Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
apic.c
Go to the documentation of this file.
1 /*
2  * Local APIC handling, local APIC timers
3  *
4  * (c) 1999, 2000, 2009 Ingo Molnar <[email protected]>
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 
58 unsigned int num_processors;
59 
60 unsigned disabled_cpus __cpuinitdata;
61 
62 /* Processor that is doing the boot up */
63 unsigned int boot_cpu_physical_apicid = -1U;
64 
65 /*
66  * The highest APIC ID seen during enumeration.
67  */
68 unsigned int max_physical_apicid;
69 
70 /*
71  * Bitmask of physically existing CPUs:
72  */
74 
75 /*
76  * Map cpu index to physical APIC ID
77  */
79 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
80 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
81 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
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  */
91 DEFINE_EARLY_PER_CPU_READ_MOSTLY(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  */
98 static int force_enable_local_apic __initdata;
99 /*
100  * APIC command line parameters
101  */
102 static int __init parse_lapic(char *arg)
103 {
104  force_enable_local_apic = 1;
105  return 0;
106 }
107 early_param("lapic", parse_lapic);
108 /* Local APIC was disabled by the BIOS and enabled by the kernel */
109 static int enabled_via_apicbase;
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  */
119 static 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 
127 static 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
137 static int apic_calibrate_pmtmr __initdata;
138 static __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 
148 #ifdef CONFIG_X86_X2APIC
149 /* x2apic enabled before OS handover */
150 int x2apic_preenabled;
151 static int x2apic_disabled;
152 static int nox2apic;
153 static __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 }
172 early_param("nox2apic", setup_nox2apic);
173 #endif
174 
175 unsigned long mp_lapic_addr;
177 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
178 static int disable_apic_timer __initdata;
179 /* Local APIC timer works in C2 */
182 
184 
185 /*
186  * Debug level, exported for io_apic.c
187  */
188 unsigned int apic_verbosity;
189 
191 
192 /* Have we found an MP table */
194 
195 static struct resource lapic_resource = {
196  .name = "Local APIC",
197  .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
198 };
199 
200 unsigned int lapic_timer_frequency = 0;
201 
202 static void apic_pm_activate(void);
203 
204 static unsigned long apic_phys;
205 
206 /*
207  * Get the LAPIC version
208  */
209 static 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  */
217 static 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  */
229 static 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  */
242 static void __init apic_disable(void)
243 {
244  pr_info("APIC: switched to apic NOOP\n");
245  apic = &apic_noop;
246 }
247 
249 {
250  while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
251  cpu_relax();
252 }
253 
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 
272 {
273  apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
274  apic_write(APIC_ICR, low);
275 }
276 
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 
291 int get_physical_broadcast(void)
292 {
293  return modern_apic() ? 0xff : 0xf;
294 }
295 #endif
296 
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  */
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  */
329 static 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())
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)) |
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 
376 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
377 
378 static 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 
385 static 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 
415 int 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 }
445 
446 /*
447  * Program the next event, relative to now
448  */
449 static 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  */
459 static 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);
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  */
495 static void lapic_timer_broadcast(const struct cpumask *mask)
496 {
497 #ifdef CONFIG_SMP
499 #endif
500 }
501 
502 
503 /*
504  * The local apic timer can be used for any function which is CPU local.
505  */
506 static 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 };
517 static 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  */
523 static 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 
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 
562 static __initdata int lapic_cal_loops = -1;
563 static __initdata long lapic_cal_t1, lapic_cal_t2;
564 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
565 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
566 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
567 
568 /*
569  * Temporary interrupt handler.
570  */
571 static 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 
599 static int __init
600 calibrate_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 
650 static 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 
664  if (lapic_timer_frequency) {
665  apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
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 
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 */
691 
692  while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
693  cpu_relax();
694 
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 
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",
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)) {
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 */
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
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  */
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 
836 {
837  setup_APIC_timer();
838 }
839 
840 /*
841  * The guts of the apic timer interrupt
842  */
843 static 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  */
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 
904 int setup_profiling_timer(unsigned int multiplier)
905 {
906  return -EINVAL;
907 }
908 
909 /*
910  * Local APIC start and shutdown
911  */
912 
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 
992 {
993  unsigned int value;
994 
995  /* APIC hasn't been mapped yet */
996  if (!x2apic_mode && !apic_phys)
997  return;
998 
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);
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  */
1030 void 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
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  */
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 
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 |
1131 }
1132 
1133 /*
1134  * An initial setup of the virtual wire mode.
1135  */
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 
1180 static 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 
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) {
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
1253 
1254  /*
1255  * Double-check whether this APIC is really registered.
1256  * This is meaningless in clustered apic mode, so we skip it.
1257  */
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 
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 
1440 {
1442 
1443  /*
1444  * Now that local APIC setup is completed for BP, configure the fault
1445  * handling for interrupt remapping.
1446  */
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  */
1456 static 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 
1463 static __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 
1489  }
1490 }
1491 
1492 void 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 
1500 void 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 
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 
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 */
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();
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 
1603 skip_x2apic:
1604  if (ret < 0) /* IR enabling failed */
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  */
1617 static 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 
1625  return 0;
1626 }
1627 #else
1628 
1629 static 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);
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)
1650  }
1651 
1652  pr_info("Found and enabled local APIC!\n");
1653  return 0;
1654 }
1655 
1656 int __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  */
1684 static 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  }
1716  return -1;
1717  } else {
1718  if (apic_verify())
1719  return -1;
1720  }
1721 
1722  apic_pm_activate();
1723 
1724  return 0;
1725 
1726 no_apic:
1727  pr_info("No local APIC present or hardware disabled\n");
1728  return -1;
1729 }
1730 #endif
1731 
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 
1780 {
1782 
1783  if (!x2apic_mode) {
1784  set_fixmap_nocache(FIX_APIC_BASE, address);
1785  apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1787  }
1788  if (boot_cpu_physical_apicid == -1U) {
1789  boot_cpu_physical_apicid = read_apic_id();
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  */
1800 
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 
1829 
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 
1858 
1859 #ifdef CONFIG_X86_IO_APIC
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  */
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  */
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();
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 
1940 
1941  irq_exit();
1942 }
1943 
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)
1966 }
1967 
1975 void 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 |
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 |
2029  value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2030  apic_write(APIC_LVT1, value);
2031 }
2032 
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)
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 
2112 {
2113  return read_apic_id();
2114 }
2115 
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 
2127  const struct cpumask *andmask,
2128  unsigned int *apicid)
2129 {
2130  unsigned int cpu;
2131 
2132  for_each_cpu_and(cpu, cpumask, andmask) {
2134  break;
2135  }
2136 
2137  if (likely(cpu < nr_cpu_ids)) {
2138  *apicid = per_cpu(x86_cpu_to_apicid, cpu);
2139  return 0;
2140  }
2141 
2142  return -EINVAL;
2143 }
2144 
2145 /*
2146  * Override the generic EOI implementation with an optimized version.
2147  * Only called during early boot when only one CPU is active and with
2148  * interrupts disabled, so we know this does not race with actual APIC driver
2149  * use.
2150  */
2151 void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2152 {
2153  struct apic **drv;
2154 
2155  for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2156  /* Should happen once for each apic */
2157  WARN_ON((*drv)->eoi_write == eoi_write);
2158  (*drv)->eoi_write = eoi_write;
2159  }
2160 }
2161 
2162 /*
2163  * Power management
2164  */
2165 #ifdef CONFIG_PM
2166 
2167 static struct {
2168  /*
2169  * 'active' is true if the local APIC was enabled by us and
2170  * not the BIOS; this signifies that we are also responsible
2171  * for disabling it before entering apm/acpi suspend
2172  */
2173  int active;
2174  /* r/w apic fields */
2175  unsigned int apic_id;
2176  unsigned int apic_taskpri;
2177  unsigned int apic_ldr;
2178  unsigned int apic_dfr;
2179  unsigned int apic_spiv;
2180  unsigned int apic_lvtt;
2181  unsigned int apic_lvtpc;
2182  unsigned int apic_lvt0;
2183  unsigned int apic_lvt1;
2184  unsigned int apic_lvterr;
2185  unsigned int apic_tmict;
2186  unsigned int apic_tdcr;
2187  unsigned int apic_thmr;
2188 } apic_pm_state;
2189 
2190 static int lapic_suspend(void)
2191 {
2192  unsigned long flags;
2193  int maxlvt;
2194 
2195  if (!apic_pm_state.active)
2196  return 0;
2197 
2198  maxlvt = lapic_get_maxlvt();
2199 
2200  apic_pm_state.apic_id = apic_read(APIC_ID);
2201  apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2202  apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2203  apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2204  apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2205  apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2206  if (maxlvt >= 4)
2207  apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2208  apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2209  apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2210  apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2211  apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2212  apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2213 #ifdef CONFIG_X86_THERMAL_VECTOR
2214  if (maxlvt >= 5)
2215  apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2216 #endif
2217 
2218  local_irq_save(flags);
2220 
2223 
2224  local_irq_restore(flags);
2225  return 0;
2226 }
2227 
2228 static void lapic_resume(void)
2229 {
2230  unsigned int l, h;
2231  unsigned long flags;
2232  int maxlvt;
2233 
2234  if (!apic_pm_state.active)
2235  return;
2236 
2237  local_irq_save(flags);
2238  if (irq_remapping_enabled) {
2239  /*
2240  * IO-APIC and PIC have their own resume routines.
2241  * We just mask them here to make sure the interrupt
2242  * subsystem is completely quiet while we enable x2apic
2243  * and interrupt-remapping.
2244  */
2246  legacy_pic->mask_all();
2247  }
2248 
2249  if (x2apic_mode)
2250  enable_x2apic();
2251  else {
2252  /*
2253  * Make sure the APICBASE points to the right address
2254  *
2255  * FIXME! This will be wrong if we ever support suspend on
2256  * SMP! We'll need to do this as part of the CPU restore!
2257  */
2258  if (boot_cpu_data.x86 >= 6) {
2259  rdmsr(MSR_IA32_APICBASE, l, h);
2260  l &= ~MSR_IA32_APICBASE_BASE;
2262  wrmsr(MSR_IA32_APICBASE, l, h);
2263  }
2264  }
2265 
2266  maxlvt = lapic_get_maxlvt();
2268  apic_write(APIC_ID, apic_pm_state.apic_id);
2269  apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2270  apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2271  apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2272  apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2273  apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2274  apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2275 #if defined(CONFIG_X86_MCE_P4THERMAL) || defined(CONFIG_X86_MCE_INTEL)
2276  if (maxlvt >= 5)
2277  apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2278 #endif
2279  if (maxlvt >= 4)
2280  apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2281  apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2282  apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2283  apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2284  apic_write(APIC_ESR, 0);
2285  apic_read(APIC_ESR);
2286  apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2287  apic_write(APIC_ESR, 0);
2288  apic_read(APIC_ESR);
2289 
2292 
2293  local_irq_restore(flags);
2294 }
2295 
2296 /*
2297  * This device has no shutdown method - fully functioning local APICs
2298  * are needed on every CPU up until machine_halt/restart/poweroff.
2299  */
2300 
2301 static struct syscore_ops lapic_syscore_ops = {
2302  .resume = lapic_resume,
2303  .suspend = lapic_suspend,
2304 };
2305 
2306 static void __cpuinit apic_pm_activate(void)
2307 {
2308  apic_pm_state.active = 1;
2309 }
2310 
2311 static int __init init_lapic_sysfs(void)
2312 {
2313  /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2314  if (cpu_has_apic)
2315  register_syscore_ops(&lapic_syscore_ops);
2316 
2317  return 0;
2318 }
2319 
2320 /* local apic needs to resume before other devices access its registers. */
2321 core_initcall(init_lapic_sysfs);
2322 
2323 #else /* CONFIG_PM */
2324 
2325 static void apic_pm_activate(void) { }
2326 
2327 #endif /* CONFIG_PM */
2328 
2329 #ifdef CONFIG_X86_64
2330 
2331 static int __cpuinit apic_cluster_num(void)
2332 {
2333  int i, clusters, zeros;
2334  unsigned id;
2335  u16 *bios_cpu_apicid;
2336  DECLARE_BITMAP(clustermap, NUM_APIC_CLUSTERS);
2337 
2338  bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
2339  bitmap_zero(clustermap, NUM_APIC_CLUSTERS);
2340 
2341  for (i = 0; i < nr_cpu_ids; i++) {
2342  /* are we being called early in kernel startup? */
2343  if (bios_cpu_apicid) {
2344  id = bios_cpu_apicid[i];
2345  } else if (i < nr_cpu_ids) {
2346  if (cpu_present(i))
2347  id = per_cpu(x86_bios_cpu_apicid, i);
2348  else
2349  continue;
2350  } else
2351  break;
2352 
2353  if (id != BAD_APICID)
2354  __set_bit(APIC_CLUSTERID(id), clustermap);
2355  }
2356 
2357  /* Problem: Partially populated chassis may not have CPUs in some of
2358  * the APIC clusters they have been allocated. Only present CPUs have
2359  * x86_bios_cpu_apicid entries, thus causing zeroes in the bitmap.
2360  * Since clusters are allocated sequentially, count zeros only if
2361  * they are bounded by ones.
2362  */
2363  clusters = 0;
2364  zeros = 0;
2365  for (i = 0; i < NUM_APIC_CLUSTERS; i++) {
2366  if (test_bit(i, clustermap)) {
2367  clusters += 1 + zeros;
2368  zeros = 0;
2369  } else
2370  ++zeros;
2371  }
2372 
2373  return clusters;
2374 }
2375 
2376 static int __cpuinitdata multi_checked;
2377 static int __cpuinitdata multi;
2378 
2379 static int __cpuinit set_multi(const struct dmi_system_id *d)
2380 {
2381  if (multi)
2382  return 0;
2383  pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2384  multi = 1;
2385  return 0;
2386 }
2387 
2388 static const __cpuinitconst struct dmi_system_id multi_dmi_table[] = {
2389  {
2390  .callback = set_multi,
2391  .ident = "IBM System Summit2",
2392  .matches = {
2393  DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2394  DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2395  },
2396  },
2397  {}
2398 };
2399 
2400 static void __cpuinit dmi_check_multi(void)
2401 {
2402  if (multi_checked)
2403  return;
2404 
2405  dmi_check_system(multi_dmi_table);
2406  multi_checked = 1;
2407 }
2408 
2409 /*
2410  * apic_is_clustered_box() -- Check if we can expect good TSC
2411  *
2412  * Thus far, the major user of this is IBM's Summit2 series:
2413  * Clustered boxes may have unsynced TSC problems if they are
2414  * multi-chassis.
2415  * Use DMI to check them
2416  */
2417 __cpuinit int apic_is_clustered_box(void)
2418 {
2419  dmi_check_multi();
2420  if (multi)
2421  return 1;
2422 
2423  if (!is_vsmp_box())
2424  return 0;
2425 
2426  /*
2427  * ScaleMP vSMPowered boxes have one cluster per board and TSCs are
2428  * not guaranteed to be synced between boards
2429  */
2430  if (apic_cluster_num() > 1)
2431  return 1;
2432 
2433  return 0;
2434 }
2435 #endif
2436 
2437 /*
2438  * APIC command line parameters
2439  */
2440 static int __init setup_disableapic(char *arg)
2441 {
2442  disable_apic = 1;
2443  setup_clear_cpu_cap(X86_FEATURE_APIC);
2444  return 0;
2445 }
2446 early_param("disableapic", setup_disableapic);
2447 
2448 /* same as disableapic, for compatibility */
2449 static int __init setup_nolapic(char *arg)
2450 {
2451  return setup_disableapic(arg);
2452 }
2453 early_param("nolapic", setup_nolapic);
2454 
2455 static int __init parse_lapic_timer_c2_ok(char *arg)
2456 {
2458  return 0;
2459 }
2460 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2461 
2462 static int __init parse_disable_apic_timer(char *arg)
2463 {
2464  disable_apic_timer = 1;
2465  return 0;
2466 }
2467 early_param("noapictimer", parse_disable_apic_timer);
2468 
2469 static int __init parse_nolapic_timer(char *arg)
2470 {
2471  disable_apic_timer = 1;
2472  return 0;
2473 }
2474 early_param("nolapic_timer", parse_nolapic_timer);
2475 
2476 static int __init apic_set_verbosity(char *arg)
2477 {
2478  if (!arg) {
2479 #ifdef CONFIG_X86_64
2480  skip_ioapic_setup = 0;
2481  return 0;
2482 #endif
2483  return -EINVAL;
2484  }
2485 
2486  if (strcmp("debug", arg) == 0)
2488  else if (strcmp("verbose", arg) == 0)
2490  else {
2491  pr_warning("APIC Verbosity level %s not recognised"
2492  " use apic=verbose or apic=debug\n", arg);
2493  return -EINVAL;
2494  }
2495 
2496  return 0;
2497 }
2498 early_param("apic", apic_set_verbosity);
2499 
2500 static int __init lapic_insert_resource(void)
2501 {
2502  if (!apic_phys)
2503  return -1;
2504 
2505  /* Put local APIC into the resource map. */
2506  lapic_resource.start = apic_phys;
2507  lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2508  insert_resource(&iomem_resource, &lapic_resource);
2509 
2510  return 0;
2511 }
2512 
2513 /*
2514  * need call insert after e820_reserve_resources()
2515  * that is using request_resource
2516  */
2517 late_initcall(lapic_insert_resource);