconst struct dmi_system_id multi_dmi_table[] = {
2353 {
2354 .callback = set_multi,
2355 .ident = "IBM System Summit2",
2356 .matches = {
2357 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2358 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2359 },
2360 },
2361 {}
2362};
2363
2364static void __cpuinit dmi_check_multi(void)
2365{
2366 if (multi_checked)
2367 return;
2368
2369 dmi_check_system(multi_dmi_table);
2370 multi_checked = 1;
2371}
2372
2373/*
2374 * apic_is_clustered_box() -- Check if we can expect good TSC
2375 *
2376 * Thus far, the major user of this is IBM's Summit2 series:
2377 * Clustered boxes may have unsynced TSC problems if they are
2378 * multi-chassis.
2379 * Use DMI to check them
2380 */
2381__cpuinit int apic_is_clustered_box(void)
2382{
2383 dmi_check_multi();
2384 if (multi)
2385 return 1;
2386
2387 if (!is_vsmp_box())
2388 return 0;
2389
2390 /*
2391 * ScaleMP vSMPowered boxes have one cluster per board and TSCs are
2392 * not guaranteed to be synced between boards
2393 */
2394 if (apic_cluster_num() > 1)
2395 return 1;
2396
2397 return 0;
2398}
2399#endif
2400
2401/*
2402 * APIC command line parameters
2403 */
2404static int __init setup_disableapic(char *arg)
2405{
2406 disable_apic = 1;
2407 setup_clear_cpu_cap(X86_FEATURE_APIC);
2408 return 0;
2409}
2410early_param("disableapic", setup_disableapic);
2411
2412/* same as disableapic, for compatibility */
2413static int __init setup_nolapic(char *arg)
2414{
2415 return setup_disableapic(arg);
2416}
2417early_param("nolapic", setup_nolapic);
2418
2419static int __init parse_lapic_timer_c2_ok(char *arg)
2420{
2421 local_apic_timer_c2_ok = 1;
2422 return 0;
2423}
2424early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2425
2426static int __init parse_disable_apic_timer(char *arg)
2427{
2428 disable_apic_timer = 1;
2429 return 0;
2430}
2431early_param("noapictimer", parse_disable_apic_timer);
2432
2433static int __init parse_nolapic_timer(char *arg)
2434{
2435 disable_apic_timer = 1;
2436 return 0;
2437}
2438early_param("nolapic_timer", parse_nolapic_timer);
2439
2440static int __init apic_set_verbosity(char *arg)
2441{
2442 if (!arg) {
2443#ifdef CONFIG_X86_64
2444 skip_ioapic_setup = 0;
2445 return 0;
2446#endif
2447 return -EINVAL;
2448 }
2449
2450 if (strcmp("debug", arg) == 0)
2451 apic_verbosity = APIC_DEBUG;
2452 else if (strcmp("verbose", arg) == 0)
2453 apic_verbosity = APIC_VERBOSE;
2454 else {
2455 pr_warning("APIC Verbosity level %s not recognised"
2456 " use apic=verbose or apic=debug\n", arg);
2457 return -EINVAL;
2458 }
2459
2460 return 0;
2461}
2462early_param("apic", apic_set_verbosity);
2463
2464static int __init lapic_insert_resource(void)
2465{
2466 if (!apic_phys)
2467 return -1;
2468
2469 /* Put local APIC into the resource map. */
2470 lapic_resource.start = apic_phys;
2471 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2472 insert_resource(&iomem_resource, &lapic_resource);
2473
2474 return 0;
2475}
2476
2477/*
2478 * need call insert after e820_reserve_resources()
2479 * that is using request_resource
2480 */
2481late_initcall(lapic_insert_resource);
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/trace/irq_vectors.h>
39#include <asm/irq_remapping.h>
40#include <asm/perf_event.h>
41#include <asm/x86_init.h>
42#include <asm/pgalloc.h>
43#include <linux/atomic.h>
44#include <asm/mpspec.h>
45#include <asm/i8259.h>
46#include <asm/proto.h>
47#include <asm/apic.h>
48#include <asm/io_apic.h>
49#include <asm/desc.h>
50#include <asm/hpet.h>
51#include <asm/idle.h>
52#include <asm/mtrr.h>
53#include <asm/time.h>
54#include <asm/smp.h>
55#include <asm/mce.h>
56#include <asm/tsc.h>
57#include <asm/hypervisor.h>
58
59unsigned int num_processors;
60
61unsigned disabled_cpus;
62
63/* Processor that is doing the boot up */
64unsigned int boot_cpu_physical_apicid = -1U;
65EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
66
67/*
68 * The highest APIC ID seen during enumeration.
69 */
70static unsigned int max_physical_apicid;
71
72/*
73 * Bitmask of physically existing CPUs:
74 */
75physid_mask_t phys_cpu_present_map;
76
77/*
78 * Processor to be disabled specified by kernel parameter
79 * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
80 * avoid undefined behaviour caused by sending INIT from AP to BSP.
81 */
82static unsigned int disabled_cpu_apicid __read_mostly = BAD_APICID;
83
84/*
85 * This variable controls which CPUs receive external NMIs. By default,
86 * external NMIs are delivered only to the BSP.
87 */
88static int apic_extnmi = APIC_EXTNMI_BSP;
89
90/*
91 * Map cpu index to physical APIC ID
92 */
93DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
94DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
95EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
96EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
97
98#ifdef CONFIG_X86_32
99
100/*
101 * On x86_32, the mapping between cpu and logical apicid may vary
102 * depending on apic in use. The following early percpu variable is
103 * used for the mapping. This is where the behaviors of x86_64 and 32
104 * actually diverge. Let's keep it ugly for now.
105 */
106DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
107
108/* Local APIC was disabled by the BIOS and enabled by the kernel */
109static int enabled_via_apicbase;
110
111/*
112 * Handle interrupt mode configuration register (IMCR).
113 * This register controls whether the interrupt signals
114 * that reach the BSP come from the master PIC or from the
115 * local APIC. Before entering Symmetric I/O Mode, either
116 * the BIOS or the operating system must switch out of
117 * PIC Mode by changing the IMCR.
118 */
119static inline void imcr_pic_to_apic(void)
120{
121 /* select IMCR register */
122 outb(0x70, 0x22);
123 /* NMI and 8259 INTR go through APIC */
124 outb(0x01, 0x23);
125}
126
127static inline void imcr_apic_to_pic(void)
128{
129 /* select IMCR register */
130 outb(0x70, 0x22);
131 /* NMI and 8259 INTR go directly to BSP */
132 outb(0x00, 0x23);
133}
134#endif
135
136/*
137 * Knob to control our willingness to enable the local APIC.
138 *
139 * +1=force-enable
140 */
141static int force_enable_local_apic __initdata;
142
143/*
144 * APIC command line parameters
145 */
146static int __init parse_lapic(char *arg)
147{
148 if (config_enabled(CONFIG_X86_32) && !arg)
149 force_enable_local_apic = 1;
150 else if (arg && !strncmp(arg, "notscdeadline", 13))
151 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
152 return 0;
153}
154early_param("lapic", parse_lapic);
155
156#ifdef CONFIG_X86_64
157static int apic_calibrate_pmtmr __initdata;
158static __init int setup_apicpmtimer(char *s)
159{
160 apic_calibrate_pmtmr = 1;
161 notsc_setup(NULL);
162 return 0;
163}
164__setup("apicpmtimer", setup_apicpmtimer);
165#endif
166
167unsigned long mp_lapic_addr;
168int disable_apic;
169/* Disable local APIC timer from the kernel commandline or via dmi quirk */
170static int disable_apic_timer __initdata;
171/* Local APIC timer works in C2 */
172int local_apic_timer_c2_ok;
173EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
174
175int first_system_vector = FIRST_SYSTEM_VECTOR;
176
177/*
178 * Debug level, exported for io_apic.c
179 */
180unsigned int apic_verbosity;
181
182int pic_mode;
183
184/* Have we found an MP table */
185int smp_found_config;
186
187static struct resource lapic_resource = {
188 .name = "Local APIC",
189 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
190};
191
192unsigned int lapic_timer_frequency = 0;
193
194static void apic_pm_activate(void);
195
196static unsigned long apic_phys;
197
198/*
199 * Get the LAPIC version
200 */
201static inline int lapic_get_version(void)
202{
203 return GET_APIC_VERSION(apic_read(APIC_LVR));
204}
205
206/*
207 * Check, if the APIC is integrated or a separate chip
208 */
209static inline int lapic_is_integrated(void)
210{
211#ifdef CONFIG_X86_64
212 return 1;
213#else
214 return APIC_INTEGRATED(lapic_get_version());
215#endif
216}
217
218/*
219 * Check, whether this is a modern or a first generation APIC
220 */
221static int modern_apic(void)
222{
223 /* AMD systems use old APIC versions, so check the CPU */
224 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
225 boot_cpu_data.x86 >= 0xf)
226 return 1;
227 return lapic_get_version() >= 0x14;
228}
229
230/*
231 * right after this call apic become NOOP driven
232 * so apic->write/read doesn't do anything
233 */
234static void __init apic_disable(void)
235{
236 pr_info("APIC: switched to apic NOOP\n");
237 apic = &apic_noop;
238}
239
240void native_apic_wait_icr_idle(void)
241{
242 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
243 cpu_relax();
244}
245
246u32 native_safe_apic_wait_icr_idle(void)
247{
248 u32 send_status;
249 int timeout;
250
251 timeout = 0;
252 do {
253 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
254 if (!send_status)
255 break;
256 inc_irq_stat(icr_read_retry_count);
257 udelay(100);
258 } while (timeout++ < 1000);
259
260 return send_status;
261}
262
263void native_apic_icr_write(u32 low, u32 id)
264{
265 unsigned long flags;
266
267 local_irq_save(flags);
268 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
269 apic_write(APIC_ICR, low);
270 local_irq_restore(flags);
271}
272
273u64 native_apic_icr_read(void)
274{
275 u32 icr1, icr2;
276
277 icr2 = apic_read(APIC_ICR2);
278 icr1 = apic_read(APIC_ICR);
279
280 return icr1 | ((u64)icr2 << 32);
281}
282
283#ifdef CONFIG_X86_32
284/**
285 * get_physical_broadcast - Get number of physical broadcast IDs
286 */
287int get_physical_broadcast(void)
288{
289 return modern_apic() ? 0xff : 0xf;
290}
291#endif
292
293/**
294 * lapic_get_maxlvt - get the maximum number of local vector table entries
295 */
296int lapic_get_maxlvt(void)
297{
298 unsigned int v;
299
300 v = apic_read(APIC_LVR);
301 /*
302 * - we always have APIC integrated on 64bit mode
303 * - 82489DXs do not report # of LVT entries
304 */
305 return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
306}
307
308/*
309 * Local APIC timer
310 */
311
312/* Clock divisor */
313#define APIC_DIVISOR 16
314#define TSC_DIVISOR 32
315
316/*
317 * This function sets up the local APIC timer, with a timeout of
318 * 'clocks' APIC bus clock. During calibration we actually call
319 * this function twice on the boot CPU, once with a bogus timeout
320 * value, second time for real. The other (noncalibrating) CPUs
321 * call this function only once, with the real, calibrated value.
322 *
323 * We do reads before writes even if unnecessary, to get around the
324 * P5 APIC double write bug.
325 */
326static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
327{
328 unsigned int lvtt_value, tmp_value;
329
330 lvtt_value = LOCAL_TIMER_VECTOR;
331 if (!oneshot)
332 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
333 else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
334 lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
335
336 if (!lapic_is_integrated())
337 lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
338
339 if (!irqen)
340 lvtt_value |= APIC_LVT_MASKED;
341
342 apic_write(APIC_LVTT, lvtt_value);
343
344 if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
345 /*
346 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
347 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
348 * According to Intel, MFENCE can do the serialization here.
349 */
350 asm volatile("mfence" : : : "memory");
351
352 printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
353 return;
354 }
355
356 /*
357 * Divide PICLK by 16
358 */
359 tmp_value = apic_read(APIC_TDCR);
360 apic_write(APIC_TDCR,
361 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
362 APIC_TDR_DIV_16);
363
364 if (!oneshot)
365 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
366}
367
368/*
369 * Setup extended LVT, AMD specific
370 *
371 * Software should use the LVT offsets the BIOS provides. The offsets
372 * are determined by the subsystems using it like those for MCE
373 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts
374 * are supported. Beginning with family 10h at least 4 offsets are
375 * available.
376 *
377 * Since the offsets must be consistent for all cores, we keep track
378 * of the LVT offsets in software and reserve the offset for the same
379 * vector also to be used on other cores. An offset is freed by
380 * setting the entry to APIC_EILVT_MASKED.
381 *
382 * If the BIOS is right, there should be no conflicts. Otherwise a
383 * "[Firmware Bug]: ..." error message is generated. However, if
384 * software does not properly determines the offsets, it is not
385 * necessarily a BIOS bug.
386 */
387
388static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
389
390static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
391{
392 return (old & APIC_EILVT_MASKED)
393 || (new == APIC_EILVT_MASKED)
394 || ((new & ~APIC_EILVT_MASKED) == old);
395}
396
397static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
398{
399 unsigned int rsvd, vector;
400
401 if (offset >= APIC_EILVT_NR_MAX)
402 return ~0;
403
404 rsvd = atomic_read(&eilvt_offsets[offset]);
405 do {
406 vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */
407 if (vector && !eilvt_entry_is_changeable(vector, new))
408 /* may not change if vectors are different */
409 return rsvd;
410 rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
411 } while (rsvd != new);
412
413 rsvd &= ~APIC_EILVT_MASKED;
414 if (rsvd && rsvd != vector)
415 pr_info("LVT offset %d assigned for vector 0x%02x\n",
416 offset, rsvd);
417
418 return new;
419}
420
421/*
422 * If mask=1, the LVT entry does not generate interrupts while mask=0
423 * enables the vector. See also the BKDGs. Must be called with
424 * preemption disabled.
425 */
426
427int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
428{
429 unsigned long reg = APIC_EILVTn(offset);
430 unsigned int new, old, reserved;
431
432 new = (mask << 16) | (msg_type << 8) | vector;
433 old = apic_read(reg);
434 reserved = reserve_eilvt_offset(offset, new);
435
436 if (reserved != new) {
437 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
438 "vector 0x%x, but the register is already in use for "
439 "vector 0x%x on another cpu\n",
440 smp_processor_id(), reg, offset, new, reserved);
441 return -EINVAL;
442 }
443
444 if (!eilvt_entry_is_changeable(old, new)) {
445 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
446 "vector 0x%x, but the register is already in use for "
447 "vector 0x%x on this cpu\n",
448 smp_processor_id(), reg, offset, new, old);
449 return -EBUSY;
450 }
451
452 apic_write(reg, new);
453
454 return 0;
455}
456EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
457
458/*
459 * Program the next event, relative to now
460 */
461static int lapic_next_event(unsigned long delta,
462 struct clock_event_device *evt)
463{
464 apic_write(APIC_TMICT, delta);
465 return 0;
466}
467
468static int lapic_next_deadline(unsigned long delta,
469 struct clock_event_device *evt)
470{
471 u64 tsc;
472
473 tsc = rdtsc();
474 wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
475 return 0;
476}
477
478static int lapic_timer_shutdown(struct clock_event_device *evt)
479{
480 unsigned int v;
481
482 /* Lapic used as dummy for broadcast ? */
483 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
484 return 0;
485
486 v = apic_read(APIC_LVTT);
487 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
488 apic_write(APIC_LVTT, v);
489 apic_write(APIC_TMICT, 0);
490 return 0;
491}
492
493static inline int
494lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
495{
496 /* Lapic used as dummy for broadcast ? */
497 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
498 return 0;
499
500 __setup_APIC_LVTT(lapic_timer_frequency, oneshot, 1);
501 return 0;
502}
503
504static int lapic_timer_set_periodic(struct clock_event_device *evt)
505{
506 return lapic_timer_set_periodic_oneshot(evt, false);
507}
508
509static int lapic_timer_set_oneshot(struct clock_event_device *evt)
510{
511 return lapic_timer_set_periodic_oneshot(evt, true);
512}
513
514/*
515 * Local APIC timer broadcast function
516 */
517static void lapic_timer_broadcast(const struct cpumask *mask)
518{
519#ifdef CONFIG_SMP
520 apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
521#endif
522}
523
524
525/*
526 * The local apic timer can be used for any function which is CPU local.
527 */
528static struct clock_event_device lapic_clockevent = {
529 .name = "lapic",
530 .features = CLOCK_EVT_FEAT_PERIODIC |
531 CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
532 | CLOCK_EVT_FEAT_DUMMY,
533 .shift = 32,
534 .set_state_shutdown = lapic_timer_shutdown,
535 .set_state_periodic = lapic_timer_set_periodic,
536 .set_state_oneshot = lapic_timer_set_oneshot,
537 .set_next_event = lapic_next_event,
538 .broadcast = lapic_timer_broadcast,
539 .rating = 100,
540 .irq = -1,
541};
542static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
543
544/*
545 * Setup the local APIC timer for this CPU. Copy the initialized values
546 * of the boot CPU and register the clock event in the framework.
547 */
548static void setup_APIC_timer(void)
549{
550 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
551
552 if (this_cpu_has(X86_FEATURE_ARAT)) {
553 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
554 /* Make LAPIC timer preferrable over percpu HPET */
555 lapic_clockevent.rating = 150;
556 }
557
558 memcpy(levt, &lapic_clockevent, sizeof(*levt));
559 levt->cpumask = cpumask_of(smp_processor_id());
560
561 if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
562 levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
563 CLOCK_EVT_FEAT_DUMMY);
564 levt->set_next_event = lapic_next_deadline;
565 clockevents_config_and_register(levt,
566 (tsc_khz / TSC_DIVISOR) * 1000,
567 0xF, ~0UL);
568 } else
569 clockevents_register_device(levt);
570}
571
572/*
573 * In this functions we calibrate APIC bus clocks to the external timer.
574 *
575 * We want to do the calibration only once since we want to have local timer
576 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
577 * frequency.
578 *
579 * This was previously done by reading the PIT/HPET and waiting for a wrap
580 * around to find out, that a tick has elapsed. I have a box, where the PIT
581 * readout is broken, so it never gets out of the wait loop again. This was
582 * also reported by others.
583 *
584 * Monitoring the jiffies value is inaccurate and the clockevents
585 * infrastructure allows us to do a simple substitution of the interrupt
586 * handler.
587 *
588 * The calibration routine also uses the pm_timer when possible, as the PIT
589 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
590 * back to normal later in the boot process).
591 */
592
593#define LAPIC_CAL_LOOPS (HZ/10)
594
595static __initdata int lapic_cal_loops = -1;
596static __initdata long lapic_cal_t1, lapic_cal_t2;
597static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
598static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
599static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
600
601/*
602 * Temporary interrupt handler.
603 */
604static void __init lapic_cal_handler(struct clock_event_device *dev)
605{
606 unsigned long long tsc = 0;
607 long tapic = apic_read(APIC_TMCCT);
608 unsigned long pm = acpi_pm_read_early();
609
610 if (cpu_has_tsc)
611 tsc = rdtsc();
612
613 switch (lapic_cal_loops++) {
614 case 0:
615 lapic_cal_t1 = tapic;
616 lapic_cal_tsc1 = tsc;
617 lapic_cal_pm1 = pm;
618 lapic_cal_j1 = jiffies;
619 break;
620
621 case LAPIC_CAL_LOOPS:
622 lapic_cal_t2 = tapic;
623 lapic_cal_tsc2 = tsc;
624 if (pm < lapic_cal_pm1)
625 pm += ACPI_PM_OVRRUN;
626 lapic_cal_pm2 = pm;
627 lapic_cal_j2 = jiffies;
628 break;
629 }
630}
631
632static int __init
633calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
634{
635 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
636 const long pm_thresh = pm_100ms / 100;
637 unsigned long mult;
638 u64 res;
639
640#ifndef CONFIG_X86_PM_TIMER
641 return -1;
642#endif
643
644 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
645
646 /* Check, if the PM timer is available */
647 if (!deltapm)
648 return -1;
649
650 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
651
652 if (deltapm > (pm_100ms - pm_thresh) &&
653 deltapm < (pm_100ms + pm_thresh)) {
654 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
655 return 0;
656 }
657
658 res = (((u64)deltapm) * mult) >> 22;
659 do_div(res, 1000000);
660 pr_warning("APIC calibration not consistent "
661 "with PM-Timer: %ldms instead of 100ms\n",(long)res);
662
663 /* Correct the lapic counter value */
664 res = (((u64)(*delta)) * pm_100ms);
665 do_div(res, deltapm);
666 pr_info("APIC delta adjusted to PM-Timer: "
667 "%lu (%ld)\n", (unsigned long)res, *delta);
668 *delta = (long)res;
669
670 /* Correct the tsc counter value */
671 if (cpu_has_tsc) {
672 res = (((u64)(*deltatsc)) * pm_100ms);
673 do_div(res, deltapm);
674 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
675 "PM-Timer: %lu (%ld)\n",
676 (unsigned long)res, *deltatsc);
677 *deltatsc = (long)res;
678 }
679
680 return 0;
681}
682
683static int __init calibrate_APIC_clock(void)
684{
685 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
686 void (*real_handler)(struct clock_event_device *dev);
687 unsigned long deltaj;
688 long delta, deltatsc;
689 int pm_referenced = 0;
690
691 /**
692 * check if lapic timer has already been calibrated by platform
693 * specific routine, such as tsc calibration code. if so, we just fill
694 * in the clockevent structure and return.
695 */
696
697 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
698 return 0;
699 } else if (lapic_timer_frequency) {
700 apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
701 lapic_timer_frequency);
702 lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
703 TICK_NSEC, lapic_clockevent.shift);
704 lapic_clockevent.max_delta_ns =
705 clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
706 lapic_clockevent.min_delta_ns =
707 clockevent_delta2ns(0xF, &lapic_clockevent);
708 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
709 return 0;
710 }
711
712 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
713 "calibrating APIC timer ...\n");
714
715 local_irq_disable();
716
717 /* Replace the global interrupt handler */
718 real_handler = global_clock_event->event_handler;
719 global_clock_event->event_handler = lapic_cal_handler;
720
721 /*
722 * Setup the APIC counter to maximum. There is no way the lapic
723 * can underflow in the 100ms detection time frame
724 */
725 __setup_APIC_LVTT(0xffffffff, 0, 0);
726
727 /* Let the interrupts run */
728 local_irq_enable();
729
730 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
731 cpu_relax();
732
733 local_irq_disable();
734
735 /* Restore the real event handler */
736 global_clock_event->event_handler = real_handler;
737
738 /* Build delta t1-t2 as apic timer counts down */
739 delta = lapic_cal_t1 - lapic_cal_t2;
740 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
741
742 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
743
744 /* we trust the PM based calibration if possible */
745 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
746 &delta, &deltatsc);
747
748 /* Calculate the scaled math multiplication factor */
749 lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
750 lapic_clockevent.shift);
751 lapic_clockevent.max_delta_ns =
752 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
753 lapic_clockevent.min_delta_ns =
754 clockevent_delta2ns(0xF, &lapic_clockevent);
755
756 lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
757
758 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
759 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
760 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
761 lapic_timer_frequency);
762
763 if (cpu_has_tsc) {
764 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
765 "%ld.%04ld MHz.\n",
766 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
767 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
768 }
769
770 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
771 "%u.%04u MHz.\n",
772 lapic_timer_frequency / (1000000 / HZ),
773 lapic_timer_frequency % (1000000 / HZ));
774
775 /*
776 * Do a sanity check on the APIC calibration result
777 */
778 if (lapic_timer_frequency < (1000000 / HZ)) {
779 local_irq_enable();
780 pr_warning("APIC frequency too slow, disabling apic timer\n");
781 return -1;
782 }
783
784 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
785
786 /*
787 * PM timer calibration failed or not turned on
788 * so lets try APIC timer based calibration
789 */
790 if (!pm_referenced) {
791 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
792
793 /*
794 * Setup the apic timer manually
795 */
796 levt->event_handler = lapic_cal_handler;
797 lapic_timer_set_periodic(levt);
798 lapic_cal_loops = -1;
799
800 /* Let the interrupts run */
801 local_irq_enable();
802
803 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
804 cpu_relax();
805
806 /* Stop the lapic timer */
807 local_irq_disable();
808 lapic_timer_shutdown(levt);
809
810 /* Jiffies delta */
811 deltaj = lapic_cal_j2 - lapic_cal_j1;
812 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
813
814 /* Check, if the jiffies result is consistent */
815 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
816 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
817 else
818 levt->features |= CLOCK_EVT_FEAT_DUMMY;
819 }
820 local_irq_enable();
821
822 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
823 pr_warning("APIC timer disabled due to verification failure\n");
824 return -1;
825 }
826
827 return 0;
828}
829
830/*
831 * Setup the boot APIC
832 *
833 * Calibrate and verify the result.
834 */
835void __init setup_boot_APIC_clock(void)
836{
837 /*
838 * The local apic timer can be disabled via the kernel
839 * commandline or from the CPU detection code. Register the lapic
840 * timer as a dummy clock event source on SMP systems, so the
841 * broadcast mechanism is used. On UP systems simply ignore it.
842 */
843 if (disable_apic_timer) {
844 pr_info("Disabling APIC timer\n");
845 /* No broadcast on UP ! */
846 if (num_possible_cpus() > 1) {
847 lapic_clockevent.mult = 1;
848 setup_APIC_timer();
849 }
850 return;
851 }
852
853 if (calibrate_APIC_clock()) {
854 /* No broadcast on UP ! */
855 if (num_possible_cpus() > 1)
856 setup_APIC_timer();
857 return;
858 }
859
860 /*
861 * If nmi_watchdog is set to IO_APIC, we need the
862 * PIT/HPET going. Otherwise register lapic as a dummy
863 * device.
864 */
865 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
866
867 /* Setup the lapic or request the broadcast */
868 setup_APIC_timer();
869}
870
871void setup_secondary_APIC_clock(void)
872{
873 setup_APIC_timer();
874}
875
876/*
877 * The guts of the apic timer interrupt
878 */
879static void local_apic_timer_interrupt(void)
880{
881 int cpu = smp_processor_id();
882 struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
883
884 /*
885 * Normally we should not be here till LAPIC has been initialized but
886 * in some cases like kdump, its possible that there is a pending LAPIC
887 * timer interrupt from previous kernel's context and is delivered in
888 * new kernel the moment interrupts are enabled.
889 *
890 * Interrupts are enabled early and LAPIC is setup much later, hence
891 * its possible that when we get here evt->event_handler is NULL.
892 * Check for event_handler being NULL and discard the interrupt as
893 * spurious.
894 */
895 if (!evt->event_handler) {
896 pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
897 /* Switch it off */
898 lapic_timer_shutdown(evt);
899 return;
900 }
901
902 /*
903 * the NMI deadlock-detector uses this.
904 */
905 inc_irq_stat(apic_timer_irqs);
906
907 evt->event_handler(evt);
908}
909
910/*
911 * Local APIC timer interrupt. This is the most natural way for doing
912 * local interrupts, but local timer interrupts can be emulated by
913 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
914 *
915 * [ if a single-CPU system runs an SMP kernel then we call the local
916 * interrupt as well. Thus we cannot inline the local irq ... ]
917 */
918__visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
919{
920 struct pt_regs *old_regs = set_irq_regs(regs);
921
922 /*
923 * NOTE! We'd better ACK the irq immediately,
924 * because timer handling can be slow.
925 *
926 * update_process_times() expects us to have done irq_enter().
927 * Besides, if we don't timer interrupts ignore the global
928 * interrupt lock, which is the WrongThing (tm) to do.
929 */
930 entering_ack_irq();
931 local_apic_timer_interrupt();
932 exiting_irq();
933
934 set_irq_regs(old_regs);
935}
936
937__visible void __irq_entry smp_trace_apic_timer_interrupt(struct pt_regs *regs)
938{
939 struct pt_regs *old_regs = set_irq_regs(regs);
940
941 /*
942 * NOTE! We'd better ACK the irq immediately,
943 * because timer handling can be slow.
944 *
945 * update_process_times() expects us to have done irq_enter().
946 * Besides, if we don't timer interrupts ignore the global
947 * interrupt lock, which is the WrongThing (tm) to do.
948 */
949 entering_ack_irq();
950 trace_local_timer_entry(LOCAL_TIMER_VECTOR);
951 local_apic_timer_interrupt();
952 trace_local_timer_exit(LOCAL_TIMER_VECTOR);
953 exiting_irq();
954
955 set_irq_regs(old_regs);
956}
957
958int setup_profiling_timer(unsigned int multiplier)
959{
960 return -EINVAL;
961}
962
963/*
964 * Local APIC start and shutdown
965 */
966
967/**
968 * clear_local_APIC - shutdown the local APIC
969 *
970 * This is called, when a CPU is disabled and before rebooting, so the state of
971 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
972 * leftovers during boot.
973 */
974void clear_local_APIC(void)
975{
976 int maxlvt;
977 u32 v;
978
979 /* APIC hasn't been mapped yet */
980 if (!x2apic_mode && !apic_phys)
981 return;
982
983 maxlvt = lapic_get_maxlvt();
984 /*
985 * Masking an LVT entry can trigger a local APIC error
986 * if the vector is zero. Mask LVTERR first to prevent this.
987 */
988 if (maxlvt >= 3) {
989 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
990 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
991 }
992 /*
993 * Careful: we have to set masks only first to deassert
994 * any level-triggered sources.
995 */
996 v = apic_read(APIC_LVTT);
997 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
998 v = apic_read(APIC_LVT0);
999 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1000 v = apic_read(APIC_LVT1);
1001 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1002 if (maxlvt >= 4) {
1003 v = apic_read(APIC_LVTPC);
1004 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1005 }
1006
1007 /* lets not touch this if we didn't frob it */
1008#ifdef CONFIG_X86_THERMAL_VECTOR
1009 if (maxlvt >= 5) {
1010 v = apic_read(APIC_LVTTHMR);
1011 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1012 }
1013#endif
1014#ifdef CONFIG_X86_MCE_INTEL
1015 if (maxlvt >= 6) {
1016 v = apic_read(APIC_LVTCMCI);
1017 if (!(v & APIC_LVT_MASKED))
1018 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1019 }
1020#endif
1021
1022 /*
1023 * Clean APIC state for other OSs:
1024 */
1025 apic_write(APIC_LVTT, APIC_LVT_MASKED);
1026 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1027 apic_write(APIC_LVT1, APIC_LVT_MASKED);
1028 if (maxlvt >= 3)
1029 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1030 if (maxlvt >= 4)
1031 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1032
1033 /* Integrated APIC (!82489DX) ? */
1034 if (lapic_is_integrated()) {
1035 if (maxlvt > 3)
1036 /* Clear ESR due to Pentium errata 3AP and 11AP */
1037 apic_write(APIC_ESR, 0);
1038 apic_read(APIC_ESR);
1039 }
1040}
1041
1042/**
1043 * disable_local_APIC - clear and disable the local APIC
1044 */
1045void disable_local_APIC(void)
1046{
1047 unsigned int value;
1048
1049 /* APIC hasn't been mapped yet */
1050 if (!x2apic_mode && !apic_phys)
1051 return;
1052
1053 clear_local_APIC();
1054
1055 /*
1056 * Disable APIC (implies clearing of registers
1057 * for 82489DX!).
1058 */
1059 value = apic_read(APIC_SPIV);
1060 value &= ~APIC_SPIV_APIC_ENABLED;
1061 apic_write(APIC_SPIV, value);
1062
1063#ifdef CONFIG_X86_32
1064 /*
1065 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1066 * restore the disabled state.
1067 */
1068 if (enabled_via_apicbase) {
1069 unsigned int l, h;
1070
1071 rdmsr(MSR_IA32_APICBASE, l, h);
1072 l &= ~MSR_IA32_APICBASE_ENABLE;
1073 wrmsr(MSR_IA32_APICBASE, l, h);
1074 }
1075#endif
1076}
1077
1078/*
1079 * If Linux enabled the LAPIC against the BIOS default disable it down before
1080 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1081 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1082 * for the case where Linux didn't enable the LAPIC.
1083 */
1084void lapic_shutdown(void)
1085{
1086 unsigned long flags;
1087
1088 if (!cpu_has_apic && !apic_from_smp_config())
1089 return;
1090
1091 local_irq_save(flags);
1092
1093#ifdef CONFIG_X86_32
1094 if (!enabled_via_apicbase)
1095 clear_local_APIC();
1096 else
1097#endif
1098 disable_local_APIC();
1099
1100
1101 local_irq_restore(flags);
1102}
1103
1104/**
1105 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1106 */
1107void __init sync_Arb_IDs(void)
1108{
1109 /*
1110 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1111 * needed on AMD.
1112 */
1113 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1114 return;
1115
1116 /*
1117 * Wait for idle.
1118 */
1119 apic_wait_icr_idle();
1120
1121 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1122 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1123 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1124}
1125
1126/*
1127 * An initial setup of the virtual wire mode.
1128 */
1129void __init init_bsp_APIC(void)
1130{
1131 unsigned int value;
1132
1133 /*
1134 * Don't do the setup now if we have a SMP BIOS as the
1135 * through-I/O-APIC virtual wire mode might be active.
1136 */
1137 if (smp_found_config || !cpu_has_apic)
1138 return;
1139
1140 /*
1141 * Do not trust the local APIC being empty at bootup.
1142 */
1143 clear_local_APIC();
1144
1145 /*
1146 * Enable APIC.
1147 */
1148 value = apic_read(APIC_SPIV);
1149 value &= ~APIC_VECTOR_MASK;
1150 value |= APIC_SPIV_APIC_ENABLED;
1151
1152#ifdef CONFIG_X86_32
1153 /* This bit is reserved on P4/Xeon and should be cleared */
1154 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1155 (boot_cpu_data.x86 == 15))
1156 value &= ~APIC_SPIV_FOCUS_DISABLED;
1157 else
1158#endif
1159 value |= APIC_SPIV_FOCUS_DISABLED;
1160 value |= SPURIOUS_APIC_VECTOR;
1161 apic_write(APIC_SPIV, value);
1162
1163 /*
1164 * Set up the virtual wire mode.
1165 */
1166 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1167 value = APIC_DM_NMI;
1168 if (!lapic_is_integrated()) /* 82489DX */
1169 value |= APIC_LVT_LEVEL_TRIGGER;
1170 if (apic_extnmi == APIC_EXTNMI_NONE)
1171 value |= APIC_LVT_MASKED;
1172 apic_write(APIC_LVT1, value);
1173}
1174
1175static void lapic_setup_esr(void)
1176{
1177 unsigned int oldvalue, value, maxlvt;
1178
1179 if (!lapic_is_integrated()) {
1180 pr_info("No ESR for 82489DX.\n");
1181 return;
1182 }
1183
1184 if (apic->disable_esr) {
1185 /*
1186 * Something untraceable is creating bad interrupts on
1187 * secondary quads ... for the moment, just leave the
1188 * ESR disabled - we can't do anything useful with the
1189 * errors anyway - mbligh
1190 */
1191 pr_info("Leaving ESR disabled.\n");
1192 return;
1193 }
1194
1195 maxlvt = lapic_get_maxlvt();
1196 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1197 apic_write(APIC_ESR, 0);
1198 oldvalue = apic_read(APIC_ESR);
1199
1200 /* enables sending errors */
1201 value = ERROR_APIC_VECTOR;
1202 apic_write(APIC_LVTERR, value);
1203
1204 /*
1205 * spec says clear errors after enabling vector.
1206 */
1207 if (maxlvt > 3)
1208 apic_write(APIC_ESR, 0);
1209 value = apic_read(APIC_ESR);
1210 if (value != oldvalue)
1211 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1212 "vector: 0x%08x after: 0x%08x\n",
1213 oldvalue, value);
1214}
1215
1216/**
1217 * setup_local_APIC - setup the local APIC
1218 *
1219 * Used to setup local APIC while initializing BSP or bringin up APs.
1220 * Always called with preemption disabled.
1221 */
1222void setup_local_APIC(void)
1223{
1224 int cpu = smp_processor_id();
1225 unsigned int value, queued;
1226 int i, j, acked = 0;
1227 unsigned long long tsc = 0, ntsc;
1228 long long max_loops = cpu_khz ? cpu_khz : 1000000;
1229
1230 if (cpu_has_tsc)
1231 tsc = rdtsc();
1232
1233 if (disable_apic) {
1234 disable_ioapic_support();
1235 return;
1236 }
1237
1238#ifdef CONFIG_X86_32
1239 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1240 if (lapic_is_integrated() && apic->disable_esr) {
1241 apic_write(APIC_ESR, 0);
1242 apic_write(APIC_ESR, 0);
1243 apic_write(APIC_ESR, 0);
1244 apic_write(APIC_ESR, 0);
1245 }
1246#endif
1247 perf_events_lapic_init();
1248
1249 /*
1250 * Double-check whether this APIC is really registered.
1251 * This is meaningless in clustered apic mode, so we skip it.
1252 */
1253 BUG_ON(!apic->apic_id_registered());
1254
1255 /*
1256 * Intel recommends to set DFR, LDR and TPR before enabling
1257 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1258 * document number 292116). So here it goes...
1259 */
1260 apic->init_apic_ldr();
1261
1262#ifdef CONFIG_X86_32
1263 /*
1264 * APIC LDR is initialized. If logical_apicid mapping was
1265 * initialized during get_smp_config(), make sure it matches the
1266 * actual value.
1267 */
1268 i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1269 WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1270 /* always use the value from LDR */
1271 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1272 logical_smp_processor_id();
1273#endif
1274
1275 /*
1276 * Set Task Priority to 'accept all'. We never change this
1277 * later on.
1278 */
1279 value = apic_read(APIC_TASKPRI);
1280 value &= ~APIC_TPRI_MASK;
1281 apic_write(APIC_TASKPRI, value);
1282
1283 /*
1284 * After a crash, we no longer service the interrupts and a pending
1285 * interrupt from previous kernel might still have ISR bit set.
1286 *
1287 * Most probably by now CPU has serviced that pending interrupt and
1288 * it might not have done the ack_APIC_irq() because it thought,
1289 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1290 * does not clear the ISR bit and cpu thinks it has already serivced
1291 * the interrupt. Hence a vector might get locked. It was noticed
1292 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1293 */
1294 do {
1295 queued = 0;
1296 for (i = APIC_ISR_NR - 1; i >= 0; i--)
1297 queued |= apic_read(APIC_IRR + i*0x10);
1298
1299 for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1300 value = apic_read(APIC_ISR + i*0x10);
1301 for (j = 31; j >= 0; j--) {
1302 if (value & (1<<j)) {
1303 ack_APIC_irq();
1304 acked++;
1305 }
1306 }
1307 }
1308 if (acked > 256) {
1309 printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1310 acked);
1311 break;
1312 }
1313 if (queued) {
1314 if (cpu_has_tsc && cpu_khz) {
1315 ntsc = rdtsc();
1316 max_loops = (cpu_khz << 10) - (ntsc - tsc);
1317 } else
1318 max_loops--;
1319 }
1320 } while (queued && max_loops > 0);
1321 WARN_ON(max_loops <= 0);
1322
1323 /*
1324 * Now that we are all set up, enable the APIC
1325 */
1326 value = apic_read(APIC_SPIV);
1327 value &= ~APIC_VECTOR_MASK;
1328 /*
1329 * Enable APIC
1330 */
1331 value |= APIC_SPIV_APIC_ENABLED;
1332
1333#ifdef CONFIG_X86_32
1334 /*
1335 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1336 * certain networking cards. If high frequency interrupts are
1337 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1338 * entry is masked/unmasked at a high rate as well then sooner or
1339 * later IOAPIC line gets 'stuck', no more interrupts are received
1340 * from the device. If focus CPU is disabled then the hang goes
1341 * away, oh well :-(
1342 *
1343 * [ This bug can be reproduced easily with a level-triggered
1344 * PCI Ne2000 networking cards and PII/PIII processors, dual
1345 * BX chipset. ]
1346 */
1347 /*
1348 * Actually disabling the focus CPU check just makes the hang less
1349 * frequent as it makes the interrupt distributon model be more
1350 * like LRU than MRU (the short-term load is more even across CPUs).
1351 * See also the comment in end_level_ioapic_irq(). --macro
1352 */
1353
1354 /*
1355 * - enable focus processor (bit==0)
1356 * - 64bit mode always use processor focus
1357 * so no need to set it
1358 */
1359 value &= ~APIC_SPIV_FOCUS_DISABLED;
1360#endif
1361
1362 /*
1363 * Set spurious IRQ vector
1364 */
1365 value |= SPURIOUS_APIC_VECTOR;
1366 apic_write(APIC_SPIV, value);
1367
1368 /*
1369 * Set up LVT0, LVT1:
1370 *
1371 * set up through-local-APIC on the BP's LINT0. This is not
1372 * strictly necessary in pure symmetric-IO mode, but sometimes
1373 * we delegate interrupts to the 8259A.
1374 */
1375 /*
1376 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1377 */
1378 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1379 if (!cpu && (pic_mode || !value)) {
1380 value = APIC_DM_EXTINT;
1381 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1382 } else {
1383 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1384 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1385 }
1386 apic_write(APIC_LVT0, value);
1387
1388 /*
1389 * Only the BSP sees the LINT1 NMI signal by default. This can be
1390 * modified by apic_extnmi= boot option.
1391 */
1392 if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1393 apic_extnmi == APIC_EXTNMI_ALL)
1394 value = APIC_DM_NMI;
1395 else
1396 value = APIC_DM_NMI | APIC_LVT_MASKED;
1397 if (!lapic_is_integrated()) /* 82489DX */
1398 value |= APIC_LVT_LEVEL_TRIGGER;
1399 apic_write(APIC_LVT1, value);
1400
1401#ifdef CONFIG_X86_MCE_INTEL
1402 /* Recheck CMCI information after local APIC is up on CPU #0 */
1403 if (!cpu)
1404 cmci_recheck();
1405#endif
1406}
1407
1408static void end_local_APIC_setup(void)
1409{
1410 lapic_setup_esr();
1411
1412#ifdef CONFIG_X86_32
1413 {
1414 unsigned int value;
1415 /* Disable the local apic timer */
1416 value = apic_read(APIC_LVTT);
1417 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1418 apic_write(APIC_LVTT, value);
1419 }
1420#endif
1421
1422 apic_pm_activate();
1423}
1424
1425/*
1426 * APIC setup function for application processors. Called from smpboot.c
1427 */
1428void apic_ap_setup(void)
1429{
1430 setup_local_APIC();
1431 end_local_APIC_setup();
1432}
1433
1434#ifdef CONFIG_X86_X2APIC
1435int x2apic_mode;
1436
1437enum {
1438 X2APIC_OFF,
1439 X2APIC_ON,
1440 X2APIC_DISABLED,
1441};
1442static int x2apic_state;
1443
1444static void __x2apic_disable(void)
1445{
1446 u64 msr;
1447
1448 if (!cpu_has_apic)
1449 return;
1450
1451 rdmsrl(MSR_IA32_APICBASE, msr);
1452 if (!(msr & X2APIC_ENABLE))
1453 return;
1454 /* Disable xapic and x2apic first and then reenable xapic mode */
1455 wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1456 wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1457 printk_once(KERN_INFO "x2apic disabled\n");
1458}
1459
1460static void __x2apic_enable(void)
1461{
1462 u64 msr;
1463
1464 rdmsrl(MSR_IA32_APICBASE, msr);
1465 if (msr & X2APIC_ENABLE)
1466 return;
1467 wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1468 printk_once(KERN_INFO "x2apic enabled\n");
1469}
1470
1471static int __init setup_nox2apic(char *str)
1472{
1473 if (x2apic_enabled()) {
1474 int apicid = native_apic_msr_read(APIC_ID);
1475
1476 if (apicid >= 255) {
1477 pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
1478 apicid);
1479 return 0;
1480 }
1481 pr_warning("x2apic already enabled.\n");
1482 __x2apic_disable();
1483 }
1484 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1485 x2apic_state = X2APIC_DISABLED;
1486 x2apic_mode = 0;
1487 return 0;
1488}
1489early_param("nox2apic", setup_nox2apic);
1490
1491/* Called from cpu_init() to enable x2apic on (secondary) cpus */
1492void x2apic_setup(void)
1493{
1494 /*
1495 * If x2apic is not in ON state, disable it if already enabled
1496 * from BIOS.
1497 */
1498 if (x2apic_state != X2APIC_ON) {
1499 __x2apic_disable();
1500 return;
1501 }
1502 __x2apic_enable();
1503}
1504
1505static __init void x2apic_disable(void)
1506{
1507 u32 x2apic_id, state = x2apic_state;
1508
1509 x2apic_mode = 0;
1510 x2apic_state = X2APIC_DISABLED;
1511
1512 if (state != X2APIC_ON)
1513 return;
1514
1515 x2apic_id = read_apic_id();
1516 if (x2apic_id >= 255)
1517 panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1518
1519 __x2apic_disable();
1520 register_lapic_address(mp_lapic_addr);
1521}
1522
1523static __init void x2apic_enable(void)
1524{
1525 if (x2apic_state != X2APIC_OFF)
1526 return;
1527
1528 x2apic_mode = 1;
1529 x2apic_state = X2APIC_ON;
1530 __x2apic_enable();
1531}
1532
1533static __init void try_to_enable_x2apic(int remap_mode)
1534{
1535 if (x2apic_state == X2APIC_DISABLED)
1536 return;
1537
1538 if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1539 /* IR is required if there is APIC ID > 255 even when running
1540 * under KVM
1541 */
1542 if (max_physical_apicid > 255 ||
1543 !hypervisor_x2apic_available()) {
1544 pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1545 x2apic_disable();
1546 return;
1547 }
1548
1549 /*
1550 * without IR all CPUs can be addressed by IOAPIC/MSI
1551 * only in physical mode
1552 */
1553 x2apic_phys = 1;
1554 }
1555 x2apic_enable();
1556}
1557
1558void __init check_x2apic(void)
1559{
1560 if (x2apic_enabled()) {
1561 pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1562 x2apic_mode = 1;
1563 x2apic_state = X2APIC_ON;
1564 } else if (!cpu_has_x2apic) {
1565 x2apic_state = X2APIC_DISABLED;
1566 }
1567}
1568#else /* CONFIG_X86_X2APIC */
1569static int __init validate_x2apic(void)
1570{
1571 if (!apic_is_x2apic_enabled())
1572 return 0;
1573 /*
1574 * Checkme: Can we simply turn off x2apic here instead of panic?
1575 */
1576 panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n");
1577}
1578early_initcall(validate_x2apic);
1579
1580static inline void try_to_enable_x2apic(int remap_mode) { }
1581static inline void __x2apic_enable(void) { }
1582#endif /* !CONFIG_X86_X2APIC */
1583
1584static int __init try_to_enable_IR(void)
1585{
1586#ifdef CONFIG_X86_IO_APIC
1587 if (!x2apic_enabled() && skip_ioapic_setup) {
1588 pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1589 return -1;
1590 }
1591#endif
1592 return irq_remapping_enable();
1593}
1594
1595void __init enable_IR_x2apic(void)
1596{
1597 unsigned long flags;
1598 int ret, ir_stat;
1599
1600 ir_stat = irq_remapping_prepare();
1601 if (ir_stat < 0 && !x2apic_supported())
1602 return;
1603
1604 ret = save_ioapic_entries();
1605 if (ret) {
1606 pr_info("Saving IO-APIC state failed: %d\n", ret);
1607 return;
1608 }
1609
1610 local_irq_save(flags);
1611 legacy_pic->mask_all();
1612 mask_ioapic_entries();
1613
1614 /* If irq_remapping_prepare() succeeded, try to enable it */
1615 if (ir_stat >= 0)
1616 ir_stat = try_to_enable_IR();
1617 /* ir_stat contains the remap mode or an error code */
1618 try_to_enable_x2apic(ir_stat);
1619
1620 if (ir_stat < 0)
1621 restore_ioapic_entries();
1622 legacy_pic->restore_mask();
1623 local_irq_restore(flags);
1624}
1625
1626#ifdef CONFIG_X86_64
1627/*
1628 * Detect and enable local APICs on non-SMP boards.
1629 * Original code written by Keir Fraser.
1630 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1631 * not correctly set up (usually the APIC timer won't work etc.)
1632 */
1633static int __init detect_init_APIC(void)
1634{
1635 if (!cpu_has_apic) {
1636 pr_info("No local APIC present\n");
1637 return -1;
1638 }
1639
1640 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1641 return 0;
1642}
1643#else
1644
1645static int __init apic_verify(void)
1646{
1647 u32 features, h, l;
1648
1649 /*
1650 * The APIC feature bit should now be enabled
1651 * in `cpuid'
1652 */
1653 features = cpuid_edx(1);
1654 if (!(features & (1 << X86_FEATURE_APIC))) {
1655 pr_warning("Could not enable APIC!\n");
1656 return -1;
1657 }
1658 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1659 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1660
1661 /* The BIOS may have set up the APIC at some other address */
1662 if (boot_cpu_data.x86 >= 6) {
1663 rdmsr(MSR_IA32_APICBASE, l, h);
1664 if (l & MSR_IA32_APICBASE_ENABLE)
1665 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1666 }
1667
1668 pr_info("Found and enabled local APIC!\n");
1669 return 0;
1670}
1671
1672int __init apic_force_enable(unsigned long addr)
1673{
1674 u32 h, l;
1675
1676 if (disable_apic)
1677 return -1;
1678
1679 /*
1680 * Some BIOSes disable the local APIC in the APIC_BASE
1681 * MSR. This can only be done in software for Intel P6 or later
1682 * and AMD K7 (Model > 1) or later.
1683 */
1684 if (boot_cpu_data.x86 >= 6) {
1685 rdmsr(MSR_IA32_APICBASE, l, h);
1686 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1687 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1688 l &= ~MSR_IA32_APICBASE_BASE;
1689 l |= MSR_IA32_APICBASE_ENABLE | addr;
1690 wrmsr(MSR_IA32_APICBASE, l, h);
1691 enabled_via_apicbase = 1;
1692 }
1693 }
1694 return apic_verify();
1695}
1696
1697/*
1698 * Detect and initialize APIC
1699 */
1700static int __init detect_init_APIC(void)
1701{
1702 /* Disabled by kernel option? */
1703 if (disable_apic)
1704 return -1;
1705
1706 switch (boot_cpu_data.x86_vendor) {
1707 case X86_VENDOR_AMD:
1708 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1709 (boot_cpu_data.x86 >= 15))
1710 break;
1711 goto no_apic;
1712 case X86_VENDOR_INTEL:
1713 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1714 (boot_cpu_data.x86 == 5 && cpu_has_apic))
1715 break;
1716 goto no_apic;
1717 default:
1718 goto no_apic;
1719 }
1720
1721 if (!cpu_has_apic) {
1722 /*
1723 * Over-ride BIOS and try to enable the local APIC only if
1724 * "lapic" specified.
1725 */
1726 if (!force_enable_local_apic) {
1727 pr_info("Local APIC disabled by BIOS -- "
1728 "you can enable it with \"lapic\"\n");
1729 return -1;
1730 }
1731 if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1732 return -1;
1733 } else {
1734 if (apic_verify())
1735 return -1;
1736 }
1737
1738 apic_pm_activate();
1739
1740 return 0;
1741
1742no_apic:
1743 pr_info("No local APIC present or hardware disabled\n");
1744 return -1;
1745}
1746#endif
1747
1748/**
1749 * init_apic_mappings - initialize APIC mappings
1750 */
1751void __init init_apic_mappings(void)
1752{
1753 unsigned int new_apicid;
1754
1755 if (x2apic_mode) {
1756 boot_cpu_physical_apicid = read_apic_id();
1757 return;
1758 }
1759
1760 /* If no local APIC can be found return early */
1761 if (!smp_found_config && detect_init_APIC()) {
1762 /* lets NOP'ify apic operations */
1763 pr_info("APIC: disable apic facility\n");
1764 apic_disable();
1765 } else {
1766 apic_phys = mp_lapic_addr;
1767
1768 /*
1769 * acpi lapic path already maps that address in
1770 * acpi_register_lapic_address()
1771 */
1772 if (!acpi_lapic && !smp_found_config)
1773 register_lapic_address(apic_phys);
1774 }
1775
1776 /*
1777 * Fetch the APIC ID of the BSP in case we have a
1778 * default configuration (or the MP table is broken).
1779 */
1780 new_apicid = read_apic_id();
1781 if (boot_cpu_physical_apicid != new_apicid) {
1782 boot_cpu_physical_apicid = new_apicid;
1783 /*
1784 * yeah -- we lie about apic_version
1785 * in case if apic was disabled via boot option
1786 * but it's not a problem for SMP compiled kernel
1787 * since smp_sanity_check is prepared for such a case
1788 * and disable smp mode
1789 */
1790 apic_version[new_apicid] =
1791 GET_APIC_VERSION(apic_read(APIC_LVR));
1792 }
1793}
1794
1795void __init register_lapic_address(unsigned long address)
1796{
1797 mp_lapic_addr = address;
1798
1799 if (!x2apic_mode) {
1800 set_fixmap_nocache(FIX_APIC_BASE, address);
1801 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1802 APIC_BASE, mp_lapic_addr);
1803 }
1804 if (boot_cpu_physical_apicid == -1U) {
1805 boot_cpu_physical_apicid = read_apic_id();
1806 apic_version[boot_cpu_physical_apicid] =
1807 GET_APIC_VERSION(apic_read(APIC_LVR));
1808 }
1809}
1810
1811int apic_version[MAX_LOCAL_APIC];
1812
1813/*
1814 * Local APIC interrupts
1815 */
1816
1817/*
1818 * This interrupt should _never_ happen with our APIC/SMP architecture
1819 */
1820static void __smp_spurious_interrupt(u8 vector)
1821{
1822 u32 v;
1823
1824 /*
1825 * Check if this really is a spurious interrupt and ACK it
1826 * if it is a vectored one. Just in case...
1827 * Spurious interrupts should not be ACKed.
1828 */
1829 v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
1830 if (v & (1 << (vector & 0x1f)))
1831 ack_APIC_irq();
1832
1833 inc_irq_stat(irq_spurious_count);
1834
1835 /* see sw-dev-man vol 3, chapter 7.4.13.5 */
1836 pr_info("spurious APIC interrupt through vector %02x on CPU#%d, "
1837 "should never happen.\n", vector, smp_processor_id());
1838}
1839
1840__visible void smp_spurious_interrupt(struct pt_regs *regs)
1841{
1842 entering_irq();
1843 __smp_spurious_interrupt(~regs->orig_ax);
1844 exiting_irq();
1845}
1846
1847__visible void smp_trace_spurious_interrupt(struct pt_regs *regs)
1848{
1849 u8 vector = ~regs->orig_ax;
1850
1851 entering_irq();
1852 trace_spurious_apic_entry(vector);
1853 __smp_spurious_interrupt(vector);
1854 trace_spurious_apic_exit(vector);
1855 exiting_irq();
1856}
1857
1858/*
1859 * This interrupt should never happen with our APIC/SMP architecture
1860 */
1861static void __smp_error_interrupt(struct pt_regs *regs)
1862{
1863 u32 v;
1864 u32 i = 0;
1865 static const char * const error_interrupt_reason[] = {
1866 "Send CS error", /* APIC Error Bit 0 */
1867 "Receive CS error", /* APIC Error Bit 1 */
1868 "Send accept error", /* APIC Error Bit 2 */
1869 "Receive accept error", /* APIC Error Bit 3 */
1870 "Redirectable IPI", /* APIC Error Bit 4 */
1871 "Send illegal vector", /* APIC Error Bit 5 */
1872 "Received illegal vector", /* APIC Error Bit 6 */
1873 "Illegal register address", /* APIC Error Bit 7 */
1874 };
1875
1876 /* First tickle the hardware, only then report what went on. -- REW */
1877 if (lapic_get_maxlvt() > 3) /* Due to the Pentium erratum 3AP. */
1878 apic_write(APIC_ESR, 0);
1879 v = apic_read(APIC_ESR);
1880 ack_APIC_irq();
1881 atomic_inc(&irq_err_count);
1882
1883 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
1884 smp_processor_id(), v);
1885
1886 v &= 0xff;
1887 while (v) {
1888 if (v & 0x1)
1889 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
1890 i++;
1891 v >>= 1;
1892 }
1893
1894 apic_printk(APIC_DEBUG, KERN_CONT "\n");
1895
1896}
1897
1898__visible void smp_error_interrupt(struct pt_regs *regs)
1899{
1900 entering_irq();
1901 __smp_error_interrupt(regs);
1902 exiting_irq();
1903}
1904
1905__visible void smp_trace_error_interrupt(struct pt_regs *regs)
1906{
1907 entering_irq();
1908 trace_error_apic_entry(ERROR_APIC_VECTOR);
1909 __smp_error_interrupt(regs);
1910 trace_error_apic_exit(ERROR_APIC_VECTOR);
1911 exiting_irq();
1912}
1913
1914/**
1915 * connect_bsp_APIC - attach the APIC to the interrupt system
1916 */
1917static void __init connect_bsp_APIC(void)
1918{
1919#ifdef CONFIG_X86_32
1920 if (pic_mode) {
1921 /*
1922 * Do not trust the local APIC being empty at bootup.
1923 */
1924 clear_local_APIC();
1925 /*
1926 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
1927 * local APIC to INT and NMI lines.
1928 */
1929 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
1930 "enabling APIC mode.\n");
1931 imcr_pic_to_apic();
1932 }
1933#endif
1934}
1935
1936/**
1937 * disconnect_bsp_APIC - detach the APIC from the interrupt system
1938 * @virt_wire_setup: indicates, whether virtual wire mode is selected
1939 *
1940 * Virtual wire mode is necessary to deliver legacy interrupts even when the
1941 * APIC is disabled.
1942 */
1943void disconnect_bsp_APIC(int virt_wire_setup)
1944{
1945 unsigned int value;
1946
1947#ifdef CONFIG_X86_32
1948 if (pic_mode) {
1949 /*
1950 * Put the board back into PIC mode (has an effect only on
1951 * certain older boards). Note that APIC interrupts, including
1952 * IPIs, won't work beyond this point! The only exception are
1953 * INIT IPIs.
1954 */
1955 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
1956 "entering PIC mode.\n");
1957 imcr_apic_to_pic();
1958 return;
1959 }
1960#endif
1961
1962 /* Go back to Virtual Wire compatibility mode */
1963
1964 /* For the spurious interrupt use vector F, and enable it */
1965 value = apic_read(APIC_SPIV);
1966 value &= ~APIC_VECTOR_MASK;
1967 value |= APIC_SPIV_APIC_ENABLED;
1968 value |= 0xf;
1969 apic_write(APIC_SPIV, value);
1970
1971 if (!virt_wire_setup) {
1972 /*
1973 * For LVT0 make it edge triggered, active high,
1974 * external and enabled
1975 */
1976 value = apic_read(APIC_LVT0);
1977 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1978 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1979 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1980 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1981 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
1982 apic_write(APIC_LVT0, value);
1983 } else {
1984 /* Disable LVT0 */
1985 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1986 }
1987
1988 /*
1989 * For LVT1 make it edge triggered, active high,
1990 * nmi and enabled
1991 */
1992 value = apic_read(APIC_LVT1);
1993 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1994 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1995 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1996 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1997 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
1998 apic_write(APIC_LVT1, value);
1999}
2000
2001int generic_processor_info(int apicid, int version)
2002{
2003 int cpu, max = nr_cpu_ids;
2004 bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2005 phys_cpu_present_map);
2006
2007 /*
2008 * boot_cpu_physical_apicid is designed to have the apicid
2009 * returned by read_apic_id(), i.e, the apicid of the
2010 * currently booting-up processor. However, on some platforms,
2011 * it is temporarily modified by the apicid reported as BSP
2012 * through MP table. Concretely:
2013 *
2014 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2015 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2016 *
2017 * This function is executed with the modified
2018 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2019 * parameter doesn't work to disable APs on kdump 2nd kernel.
2020 *
2021 * Since fixing handling of boot_cpu_physical_apicid requires
2022 * another discussion and tests on each platform, we leave it
2023 * for now and here we use read_apic_id() directly in this
2024 * function, generic_processor_info().
2025 */
2026 if (disabled_cpu_apicid != BAD_APICID &&
2027 disabled_cpu_apicid != read_apic_id() &&
2028 disabled_cpu_apicid == apicid) {
2029 int thiscpu = num_processors + disabled_cpus;
2030
2031 pr_warning("APIC: Disabling requested cpu."
2032 " Processor %d/0x%x ignored.\n",
2033 thiscpu, apicid);
2034
2035 disabled_cpus++;
2036 return -ENODEV;
2037 }
2038
2039 /*
2040 * If boot cpu has not been detected yet, then only allow upto
2041 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2042 */
2043 if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2044 apicid != boot_cpu_physical_apicid) {
2045 int thiscpu = max + disabled_cpus - 1;
2046
2047 pr_warning(
2048 "ACPI: NR_CPUS/possible_cpus limit of %i almost"
2049 " reached. Keeping one slot for boot cpu."
2050 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2051
2052 disabled_cpus++;
2053 return -ENODEV;
2054 }
2055
2056 if (num_processors >= nr_cpu_ids) {
2057 int thiscpu = max + disabled_cpus;
2058
2059 pr_warning(
2060 "ACPI: NR_CPUS/possible_cpus limit of %i reached."
2061 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2062
2063 disabled_cpus++;
2064 return -EINVAL;
2065 }
2066
2067 num_processors++;
2068 if (apicid == boot_cpu_physical_apicid) {
2069 /*
2070 * x86_bios_cpu_apicid is required to have processors listed
2071 * in same order as logical cpu numbers. Hence the first
2072 * entry is BSP, and so on.
2073 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2074 * for BSP.
2075 */
2076 cpu = 0;
2077 } else
2078 cpu = cpumask_next_zero(-1, cpu_present_mask);
2079
2080 /*
2081 * This can happen on physical hotplug. The sanity check at boot time
2082 * is done from native_smp_prepare_cpus() after num_possible_cpus() is
2083 * established.
2084 */
2085 if (topology_update_package_map(apicid, cpu) < 0) {
2086 int thiscpu = max + disabled_cpus;
2087
2088 pr_warning("ACPI: Package limit reached. Processor %d/0x%x ignored.\n",
2089 thiscpu, apicid);
2090 disabled_cpus++;
2091 return -ENOSPC;
2092 }
2093
2094 /*
2095 * Validate version
2096 */
2097 if (version == 0x0) {
2098 pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2099 cpu, apicid);
2100 version = 0x10;
2101 }
2102 apic_version[apicid] = version;
2103
2104 if (version != apic_version[boot_cpu_physical_apicid]) {
2105 pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2106 apic_version[boot_cpu_physical_apicid], cpu, version);
2107 }
2108
2109 physid_set(apicid, phys_cpu_present_map);
2110 if (apicid > max_physical_apicid)
2111 max_physical_apicid = apicid;
2112
2113#if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2114 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2115 early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2116#endif
2117#ifdef CONFIG_X86_32
2118 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2119 apic->x86_32_early_logical_apicid(cpu);
2120#endif
2121 set_cpu_possible(cpu, true);
2122 set_cpu_present(cpu, true);
2123
2124 return cpu;
2125}
2126
2127int hard_smp_processor_id(void)
2128{
2129 return read_apic_id();
2130}
2131
2132void default_init_apic_ldr(void)
2133{
2134 unsigned long val;
2135
2136 apic_write(APIC_DFR, APIC_DFR_VALUE);
2137 val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
2138 val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
2139 apic_write(APIC_LDR, val);
2140}
2141
2142int default_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
2143 const struct cpumask *andmask,
2144 unsigned int *apicid)
2145{
2146 unsigned int cpu;
2147
2148 for_each_cpu_and(cpu, cpumask, andmask) {
2149 if (cpumask_test_cpu(cpu, cpu_online_mask))
2150 break;
2151 }
2152
2153 if (likely(cpu < nr_cpu_ids)) {
2154 *apicid = per_cpu(x86_cpu_to_apicid, cpu);
2155 return 0;
2156 }
2157
2158 return -EINVAL;
2159}
2160
2161/*
2162 * Override the generic EOI implementation with an optimized version.
2163 * Only called during early boot when only one CPU is active and with
2164 * interrupts disabled, so we know this does not race with actual APIC driver
2165 * use.
2166 */
2167void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2168{
2169 struct apic **drv;
2170
2171 for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2172 /* Should happen once for each apic */
2173 WARN_ON((*drv)->eoi_write == eoi_write);
2174 (*drv)->eoi_write = eoi_write;
2175 }
2176}
2177
2178static void __init apic_bsp_up_setup(void)
2179{
2180#ifdef CONFIG_X86_64
2181 apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
2182#else
2183 /*
2184 * Hack: In case of kdump, after a crash, kernel might be booting
2185 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2186 * might be zero if read from MP tables. Get it from LAPIC.
2187 */
2188# ifdef CONFIG_CRASH_DUMP
2189 boot_cpu_physical_apicid = read_apic_id();
2190# endif
2191#endif
2192 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2193}
2194
2195/**
2196 * apic_bsp_setup - Setup function for local apic and io-apic
2197 * @upmode: Force UP mode (for APIC_init_uniprocessor)
2198 *
2199 * Returns:
2200 * apic_id of BSP APIC
2201 */
2202int __init apic_bsp_setup(bool upmode)
2203{
2204 int id;
2205
2206 connect_bsp_APIC();
2207 if (upmode)
2208 apic_bsp_up_setup();
2209 setup_local_APIC();
2210
2211 if (x2apic_mode)
2212 id = apic_read(APIC_LDR);
2213 else
2214 id = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
2215
2216 enable_IO_APIC();
2217 end_local_APIC_setup();
2218 irq_remap_enable_fault_handling();
2219 setup_IO_APIC();
2220 /* Setup local timer */
2221 x86_init.timers.setup_percpu_clockev();
2222 return id;
2223}
2224
2225/*
2226 * This initializes the IO-APIC and APIC hardware if this is
2227 * a UP kernel.
2228 */
2229int __init APIC_init_uniprocessor(void)
2230{
2231 if (disable_apic) {
2232 pr_info("Apic disabled\n");
2233 return -1;
2234 }
2235#ifdef CONFIG_X86_64
2236 if (!cpu_has_apic) {
2237 disable_apic = 1;
2238 pr_info("Apic disabled by BIOS\n");
2239 return -1;
2240 }
2241#else
2242 if (!smp_found_config && !cpu_has_apic)
2243 return -1;
2244
2245 /*
2246 * Complain if the BIOS pretends there is one.
2247 */
2248 if (!cpu_has_apic &&
2249 APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
2250 pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
2251 boot_cpu_physical_apicid);
2252 return -1;
2253 }
2254#endif
2255
2256 if (!smp_found_config)
2257 disable_ioapic_support();
2258
2259 default_setup_apic_routing();
2260 apic_bsp_setup(true);
2261 return 0;
2262}
2263
2264#ifdef CONFIG_UP_LATE_INIT
2265void __init up_late_init(void)
2266{
2267 APIC_init_uniprocessor();
2268}
2269#endif
2270
2271/*
2272 * Power management
2273 */
2274#ifdef CONFIG_PM
2275
2276static struct {
2277 /*
2278 * 'active' is true if the local APIC was enabled by us and
2279 * not the BIOS; this signifies that we are also responsible
2280 * for disabling it before entering apm/acpi suspend
2281 */
2282 int active;
2283 /* r/w apic fields */
2284 unsigned int apic_id;
2285 unsigned int apic_taskpri;
2286 unsigned int apic_ldr;
2287 unsigned int apic_dfr;
2288 unsigned int apic_spiv;
2289 unsigned int apic_lvtt;
2290 unsigned int apic_lvtpc;
2291 unsigned int apic_lvt0;
2292 unsigned int apic_lvt1;
2293 unsigned int apic_lvterr;
2294 unsigned int apic_tmict;
2295 unsigned int apic_tdcr;
2296 unsigned int apic_thmr;
2297 unsigned int apic_cmci;
2298} apic_pm_state;
2299
2300static int lapic_suspend(void)
2301{
2302 unsigned long flags;
2303 int maxlvt;
2304
2305 if (!apic_pm_state.active)
2306 return 0;
2307
2308 maxlvt = lapic_get_maxlvt();
2309
2310 apic_pm_state.apic_id = apic_read(APIC_ID);
2311 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2312 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2313 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2314 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2315 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2316 if (maxlvt >= 4)
2317 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2318 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2319 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2320 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2321 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2322 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2323#ifdef CONFIG_X86_THERMAL_VECTOR
2324 if (maxlvt >= 5)
2325 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2326#endif
2327#ifdef CONFIG_X86_MCE_INTEL
2328 if (maxlvt >= 6)
2329 apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2330#endif
2331
2332 local_irq_save(flags);
2333 disable_local_APIC();
2334
2335 irq_remapping_disable();
2336
2337 local_irq_restore(flags);
2338 return 0;
2339}
2340
2341static void lapic_resume(void)
2342{
2343 unsigned int l, h;
2344 unsigned long flags;
2345 int maxlvt;
2346
2347 if (!apic_pm_state.active)
2348 return;
2349
2350 local_irq_save(flags);
2351
2352 /*
2353 * IO-APIC and PIC have their own resume routines.
2354 * We just mask them here to make sure the interrupt
2355 * subsystem is completely quiet while we enable x2apic
2356 * and interrupt-remapping.
2357 */
2358 mask_ioapic_entries();
2359 legacy_pic->mask_all();
2360
2361 if (x2apic_mode) {
2362 __x2apic_enable();
2363 } else {
2364 /*
2365 * Make sure the APICBASE points to the right address
2366 *
2367 * FIXME! This will be wrong if we ever support suspend on
2368 * SMP! We'll need to do this as part of the CPU restore!
2369 */
2370 if (boot_cpu_data.x86 >= 6) {
2371 rdmsr(MSR_IA32_APICBASE, l, h);
2372 l &= ~MSR_IA32_APICBASE_BASE;
2373 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2374 wrmsr(MSR_IA32_APICBASE, l, h);
2375 }
2376 }
2377
2378 maxlvt = lapic_get_maxlvt();
2379 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2380 apic_write(APIC_ID, apic_pm_state.apic_id);
2381 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2382 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2383 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2384 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2385 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2386 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2387#ifdef CONFIG_X86_THERMAL_VECTOR
2388 if (maxlvt >= 5)
2389 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2390#endif
2391#ifdef CONFIG_X86_MCE_INTEL
2392 if (maxlvt >= 6)
2393 apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2394#endif
2395 if (maxlvt >= 4)
2396 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2397 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2398 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2399 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2400 apic_write(APIC_ESR, 0);
2401 apic_read(APIC_ESR);
2402 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2403 apic_write(APIC_ESR, 0);
2404 apic_read(APIC_ESR);
2405
2406 irq_remapping_reenable(x2apic_mode);
2407
2408 local_irq_restore(flags);
2409}
2410
2411/*
2412 * This device has no shutdown method - fully functioning local APICs
2413 * are needed on every CPU up until machine_halt/restart/poweroff.
2414 */
2415
2416static struct syscore_ops lapic_syscore_ops = {
2417 .resume = lapic_resume,
2418 .suspend = lapic_suspend,
2419};
2420
2421static void apic_pm_activate(void)
2422{
2423 apic_pm_state.active = 1;
2424}
2425
2426static int __init init_lapic_sysfs(void)
2427{
2428 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2429 if (cpu_has_apic)
2430 register_syscore_ops(&lapic_syscore_ops);
2431
2432 return 0;
2433}
2434
2435/* local apic needs to resume before other devices access its registers. */
2436core_initcall(init_lapic_sysfs);
2437
2438#else /* CONFIG_PM */
2439
2440static void apic_pm_activate(void) { }
2441
2442#endif /* CONFIG_PM */
2443
2444#ifdef CONFIG_X86_64
2445
2446static int multi_checked;
2447static int multi;
2448
2449static int set_multi(const struct dmi_system_id *d)
2450{
2451 if (multi)
2452 return 0;
2453 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2454 multi = 1;
2455 return 0;
2456}
2457
2458static const struct dmi_system_id multi_dmi_table[] = {
2459 {
2460 .callback = set_multi,
2461 .ident = "IBM System Summit2",
2462 .matches = {
2463 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2464 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2465 },
2466 },
2467 {}
2468};
2469
2470static void dmi_check_multi(void)
2471{
2472 if (multi_checked)
2473 return;
2474
2475 dmi_check_system(multi_dmi_table);
2476 multi_checked = 1;
2477}
2478
2479/*
2480 * apic_is_clustered_box() -- Check if we can expect good TSC
2481 *
2482 * Thus far, the major user of this is IBM's Summit2 series:
2483 * Clustered boxes may have unsynced TSC problems if they are
2484 * multi-chassis.
2485 * Use DMI to check them
2486 */
2487int apic_is_clustered_box(void)
2488{
2489 dmi_check_multi();
2490 return multi;
2491}
2492#endif
2493
2494/*
2495 * APIC command line parameters
2496 */
2497static int __init setup_disableapic(char *arg)
2498{
2499 disable_apic = 1;
2500 setup_clear_cpu_cap(X86_FEATURE_APIC);
2501 return 0;
2502}
2503early_param("disableapic", setup_disableapic);
2504
2505/* same as disableapic, for compatibility */
2506static int __init setup_nolapic(char *arg)
2507{
2508 return setup_disableapic(arg);
2509}
2510early_param("nolapic", setup_nolapic);
2511
2512static int __init parse_lapic_timer_c2_ok(char *arg)
2513{
2514 local_apic_timer_c2_ok = 1;
2515 return 0;
2516}
2517early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2518
2519static int __init parse_disable_apic_timer(char *arg)
2520{
2521 disable_apic_timer = 1;
2522 return 0;
2523}
2524early_param("noapictimer", parse_disable_apic_timer);
2525
2526static int __init parse_nolapic_timer(char *arg)
2527{
2528 disable_apic_timer = 1;
2529 return 0;
2530}
2531early_param("nolapic_timer", parse_nolapic_timer);
2532
2533static int __init apic_set_verbosity(char *arg)
2534{
2535 if (!arg) {
2536#ifdef CONFIG_X86_64
2537 skip_ioapic_setup = 0;
2538 return 0;
2539#endif
2540 return -EINVAL;
2541 }
2542
2543 if (strcmp("debug", arg) == 0)
2544 apic_verbosity = APIC_DEBUG;
2545 else if (strcmp("verbose", arg) == 0)
2546 apic_verbosity = APIC_VERBOSE;
2547 else {
2548 pr_warning("APIC Verbosity level %s not recognised"
2549 " use apic=verbose or apic=debug\n", arg);
2550 return -EINVAL;
2551 }
2552
2553 return 0;
2554}
2555early_param("apic", apic_set_verbosity);
2556
2557static int __init lapic_insert_resource(void)
2558{
2559 if (!apic_phys)
2560 return -1;
2561
2562 /* Put local APIC into the resource map. */
2563 lapic_resource.start = apic_phys;
2564 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2565 insert_resource(&iomem_resource, &lapic_resource);
2566
2567 return 0;
2568}
2569
2570/*
2571 * need call insert after e820_reserve_resources()
2572 * that is using request_resource
2573 */
2574late_initcall(lapic_insert_resource);
2575
2576static int __init apic_set_disabled_cpu_apicid(char *arg)
2577{
2578 if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2579 return -EINVAL;
2580
2581 return 0;
2582}
2583early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2584
2585static int __init apic_set_extnmi(char *arg)
2586{
2587 if (!arg)
2588 return -EINVAL;
2589
2590 if (!strncmp("all", arg, 3))
2591 apic_extnmi = APIC_EXTNMI_ALL;
2592 else if (!strncmp("none", arg, 4))
2593 apic_extnmi = APIC_EXTNMI_NONE;
2594 else if (!strncmp("bsp", arg, 3))
2595 apic_extnmi = APIC_EXTNMI_BSP;
2596 else {
2597 pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2598 return -EINVAL;
2599 }
2600
2601 return 0;
2602}
2603early_param("apic_extnmi", apic_set_extnmi);