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/export.h>
27#include <linux/syscore_ops.h>
28#include <linux/delay.h>
29#include <linux/timex.h>
30#include <linux/i8253.h>
31#include <linux/dmar.h>
32#include <linux/init.h>
33#include <linux/cpu.h>
34#include <linux/dmi.h>
35#include <linux/smp.h>
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/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#include <asm/cpu_device_id.h>
58#include <asm/intel-family.h>
59
60unsigned int num_processors;
61
62unsigned disabled_cpus;
63
64/* Processor that is doing the boot up */
65unsigned int boot_cpu_physical_apicid = -1U;
66EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
67
68u8 boot_cpu_apic_version;
69
70/*
71 * The highest APIC ID seen during enumeration.
72 */
73static unsigned int max_physical_apicid;
74
75/*
76 * Bitmask of physically existing CPUs:
77 */
78physid_mask_t phys_cpu_present_map;
79
80/*
81 * Processor to be disabled specified by kernel parameter
82 * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
83 * avoid undefined behaviour caused by sending INIT from AP to BSP.
84 */
85static unsigned int disabled_cpu_apicid __read_mostly = BAD_APICID;
86
87/*
88 * This variable controls which CPUs receive external NMIs. By default,
89 * external NMIs are delivered only to the BSP.
90 */
91static int apic_extnmi = APIC_EXTNMI_BSP;
92
93/*
94 * Map cpu index to physical APIC ID
95 */
96DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
97DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
98DEFINE_EARLY_PER_CPU_READ_MOSTLY(u32, x86_cpu_to_acpiid, U32_MAX);
99EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
100EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
101EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_acpiid);
102
103#ifdef CONFIG_X86_32
104
105/*
106 * On x86_32, the mapping between cpu and logical apicid may vary
107 * depending on apic in use. The following early percpu variable is
108 * used for the mapping. This is where the behaviors of x86_64 and 32
109 * actually diverge. Let's keep it ugly for now.
110 */
111DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
112
113/* Local APIC was disabled by the BIOS and enabled by the kernel */
114static int enabled_via_apicbase;
115
116/*
117 * Handle interrupt mode configuration register (IMCR).
118 * This register controls whether the interrupt signals
119 * that reach the BSP come from the master PIC or from the
120 * local APIC. Before entering Symmetric I/O Mode, either
121 * the BIOS or the operating system must switch out of
122 * PIC Mode by changing the IMCR.
123 */
124static inline void imcr_pic_to_apic(void)
125{
126 /* select IMCR register */
127 outb(0x70, 0x22);
128 /* NMI and 8259 INTR go through APIC */
129 outb(0x01, 0x23);
130}
131
132static inline void imcr_apic_to_pic(void)
133{
134 /* select IMCR register */
135 outb(0x70, 0x22);
136 /* NMI and 8259 INTR go directly to BSP */
137 outb(0x00, 0x23);
138}
139#endif
140
141/*
142 * Knob to control our willingness to enable the local APIC.
143 *
144 * +1=force-enable
145 */
146static int force_enable_local_apic __initdata;
147
148/*
149 * APIC command line parameters
150 */
151static int __init parse_lapic(char *arg)
152{
153 if (IS_ENABLED(CONFIG_X86_32) && !arg)
154 force_enable_local_apic = 1;
155 else if (arg && !strncmp(arg, "notscdeadline", 13))
156 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
157 return 0;
158}
159early_param("lapic", parse_lapic);
160
161#ifdef CONFIG_X86_64
162static int apic_calibrate_pmtmr __initdata;
163static __init int setup_apicpmtimer(char *s)
164{
165 apic_calibrate_pmtmr = 1;
166 notsc_setup(NULL);
167 return 0;
168}
169__setup("apicpmtimer", setup_apicpmtimer);
170#endif
171
172unsigned long mp_lapic_addr;
173int disable_apic;
174/* Disable local APIC timer from the kernel commandline or via dmi quirk */
175static int disable_apic_timer __initdata;
176/* Local APIC timer works in C2 */
177int local_apic_timer_c2_ok;
178EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
179
180/*
181 * Debug level, exported for io_apic.c
182 */
183unsigned int apic_verbosity;
184
185int pic_mode;
186
187/* Have we found an MP table */
188int smp_found_config;
189
190static struct resource lapic_resource = {
191 .name = "Local APIC",
192 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
193};
194
195unsigned int lapic_timer_frequency = 0;
196
197static void apic_pm_activate(void);
198
199static unsigned long apic_phys;
200
201/*
202 * Get the LAPIC version
203 */
204static inline int lapic_get_version(void)
205{
206 return GET_APIC_VERSION(apic_read(APIC_LVR));
207}
208
209/*
210 * Check, if the APIC is integrated or a separate chip
211 */
212static inline int lapic_is_integrated(void)
213{
214 return APIC_INTEGRATED(lapic_get_version());
215}
216
217/*
218 * Check, whether this is a modern or a first generation APIC
219 */
220static int modern_apic(void)
221{
222 /* AMD systems use old APIC versions, so check the CPU */
223 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
224 boot_cpu_data.x86 >= 0xf)
225 return 1;
226 return lapic_get_version() >= 0x14;
227}
228
229/*
230 * right after this call apic become NOOP driven
231 * so apic->write/read doesn't do anything
232 */
233static void __init apic_disable(void)
234{
235 pr_info("APIC: switched to apic NOOP\n");
236 apic = &apic_noop;
237}
238
239void native_apic_wait_icr_idle(void)
240{
241 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
242 cpu_relax();
243}
244
245u32 native_safe_apic_wait_icr_idle(void)
246{
247 u32 send_status;
248 int timeout;
249
250 timeout = 0;
251 do {
252 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
253 if (!send_status)
254 break;
255 inc_irq_stat(icr_read_retry_count);
256 udelay(100);
257 } while (timeout++ < 1000);
258
259 return send_status;
260}
261
262void native_apic_icr_write(u32 low, u32 id)
263{
264 unsigned long flags;
265
266 local_irq_save(flags);
267 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
268 apic_write(APIC_ICR, low);
269 local_irq_restore(flags);
270}
271
272u64 native_apic_icr_read(void)
273{
274 u32 icr1, icr2;
275
276 icr2 = apic_read(APIC_ICR2);
277 icr1 = apic_read(APIC_ICR);
278
279 return icr1 | ((u64)icr2 << 32);
280}
281
282#ifdef CONFIG_X86_32
283/**
284 * get_physical_broadcast - Get number of physical broadcast IDs
285 */
286int get_physical_broadcast(void)
287{
288 return modern_apic() ? 0xff : 0xf;
289}
290#endif
291
292/**
293 * lapic_get_maxlvt - get the maximum number of local vector table entries
294 */
295int lapic_get_maxlvt(void)
296{
297 /*
298 * - we always have APIC integrated on 64bit mode
299 * - 82489DXs do not report # of LVT entries
300 */
301 return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
302}
303
304/*
305 * Local APIC timer
306 */
307
308/* Clock divisor */
309#define APIC_DIVISOR 16
310#define TSC_DIVISOR 8
311
312/*
313 * This function sets up the local APIC timer, with a timeout of
314 * 'clocks' APIC bus clock. During calibration we actually call
315 * this function twice on the boot CPU, once with a bogus timeout
316 * value, second time for real. The other (noncalibrating) CPUs
317 * call this function only once, with the real, calibrated value.
318 *
319 * We do reads before writes even if unnecessary, to get around the
320 * P5 APIC double write bug.
321 */
322static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
323{
324 unsigned int lvtt_value, tmp_value;
325
326 lvtt_value = LOCAL_TIMER_VECTOR;
327 if (!oneshot)
328 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
329 else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
330 lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
331
332 if (!lapic_is_integrated())
333 lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
334
335 if (!irqen)
336 lvtt_value |= APIC_LVT_MASKED;
337
338 apic_write(APIC_LVTT, lvtt_value);
339
340 if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
341 /*
342 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
343 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
344 * According to Intel, MFENCE can do the serialization here.
345 */
346 asm volatile("mfence" : : : "memory");
347
348 printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
349 return;
350 }
351
352 /*
353 * Divide PICLK by 16
354 */
355 tmp_value = apic_read(APIC_TDCR);
356 apic_write(APIC_TDCR,
357 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
358 APIC_TDR_DIV_16);
359
360 if (!oneshot)
361 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
362}
363
364/*
365 * Setup extended LVT, AMD specific
366 *
367 * Software should use the LVT offsets the BIOS provides. The offsets
368 * are determined by the subsystems using it like those for MCE
369 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts
370 * are supported. Beginning with family 10h at least 4 offsets are
371 * available.
372 *
373 * Since the offsets must be consistent for all cores, we keep track
374 * of the LVT offsets in software and reserve the offset for the same
375 * vector also to be used on other cores. An offset is freed by
376 * setting the entry to APIC_EILVT_MASKED.
377 *
378 * If the BIOS is right, there should be no conflicts. Otherwise a
379 * "[Firmware Bug]: ..." error message is generated. However, if
380 * software does not properly determines the offsets, it is not
381 * necessarily a BIOS bug.
382 */
383
384static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
385
386static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
387{
388 return (old & APIC_EILVT_MASKED)
389 || (new == APIC_EILVT_MASKED)
390 || ((new & ~APIC_EILVT_MASKED) == old);
391}
392
393static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
394{
395 unsigned int rsvd, vector;
396
397 if (offset >= APIC_EILVT_NR_MAX)
398 return ~0;
399
400 rsvd = atomic_read(&eilvt_offsets[offset]);
401 do {
402 vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */
403 if (vector && !eilvt_entry_is_changeable(vector, new))
404 /* may not change if vectors are different */
405 return rsvd;
406 rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
407 } while (rsvd != new);
408
409 rsvd &= ~APIC_EILVT_MASKED;
410 if (rsvd && rsvd != vector)
411 pr_info("LVT offset %d assigned for vector 0x%02x\n",
412 offset, rsvd);
413
414 return new;
415}
416
417/*
418 * If mask=1, the LVT entry does not generate interrupts while mask=0
419 * enables the vector. See also the BKDGs. Must be called with
420 * preemption disabled.
421 */
422
423int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
424{
425 unsigned long reg = APIC_EILVTn(offset);
426 unsigned int new, old, reserved;
427
428 new = (mask << 16) | (msg_type << 8) | vector;
429 old = apic_read(reg);
430 reserved = reserve_eilvt_offset(offset, new);
431
432 if (reserved != 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 another cpu\n",
436 smp_processor_id(), reg, offset, new, reserved);
437 return -EINVAL;
438 }
439
440 if (!eilvt_entry_is_changeable(old, new)) {
441 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
442 "vector 0x%x, but the register is already in use for "
443 "vector 0x%x on this cpu\n",
444 smp_processor_id(), reg, offset, new, old);
445 return -EBUSY;
446 }
447
448 apic_write(reg, new);
449
450 return 0;
451}
452EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
453
454/*
455 * Program the next event, relative to now
456 */
457static int lapic_next_event(unsigned long delta,
458 struct clock_event_device *evt)
459{
460 apic_write(APIC_TMICT, delta);
461 return 0;
462}
463
464static int lapic_next_deadline(unsigned long delta,
465 struct clock_event_device *evt)
466{
467 u64 tsc;
468
469 tsc = rdtsc();
470 wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
471 return 0;
472}
473
474static int lapic_timer_shutdown(struct clock_event_device *evt)
475{
476 unsigned int v;
477
478 /* Lapic used as dummy for broadcast ? */
479 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
480 return 0;
481
482 v = apic_read(APIC_LVTT);
483 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
484 apic_write(APIC_LVTT, v);
485 apic_write(APIC_TMICT, 0);
486 return 0;
487}
488
489static inline int
490lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
491{
492 /* Lapic used as dummy for broadcast ? */
493 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
494 return 0;
495
496 __setup_APIC_LVTT(lapic_timer_frequency, oneshot, 1);
497 return 0;
498}
499
500static int lapic_timer_set_periodic(struct clock_event_device *evt)
501{
502 return lapic_timer_set_periodic_oneshot(evt, false);
503}
504
505static int lapic_timer_set_oneshot(struct clock_event_device *evt)
506{
507 return lapic_timer_set_periodic_oneshot(evt, true);
508}
509
510/*
511 * Local APIC timer broadcast function
512 */
513static void lapic_timer_broadcast(const struct cpumask *mask)
514{
515#ifdef CONFIG_SMP
516 apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
517#endif
518}
519
520
521/*
522 * The local apic timer can be used for any function which is CPU local.
523 */
524static struct clock_event_device lapic_clockevent = {
525 .name = "lapic",
526 .features = CLOCK_EVT_FEAT_PERIODIC |
527 CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
528 | CLOCK_EVT_FEAT_DUMMY,
529 .shift = 32,
530 .set_state_shutdown = lapic_timer_shutdown,
531 .set_state_periodic = lapic_timer_set_periodic,
532 .set_state_oneshot = lapic_timer_set_oneshot,
533 .set_state_oneshot_stopped = lapic_timer_shutdown,
534 .set_next_event = lapic_next_event,
535 .broadcast = lapic_timer_broadcast,
536 .rating = 100,
537 .irq = -1,
538};
539static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
540
541#define DEADLINE_MODEL_MATCH_FUNC(model, func) \
542 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&func }
543
544#define DEADLINE_MODEL_MATCH_REV(model, rev) \
545 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)rev }
546
547static u32 hsx_deadline_rev(void)
548{
549 switch (boot_cpu_data.x86_stepping) {
550 case 0x02: return 0x3a; /* EP */
551 case 0x04: return 0x0f; /* EX */
552 }
553
554 return ~0U;
555}
556
557static u32 bdx_deadline_rev(void)
558{
559 switch (boot_cpu_data.x86_stepping) {
560 case 0x02: return 0x00000011;
561 case 0x03: return 0x0700000e;
562 case 0x04: return 0x0f00000c;
563 case 0x05: return 0x0e000003;
564 }
565
566 return ~0U;
567}
568
569static u32 skx_deadline_rev(void)
570{
571 switch (boot_cpu_data.x86_stepping) {
572 case 0x03: return 0x01000136;
573 case 0x04: return 0x02000014;
574 }
575
576 return ~0U;
577}
578
579static const struct x86_cpu_id deadline_match[] = {
580 DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_HASWELL_X, hsx_deadline_rev),
581 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_X, 0x0b000020),
582 DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_BROADWELL_XEON_D, bdx_deadline_rev),
583 DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_SKYLAKE_X, skx_deadline_rev),
584
585 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_CORE, 0x22),
586 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_ULT, 0x20),
587 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_GT3E, 0x17),
588
589 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_CORE, 0x25),
590 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_GT3E, 0x17),
591
592 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_MOBILE, 0xb2),
593 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_DESKTOP, 0xb2),
594
595 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_MOBILE, 0x52),
596 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_DESKTOP, 0x52),
597
598 {},
599};
600
601static void apic_check_deadline_errata(void)
602{
603 const struct x86_cpu_id *m;
604 u32 rev;
605
606 if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER) ||
607 boot_cpu_has(X86_FEATURE_HYPERVISOR))
608 return;
609
610 m = x86_match_cpu(deadline_match);
611 if (!m)
612 return;
613
614 /*
615 * Function pointers will have the MSB set due to address layout,
616 * immediate revisions will not.
617 */
618 if ((long)m->driver_data < 0)
619 rev = ((u32 (*)(void))(m->driver_data))();
620 else
621 rev = (u32)m->driver_data;
622
623 if (boot_cpu_data.microcode >= rev)
624 return;
625
626 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
627 pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
628 "please update microcode to version: 0x%x (or later)\n", rev);
629}
630
631/*
632 * Setup the local APIC timer for this CPU. Copy the initialized values
633 * of the boot CPU and register the clock event in the framework.
634 */
635static void setup_APIC_timer(void)
636{
637 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
638
639 if (this_cpu_has(X86_FEATURE_ARAT)) {
640 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
641 /* Make LAPIC timer preferrable over percpu HPET */
642 lapic_clockevent.rating = 150;
643 }
644
645 memcpy(levt, &lapic_clockevent, sizeof(*levt));
646 levt->cpumask = cpumask_of(smp_processor_id());
647
648 if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
649 levt->name = "lapic-deadline";
650 levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
651 CLOCK_EVT_FEAT_DUMMY);
652 levt->set_next_event = lapic_next_deadline;
653 clockevents_config_and_register(levt,
654 tsc_khz * (1000 / TSC_DIVISOR),
655 0xF, ~0UL);
656 } else
657 clockevents_register_device(levt);
658}
659
660/*
661 * Install the updated TSC frequency from recalibration at the TSC
662 * deadline clockevent devices.
663 */
664static void __lapic_update_tsc_freq(void *info)
665{
666 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
667
668 if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
669 return;
670
671 clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
672}
673
674void lapic_update_tsc_freq(void)
675{
676 /*
677 * The clockevent device's ->mult and ->shift can both be
678 * changed. In order to avoid races, schedule the frequency
679 * update code on each CPU.
680 */
681 on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
682}
683
684/*
685 * In this functions we calibrate APIC bus clocks to the external timer.
686 *
687 * We want to do the calibration only once since we want to have local timer
688 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
689 * frequency.
690 *
691 * This was previously done by reading the PIT/HPET and waiting for a wrap
692 * around to find out, that a tick has elapsed. I have a box, where the PIT
693 * readout is broken, so it never gets out of the wait loop again. This was
694 * also reported by others.
695 *
696 * Monitoring the jiffies value is inaccurate and the clockevents
697 * infrastructure allows us to do a simple substitution of the interrupt
698 * handler.
699 *
700 * The calibration routine also uses the pm_timer when possible, as the PIT
701 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
702 * back to normal later in the boot process).
703 */
704
705#define LAPIC_CAL_LOOPS (HZ/10)
706
707static __initdata int lapic_cal_loops = -1;
708static __initdata long lapic_cal_t1, lapic_cal_t2;
709static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
710static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
711static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
712
713/*
714 * Temporary interrupt handler.
715 */
716static void __init lapic_cal_handler(struct clock_event_device *dev)
717{
718 unsigned long long tsc = 0;
719 long tapic = apic_read(APIC_TMCCT);
720 unsigned long pm = acpi_pm_read_early();
721
722 if (boot_cpu_has(X86_FEATURE_TSC))
723 tsc = rdtsc();
724
725 switch (lapic_cal_loops++) {
726 case 0:
727 lapic_cal_t1 = tapic;
728 lapic_cal_tsc1 = tsc;
729 lapic_cal_pm1 = pm;
730 lapic_cal_j1 = jiffies;
731 break;
732
733 case LAPIC_CAL_LOOPS:
734 lapic_cal_t2 = tapic;
735 lapic_cal_tsc2 = tsc;
736 if (pm < lapic_cal_pm1)
737 pm += ACPI_PM_OVRRUN;
738 lapic_cal_pm2 = pm;
739 lapic_cal_j2 = jiffies;
740 break;
741 }
742}
743
744static int __init
745calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
746{
747 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
748 const long pm_thresh = pm_100ms / 100;
749 unsigned long mult;
750 u64 res;
751
752#ifndef CONFIG_X86_PM_TIMER
753 return -1;
754#endif
755
756 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
757
758 /* Check, if the PM timer is available */
759 if (!deltapm)
760 return -1;
761
762 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
763
764 if (deltapm > (pm_100ms - pm_thresh) &&
765 deltapm < (pm_100ms + pm_thresh)) {
766 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
767 return 0;
768 }
769
770 res = (((u64)deltapm) * mult) >> 22;
771 do_div(res, 1000000);
772 pr_warning("APIC calibration not consistent "
773 "with PM-Timer: %ldms instead of 100ms\n",(long)res);
774
775 /* Correct the lapic counter value */
776 res = (((u64)(*delta)) * pm_100ms);
777 do_div(res, deltapm);
778 pr_info("APIC delta adjusted to PM-Timer: "
779 "%lu (%ld)\n", (unsigned long)res, *delta);
780 *delta = (long)res;
781
782 /* Correct the tsc counter value */
783 if (boot_cpu_has(X86_FEATURE_TSC)) {
784 res = (((u64)(*deltatsc)) * pm_100ms);
785 do_div(res, deltapm);
786 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
787 "PM-Timer: %lu (%ld)\n",
788 (unsigned long)res, *deltatsc);
789 *deltatsc = (long)res;
790 }
791
792 return 0;
793}
794
795static int __init calibrate_APIC_clock(void)
796{
797 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
798 void (*real_handler)(struct clock_event_device *dev);
799 unsigned long deltaj;
800 long delta, deltatsc;
801 int pm_referenced = 0;
802
803 /**
804 * check if lapic timer has already been calibrated by platform
805 * specific routine, such as tsc calibration code. if so, we just fill
806 * in the clockevent structure and return.
807 */
808
809 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
810 return 0;
811 } else if (lapic_timer_frequency) {
812 apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
813 lapic_timer_frequency);
814 lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
815 TICK_NSEC, lapic_clockevent.shift);
816 lapic_clockevent.max_delta_ns =
817 clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
818 lapic_clockevent.max_delta_ticks = 0x7FFFFF;
819 lapic_clockevent.min_delta_ns =
820 clockevent_delta2ns(0xF, &lapic_clockevent);
821 lapic_clockevent.min_delta_ticks = 0xF;
822 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
823 return 0;
824 }
825
826 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
827 "calibrating APIC timer ...\n");
828
829 local_irq_disable();
830
831 /* Replace the global interrupt handler */
832 real_handler = global_clock_event->event_handler;
833 global_clock_event->event_handler = lapic_cal_handler;
834
835 /*
836 * Setup the APIC counter to maximum. There is no way the lapic
837 * can underflow in the 100ms detection time frame
838 */
839 __setup_APIC_LVTT(0xffffffff, 0, 0);
840
841 /* Let the interrupts run */
842 local_irq_enable();
843
844 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
845 cpu_relax();
846
847 local_irq_disable();
848
849 /* Restore the real event handler */
850 global_clock_event->event_handler = real_handler;
851
852 /* Build delta t1-t2 as apic timer counts down */
853 delta = lapic_cal_t1 - lapic_cal_t2;
854 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
855
856 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
857
858 /* we trust the PM based calibration if possible */
859 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
860 &delta, &deltatsc);
861
862 /* Calculate the scaled math multiplication factor */
863 lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
864 lapic_clockevent.shift);
865 lapic_clockevent.max_delta_ns =
866 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
867 lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
868 lapic_clockevent.min_delta_ns =
869 clockevent_delta2ns(0xF, &lapic_clockevent);
870 lapic_clockevent.min_delta_ticks = 0xF;
871
872 lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
873
874 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
875 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
876 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
877 lapic_timer_frequency);
878
879 if (boot_cpu_has(X86_FEATURE_TSC)) {
880 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
881 "%ld.%04ld MHz.\n",
882 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
883 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
884 }
885
886 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
887 "%u.%04u MHz.\n",
888 lapic_timer_frequency / (1000000 / HZ),
889 lapic_timer_frequency % (1000000 / HZ));
890
891 /*
892 * Do a sanity check on the APIC calibration result
893 */
894 if (lapic_timer_frequency < (1000000 / HZ)) {
895 local_irq_enable();
896 pr_warning("APIC frequency too slow, disabling apic timer\n");
897 return -1;
898 }
899
900 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
901
902 /*
903 * PM timer calibration failed or not turned on
904 * so lets try APIC timer based calibration
905 */
906 if (!pm_referenced) {
907 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
908
909 /*
910 * Setup the apic timer manually
911 */
912 levt->event_handler = lapic_cal_handler;
913 lapic_timer_set_periodic(levt);
914 lapic_cal_loops = -1;
915
916 /* Let the interrupts run */
917 local_irq_enable();
918
919 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
920 cpu_relax();
921
922 /* Stop the lapic timer */
923 local_irq_disable();
924 lapic_timer_shutdown(levt);
925
926 /* Jiffies delta */
927 deltaj = lapic_cal_j2 - lapic_cal_j1;
928 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
929
930 /* Check, if the jiffies result is consistent */
931 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
932 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
933 else
934 levt->features |= CLOCK_EVT_FEAT_DUMMY;
935 }
936 local_irq_enable();
937
938 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
939 pr_warning("APIC timer disabled due to verification failure\n");
940 return -1;
941 }
942
943 return 0;
944}
945
946/*
947 * Setup the boot APIC
948 *
949 * Calibrate and verify the result.
950 */
951void __init setup_boot_APIC_clock(void)
952{
953 /*
954 * The local apic timer can be disabled via the kernel
955 * commandline or from the CPU detection code. Register the lapic
956 * timer as a dummy clock event source on SMP systems, so the
957 * broadcast mechanism is used. On UP systems simply ignore it.
958 */
959 if (disable_apic_timer) {
960 pr_info("Disabling APIC timer\n");
961 /* No broadcast on UP ! */
962 if (num_possible_cpus() > 1) {
963 lapic_clockevent.mult = 1;
964 setup_APIC_timer();
965 }
966 return;
967 }
968
969 if (calibrate_APIC_clock()) {
970 /* No broadcast on UP ! */
971 if (num_possible_cpus() > 1)
972 setup_APIC_timer();
973 return;
974 }
975
976 /*
977 * If nmi_watchdog is set to IO_APIC, we need the
978 * PIT/HPET going. Otherwise register lapic as a dummy
979 * device.
980 */
981 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
982
983 /* Setup the lapic or request the broadcast */
984 setup_APIC_timer();
985 amd_e400_c1e_apic_setup();
986}
987
988void setup_secondary_APIC_clock(void)
989{
990 setup_APIC_timer();
991 amd_e400_c1e_apic_setup();
992}
993
994/*
995 * The guts of the apic timer interrupt
996 */
997static void local_apic_timer_interrupt(void)
998{
999 struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1000
1001 /*
1002 * Normally we should not be here till LAPIC has been initialized but
1003 * in some cases like kdump, its possible that there is a pending LAPIC
1004 * timer interrupt from previous kernel's context and is delivered in
1005 * new kernel the moment interrupts are enabled.
1006 *
1007 * Interrupts are enabled early and LAPIC is setup much later, hence
1008 * its possible that when we get here evt->event_handler is NULL.
1009 * Check for event_handler being NULL and discard the interrupt as
1010 * spurious.
1011 */
1012 if (!evt->event_handler) {
1013 pr_warning("Spurious LAPIC timer interrupt on cpu %d\n",
1014 smp_processor_id());
1015 /* Switch it off */
1016 lapic_timer_shutdown(evt);
1017 return;
1018 }
1019
1020 /*
1021 * the NMI deadlock-detector uses this.
1022 */
1023 inc_irq_stat(apic_timer_irqs);
1024
1025 evt->event_handler(evt);
1026}
1027
1028/*
1029 * Local APIC timer interrupt. This is the most natural way for doing
1030 * local interrupts, but local timer interrupts can be emulated by
1031 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1032 *
1033 * [ if a single-CPU system runs an SMP kernel then we call the local
1034 * interrupt as well. Thus we cannot inline the local irq ... ]
1035 */
1036__visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
1037{
1038 struct pt_regs *old_regs = set_irq_regs(regs);
1039
1040 /*
1041 * NOTE! We'd better ACK the irq immediately,
1042 * because timer handling can be slow.
1043 *
1044 * update_process_times() expects us to have done irq_enter().
1045 * Besides, if we don't timer interrupts ignore the global
1046 * interrupt lock, which is the WrongThing (tm) to do.
1047 */
1048 entering_ack_irq();
1049 trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1050 local_apic_timer_interrupt();
1051 trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1052 exiting_irq();
1053
1054 set_irq_regs(old_regs);
1055}
1056
1057int setup_profiling_timer(unsigned int multiplier)
1058{
1059 return -EINVAL;
1060}
1061
1062/*
1063 * Local APIC start and shutdown
1064 */
1065
1066/**
1067 * clear_local_APIC - shutdown the local APIC
1068 *
1069 * This is called, when a CPU is disabled and before rebooting, so the state of
1070 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1071 * leftovers during boot.
1072 */
1073void clear_local_APIC(void)
1074{
1075 int maxlvt;
1076 u32 v;
1077
1078 /* APIC hasn't been mapped yet */
1079 if (!x2apic_mode && !apic_phys)
1080 return;
1081
1082 maxlvt = lapic_get_maxlvt();
1083 /*
1084 * Masking an LVT entry can trigger a local APIC error
1085 * if the vector is zero. Mask LVTERR first to prevent this.
1086 */
1087 if (maxlvt >= 3) {
1088 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1089 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1090 }
1091 /*
1092 * Careful: we have to set masks only first to deassert
1093 * any level-triggered sources.
1094 */
1095 v = apic_read(APIC_LVTT);
1096 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1097 v = apic_read(APIC_LVT0);
1098 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1099 v = apic_read(APIC_LVT1);
1100 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1101 if (maxlvt >= 4) {
1102 v = apic_read(APIC_LVTPC);
1103 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1104 }
1105
1106 /* lets not touch this if we didn't frob it */
1107#ifdef CONFIG_X86_THERMAL_VECTOR
1108 if (maxlvt >= 5) {
1109 v = apic_read(APIC_LVTTHMR);
1110 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1111 }
1112#endif
1113#ifdef CONFIG_X86_MCE_INTEL
1114 if (maxlvt >= 6) {
1115 v = apic_read(APIC_LVTCMCI);
1116 if (!(v & APIC_LVT_MASKED))
1117 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1118 }
1119#endif
1120
1121 /*
1122 * Clean APIC state for other OSs:
1123 */
1124 apic_write(APIC_LVTT, APIC_LVT_MASKED);
1125 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1126 apic_write(APIC_LVT1, APIC_LVT_MASKED);
1127 if (maxlvt >= 3)
1128 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1129 if (maxlvt >= 4)
1130 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1131
1132 /* Integrated APIC (!82489DX) ? */
1133 if (lapic_is_integrated()) {
1134 if (maxlvt > 3)
1135 /* Clear ESR due to Pentium errata 3AP and 11AP */
1136 apic_write(APIC_ESR, 0);
1137 apic_read(APIC_ESR);
1138 }
1139}
1140
1141/**
1142 * disable_local_APIC - clear and disable the local APIC
1143 */
1144void disable_local_APIC(void)
1145{
1146 unsigned int value;
1147
1148 /* APIC hasn't been mapped yet */
1149 if (!x2apic_mode && !apic_phys)
1150 return;
1151
1152 clear_local_APIC();
1153
1154 /*
1155 * Disable APIC (implies clearing of registers
1156 * for 82489DX!).
1157 */
1158 value = apic_read(APIC_SPIV);
1159 value &= ~APIC_SPIV_APIC_ENABLED;
1160 apic_write(APIC_SPIV, value);
1161
1162#ifdef CONFIG_X86_32
1163 /*
1164 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1165 * restore the disabled state.
1166 */
1167 if (enabled_via_apicbase) {
1168 unsigned int l, h;
1169
1170 rdmsr(MSR_IA32_APICBASE, l, h);
1171 l &= ~MSR_IA32_APICBASE_ENABLE;
1172 wrmsr(MSR_IA32_APICBASE, l, h);
1173 }
1174#endif
1175}
1176
1177/*
1178 * If Linux enabled the LAPIC against the BIOS default disable it down before
1179 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1180 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1181 * for the case where Linux didn't enable the LAPIC.
1182 */
1183void lapic_shutdown(void)
1184{
1185 unsigned long flags;
1186
1187 if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1188 return;
1189
1190 local_irq_save(flags);
1191
1192#ifdef CONFIG_X86_32
1193 if (!enabled_via_apicbase)
1194 clear_local_APIC();
1195 else
1196#endif
1197 disable_local_APIC();
1198
1199
1200 local_irq_restore(flags);
1201}
1202
1203/**
1204 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1205 */
1206void __init sync_Arb_IDs(void)
1207{
1208 /*
1209 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1210 * needed on AMD.
1211 */
1212 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1213 return;
1214
1215 /*
1216 * Wait for idle.
1217 */
1218 apic_wait_icr_idle();
1219
1220 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1221 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1222 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1223}
1224
1225enum apic_intr_mode_id apic_intr_mode;
1226
1227static int __init apic_intr_mode_select(void)
1228{
1229 /* Check kernel option */
1230 if (disable_apic) {
1231 pr_info("APIC disabled via kernel command line\n");
1232 return APIC_PIC;
1233 }
1234
1235 /* Check BIOS */
1236#ifdef CONFIG_X86_64
1237 /* On 64-bit, the APIC must be integrated, Check local APIC only */
1238 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1239 disable_apic = 1;
1240 pr_info("APIC disabled by BIOS\n");
1241 return APIC_PIC;
1242 }
1243#else
1244 /* On 32-bit, the APIC may be integrated APIC or 82489DX */
1245
1246 /* Neither 82489DX nor integrated APIC ? */
1247 if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1248 disable_apic = 1;
1249 return APIC_PIC;
1250 }
1251
1252 /* If the BIOS pretends there is an integrated APIC ? */
1253 if (!boot_cpu_has(X86_FEATURE_APIC) &&
1254 APIC_INTEGRATED(boot_cpu_apic_version)) {
1255 disable_apic = 1;
1256 pr_err(FW_BUG "Local APIC %d not detected, force emulation\n",
1257 boot_cpu_physical_apicid);
1258 return APIC_PIC;
1259 }
1260#endif
1261
1262 /* Check MP table or ACPI MADT configuration */
1263 if (!smp_found_config) {
1264 disable_ioapic_support();
1265 if (!acpi_lapic) {
1266 pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1267 return APIC_VIRTUAL_WIRE_NO_CONFIG;
1268 }
1269 return APIC_VIRTUAL_WIRE;
1270 }
1271
1272#ifdef CONFIG_SMP
1273 /* If SMP should be disabled, then really disable it! */
1274 if (!setup_max_cpus) {
1275 pr_info("APIC: SMP mode deactivated\n");
1276 return APIC_SYMMETRIC_IO_NO_ROUTING;
1277 }
1278
1279 if (read_apic_id() != boot_cpu_physical_apicid) {
1280 panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1281 read_apic_id(), boot_cpu_physical_apicid);
1282 /* Or can we switch back to PIC here? */
1283 }
1284#endif
1285
1286 return APIC_SYMMETRIC_IO;
1287}
1288
1289/*
1290 * An initial setup of the virtual wire mode.
1291 */
1292void __init init_bsp_APIC(void)
1293{
1294 unsigned int value;
1295
1296 /*
1297 * Don't do the setup now if we have a SMP BIOS as the
1298 * through-I/O-APIC virtual wire mode might be active.
1299 */
1300 if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1301 return;
1302
1303 /*
1304 * Do not trust the local APIC being empty at bootup.
1305 */
1306 clear_local_APIC();
1307
1308 /*
1309 * Enable APIC.
1310 */
1311 value = apic_read(APIC_SPIV);
1312 value &= ~APIC_VECTOR_MASK;
1313 value |= APIC_SPIV_APIC_ENABLED;
1314
1315#ifdef CONFIG_X86_32
1316 /* This bit is reserved on P4/Xeon and should be cleared */
1317 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1318 (boot_cpu_data.x86 == 15))
1319 value &= ~APIC_SPIV_FOCUS_DISABLED;
1320 else
1321#endif
1322 value |= APIC_SPIV_FOCUS_DISABLED;
1323 value |= SPURIOUS_APIC_VECTOR;
1324 apic_write(APIC_SPIV, value);
1325
1326 /*
1327 * Set up the virtual wire mode.
1328 */
1329 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1330 value = APIC_DM_NMI;
1331 if (!lapic_is_integrated()) /* 82489DX */
1332 value |= APIC_LVT_LEVEL_TRIGGER;
1333 if (apic_extnmi == APIC_EXTNMI_NONE)
1334 value |= APIC_LVT_MASKED;
1335 apic_write(APIC_LVT1, value);
1336}
1337
1338/* Init the interrupt delivery mode for the BSP */
1339void __init apic_intr_mode_init(void)
1340{
1341 bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1342
1343 apic_intr_mode = apic_intr_mode_select();
1344
1345 switch (apic_intr_mode) {
1346 case APIC_PIC:
1347 pr_info("APIC: Keep in PIC mode(8259)\n");
1348 return;
1349 case APIC_VIRTUAL_WIRE:
1350 pr_info("APIC: Switch to virtual wire mode setup\n");
1351 default_setup_apic_routing();
1352 break;
1353 case APIC_VIRTUAL_WIRE_NO_CONFIG:
1354 pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1355 upmode = true;
1356 default_setup_apic_routing();
1357 break;
1358 case APIC_SYMMETRIC_IO:
1359 pr_info("APIC: Switch to symmetric I/O mode setup\n");
1360 default_setup_apic_routing();
1361 break;
1362 case APIC_SYMMETRIC_IO_NO_ROUTING:
1363 pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1364 break;
1365 }
1366
1367 apic_bsp_setup(upmode);
1368}
1369
1370static void lapic_setup_esr(void)
1371{
1372 unsigned int oldvalue, value, maxlvt;
1373
1374 if (!lapic_is_integrated()) {
1375 pr_info("No ESR for 82489DX.\n");
1376 return;
1377 }
1378
1379 if (apic->disable_esr) {
1380 /*
1381 * Something untraceable is creating bad interrupts on
1382 * secondary quads ... for the moment, just leave the
1383 * ESR disabled - we can't do anything useful with the
1384 * errors anyway - mbligh
1385 */
1386 pr_info("Leaving ESR disabled.\n");
1387 return;
1388 }
1389
1390 maxlvt = lapic_get_maxlvt();
1391 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1392 apic_write(APIC_ESR, 0);
1393 oldvalue = apic_read(APIC_ESR);
1394
1395 /* enables sending errors */
1396 value = ERROR_APIC_VECTOR;
1397 apic_write(APIC_LVTERR, value);
1398
1399 /*
1400 * spec says clear errors after enabling vector.
1401 */
1402 if (maxlvt > 3)
1403 apic_write(APIC_ESR, 0);
1404 value = apic_read(APIC_ESR);
1405 if (value != oldvalue)
1406 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1407 "vector: 0x%08x after: 0x%08x\n",
1408 oldvalue, value);
1409}
1410
1411static void apic_pending_intr_clear(void)
1412{
1413 long long max_loops = cpu_khz ? cpu_khz : 1000000;
1414 unsigned long long tsc = 0, ntsc;
1415 unsigned int queued;
1416 unsigned long value;
1417 int i, j, acked = 0;
1418
1419 if (boot_cpu_has(X86_FEATURE_TSC))
1420 tsc = rdtsc();
1421 /*
1422 * After a crash, we no longer service the interrupts and a pending
1423 * interrupt from previous kernel might still have ISR bit set.
1424 *
1425 * Most probably by now CPU has serviced that pending interrupt and
1426 * it might not have done the ack_APIC_irq() because it thought,
1427 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1428 * does not clear the ISR bit and cpu thinks it has already serivced
1429 * the interrupt. Hence a vector might get locked. It was noticed
1430 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1431 */
1432 do {
1433 queued = 0;
1434 for (i = APIC_ISR_NR - 1; i >= 0; i--)
1435 queued |= apic_read(APIC_IRR + i*0x10);
1436
1437 for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1438 value = apic_read(APIC_ISR + i*0x10);
1439 for_each_set_bit(j, &value, 32) {
1440 ack_APIC_irq();
1441 acked++;
1442 }
1443 }
1444 if (acked > 256) {
1445 pr_err("LAPIC pending interrupts after %d EOI\n", acked);
1446 break;
1447 }
1448 if (queued) {
1449 if (boot_cpu_has(X86_FEATURE_TSC) && cpu_khz) {
1450 ntsc = rdtsc();
1451 max_loops = (cpu_khz << 10) - (ntsc - tsc);
1452 } else {
1453 max_loops--;
1454 }
1455 }
1456 } while (queued && max_loops > 0);
1457 WARN_ON(max_loops <= 0);
1458}
1459
1460/**
1461 * setup_local_APIC - setup the local APIC
1462 *
1463 * Used to setup local APIC while initializing BSP or bringing up APs.
1464 * Always called with preemption disabled.
1465 */
1466static void setup_local_APIC(void)
1467{
1468 int cpu = smp_processor_id();
1469 unsigned int value;
1470#ifdef CONFIG_X86_32
1471 int logical_apicid, ldr_apicid;
1472#endif
1473
1474
1475 if (disable_apic) {
1476 disable_ioapic_support();
1477 return;
1478 }
1479
1480#ifdef CONFIG_X86_32
1481 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1482 if (lapic_is_integrated() && apic->disable_esr) {
1483 apic_write(APIC_ESR, 0);
1484 apic_write(APIC_ESR, 0);
1485 apic_write(APIC_ESR, 0);
1486 apic_write(APIC_ESR, 0);
1487 }
1488#endif
1489 perf_events_lapic_init();
1490
1491 /*
1492 * Double-check whether this APIC is really registered.
1493 * This is meaningless in clustered apic mode, so we skip it.
1494 */
1495 BUG_ON(!apic->apic_id_registered());
1496
1497 /*
1498 * Intel recommends to set DFR, LDR and TPR before enabling
1499 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1500 * document number 292116). So here it goes...
1501 */
1502 apic->init_apic_ldr();
1503
1504#ifdef CONFIG_X86_32
1505 /*
1506 * APIC LDR is initialized. If logical_apicid mapping was
1507 * initialized during get_smp_config(), make sure it matches the
1508 * actual value.
1509 */
1510 logical_apicid = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1511 ldr_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1512 WARN_ON(logical_apicid != BAD_APICID && logical_apicid != ldr_apicid);
1513 /* always use the value from LDR */
1514 early_per_cpu(x86_cpu_to_logical_apicid, cpu) = ldr_apicid;
1515#endif
1516
1517 /*
1518 * Set Task Priority to 'accept all'. We never change this
1519 * later on.
1520 */
1521 value = apic_read(APIC_TASKPRI);
1522 value &= ~APIC_TPRI_MASK;
1523 apic_write(APIC_TASKPRI, value);
1524
1525 apic_pending_intr_clear();
1526
1527 /*
1528 * Now that we are all set up, enable the APIC
1529 */
1530 value = apic_read(APIC_SPIV);
1531 value &= ~APIC_VECTOR_MASK;
1532 /*
1533 * Enable APIC
1534 */
1535 value |= APIC_SPIV_APIC_ENABLED;
1536
1537#ifdef CONFIG_X86_32
1538 /*
1539 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1540 * certain networking cards. If high frequency interrupts are
1541 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1542 * entry is masked/unmasked at a high rate as well then sooner or
1543 * later IOAPIC line gets 'stuck', no more interrupts are received
1544 * from the device. If focus CPU is disabled then the hang goes
1545 * away, oh well :-(
1546 *
1547 * [ This bug can be reproduced easily with a level-triggered
1548 * PCI Ne2000 networking cards and PII/PIII processors, dual
1549 * BX chipset. ]
1550 */
1551 /*
1552 * Actually disabling the focus CPU check just makes the hang less
1553 * frequent as it makes the interrupt distributon model be more
1554 * like LRU than MRU (the short-term load is more even across CPUs).
1555 */
1556
1557 /*
1558 * - enable focus processor (bit==0)
1559 * - 64bit mode always use processor focus
1560 * so no need to set it
1561 */
1562 value &= ~APIC_SPIV_FOCUS_DISABLED;
1563#endif
1564
1565 /*
1566 * Set spurious IRQ vector
1567 */
1568 value |= SPURIOUS_APIC_VECTOR;
1569 apic_write(APIC_SPIV, value);
1570
1571 /*
1572 * Set up LVT0, LVT1:
1573 *
1574 * set up through-local-APIC on the boot CPU's LINT0. This is not
1575 * strictly necessary in pure symmetric-IO mode, but sometimes
1576 * we delegate interrupts to the 8259A.
1577 */
1578 /*
1579 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1580 */
1581 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1582 if (!cpu && (pic_mode || !value || skip_ioapic_setup)) {
1583 value = APIC_DM_EXTINT;
1584 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1585 } else {
1586 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1587 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1588 }
1589 apic_write(APIC_LVT0, value);
1590
1591 /*
1592 * Only the BSP sees the LINT1 NMI signal by default. This can be
1593 * modified by apic_extnmi= boot option.
1594 */
1595 if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1596 apic_extnmi == APIC_EXTNMI_ALL)
1597 value = APIC_DM_NMI;
1598 else
1599 value = APIC_DM_NMI | APIC_LVT_MASKED;
1600
1601 /* Is 82489DX ? */
1602 if (!lapic_is_integrated())
1603 value |= APIC_LVT_LEVEL_TRIGGER;
1604 apic_write(APIC_LVT1, value);
1605
1606#ifdef CONFIG_X86_MCE_INTEL
1607 /* Recheck CMCI information after local APIC is up on CPU #0 */
1608 if (!cpu)
1609 cmci_recheck();
1610#endif
1611}
1612
1613static void end_local_APIC_setup(void)
1614{
1615 lapic_setup_esr();
1616
1617#ifdef CONFIG_X86_32
1618 {
1619 unsigned int value;
1620 /* Disable the local apic timer */
1621 value = apic_read(APIC_LVTT);
1622 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1623 apic_write(APIC_LVTT, value);
1624 }
1625#endif
1626
1627 apic_pm_activate();
1628}
1629
1630/*
1631 * APIC setup function for application processors. Called from smpboot.c
1632 */
1633void apic_ap_setup(void)
1634{
1635 setup_local_APIC();
1636 end_local_APIC_setup();
1637}
1638
1639#ifdef CONFIG_X86_X2APIC
1640int x2apic_mode;
1641
1642enum {
1643 X2APIC_OFF,
1644 X2APIC_ON,
1645 X2APIC_DISABLED,
1646};
1647static int x2apic_state;
1648
1649static void __x2apic_disable(void)
1650{
1651 u64 msr;
1652
1653 if (!boot_cpu_has(X86_FEATURE_APIC))
1654 return;
1655
1656 rdmsrl(MSR_IA32_APICBASE, msr);
1657 if (!(msr & X2APIC_ENABLE))
1658 return;
1659 /* Disable xapic and x2apic first and then reenable xapic mode */
1660 wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1661 wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1662 printk_once(KERN_INFO "x2apic disabled\n");
1663}
1664
1665static void __x2apic_enable(void)
1666{
1667 u64 msr;
1668
1669 rdmsrl(MSR_IA32_APICBASE, msr);
1670 if (msr & X2APIC_ENABLE)
1671 return;
1672 wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1673 printk_once(KERN_INFO "x2apic enabled\n");
1674}
1675
1676static int __init setup_nox2apic(char *str)
1677{
1678 if (x2apic_enabled()) {
1679 int apicid = native_apic_msr_read(APIC_ID);
1680
1681 if (apicid >= 255) {
1682 pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
1683 apicid);
1684 return 0;
1685 }
1686 pr_warning("x2apic already enabled.\n");
1687 __x2apic_disable();
1688 }
1689 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1690 x2apic_state = X2APIC_DISABLED;
1691 x2apic_mode = 0;
1692 return 0;
1693}
1694early_param("nox2apic", setup_nox2apic);
1695
1696/* Called from cpu_init() to enable x2apic on (secondary) cpus */
1697void x2apic_setup(void)
1698{
1699 /*
1700 * If x2apic is not in ON state, disable it if already enabled
1701 * from BIOS.
1702 */
1703 if (x2apic_state != X2APIC_ON) {
1704 __x2apic_disable();
1705 return;
1706 }
1707 __x2apic_enable();
1708}
1709
1710static __init void x2apic_disable(void)
1711{
1712 u32 x2apic_id, state = x2apic_state;
1713
1714 x2apic_mode = 0;
1715 x2apic_state = X2APIC_DISABLED;
1716
1717 if (state != X2APIC_ON)
1718 return;
1719
1720 x2apic_id = read_apic_id();
1721 if (x2apic_id >= 255)
1722 panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1723
1724 __x2apic_disable();
1725 register_lapic_address(mp_lapic_addr);
1726}
1727
1728static __init void x2apic_enable(void)
1729{
1730 if (x2apic_state != X2APIC_OFF)
1731 return;
1732
1733 x2apic_mode = 1;
1734 x2apic_state = X2APIC_ON;
1735 __x2apic_enable();
1736}
1737
1738static __init void try_to_enable_x2apic(int remap_mode)
1739{
1740 if (x2apic_state == X2APIC_DISABLED)
1741 return;
1742
1743 if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1744 /* IR is required if there is APIC ID > 255 even when running
1745 * under KVM
1746 */
1747 if (max_physical_apicid > 255 ||
1748 !x86_init.hyper.x2apic_available()) {
1749 pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1750 x2apic_disable();
1751 return;
1752 }
1753
1754 /*
1755 * without IR all CPUs can be addressed by IOAPIC/MSI
1756 * only in physical mode
1757 */
1758 x2apic_phys = 1;
1759 }
1760 x2apic_enable();
1761}
1762
1763void __init check_x2apic(void)
1764{
1765 if (x2apic_enabled()) {
1766 pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1767 x2apic_mode = 1;
1768 x2apic_state = X2APIC_ON;
1769 } else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1770 x2apic_state = X2APIC_DISABLED;
1771 }
1772}
1773#else /* CONFIG_X86_X2APIC */
1774static int __init validate_x2apic(void)
1775{
1776 if (!apic_is_x2apic_enabled())
1777 return 0;
1778 /*
1779 * Checkme: Can we simply turn off x2apic here instead of panic?
1780 */
1781 panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n");
1782}
1783early_initcall(validate_x2apic);
1784
1785static inline void try_to_enable_x2apic(int remap_mode) { }
1786static inline void __x2apic_enable(void) { }
1787#endif /* !CONFIG_X86_X2APIC */
1788
1789void __init enable_IR_x2apic(void)
1790{
1791 unsigned long flags;
1792 int ret, ir_stat;
1793
1794 if (skip_ioapic_setup) {
1795 pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1796 return;
1797 }
1798
1799 ir_stat = irq_remapping_prepare();
1800 if (ir_stat < 0 && !x2apic_supported())
1801 return;
1802
1803 ret = save_ioapic_entries();
1804 if (ret) {
1805 pr_info("Saving IO-APIC state failed: %d\n", ret);
1806 return;
1807 }
1808
1809 local_irq_save(flags);
1810 legacy_pic->mask_all();
1811 mask_ioapic_entries();
1812
1813 /* If irq_remapping_prepare() succeeded, try to enable it */
1814 if (ir_stat >= 0)
1815 ir_stat = irq_remapping_enable();
1816 /* ir_stat contains the remap mode or an error code */
1817 try_to_enable_x2apic(ir_stat);
1818
1819 if (ir_stat < 0)
1820 restore_ioapic_entries();
1821 legacy_pic->restore_mask();
1822 local_irq_restore(flags);
1823}
1824
1825#ifdef CONFIG_X86_64
1826/*
1827 * Detect and enable local APICs on non-SMP boards.
1828 * Original code written by Keir Fraser.
1829 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1830 * not correctly set up (usually the APIC timer won't work etc.)
1831 */
1832static int __init detect_init_APIC(void)
1833{
1834 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1835 pr_info("No local APIC present\n");
1836 return -1;
1837 }
1838
1839 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1840 return 0;
1841}
1842#else
1843
1844static int __init apic_verify(void)
1845{
1846 u32 features, h, l;
1847
1848 /*
1849 * The APIC feature bit should now be enabled
1850 * in `cpuid'
1851 */
1852 features = cpuid_edx(1);
1853 if (!(features & (1 << X86_FEATURE_APIC))) {
1854 pr_warning("Could not enable APIC!\n");
1855 return -1;
1856 }
1857 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1858 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1859
1860 /* The BIOS may have set up the APIC at some other address */
1861 if (boot_cpu_data.x86 >= 6) {
1862 rdmsr(MSR_IA32_APICBASE, l, h);
1863 if (l & MSR_IA32_APICBASE_ENABLE)
1864 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1865 }
1866
1867 pr_info("Found and enabled local APIC!\n");
1868 return 0;
1869}
1870
1871int __init apic_force_enable(unsigned long addr)
1872{
1873 u32 h, l;
1874
1875 if (disable_apic)
1876 return -1;
1877
1878 /*
1879 * Some BIOSes disable the local APIC in the APIC_BASE
1880 * MSR. This can only be done in software for Intel P6 or later
1881 * and AMD K7 (Model > 1) or later.
1882 */
1883 if (boot_cpu_data.x86 >= 6) {
1884 rdmsr(MSR_IA32_APICBASE, l, h);
1885 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1886 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1887 l &= ~MSR_IA32_APICBASE_BASE;
1888 l |= MSR_IA32_APICBASE_ENABLE | addr;
1889 wrmsr(MSR_IA32_APICBASE, l, h);
1890 enabled_via_apicbase = 1;
1891 }
1892 }
1893 return apic_verify();
1894}
1895
1896/*
1897 * Detect and initialize APIC
1898 */
1899static int __init detect_init_APIC(void)
1900{
1901 /* Disabled by kernel option? */
1902 if (disable_apic)
1903 return -1;
1904
1905 switch (boot_cpu_data.x86_vendor) {
1906 case X86_VENDOR_AMD:
1907 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1908 (boot_cpu_data.x86 >= 15))
1909 break;
1910 goto no_apic;
1911 case X86_VENDOR_INTEL:
1912 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1913 (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
1914 break;
1915 goto no_apic;
1916 default:
1917 goto no_apic;
1918 }
1919
1920 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1921 /*
1922 * Over-ride BIOS and try to enable the local APIC only if
1923 * "lapic" specified.
1924 */
1925 if (!force_enable_local_apic) {
1926 pr_info("Local APIC disabled by BIOS -- "
1927 "you can enable it with \"lapic\"\n");
1928 return -1;
1929 }
1930 if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1931 return -1;
1932 } else {
1933 if (apic_verify())
1934 return -1;
1935 }
1936
1937 apic_pm_activate();
1938
1939 return 0;
1940
1941no_apic:
1942 pr_info("No local APIC present or hardware disabled\n");
1943 return -1;
1944}
1945#endif
1946
1947/**
1948 * init_apic_mappings - initialize APIC mappings
1949 */
1950void __init init_apic_mappings(void)
1951{
1952 unsigned int new_apicid;
1953
1954 apic_check_deadline_errata();
1955
1956 if (x2apic_mode) {
1957 boot_cpu_physical_apicid = read_apic_id();
1958 return;
1959 }
1960
1961 /* If no local APIC can be found return early */
1962 if (!smp_found_config && detect_init_APIC()) {
1963 /* lets NOP'ify apic operations */
1964 pr_info("APIC: disable apic facility\n");
1965 apic_disable();
1966 } else {
1967 apic_phys = mp_lapic_addr;
1968
1969 /*
1970 * If the system has ACPI MADT tables or MP info, the LAPIC
1971 * address is already registered.
1972 */
1973 if (!acpi_lapic && !smp_found_config)
1974 register_lapic_address(apic_phys);
1975 }
1976
1977 /*
1978 * Fetch the APIC ID of the BSP in case we have a
1979 * default configuration (or the MP table is broken).
1980 */
1981 new_apicid = read_apic_id();
1982 if (boot_cpu_physical_apicid != new_apicid) {
1983 boot_cpu_physical_apicid = new_apicid;
1984 /*
1985 * yeah -- we lie about apic_version
1986 * in case if apic was disabled via boot option
1987 * but it's not a problem for SMP compiled kernel
1988 * since apic_intr_mode_select is prepared for such
1989 * a case and disable smp mode
1990 */
1991 boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
1992 }
1993}
1994
1995void __init register_lapic_address(unsigned long address)
1996{
1997 mp_lapic_addr = address;
1998
1999 if (!x2apic_mode) {
2000 set_fixmap_nocache(FIX_APIC_BASE, address);
2001 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
2002 APIC_BASE, address);
2003 }
2004 if (boot_cpu_physical_apicid == -1U) {
2005 boot_cpu_physical_apicid = read_apic_id();
2006 boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2007 }
2008}
2009
2010/*
2011 * Local APIC interrupts
2012 */
2013
2014/*
2015 * This interrupt should _never_ happen with our APIC/SMP architecture
2016 */
2017__visible void __irq_entry smp_spurious_interrupt(struct pt_regs *regs)
2018{
2019 u8 vector = ~regs->orig_ax;
2020 u32 v;
2021
2022 entering_irq();
2023 trace_spurious_apic_entry(vector);
2024
2025 /*
2026 * Check if this really is a spurious interrupt and ACK it
2027 * if it is a vectored one. Just in case...
2028 * Spurious interrupts should not be ACKed.
2029 */
2030 v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2031 if (v & (1 << (vector & 0x1f)))
2032 ack_APIC_irq();
2033
2034 inc_irq_stat(irq_spurious_count);
2035
2036 /* see sw-dev-man vol 3, chapter 7.4.13.5 */
2037 pr_info("spurious APIC interrupt through vector %02x on CPU#%d, "
2038 "should never happen.\n", vector, smp_processor_id());
2039
2040 trace_spurious_apic_exit(vector);
2041 exiting_irq();
2042}
2043
2044/*
2045 * This interrupt should never happen with our APIC/SMP architecture
2046 */
2047__visible void __irq_entry smp_error_interrupt(struct pt_regs *regs)
2048{
2049 static const char * const error_interrupt_reason[] = {
2050 "Send CS error", /* APIC Error Bit 0 */
2051 "Receive CS error", /* APIC Error Bit 1 */
2052 "Send accept error", /* APIC Error Bit 2 */
2053 "Receive accept error", /* APIC Error Bit 3 */
2054 "Redirectable IPI", /* APIC Error Bit 4 */
2055 "Send illegal vector", /* APIC Error Bit 5 */
2056 "Received illegal vector", /* APIC Error Bit 6 */
2057 "Illegal register address", /* APIC Error Bit 7 */
2058 };
2059 u32 v, i = 0;
2060
2061 entering_irq();
2062 trace_error_apic_entry(ERROR_APIC_VECTOR);
2063
2064 /* First tickle the hardware, only then report what went on. -- REW */
2065 if (lapic_get_maxlvt() > 3) /* Due to the Pentium erratum 3AP. */
2066 apic_write(APIC_ESR, 0);
2067 v = apic_read(APIC_ESR);
2068 ack_APIC_irq();
2069 atomic_inc(&irq_err_count);
2070
2071 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2072 smp_processor_id(), v);
2073
2074 v &= 0xff;
2075 while (v) {
2076 if (v & 0x1)
2077 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2078 i++;
2079 v >>= 1;
2080 }
2081
2082 apic_printk(APIC_DEBUG, KERN_CONT "\n");
2083
2084 trace_error_apic_exit(ERROR_APIC_VECTOR);
2085 exiting_irq();
2086}
2087
2088/**
2089 * connect_bsp_APIC - attach the APIC to the interrupt system
2090 */
2091static void __init connect_bsp_APIC(void)
2092{
2093#ifdef CONFIG_X86_32
2094 if (pic_mode) {
2095 /*
2096 * Do not trust the local APIC being empty at bootup.
2097 */
2098 clear_local_APIC();
2099 /*
2100 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
2101 * local APIC to INT and NMI lines.
2102 */
2103 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2104 "enabling APIC mode.\n");
2105 imcr_pic_to_apic();
2106 }
2107#endif
2108}
2109
2110/**
2111 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2112 * @virt_wire_setup: indicates, whether virtual wire mode is selected
2113 *
2114 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2115 * APIC is disabled.
2116 */
2117void disconnect_bsp_APIC(int virt_wire_setup)
2118{
2119 unsigned int value;
2120
2121#ifdef CONFIG_X86_32
2122 if (pic_mode) {
2123 /*
2124 * Put the board back into PIC mode (has an effect only on
2125 * certain older boards). Note that APIC interrupts, including
2126 * IPIs, won't work beyond this point! The only exception are
2127 * INIT IPIs.
2128 */
2129 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2130 "entering PIC mode.\n");
2131 imcr_apic_to_pic();
2132 return;
2133 }
2134#endif
2135
2136 /* Go back to Virtual Wire compatibility mode */
2137
2138 /* For the spurious interrupt use vector F, and enable it */
2139 value = apic_read(APIC_SPIV);
2140 value &= ~APIC_VECTOR_MASK;
2141 value |= APIC_SPIV_APIC_ENABLED;
2142 value |= 0xf;
2143 apic_write(APIC_SPIV, value);
2144
2145 if (!virt_wire_setup) {
2146 /*
2147 * For LVT0 make it edge triggered, active high,
2148 * external and enabled
2149 */
2150 value = apic_read(APIC_LVT0);
2151 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2152 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2153 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2154 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2155 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2156 apic_write(APIC_LVT0, value);
2157 } else {
2158 /* Disable LVT0 */
2159 apic_write(APIC_LVT0, APIC_LVT_MASKED);
2160 }
2161
2162 /*
2163 * For LVT1 make it edge triggered, active high,
2164 * nmi and enabled
2165 */
2166 value = apic_read(APIC_LVT1);
2167 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2168 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2169 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2170 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2171 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2172 apic_write(APIC_LVT1, value);
2173}
2174
2175/*
2176 * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2177 * contiguously, it equals to current allocated max logical CPU ID plus 1.
2178 * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2179 * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2180 *
2181 * NOTE: Reserve 0 for BSP.
2182 */
2183static int nr_logical_cpuids = 1;
2184
2185/*
2186 * Used to store mapping between logical CPU IDs and APIC IDs.
2187 */
2188static int cpuid_to_apicid[] = {
2189 [0 ... NR_CPUS - 1] = -1,
2190};
2191
2192/*
2193 * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2194 * and cpuid_to_apicid[] synchronized.
2195 */
2196static int allocate_logical_cpuid(int apicid)
2197{
2198 int i;
2199
2200 /*
2201 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2202 * check if the kernel has allocated a cpuid for it.
2203 */
2204 for (i = 0; i < nr_logical_cpuids; i++) {
2205 if (cpuid_to_apicid[i] == apicid)
2206 return i;
2207 }
2208
2209 /* Allocate a new cpuid. */
2210 if (nr_logical_cpuids >= nr_cpu_ids) {
2211 WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %u reached. "
2212 "Processor %d/0x%x and the rest are ignored.\n",
2213 nr_cpu_ids, nr_logical_cpuids, apicid);
2214 return -EINVAL;
2215 }
2216
2217 cpuid_to_apicid[nr_logical_cpuids] = apicid;
2218 return nr_logical_cpuids++;
2219}
2220
2221int generic_processor_info(int apicid, int version)
2222{
2223 int cpu, max = nr_cpu_ids;
2224 bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2225 phys_cpu_present_map);
2226
2227 /*
2228 * boot_cpu_physical_apicid is designed to have the apicid
2229 * returned by read_apic_id(), i.e, the apicid of the
2230 * currently booting-up processor. However, on some platforms,
2231 * it is temporarily modified by the apicid reported as BSP
2232 * through MP table. Concretely:
2233 *
2234 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2235 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2236 *
2237 * This function is executed with the modified
2238 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2239 * parameter doesn't work to disable APs on kdump 2nd kernel.
2240 *
2241 * Since fixing handling of boot_cpu_physical_apicid requires
2242 * another discussion and tests on each platform, we leave it
2243 * for now and here we use read_apic_id() directly in this
2244 * function, generic_processor_info().
2245 */
2246 if (disabled_cpu_apicid != BAD_APICID &&
2247 disabled_cpu_apicid != read_apic_id() &&
2248 disabled_cpu_apicid == apicid) {
2249 int thiscpu = num_processors + disabled_cpus;
2250
2251 pr_warning("APIC: Disabling requested cpu."
2252 " Processor %d/0x%x ignored.\n",
2253 thiscpu, apicid);
2254
2255 disabled_cpus++;
2256 return -ENODEV;
2257 }
2258
2259 /*
2260 * If boot cpu has not been detected yet, then only allow upto
2261 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2262 */
2263 if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2264 apicid != boot_cpu_physical_apicid) {
2265 int thiscpu = max + disabled_cpus - 1;
2266
2267 pr_warning(
2268 "APIC: NR_CPUS/possible_cpus limit of %i almost"
2269 " reached. Keeping one slot for boot cpu."
2270 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2271
2272 disabled_cpus++;
2273 return -ENODEV;
2274 }
2275
2276 if (num_processors >= nr_cpu_ids) {
2277 int thiscpu = max + disabled_cpus;
2278
2279 pr_warning("APIC: NR_CPUS/possible_cpus limit of %i "
2280 "reached. Processor %d/0x%x ignored.\n",
2281 max, thiscpu, apicid);
2282
2283 disabled_cpus++;
2284 return -EINVAL;
2285 }
2286
2287 if (apicid == boot_cpu_physical_apicid) {
2288 /*
2289 * x86_bios_cpu_apicid is required to have processors listed
2290 * in same order as logical cpu numbers. Hence the first
2291 * entry is BSP, and so on.
2292 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2293 * for BSP.
2294 */
2295 cpu = 0;
2296
2297 /* Logical cpuid 0 is reserved for BSP. */
2298 cpuid_to_apicid[0] = apicid;
2299 } else {
2300 cpu = allocate_logical_cpuid(apicid);
2301 if (cpu < 0) {
2302 disabled_cpus++;
2303 return -EINVAL;
2304 }
2305 }
2306
2307 /*
2308 * Validate version
2309 */
2310 if (version == 0x0) {
2311 pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2312 cpu, apicid);
2313 version = 0x10;
2314 }
2315
2316 if (version != boot_cpu_apic_version) {
2317 pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2318 boot_cpu_apic_version, cpu, version);
2319 }
2320
2321 if (apicid > max_physical_apicid)
2322 max_physical_apicid = apicid;
2323
2324#if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2325 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2326 early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2327#endif
2328#ifdef CONFIG_X86_32
2329 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2330 apic->x86_32_early_logical_apicid(cpu);
2331#endif
2332 set_cpu_possible(cpu, true);
2333 physid_set(apicid, phys_cpu_present_map);
2334 set_cpu_present(cpu, true);
2335 num_processors++;
2336
2337 return cpu;
2338}
2339
2340int hard_smp_processor_id(void)
2341{
2342 return read_apic_id();
2343}
2344
2345/*
2346 * Override the generic EOI implementation with an optimized version.
2347 * Only called during early boot when only one CPU is active and with
2348 * interrupts disabled, so we know this does not race with actual APIC driver
2349 * use.
2350 */
2351void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2352{
2353 struct apic **drv;
2354
2355 for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2356 /* Should happen once for each apic */
2357 WARN_ON((*drv)->eoi_write == eoi_write);
2358 (*drv)->native_eoi_write = (*drv)->eoi_write;
2359 (*drv)->eoi_write = eoi_write;
2360 }
2361}
2362
2363static void __init apic_bsp_up_setup(void)
2364{
2365#ifdef CONFIG_X86_64
2366 apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2367#else
2368 /*
2369 * Hack: In case of kdump, after a crash, kernel might be booting
2370 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2371 * might be zero if read from MP tables. Get it from LAPIC.
2372 */
2373# ifdef CONFIG_CRASH_DUMP
2374 boot_cpu_physical_apicid = read_apic_id();
2375# endif
2376#endif
2377 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2378}
2379
2380/**
2381 * apic_bsp_setup - Setup function for local apic and io-apic
2382 * @upmode: Force UP mode (for APIC_init_uniprocessor)
2383 *
2384 * Returns:
2385 * apic_id of BSP APIC
2386 */
2387void __init apic_bsp_setup(bool upmode)
2388{
2389 connect_bsp_APIC();
2390 if (upmode)
2391 apic_bsp_up_setup();
2392 setup_local_APIC();
2393
2394 enable_IO_APIC();
2395 end_local_APIC_setup();
2396 irq_remap_enable_fault_handling();
2397 setup_IO_APIC();
2398}
2399
2400#ifdef CONFIG_UP_LATE_INIT
2401void __init up_late_init(void)
2402{
2403 if (apic_intr_mode == APIC_PIC)
2404 return;
2405
2406 /* Setup local timer */
2407 x86_init.timers.setup_percpu_clockev();
2408}
2409#endif
2410
2411/*
2412 * Power management
2413 */
2414#ifdef CONFIG_PM
2415
2416static struct {
2417 /*
2418 * 'active' is true if the local APIC was enabled by us and
2419 * not the BIOS; this signifies that we are also responsible
2420 * for disabling it before entering apm/acpi suspend
2421 */
2422 int active;
2423 /* r/w apic fields */
2424 unsigned int apic_id;
2425 unsigned int apic_taskpri;
2426 unsigned int apic_ldr;
2427 unsigned int apic_dfr;
2428 unsigned int apic_spiv;
2429 unsigned int apic_lvtt;
2430 unsigned int apic_lvtpc;
2431 unsigned int apic_lvt0;
2432 unsigned int apic_lvt1;
2433 unsigned int apic_lvterr;
2434 unsigned int apic_tmict;
2435 unsigned int apic_tdcr;
2436 unsigned int apic_thmr;
2437 unsigned int apic_cmci;
2438} apic_pm_state;
2439
2440static int lapic_suspend(void)
2441{
2442 unsigned long flags;
2443 int maxlvt;
2444
2445 if (!apic_pm_state.active)
2446 return 0;
2447
2448 maxlvt = lapic_get_maxlvt();
2449
2450 apic_pm_state.apic_id = apic_read(APIC_ID);
2451 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2452 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2453 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2454 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2455 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2456 if (maxlvt >= 4)
2457 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2458 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2459 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2460 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2461 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2462 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2463#ifdef CONFIG_X86_THERMAL_VECTOR
2464 if (maxlvt >= 5)
2465 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2466#endif
2467#ifdef CONFIG_X86_MCE_INTEL
2468 if (maxlvt >= 6)
2469 apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2470#endif
2471
2472 local_irq_save(flags);
2473 disable_local_APIC();
2474
2475 irq_remapping_disable();
2476
2477 local_irq_restore(flags);
2478 return 0;
2479}
2480
2481static void lapic_resume(void)
2482{
2483 unsigned int l, h;
2484 unsigned long flags;
2485 int maxlvt;
2486
2487 if (!apic_pm_state.active)
2488 return;
2489
2490 local_irq_save(flags);
2491
2492 /*
2493 * IO-APIC and PIC have their own resume routines.
2494 * We just mask them here to make sure the interrupt
2495 * subsystem is completely quiet while we enable x2apic
2496 * and interrupt-remapping.
2497 */
2498 mask_ioapic_entries();
2499 legacy_pic->mask_all();
2500
2501 if (x2apic_mode) {
2502 __x2apic_enable();
2503 } else {
2504 /*
2505 * Make sure the APICBASE points to the right address
2506 *
2507 * FIXME! This will be wrong if we ever support suspend on
2508 * SMP! We'll need to do this as part of the CPU restore!
2509 */
2510 if (boot_cpu_data.x86 >= 6) {
2511 rdmsr(MSR_IA32_APICBASE, l, h);
2512 l &= ~MSR_IA32_APICBASE_BASE;
2513 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2514 wrmsr(MSR_IA32_APICBASE, l, h);
2515 }
2516 }
2517
2518 maxlvt = lapic_get_maxlvt();
2519 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2520 apic_write(APIC_ID, apic_pm_state.apic_id);
2521 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2522 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2523 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2524 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2525 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2526 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2527#ifdef CONFIG_X86_THERMAL_VECTOR
2528 if (maxlvt >= 5)
2529 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2530#endif
2531#ifdef CONFIG_X86_MCE_INTEL
2532 if (maxlvt >= 6)
2533 apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2534#endif
2535 if (maxlvt >= 4)
2536 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2537 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2538 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2539 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2540 apic_write(APIC_ESR, 0);
2541 apic_read(APIC_ESR);
2542 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2543 apic_write(APIC_ESR, 0);
2544 apic_read(APIC_ESR);
2545
2546 irq_remapping_reenable(x2apic_mode);
2547
2548 local_irq_restore(flags);
2549}
2550
2551/*
2552 * This device has no shutdown method - fully functioning local APICs
2553 * are needed on every CPU up until machine_halt/restart/poweroff.
2554 */
2555
2556static struct syscore_ops lapic_syscore_ops = {
2557 .resume = lapic_resume,
2558 .suspend = lapic_suspend,
2559};
2560
2561static void apic_pm_activate(void)
2562{
2563 apic_pm_state.active = 1;
2564}
2565
2566static int __init init_lapic_sysfs(void)
2567{
2568 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2569 if (boot_cpu_has(X86_FEATURE_APIC))
2570 register_syscore_ops(&lapic_syscore_ops);
2571
2572 return 0;
2573}
2574
2575/* local apic needs to resume before other devices access its registers. */
2576core_initcall(init_lapic_sysfs);
2577
2578#else /* CONFIG_PM */
2579
2580static void apic_pm_activate(void) { }
2581
2582#endif /* CONFIG_PM */
2583
2584#ifdef CONFIG_X86_64
2585
2586static int multi_checked;
2587static int multi;
2588
2589static int set_multi(const struct dmi_system_id *d)
2590{
2591 if (multi)
2592 return 0;
2593 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2594 multi = 1;
2595 return 0;
2596}
2597
2598static const struct dmi_system_id multi_dmi_table[] = {
2599 {
2600 .callback = set_multi,
2601 .ident = "IBM System Summit2",
2602 .matches = {
2603 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2604 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2605 },
2606 },
2607 {}
2608};
2609
2610static void dmi_check_multi(void)
2611{
2612 if (multi_checked)
2613 return;
2614
2615 dmi_check_system(multi_dmi_table);
2616 multi_checked = 1;
2617}
2618
2619/*
2620 * apic_is_clustered_box() -- Check if we can expect good TSC
2621 *
2622 * Thus far, the major user of this is IBM's Summit2 series:
2623 * Clustered boxes may have unsynced TSC problems if they are
2624 * multi-chassis.
2625 * Use DMI to check them
2626 */
2627int apic_is_clustered_box(void)
2628{
2629 dmi_check_multi();
2630 return multi;
2631}
2632#endif
2633
2634/*
2635 * APIC command line parameters
2636 */
2637static int __init setup_disableapic(char *arg)
2638{
2639 disable_apic = 1;
2640 setup_clear_cpu_cap(X86_FEATURE_APIC);
2641 return 0;
2642}
2643early_param("disableapic", setup_disableapic);
2644
2645/* same as disableapic, for compatibility */
2646static int __init setup_nolapic(char *arg)
2647{
2648 return setup_disableapic(arg);
2649}
2650early_param("nolapic", setup_nolapic);
2651
2652static int __init parse_lapic_timer_c2_ok(char *arg)
2653{
2654 local_apic_timer_c2_ok = 1;
2655 return 0;
2656}
2657early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2658
2659static int __init parse_disable_apic_timer(char *arg)
2660{
2661 disable_apic_timer = 1;
2662 return 0;
2663}
2664early_param("noapictimer", parse_disable_apic_timer);
2665
2666static int __init parse_nolapic_timer(char *arg)
2667{
2668 disable_apic_timer = 1;
2669 return 0;
2670}
2671early_param("nolapic_timer", parse_nolapic_timer);
2672
2673static int __init apic_set_verbosity(char *arg)
2674{
2675 if (!arg) {
2676#ifdef CONFIG_X86_64
2677 skip_ioapic_setup = 0;
2678 return 0;
2679#endif
2680 return -EINVAL;
2681 }
2682
2683 if (strcmp("debug", arg) == 0)
2684 apic_verbosity = APIC_DEBUG;
2685 else if (strcmp("verbose", arg) == 0)
2686 apic_verbosity = APIC_VERBOSE;
2687#ifdef CONFIG_X86_64
2688 else {
2689 pr_warning("APIC Verbosity level %s not recognised"
2690 " use apic=verbose or apic=debug\n", arg);
2691 return -EINVAL;
2692 }
2693#endif
2694
2695 return 0;
2696}
2697early_param("apic", apic_set_verbosity);
2698
2699static int __init lapic_insert_resource(void)
2700{
2701 if (!apic_phys)
2702 return -1;
2703
2704 /* Put local APIC into the resource map. */
2705 lapic_resource.start = apic_phys;
2706 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2707 insert_resource(&iomem_resource, &lapic_resource);
2708
2709 return 0;
2710}
2711
2712/*
2713 * need call insert after e820__reserve_resources()
2714 * that is using request_resource
2715 */
2716late_initcall(lapic_insert_resource);
2717
2718static int __init apic_set_disabled_cpu_apicid(char *arg)
2719{
2720 if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2721 return -EINVAL;
2722
2723 return 0;
2724}
2725early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2726
2727static int __init apic_set_extnmi(char *arg)
2728{
2729 if (!arg)
2730 return -EINVAL;
2731
2732 if (!strncmp("all", arg, 3))
2733 apic_extnmi = APIC_EXTNMI_ALL;
2734 else if (!strncmp("none", arg, 4))
2735 apic_extnmi = APIC_EXTNMI_NONE;
2736 else if (!strncmp("bsp", arg, 3))
2737 apic_extnmi = APIC_EXTNMI_BSP;
2738 else {
2739 pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2740 return -EINVAL;
2741 }
2742
2743 return 0;
2744}
2745early_param("apic_extnmi", apic_set_extnmi);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Local APIC handling, local APIC timers
4 *
5 * (c) 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
6 *
7 * Fixes
8 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
9 * thanks to Eric Gilmore
10 * and Rolf G. Tews
11 * for testing these extensively.
12 * Maciej W. Rozycki : Various updates and fixes.
13 * Mikael Pettersson : Power Management for UP-APIC.
14 * Pavel Machek and
15 * Mikael Pettersson : PM converted to driver model.
16 */
17
18#include <linux/perf_event.h>
19#include <linux/kernel_stat.h>
20#include <linux/mc146818rtc.h>
21#include <linux/acpi_pmtmr.h>
22#include <linux/clockchips.h>
23#include <linux/interrupt.h>
24#include <linux/memblock.h>
25#include <linux/ftrace.h>
26#include <linux/ioport.h>
27#include <linux/export.h>
28#include <linux/syscore_ops.h>
29#include <linux/delay.h>
30#include <linux/timex.h>
31#include <linux/i8253.h>
32#include <linux/dmar.h>
33#include <linux/init.h>
34#include <linux/cpu.h>
35#include <linux/dmi.h>
36#include <linux/smp.h>
37#include <linux/mm.h>
38
39#include <xen/xen.h>
40
41#include <asm/trace/irq_vectors.h>
42#include <asm/irq_remapping.h>
43#include <asm/pc-conf-reg.h>
44#include <asm/perf_event.h>
45#include <asm/x86_init.h>
46#include <linux/atomic.h>
47#include <asm/barrier.h>
48#include <asm/mpspec.h>
49#include <asm/i8259.h>
50#include <asm/proto.h>
51#include <asm/traps.h>
52#include <asm/apic.h>
53#include <asm/acpi.h>
54#include <asm/io_apic.h>
55#include <asm/desc.h>
56#include <asm/hpet.h>
57#include <asm/mtrr.h>
58#include <asm/time.h>
59#include <asm/smp.h>
60#include <asm/mce.h>
61#include <asm/tsc.h>
62#include <asm/hypervisor.h>
63#include <asm/cpu_device_id.h>
64#include <asm/intel-family.h>
65#include <asm/irq_regs.h>
66#include <asm/cpu.h>
67
68#include "local.h"
69
70unsigned int num_processors;
71
72unsigned disabled_cpus;
73
74/* Processor that is doing the boot up */
75u32 boot_cpu_physical_apicid __ro_after_init = BAD_APICID;
76EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
77
78u8 boot_cpu_apic_version __ro_after_init;
79
80/*
81 * Bitmask of physically existing CPUs:
82 */
83physid_mask_t phys_cpu_present_map;
84
85/*
86 * Processor to be disabled specified by kernel parameter
87 * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
88 * avoid undefined behaviour caused by sending INIT from AP to BSP.
89 */
90static u32 disabled_cpu_apicid __ro_after_init = BAD_APICID;
91
92/*
93 * This variable controls which CPUs receive external NMIs. By default,
94 * external NMIs are delivered only to the BSP.
95 */
96static int apic_extnmi __ro_after_init = APIC_EXTNMI_BSP;
97
98/*
99 * Hypervisor supports 15 bits of APIC ID in MSI Extended Destination ID
100 */
101static bool virt_ext_dest_id __ro_after_init;
102
103/* For parallel bootup. */
104unsigned long apic_mmio_base __ro_after_init;
105
106static inline bool apic_accessible(void)
107{
108 return x2apic_mode || apic_mmio_base;
109}
110
111/*
112 * Map cpu index to physical APIC ID
113 */
114DEFINE_EARLY_PER_CPU_READ_MOSTLY(u32, x86_cpu_to_apicid, BAD_APICID);
115DEFINE_EARLY_PER_CPU_READ_MOSTLY(u32, x86_cpu_to_acpiid, U32_MAX);
116EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
117EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_acpiid);
118
119#ifdef CONFIG_X86_32
120/* Local APIC was disabled by the BIOS and enabled by the kernel */
121static int enabled_via_apicbase __ro_after_init;
122
123/*
124 * Handle interrupt mode configuration register (IMCR).
125 * This register controls whether the interrupt signals
126 * that reach the BSP come from the master PIC or from the
127 * local APIC. Before entering Symmetric I/O Mode, either
128 * the BIOS or the operating system must switch out of
129 * PIC Mode by changing the IMCR.
130 */
131static inline void imcr_pic_to_apic(void)
132{
133 /* NMI and 8259 INTR go through APIC */
134 pc_conf_set(PC_CONF_MPS_IMCR, 0x01);
135}
136
137static inline void imcr_apic_to_pic(void)
138{
139 /* NMI and 8259 INTR go directly to BSP */
140 pc_conf_set(PC_CONF_MPS_IMCR, 0x00);
141}
142#endif
143
144/*
145 * Knob to control our willingness to enable the local APIC.
146 *
147 * +1=force-enable
148 */
149static int force_enable_local_apic __initdata;
150
151/*
152 * APIC command line parameters
153 */
154static int __init parse_lapic(char *arg)
155{
156 if (IS_ENABLED(CONFIG_X86_32) && !arg)
157 force_enable_local_apic = 1;
158 else if (arg && !strncmp(arg, "notscdeadline", 13))
159 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
160 return 0;
161}
162early_param("lapic", parse_lapic);
163
164#ifdef CONFIG_X86_64
165static int apic_calibrate_pmtmr __initdata;
166static __init int setup_apicpmtimer(char *s)
167{
168 apic_calibrate_pmtmr = 1;
169 notsc_setup(NULL);
170 return 1;
171}
172__setup("apicpmtimer", setup_apicpmtimer);
173#endif
174
175static unsigned long mp_lapic_addr __ro_after_init;
176bool apic_is_disabled __ro_after_init;
177/* Disable local APIC timer from the kernel commandline or via dmi quirk */
178static int disable_apic_timer __initdata;
179/* Local APIC timer works in C2 */
180int local_apic_timer_c2_ok __ro_after_init;
181EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
182
183/*
184 * Debug level, exported for io_apic.c
185 */
186int apic_verbosity __ro_after_init;
187
188int pic_mode __ro_after_init;
189
190/* Have we found an MP table */
191int smp_found_config __ro_after_init;
192
193static struct resource lapic_resource = {
194 .name = "Local APIC",
195 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
196};
197
198unsigned int lapic_timer_period = 0;
199
200static void apic_pm_activate(void);
201
202/*
203 * Get the LAPIC version
204 */
205static inline int lapic_get_version(void)
206{
207 return GET_APIC_VERSION(apic_read(APIC_LVR));
208}
209
210/*
211 * Check, if the APIC is integrated or a separate chip
212 */
213static inline int lapic_is_integrated(void)
214{
215 return APIC_INTEGRATED(lapic_get_version());
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
228 /* Hygon systems use modern APIC */
229 if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
230 return 1;
231
232 return lapic_get_version() >= 0x14;
233}
234
235/*
236 * right after this call apic become NOOP driven
237 * so apic->write/read doesn't do anything
238 */
239static void __init apic_disable(void)
240{
241 apic_install_driver(&apic_noop);
242}
243
244void native_apic_icr_write(u32 low, u32 id)
245{
246 unsigned long flags;
247
248 local_irq_save(flags);
249 apic_write(APIC_ICR2, SET_XAPIC_DEST_FIELD(id));
250 apic_write(APIC_ICR, low);
251 local_irq_restore(flags);
252}
253
254u64 native_apic_icr_read(void)
255{
256 u32 icr1, icr2;
257
258 icr2 = apic_read(APIC_ICR2);
259 icr1 = apic_read(APIC_ICR);
260
261 return icr1 | ((u64)icr2 << 32);
262}
263
264#ifdef CONFIG_X86_32
265/**
266 * get_physical_broadcast - Get number of physical broadcast IDs
267 */
268int get_physical_broadcast(void)
269{
270 return modern_apic() ? 0xff : 0xf;
271}
272#endif
273
274/**
275 * lapic_get_maxlvt - get the maximum number of local vector table entries
276 */
277int lapic_get_maxlvt(void)
278{
279 /*
280 * - we always have APIC integrated on 64bit mode
281 * - 82489DXs do not report # of LVT entries
282 */
283 return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
284}
285
286/*
287 * Local APIC timer
288 */
289
290/* Clock divisor */
291#define APIC_DIVISOR 16
292#define TSC_DIVISOR 8
293
294/* i82489DX specific */
295#define I82489DX_BASE_DIVIDER (((0x2) << 18))
296
297/*
298 * This function sets up the local APIC timer, with a timeout of
299 * 'clocks' APIC bus clock. During calibration we actually call
300 * this function twice on the boot CPU, once with a bogus timeout
301 * value, second time for real. The other (noncalibrating) CPUs
302 * call this function only once, with the real, calibrated value.
303 *
304 * We do reads before writes even if unnecessary, to get around the
305 * P5 APIC double write bug.
306 */
307static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
308{
309 unsigned int lvtt_value, tmp_value;
310
311 lvtt_value = LOCAL_TIMER_VECTOR;
312 if (!oneshot)
313 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
314 else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
315 lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
316
317 /*
318 * The i82489DX APIC uses bit 18 and 19 for the base divider. This
319 * overlaps with bit 18 on integrated APICs, but is not documented
320 * in the SDM. No problem though. i82489DX equipped systems do not
321 * have TSC deadline timer.
322 */
323 if (!lapic_is_integrated())
324 lvtt_value |= I82489DX_BASE_DIVIDER;
325
326 if (!irqen)
327 lvtt_value |= APIC_LVT_MASKED;
328
329 apic_write(APIC_LVTT, lvtt_value);
330
331 if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
332 /*
333 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
334 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
335 * According to Intel, MFENCE can do the serialization here.
336 */
337 asm volatile("mfence" : : : "memory");
338 return;
339 }
340
341 /*
342 * Divide PICLK by 16
343 */
344 tmp_value = apic_read(APIC_TDCR);
345 apic_write(APIC_TDCR,
346 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
347 APIC_TDR_DIV_16);
348
349 if (!oneshot)
350 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
351}
352
353/*
354 * Setup extended LVT, AMD specific
355 *
356 * Software should use the LVT offsets the BIOS provides. The offsets
357 * are determined by the subsystems using it like those for MCE
358 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts
359 * are supported. Beginning with family 10h at least 4 offsets are
360 * available.
361 *
362 * Since the offsets must be consistent for all cores, we keep track
363 * of the LVT offsets in software and reserve the offset for the same
364 * vector also to be used on other cores. An offset is freed by
365 * setting the entry to APIC_EILVT_MASKED.
366 *
367 * If the BIOS is right, there should be no conflicts. Otherwise a
368 * "[Firmware Bug]: ..." error message is generated. However, if
369 * software does not properly determines the offsets, it is not
370 * necessarily a BIOS bug.
371 */
372
373static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
374
375static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
376{
377 return (old & APIC_EILVT_MASKED)
378 || (new == APIC_EILVT_MASKED)
379 || ((new & ~APIC_EILVT_MASKED) == old);
380}
381
382static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
383{
384 unsigned int rsvd, vector;
385
386 if (offset >= APIC_EILVT_NR_MAX)
387 return ~0;
388
389 rsvd = atomic_read(&eilvt_offsets[offset]);
390 do {
391 vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */
392 if (vector && !eilvt_entry_is_changeable(vector, new))
393 /* may not change if vectors are different */
394 return rsvd;
395 } while (!atomic_try_cmpxchg(&eilvt_offsets[offset], &rsvd, new));
396
397 rsvd = new & ~APIC_EILVT_MASKED;
398 if (rsvd && rsvd != vector)
399 pr_info("LVT offset %d assigned for vector 0x%02x\n",
400 offset, rsvd);
401
402 return new;
403}
404
405/*
406 * If mask=1, the LVT entry does not generate interrupts while mask=0
407 * enables the vector. See also the BKDGs. Must be called with
408 * preemption disabled.
409 */
410
411int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
412{
413 unsigned long reg = APIC_EILVTn(offset);
414 unsigned int new, old, reserved;
415
416 new = (mask << 16) | (msg_type << 8) | vector;
417 old = apic_read(reg);
418 reserved = reserve_eilvt_offset(offset, new);
419
420 if (reserved != new) {
421 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
422 "vector 0x%x, but the register is already in use for "
423 "vector 0x%x on another cpu\n",
424 smp_processor_id(), reg, offset, new, reserved);
425 return -EINVAL;
426 }
427
428 if (!eilvt_entry_is_changeable(old, new)) {
429 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
430 "vector 0x%x, but the register is already in use for "
431 "vector 0x%x on this cpu\n",
432 smp_processor_id(), reg, offset, new, old);
433 return -EBUSY;
434 }
435
436 apic_write(reg, new);
437
438 return 0;
439}
440EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
441
442/*
443 * Program the next event, relative to now
444 */
445static int lapic_next_event(unsigned long delta,
446 struct clock_event_device *evt)
447{
448 apic_write(APIC_TMICT, delta);
449 return 0;
450}
451
452static int lapic_next_deadline(unsigned long delta,
453 struct clock_event_device *evt)
454{
455 u64 tsc;
456
457 /* This MSR is special and need a special fence: */
458 weak_wrmsr_fence();
459
460 tsc = rdtsc();
461 wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
462 return 0;
463}
464
465static int lapic_timer_shutdown(struct clock_event_device *evt)
466{
467 unsigned int v;
468
469 /* Lapic used as dummy for broadcast ? */
470 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
471 return 0;
472
473 v = apic_read(APIC_LVTT);
474 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
475 apic_write(APIC_LVTT, v);
476 apic_write(APIC_TMICT, 0);
477 return 0;
478}
479
480static inline int
481lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
482{
483 /* Lapic used as dummy for broadcast ? */
484 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
485 return 0;
486
487 __setup_APIC_LVTT(lapic_timer_period, oneshot, 1);
488 return 0;
489}
490
491static int lapic_timer_set_periodic(struct clock_event_device *evt)
492{
493 return lapic_timer_set_periodic_oneshot(evt, false);
494}
495
496static int lapic_timer_set_oneshot(struct clock_event_device *evt)
497{
498 return lapic_timer_set_periodic_oneshot(evt, true);
499}
500
501/*
502 * Local APIC timer broadcast function
503 */
504static void lapic_timer_broadcast(const struct cpumask *mask)
505{
506#ifdef CONFIG_SMP
507 __apic_send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
508#endif
509}
510
511
512/*
513 * The local apic timer can be used for any function which is CPU local.
514 */
515static struct clock_event_device lapic_clockevent = {
516 .name = "lapic",
517 .features = CLOCK_EVT_FEAT_PERIODIC |
518 CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
519 | CLOCK_EVT_FEAT_DUMMY,
520 .shift = 32,
521 .set_state_shutdown = lapic_timer_shutdown,
522 .set_state_periodic = lapic_timer_set_periodic,
523 .set_state_oneshot = lapic_timer_set_oneshot,
524 .set_state_oneshot_stopped = lapic_timer_shutdown,
525 .set_next_event = lapic_next_event,
526 .broadcast = lapic_timer_broadcast,
527 .rating = 100,
528 .irq = -1,
529};
530static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
531
532static const struct x86_cpu_id deadline_match[] __initconst = {
533 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x2, 0x2), 0x3a), /* EP */
534 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x4, 0x4), 0x0f), /* EX */
535
536 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_X, 0x0b000020),
537
538 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x2, 0x2), 0x00000011),
539 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x3, 0x3), 0x0700000e),
540 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x4, 0x4), 0x0f00000c),
541 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x5, 0x5), 0x0e000003),
542
543 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x3, 0x3), 0x01000136),
544 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x4, 0x4), 0x02000014),
545 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x5, 0xf), 0),
546
547 X86_MATCH_INTEL_FAM6_MODEL( HASWELL, 0x22),
548 X86_MATCH_INTEL_FAM6_MODEL( HASWELL_L, 0x20),
549 X86_MATCH_INTEL_FAM6_MODEL( HASWELL_G, 0x17),
550
551 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL, 0x25),
552 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_G, 0x17),
553
554 X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE_L, 0xb2),
555 X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE, 0xb2),
556
557 X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE_L, 0x52),
558 X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE, 0x52),
559
560 {},
561};
562
563static __init bool apic_validate_deadline_timer(void)
564{
565 const struct x86_cpu_id *m;
566 u32 rev;
567
568 if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
569 return false;
570 if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
571 return true;
572
573 m = x86_match_cpu(deadline_match);
574 if (!m)
575 return true;
576
577 rev = (u32)m->driver_data;
578
579 if (boot_cpu_data.microcode >= rev)
580 return true;
581
582 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
583 pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
584 "please update microcode to version: 0x%x (or later)\n", rev);
585 return false;
586}
587
588/*
589 * Setup the local APIC timer for this CPU. Copy the initialized values
590 * of the boot CPU and register the clock event in the framework.
591 */
592static void setup_APIC_timer(void)
593{
594 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
595
596 if (this_cpu_has(X86_FEATURE_ARAT)) {
597 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
598 /* Make LAPIC timer preferable over percpu HPET */
599 lapic_clockevent.rating = 150;
600 }
601
602 memcpy(levt, &lapic_clockevent, sizeof(*levt));
603 levt->cpumask = cpumask_of(smp_processor_id());
604
605 if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
606 levt->name = "lapic-deadline";
607 levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
608 CLOCK_EVT_FEAT_DUMMY);
609 levt->set_next_event = lapic_next_deadline;
610 clockevents_config_and_register(levt,
611 tsc_khz * (1000 / TSC_DIVISOR),
612 0xF, ~0UL);
613 } else
614 clockevents_register_device(levt);
615}
616
617/*
618 * Install the updated TSC frequency from recalibration at the TSC
619 * deadline clockevent devices.
620 */
621static void __lapic_update_tsc_freq(void *info)
622{
623 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
624
625 if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
626 return;
627
628 clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
629}
630
631void lapic_update_tsc_freq(void)
632{
633 /*
634 * The clockevent device's ->mult and ->shift can both be
635 * changed. In order to avoid races, schedule the frequency
636 * update code on each CPU.
637 */
638 on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
639}
640
641/*
642 * In this functions we calibrate APIC bus clocks to the external timer.
643 *
644 * We want to do the calibration only once since we want to have local timer
645 * irqs synchronous. CPUs connected by the same APIC bus have the very same bus
646 * frequency.
647 *
648 * This was previously done by reading the PIT/HPET and waiting for a wrap
649 * around to find out, that a tick has elapsed. I have a box, where the PIT
650 * readout is broken, so it never gets out of the wait loop again. This was
651 * also reported by others.
652 *
653 * Monitoring the jiffies value is inaccurate and the clockevents
654 * infrastructure allows us to do a simple substitution of the interrupt
655 * handler.
656 *
657 * The calibration routine also uses the pm_timer when possible, as the PIT
658 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
659 * back to normal later in the boot process).
660 */
661
662#define LAPIC_CAL_LOOPS (HZ/10)
663
664static __initdata int lapic_cal_loops = -1;
665static __initdata long lapic_cal_t1, lapic_cal_t2;
666static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
667static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
668static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
669
670/*
671 * Temporary interrupt handler and polled calibration function.
672 */
673static void __init lapic_cal_handler(struct clock_event_device *dev)
674{
675 unsigned long long tsc = 0;
676 long tapic = apic_read(APIC_TMCCT);
677 unsigned long pm = acpi_pm_read_early();
678
679 if (boot_cpu_has(X86_FEATURE_TSC))
680 tsc = rdtsc();
681
682 switch (lapic_cal_loops++) {
683 case 0:
684 lapic_cal_t1 = tapic;
685 lapic_cal_tsc1 = tsc;
686 lapic_cal_pm1 = pm;
687 lapic_cal_j1 = jiffies;
688 break;
689
690 case LAPIC_CAL_LOOPS:
691 lapic_cal_t2 = tapic;
692 lapic_cal_tsc2 = tsc;
693 if (pm < lapic_cal_pm1)
694 pm += ACPI_PM_OVRRUN;
695 lapic_cal_pm2 = pm;
696 lapic_cal_j2 = jiffies;
697 break;
698 }
699}
700
701static int __init
702calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
703{
704 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
705 const long pm_thresh = pm_100ms / 100;
706 unsigned long mult;
707 u64 res;
708
709#ifndef CONFIG_X86_PM_TIMER
710 return -1;
711#endif
712
713 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
714
715 /* Check, if the PM timer is available */
716 if (!deltapm)
717 return -1;
718
719 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
720
721 if (deltapm > (pm_100ms - pm_thresh) &&
722 deltapm < (pm_100ms + pm_thresh)) {
723 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
724 return 0;
725 }
726
727 res = (((u64)deltapm) * mult) >> 22;
728 do_div(res, 1000000);
729 pr_warn("APIC calibration not consistent "
730 "with PM-Timer: %ldms instead of 100ms\n", (long)res);
731
732 /* Correct the lapic counter value */
733 res = (((u64)(*delta)) * pm_100ms);
734 do_div(res, deltapm);
735 pr_info("APIC delta adjusted to PM-Timer: "
736 "%lu (%ld)\n", (unsigned long)res, *delta);
737 *delta = (long)res;
738
739 /* Correct the tsc counter value */
740 if (boot_cpu_has(X86_FEATURE_TSC)) {
741 res = (((u64)(*deltatsc)) * pm_100ms);
742 do_div(res, deltapm);
743 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
744 "PM-Timer: %lu (%ld)\n",
745 (unsigned long)res, *deltatsc);
746 *deltatsc = (long)res;
747 }
748
749 return 0;
750}
751
752static int __init lapic_init_clockevent(void)
753{
754 if (!lapic_timer_period)
755 return -1;
756
757 /* Calculate the scaled math multiplication factor */
758 lapic_clockevent.mult = div_sc(lapic_timer_period/APIC_DIVISOR,
759 TICK_NSEC, lapic_clockevent.shift);
760 lapic_clockevent.max_delta_ns =
761 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
762 lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
763 lapic_clockevent.min_delta_ns =
764 clockevent_delta2ns(0xF, &lapic_clockevent);
765 lapic_clockevent.min_delta_ticks = 0xF;
766
767 return 0;
768}
769
770bool __init apic_needs_pit(void)
771{
772 /*
773 * If the frequencies are not known, PIT is required for both TSC
774 * and apic timer calibration.
775 */
776 if (!tsc_khz || !cpu_khz)
777 return true;
778
779 /* Is there an APIC at all or is it disabled? */
780 if (!boot_cpu_has(X86_FEATURE_APIC) || apic_is_disabled)
781 return true;
782
783 /*
784 * If interrupt delivery mode is legacy PIC or virtual wire without
785 * configuration, the local APIC timer won't be set up. Make sure
786 * that the PIT is initialized.
787 */
788 if (apic_intr_mode == APIC_PIC ||
789 apic_intr_mode == APIC_VIRTUAL_WIRE_NO_CONFIG)
790 return true;
791
792 /* Virt guests may lack ARAT, but still have DEADLINE */
793 if (!boot_cpu_has(X86_FEATURE_ARAT))
794 return true;
795
796 /* Deadline timer is based on TSC so no further PIT action required */
797 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
798 return false;
799
800 /* APIC timer disabled? */
801 if (disable_apic_timer)
802 return true;
803 /*
804 * The APIC timer frequency is known already, no PIT calibration
805 * required. If unknown, let the PIT be initialized.
806 */
807 return lapic_timer_period == 0;
808}
809
810static int __init calibrate_APIC_clock(void)
811{
812 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
813 u64 tsc_perj = 0, tsc_start = 0;
814 unsigned long jif_start;
815 unsigned long deltaj;
816 long delta, deltatsc;
817 int pm_referenced = 0;
818
819 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
820 return 0;
821
822 /*
823 * Check if lapic timer has already been calibrated by platform
824 * specific routine, such as tsc calibration code. If so just fill
825 * in the clockevent structure and return.
826 */
827 if (!lapic_init_clockevent()) {
828 apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
829 lapic_timer_period);
830 /*
831 * Direct calibration methods must have an always running
832 * local APIC timer, no need for broadcast timer.
833 */
834 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
835 return 0;
836 }
837
838 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
839 "calibrating APIC timer ...\n");
840
841 /*
842 * There are platforms w/o global clockevent devices. Instead of
843 * making the calibration conditional on that, use a polling based
844 * approach everywhere.
845 */
846 local_irq_disable();
847
848 /*
849 * Setup the APIC counter to maximum. There is no way the lapic
850 * can underflow in the 100ms detection time frame
851 */
852 __setup_APIC_LVTT(0xffffffff, 0, 0);
853
854 /*
855 * Methods to terminate the calibration loop:
856 * 1) Global clockevent if available (jiffies)
857 * 2) TSC if available and frequency is known
858 */
859 jif_start = READ_ONCE(jiffies);
860
861 if (tsc_khz) {
862 tsc_start = rdtsc();
863 tsc_perj = div_u64((u64)tsc_khz * 1000, HZ);
864 }
865
866 /*
867 * Enable interrupts so the tick can fire, if a global
868 * clockevent device is available
869 */
870 local_irq_enable();
871
872 while (lapic_cal_loops <= LAPIC_CAL_LOOPS) {
873 /* Wait for a tick to elapse */
874 while (1) {
875 if (tsc_khz) {
876 u64 tsc_now = rdtsc();
877 if ((tsc_now - tsc_start) >= tsc_perj) {
878 tsc_start += tsc_perj;
879 break;
880 }
881 } else {
882 unsigned long jif_now = READ_ONCE(jiffies);
883
884 if (time_after(jif_now, jif_start)) {
885 jif_start = jif_now;
886 break;
887 }
888 }
889 cpu_relax();
890 }
891
892 /* Invoke the calibration routine */
893 local_irq_disable();
894 lapic_cal_handler(NULL);
895 local_irq_enable();
896 }
897
898 local_irq_disable();
899
900 /* Build delta t1-t2 as apic timer counts down */
901 delta = lapic_cal_t1 - lapic_cal_t2;
902 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
903
904 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
905
906 /* we trust the PM based calibration if possible */
907 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
908 &delta, &deltatsc);
909
910 lapic_timer_period = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
911 lapic_init_clockevent();
912
913 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
914 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
915 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
916 lapic_timer_period);
917
918 if (boot_cpu_has(X86_FEATURE_TSC)) {
919 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
920 "%ld.%04ld MHz.\n",
921 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
922 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
923 }
924
925 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
926 "%u.%04u MHz.\n",
927 lapic_timer_period / (1000000 / HZ),
928 lapic_timer_period % (1000000 / HZ));
929
930 /*
931 * Do a sanity check on the APIC calibration result
932 */
933 if (lapic_timer_period < (1000000 / HZ)) {
934 local_irq_enable();
935 pr_warn("APIC frequency too slow, disabling apic timer\n");
936 return -1;
937 }
938
939 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
940
941 /*
942 * PM timer calibration failed or not turned on so lets try APIC
943 * timer based calibration, if a global clockevent device is
944 * available.
945 */
946 if (!pm_referenced && global_clock_event) {
947 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
948
949 /*
950 * Setup the apic timer manually
951 */
952 levt->event_handler = lapic_cal_handler;
953 lapic_timer_set_periodic(levt);
954 lapic_cal_loops = -1;
955
956 /* Let the interrupts run */
957 local_irq_enable();
958
959 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
960 cpu_relax();
961
962 /* Stop the lapic timer */
963 local_irq_disable();
964 lapic_timer_shutdown(levt);
965
966 /* Jiffies delta */
967 deltaj = lapic_cal_j2 - lapic_cal_j1;
968 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
969
970 /* Check, if the jiffies result is consistent */
971 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
972 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
973 else
974 levt->features |= CLOCK_EVT_FEAT_DUMMY;
975 }
976 local_irq_enable();
977
978 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
979 pr_warn("APIC timer disabled due to verification failure\n");
980 return -1;
981 }
982
983 return 0;
984}
985
986/*
987 * Setup the boot APIC
988 *
989 * Calibrate and verify the result.
990 */
991void __init setup_boot_APIC_clock(void)
992{
993 /*
994 * The local apic timer can be disabled via the kernel
995 * commandline or from the CPU detection code. Register the lapic
996 * timer as a dummy clock event source on SMP systems, so the
997 * broadcast mechanism is used. On UP systems simply ignore it.
998 */
999 if (disable_apic_timer) {
1000 pr_info("Disabling APIC timer\n");
1001 /* No broadcast on UP ! */
1002 if (num_possible_cpus() > 1) {
1003 lapic_clockevent.mult = 1;
1004 setup_APIC_timer();
1005 }
1006 return;
1007 }
1008
1009 if (calibrate_APIC_clock()) {
1010 /* No broadcast on UP ! */
1011 if (num_possible_cpus() > 1)
1012 setup_APIC_timer();
1013 return;
1014 }
1015
1016 /*
1017 * If nmi_watchdog is set to IO_APIC, we need the
1018 * PIT/HPET going. Otherwise register lapic as a dummy
1019 * device.
1020 */
1021 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
1022
1023 /* Setup the lapic or request the broadcast */
1024 setup_APIC_timer();
1025 amd_e400_c1e_apic_setup();
1026}
1027
1028void setup_secondary_APIC_clock(void)
1029{
1030 setup_APIC_timer();
1031 amd_e400_c1e_apic_setup();
1032}
1033
1034/*
1035 * The guts of the apic timer interrupt
1036 */
1037static void local_apic_timer_interrupt(void)
1038{
1039 struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1040
1041 /*
1042 * Normally we should not be here till LAPIC has been initialized but
1043 * in some cases like kdump, its possible that there is a pending LAPIC
1044 * timer interrupt from previous kernel's context and is delivered in
1045 * new kernel the moment interrupts are enabled.
1046 *
1047 * Interrupts are enabled early and LAPIC is setup much later, hence
1048 * its possible that when we get here evt->event_handler is NULL.
1049 * Check for event_handler being NULL and discard the interrupt as
1050 * spurious.
1051 */
1052 if (!evt->event_handler) {
1053 pr_warn("Spurious LAPIC timer interrupt on cpu %d\n",
1054 smp_processor_id());
1055 /* Switch it off */
1056 lapic_timer_shutdown(evt);
1057 return;
1058 }
1059
1060 /*
1061 * the NMI deadlock-detector uses this.
1062 */
1063 inc_irq_stat(apic_timer_irqs);
1064
1065 evt->event_handler(evt);
1066}
1067
1068/*
1069 * Local APIC timer interrupt. This is the most natural way for doing
1070 * local interrupts, but local timer interrupts can be emulated by
1071 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1072 *
1073 * [ if a single-CPU system runs an SMP kernel then we call the local
1074 * interrupt as well. Thus we cannot inline the local irq ... ]
1075 */
1076DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt)
1077{
1078 struct pt_regs *old_regs = set_irq_regs(regs);
1079
1080 apic_eoi();
1081 trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1082 local_apic_timer_interrupt();
1083 trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1084
1085 set_irq_regs(old_regs);
1086}
1087
1088/*
1089 * Local APIC start and shutdown
1090 */
1091
1092/**
1093 * clear_local_APIC - shutdown the local APIC
1094 *
1095 * This is called, when a CPU is disabled and before rebooting, so the state of
1096 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1097 * leftovers during boot.
1098 */
1099void clear_local_APIC(void)
1100{
1101 int maxlvt;
1102 u32 v;
1103
1104 if (!apic_accessible())
1105 return;
1106
1107 maxlvt = lapic_get_maxlvt();
1108 /*
1109 * Masking an LVT entry can trigger a local APIC error
1110 * if the vector is zero. Mask LVTERR first to prevent this.
1111 */
1112 if (maxlvt >= 3) {
1113 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1114 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1115 }
1116 /*
1117 * Careful: we have to set masks only first to deassert
1118 * any level-triggered sources.
1119 */
1120 v = apic_read(APIC_LVTT);
1121 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1122 v = apic_read(APIC_LVT0);
1123 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1124 v = apic_read(APIC_LVT1);
1125 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1126 if (maxlvt >= 4) {
1127 v = apic_read(APIC_LVTPC);
1128 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1129 }
1130
1131 /* lets not touch this if we didn't frob it */
1132#ifdef CONFIG_X86_THERMAL_VECTOR
1133 if (maxlvt >= 5) {
1134 v = apic_read(APIC_LVTTHMR);
1135 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1136 }
1137#endif
1138#ifdef CONFIG_X86_MCE_INTEL
1139 if (maxlvt >= 6) {
1140 v = apic_read(APIC_LVTCMCI);
1141 if (!(v & APIC_LVT_MASKED))
1142 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1143 }
1144#endif
1145
1146 /*
1147 * Clean APIC state for other OSs:
1148 */
1149 apic_write(APIC_LVTT, APIC_LVT_MASKED);
1150 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1151 apic_write(APIC_LVT1, APIC_LVT_MASKED);
1152 if (maxlvt >= 3)
1153 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1154 if (maxlvt >= 4)
1155 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1156
1157 /* Integrated APIC (!82489DX) ? */
1158 if (lapic_is_integrated()) {
1159 if (maxlvt > 3)
1160 /* Clear ESR due to Pentium errata 3AP and 11AP */
1161 apic_write(APIC_ESR, 0);
1162 apic_read(APIC_ESR);
1163 }
1164}
1165
1166/**
1167 * apic_soft_disable - Clears and software disables the local APIC on hotplug
1168 *
1169 * Contrary to disable_local_APIC() this does not touch the enable bit in
1170 * MSR_IA32_APICBASE. Clearing that bit on systems based on the 3 wire APIC
1171 * bus would require a hardware reset as the APIC would lose track of bus
1172 * arbitration. On systems with FSB delivery APICBASE could be disabled,
1173 * but it has to be guaranteed that no interrupt is sent to the APIC while
1174 * in that state and it's not clear from the SDM whether it still responds
1175 * to INIT/SIPI messages. Stay on the safe side and use software disable.
1176 */
1177void apic_soft_disable(void)
1178{
1179 u32 value;
1180
1181 clear_local_APIC();
1182
1183 /* Soft disable APIC (implies clearing of registers for 82489DX!). */
1184 value = apic_read(APIC_SPIV);
1185 value &= ~APIC_SPIV_APIC_ENABLED;
1186 apic_write(APIC_SPIV, value);
1187}
1188
1189/**
1190 * disable_local_APIC - clear and disable the local APIC
1191 */
1192void disable_local_APIC(void)
1193{
1194 if (!apic_accessible())
1195 return;
1196
1197 apic_soft_disable();
1198
1199#ifdef CONFIG_X86_32
1200 /*
1201 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1202 * restore the disabled state.
1203 */
1204 if (enabled_via_apicbase) {
1205 unsigned int l, h;
1206
1207 rdmsr(MSR_IA32_APICBASE, l, h);
1208 l &= ~MSR_IA32_APICBASE_ENABLE;
1209 wrmsr(MSR_IA32_APICBASE, l, h);
1210 }
1211#endif
1212}
1213
1214/*
1215 * If Linux enabled the LAPIC against the BIOS default disable it down before
1216 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1217 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1218 * for the case where Linux didn't enable the LAPIC.
1219 */
1220void lapic_shutdown(void)
1221{
1222 unsigned long flags;
1223
1224 if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1225 return;
1226
1227 local_irq_save(flags);
1228
1229#ifdef CONFIG_X86_32
1230 if (!enabled_via_apicbase)
1231 clear_local_APIC();
1232 else
1233#endif
1234 disable_local_APIC();
1235
1236
1237 local_irq_restore(flags);
1238}
1239
1240/**
1241 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1242 */
1243void __init sync_Arb_IDs(void)
1244{
1245 /*
1246 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1247 * needed on AMD.
1248 */
1249 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1250 return;
1251
1252 /*
1253 * Wait for idle.
1254 */
1255 apic_wait_icr_idle();
1256
1257 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1258 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1259 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1260}
1261
1262enum apic_intr_mode_id apic_intr_mode __ro_after_init;
1263
1264static int __init __apic_intr_mode_select(void)
1265{
1266 /* Check kernel option */
1267 if (apic_is_disabled) {
1268 pr_info("APIC disabled via kernel command line\n");
1269 return APIC_PIC;
1270 }
1271
1272 /* Check BIOS */
1273#ifdef CONFIG_X86_64
1274 /* On 64-bit, the APIC must be integrated, Check local APIC only */
1275 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1276 apic_is_disabled = true;
1277 pr_info("APIC disabled by BIOS\n");
1278 return APIC_PIC;
1279 }
1280#else
1281 /* On 32-bit, the APIC may be integrated APIC or 82489DX */
1282
1283 /* Neither 82489DX nor integrated APIC ? */
1284 if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1285 apic_is_disabled = true;
1286 return APIC_PIC;
1287 }
1288
1289 /* If the BIOS pretends there is an integrated APIC ? */
1290 if (!boot_cpu_has(X86_FEATURE_APIC) &&
1291 APIC_INTEGRATED(boot_cpu_apic_version)) {
1292 apic_is_disabled = true;
1293 pr_err(FW_BUG "Local APIC not detected, force emulation\n");
1294 return APIC_PIC;
1295 }
1296#endif
1297
1298 /* Check MP table or ACPI MADT configuration */
1299 if (!smp_found_config) {
1300 disable_ioapic_support();
1301 if (!acpi_lapic) {
1302 pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1303 return APIC_VIRTUAL_WIRE_NO_CONFIG;
1304 }
1305 return APIC_VIRTUAL_WIRE;
1306 }
1307
1308#ifdef CONFIG_SMP
1309 /* If SMP should be disabled, then really disable it! */
1310 if (!setup_max_cpus) {
1311 pr_info("APIC: SMP mode deactivated\n");
1312 return APIC_SYMMETRIC_IO_NO_ROUTING;
1313 }
1314#endif
1315
1316 return APIC_SYMMETRIC_IO;
1317}
1318
1319/* Select the interrupt delivery mode for the BSP */
1320void __init apic_intr_mode_select(void)
1321{
1322 apic_intr_mode = __apic_intr_mode_select();
1323}
1324
1325/*
1326 * An initial setup of the virtual wire mode.
1327 */
1328void __init init_bsp_APIC(void)
1329{
1330 unsigned int value;
1331
1332 /*
1333 * Don't do the setup now if we have a SMP BIOS as the
1334 * through-I/O-APIC virtual wire mode might be active.
1335 */
1336 if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1337 return;
1338
1339 /*
1340 * Do not trust the local APIC being empty at bootup.
1341 */
1342 clear_local_APIC();
1343
1344 /*
1345 * Enable APIC.
1346 */
1347 value = apic_read(APIC_SPIV);
1348 value &= ~APIC_VECTOR_MASK;
1349 value |= APIC_SPIV_APIC_ENABLED;
1350
1351#ifdef CONFIG_X86_32
1352 /* This bit is reserved on P4/Xeon and should be cleared */
1353 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1354 (boot_cpu_data.x86 == 15))
1355 value &= ~APIC_SPIV_FOCUS_DISABLED;
1356 else
1357#endif
1358 value |= APIC_SPIV_FOCUS_DISABLED;
1359 value |= SPURIOUS_APIC_VECTOR;
1360 apic_write(APIC_SPIV, value);
1361
1362 /*
1363 * Set up the virtual wire mode.
1364 */
1365 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1366 value = APIC_DM_NMI;
1367 if (!lapic_is_integrated()) /* 82489DX */
1368 value |= APIC_LVT_LEVEL_TRIGGER;
1369 if (apic_extnmi == APIC_EXTNMI_NONE)
1370 value |= APIC_LVT_MASKED;
1371 apic_write(APIC_LVT1, value);
1372}
1373
1374static void __init apic_bsp_setup(bool upmode);
1375
1376/* Init the interrupt delivery mode for the BSP */
1377void __init apic_intr_mode_init(void)
1378{
1379 bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1380
1381 switch (apic_intr_mode) {
1382 case APIC_PIC:
1383 pr_info("APIC: Keep in PIC mode(8259)\n");
1384 return;
1385 case APIC_VIRTUAL_WIRE:
1386 pr_info("APIC: Switch to virtual wire mode setup\n");
1387 break;
1388 case APIC_VIRTUAL_WIRE_NO_CONFIG:
1389 pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1390 upmode = true;
1391 break;
1392 case APIC_SYMMETRIC_IO:
1393 pr_info("APIC: Switch to symmetric I/O mode setup\n");
1394 break;
1395 case APIC_SYMMETRIC_IO_NO_ROUTING:
1396 pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1397 break;
1398 }
1399
1400 x86_64_probe_apic();
1401
1402 x86_32_install_bigsmp();
1403
1404 if (x86_platform.apic_post_init)
1405 x86_platform.apic_post_init();
1406
1407 apic_bsp_setup(upmode);
1408}
1409
1410static void lapic_setup_esr(void)
1411{
1412 unsigned int oldvalue, value, maxlvt;
1413
1414 if (!lapic_is_integrated()) {
1415 pr_info("No ESR for 82489DX.\n");
1416 return;
1417 }
1418
1419 if (apic->disable_esr) {
1420 /*
1421 * Something untraceable is creating bad interrupts on
1422 * secondary quads ... for the moment, just leave the
1423 * ESR disabled - we can't do anything useful with the
1424 * errors anyway - mbligh
1425 */
1426 pr_info("Leaving ESR disabled.\n");
1427 return;
1428 }
1429
1430 maxlvt = lapic_get_maxlvt();
1431 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1432 apic_write(APIC_ESR, 0);
1433 oldvalue = apic_read(APIC_ESR);
1434
1435 /* enables sending errors */
1436 value = ERROR_APIC_VECTOR;
1437 apic_write(APIC_LVTERR, value);
1438
1439 /*
1440 * spec says clear errors after enabling vector.
1441 */
1442 if (maxlvt > 3)
1443 apic_write(APIC_ESR, 0);
1444 value = apic_read(APIC_ESR);
1445 if (value != oldvalue)
1446 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1447 "vector: 0x%08x after: 0x%08x\n",
1448 oldvalue, value);
1449}
1450
1451#define APIC_IR_REGS APIC_ISR_NR
1452#define APIC_IR_BITS (APIC_IR_REGS * 32)
1453#define APIC_IR_MAPSIZE (APIC_IR_BITS / BITS_PER_LONG)
1454
1455union apic_ir {
1456 unsigned long map[APIC_IR_MAPSIZE];
1457 u32 regs[APIC_IR_REGS];
1458};
1459
1460static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr)
1461{
1462 int i, bit;
1463
1464 /* Read the IRRs */
1465 for (i = 0; i < APIC_IR_REGS; i++)
1466 irr->regs[i] = apic_read(APIC_IRR + i * 0x10);
1467
1468 /* Read the ISRs */
1469 for (i = 0; i < APIC_IR_REGS; i++)
1470 isr->regs[i] = apic_read(APIC_ISR + i * 0x10);
1471
1472 /*
1473 * If the ISR map is not empty. ACK the APIC and run another round
1474 * to verify whether a pending IRR has been unblocked and turned
1475 * into a ISR.
1476 */
1477 if (!bitmap_empty(isr->map, APIC_IR_BITS)) {
1478 /*
1479 * There can be multiple ISR bits set when a high priority
1480 * interrupt preempted a lower priority one. Issue an ACK
1481 * per set bit.
1482 */
1483 for_each_set_bit(bit, isr->map, APIC_IR_BITS)
1484 apic_eoi();
1485 return true;
1486 }
1487
1488 return !bitmap_empty(irr->map, APIC_IR_BITS);
1489}
1490
1491/*
1492 * After a crash, we no longer service the interrupts and a pending
1493 * interrupt from previous kernel might still have ISR bit set.
1494 *
1495 * Most probably by now the CPU has serviced that pending interrupt and it
1496 * might not have done the apic_eoi() because it thought, interrupt
1497 * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear
1498 * the ISR bit and cpu thinks it has already serviced the interrupt. Hence
1499 * a vector might get locked. It was noticed for timer irq (vector
1500 * 0x31). Issue an extra EOI to clear ISR.
1501 *
1502 * If there are pending IRR bits they turn into ISR bits after a higher
1503 * priority ISR bit has been acked.
1504 */
1505static void apic_pending_intr_clear(void)
1506{
1507 union apic_ir irr, isr;
1508 unsigned int i;
1509
1510 /* 512 loops are way oversized and give the APIC a chance to obey. */
1511 for (i = 0; i < 512; i++) {
1512 if (!apic_check_and_ack(&irr, &isr))
1513 return;
1514 }
1515 /* Dump the IRR/ISR content if that failed */
1516 pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map);
1517}
1518
1519/**
1520 * setup_local_APIC - setup the local APIC
1521 *
1522 * Used to setup local APIC while initializing BSP or bringing up APs.
1523 * Always called with preemption disabled.
1524 */
1525static void setup_local_APIC(void)
1526{
1527 int cpu = smp_processor_id();
1528 unsigned int value;
1529
1530 if (apic_is_disabled) {
1531 disable_ioapic_support();
1532 return;
1533 }
1534
1535 /*
1536 * If this comes from kexec/kcrash the APIC might be enabled in
1537 * SPIV. Soft disable it before doing further initialization.
1538 */
1539 value = apic_read(APIC_SPIV);
1540 value &= ~APIC_SPIV_APIC_ENABLED;
1541 apic_write(APIC_SPIV, value);
1542
1543#ifdef CONFIG_X86_32
1544 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1545 if (lapic_is_integrated() && apic->disable_esr) {
1546 apic_write(APIC_ESR, 0);
1547 apic_write(APIC_ESR, 0);
1548 apic_write(APIC_ESR, 0);
1549 apic_write(APIC_ESR, 0);
1550 }
1551#endif
1552 /* Validate that the APIC is registered if required */
1553 BUG_ON(apic->apic_id_registered && !apic->apic_id_registered());
1554
1555 /*
1556 * Intel recommends to set DFR, LDR and TPR before enabling
1557 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1558 * document number 292116).
1559 *
1560 * Except for APICs which operate in physical destination mode.
1561 */
1562 if (apic->init_apic_ldr)
1563 apic->init_apic_ldr();
1564
1565 /*
1566 * Set Task Priority to 'accept all except vectors 0-31'. An APIC
1567 * vector in the 16-31 range could be delivered if TPR == 0, but we
1568 * would think it's an exception and terrible things will happen. We
1569 * never change this later on.
1570 */
1571 value = apic_read(APIC_TASKPRI);
1572 value &= ~APIC_TPRI_MASK;
1573 value |= 0x10;
1574 apic_write(APIC_TASKPRI, value);
1575
1576 /* Clear eventually stale ISR/IRR bits */
1577 apic_pending_intr_clear();
1578
1579 /*
1580 * Now that we are all set up, enable the APIC
1581 */
1582 value = apic_read(APIC_SPIV);
1583 value &= ~APIC_VECTOR_MASK;
1584 /*
1585 * Enable APIC
1586 */
1587 value |= APIC_SPIV_APIC_ENABLED;
1588
1589#ifdef CONFIG_X86_32
1590 /*
1591 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1592 * certain networking cards. If high frequency interrupts are
1593 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1594 * entry is masked/unmasked at a high rate as well then sooner or
1595 * later IOAPIC line gets 'stuck', no more interrupts are received
1596 * from the device. If focus CPU is disabled then the hang goes
1597 * away, oh well :-(
1598 *
1599 * [ This bug can be reproduced easily with a level-triggered
1600 * PCI Ne2000 networking cards and PII/PIII processors, dual
1601 * BX chipset. ]
1602 */
1603 /*
1604 * Actually disabling the focus CPU check just makes the hang less
1605 * frequent as it makes the interrupt distribution model be more
1606 * like LRU than MRU (the short-term load is more even across CPUs).
1607 */
1608
1609 /*
1610 * - enable focus processor (bit==0)
1611 * - 64bit mode always use processor focus
1612 * so no need to set it
1613 */
1614 value &= ~APIC_SPIV_FOCUS_DISABLED;
1615#endif
1616
1617 /*
1618 * Set spurious IRQ vector
1619 */
1620 value |= SPURIOUS_APIC_VECTOR;
1621 apic_write(APIC_SPIV, value);
1622
1623 perf_events_lapic_init();
1624
1625 /*
1626 * Set up LVT0, LVT1:
1627 *
1628 * set up through-local-APIC on the boot CPU's LINT0. This is not
1629 * strictly necessary in pure symmetric-IO mode, but sometimes
1630 * we delegate interrupts to the 8259A.
1631 */
1632 /*
1633 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1634 */
1635 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1636 if (!cpu && (pic_mode || !value || ioapic_is_disabled)) {
1637 value = APIC_DM_EXTINT;
1638 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1639 } else {
1640 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1641 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1642 }
1643 apic_write(APIC_LVT0, value);
1644
1645 /*
1646 * Only the BSP sees the LINT1 NMI signal by default. This can be
1647 * modified by apic_extnmi= boot option.
1648 */
1649 if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1650 apic_extnmi == APIC_EXTNMI_ALL)
1651 value = APIC_DM_NMI;
1652 else
1653 value = APIC_DM_NMI | APIC_LVT_MASKED;
1654
1655 /* Is 82489DX ? */
1656 if (!lapic_is_integrated())
1657 value |= APIC_LVT_LEVEL_TRIGGER;
1658 apic_write(APIC_LVT1, value);
1659
1660#ifdef CONFIG_X86_MCE_INTEL
1661 /* Recheck CMCI information after local APIC is up on CPU #0 */
1662 if (!cpu)
1663 cmci_recheck();
1664#endif
1665}
1666
1667static void end_local_APIC_setup(void)
1668{
1669 lapic_setup_esr();
1670
1671#ifdef CONFIG_X86_32
1672 {
1673 unsigned int value;
1674 /* Disable the local apic timer */
1675 value = apic_read(APIC_LVTT);
1676 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1677 apic_write(APIC_LVTT, value);
1678 }
1679#endif
1680
1681 apic_pm_activate();
1682}
1683
1684/*
1685 * APIC setup function for application processors. Called from smpboot.c
1686 */
1687void apic_ap_setup(void)
1688{
1689 setup_local_APIC();
1690 end_local_APIC_setup();
1691}
1692
1693static __init void cpu_set_boot_apic(void);
1694
1695static __init void apic_read_boot_cpu_id(bool x2apic)
1696{
1697 /*
1698 * This can be invoked from check_x2apic() before the APIC has been
1699 * selected. But that code knows for sure that the BIOS enabled
1700 * X2APIC.
1701 */
1702 if (x2apic) {
1703 boot_cpu_physical_apicid = native_apic_msr_read(APIC_ID);
1704 boot_cpu_apic_version = GET_APIC_VERSION(native_apic_msr_read(APIC_LVR));
1705 } else {
1706 boot_cpu_physical_apicid = read_apic_id();
1707 boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
1708 }
1709 cpu_set_boot_apic();
1710}
1711
1712#ifdef CONFIG_X86_X2APIC
1713int x2apic_mode;
1714EXPORT_SYMBOL_GPL(x2apic_mode);
1715
1716enum {
1717 X2APIC_OFF,
1718 X2APIC_DISABLED,
1719 /* All states below here have X2APIC enabled */
1720 X2APIC_ON,
1721 X2APIC_ON_LOCKED
1722};
1723static int x2apic_state;
1724
1725static bool x2apic_hw_locked(void)
1726{
1727 u64 ia32_cap;
1728 u64 msr;
1729
1730 ia32_cap = x86_read_arch_cap_msr();
1731 if (ia32_cap & ARCH_CAP_XAPIC_DISABLE) {
1732 rdmsrl(MSR_IA32_XAPIC_DISABLE_STATUS, msr);
1733 return (msr & LEGACY_XAPIC_DISABLED);
1734 }
1735 return false;
1736}
1737
1738static void __x2apic_disable(void)
1739{
1740 u64 msr;
1741
1742 if (!boot_cpu_has(X86_FEATURE_APIC))
1743 return;
1744
1745 rdmsrl(MSR_IA32_APICBASE, msr);
1746 if (!(msr & X2APIC_ENABLE))
1747 return;
1748 /* Disable xapic and x2apic first and then reenable xapic mode */
1749 wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1750 wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1751 printk_once(KERN_INFO "x2apic disabled\n");
1752}
1753
1754static void __x2apic_enable(void)
1755{
1756 u64 msr;
1757
1758 rdmsrl(MSR_IA32_APICBASE, msr);
1759 if (msr & X2APIC_ENABLE)
1760 return;
1761 wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1762 printk_once(KERN_INFO "x2apic enabled\n");
1763}
1764
1765static int __init setup_nox2apic(char *str)
1766{
1767 if (x2apic_enabled()) {
1768 u32 apicid = native_apic_msr_read(APIC_ID);
1769
1770 if (apicid >= 255) {
1771 pr_warn("Apicid: %08x, cannot enforce nox2apic\n",
1772 apicid);
1773 return 0;
1774 }
1775 if (x2apic_hw_locked()) {
1776 pr_warn("APIC locked in x2apic mode, can't disable\n");
1777 return 0;
1778 }
1779 pr_warn("x2apic already enabled.\n");
1780 __x2apic_disable();
1781 }
1782 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1783 x2apic_state = X2APIC_DISABLED;
1784 x2apic_mode = 0;
1785 return 0;
1786}
1787early_param("nox2apic", setup_nox2apic);
1788
1789/* Called from cpu_init() to enable x2apic on (secondary) cpus */
1790void x2apic_setup(void)
1791{
1792 /*
1793 * Try to make the AP's APIC state match that of the BSP, but if the
1794 * BSP is unlocked and the AP is locked then there is a state mismatch.
1795 * Warn about the mismatch in case a GP fault occurs due to a locked AP
1796 * trying to be turned off.
1797 */
1798 if (x2apic_state != X2APIC_ON_LOCKED && x2apic_hw_locked())
1799 pr_warn("x2apic lock mismatch between BSP and AP.\n");
1800 /*
1801 * If x2apic is not in ON or LOCKED state, disable it if already enabled
1802 * from BIOS.
1803 */
1804 if (x2apic_state < X2APIC_ON) {
1805 __x2apic_disable();
1806 return;
1807 }
1808 __x2apic_enable();
1809}
1810
1811static __init void apic_set_fixmap(void);
1812
1813static __init void x2apic_disable(void)
1814{
1815 u32 x2apic_id, state = x2apic_state;
1816
1817 x2apic_mode = 0;
1818 x2apic_state = X2APIC_DISABLED;
1819
1820 if (state != X2APIC_ON)
1821 return;
1822
1823 x2apic_id = read_apic_id();
1824 if (x2apic_id >= 255)
1825 panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1826
1827 if (x2apic_hw_locked()) {
1828 pr_warn("Cannot disable locked x2apic, id: %08x\n", x2apic_id);
1829 return;
1830 }
1831
1832 __x2apic_disable();
1833 apic_set_fixmap();
1834}
1835
1836static __init void x2apic_enable(void)
1837{
1838 if (x2apic_state != X2APIC_OFF)
1839 return;
1840
1841 x2apic_mode = 1;
1842 x2apic_state = X2APIC_ON;
1843 __x2apic_enable();
1844}
1845
1846static __init void try_to_enable_x2apic(int remap_mode)
1847{
1848 if (x2apic_state == X2APIC_DISABLED)
1849 return;
1850
1851 if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1852 u32 apic_limit = 255;
1853
1854 /*
1855 * Using X2APIC without IR is not architecturally supported
1856 * on bare metal but may be supported in guests.
1857 */
1858 if (!x86_init.hyper.x2apic_available()) {
1859 pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1860 x2apic_disable();
1861 return;
1862 }
1863
1864 /*
1865 * If the hypervisor supports extended destination ID in
1866 * MSI, that increases the maximum APIC ID that can be
1867 * used for non-remapped IRQ domains.
1868 */
1869 if (x86_init.hyper.msi_ext_dest_id()) {
1870 virt_ext_dest_id = 1;
1871 apic_limit = 32767;
1872 }
1873
1874 /*
1875 * Without IR, all CPUs can be addressed by IOAPIC/MSI only
1876 * in physical mode, and CPUs with an APIC ID that cannot
1877 * be addressed must not be brought online.
1878 */
1879 x2apic_set_max_apicid(apic_limit);
1880 x2apic_phys = 1;
1881 }
1882 x2apic_enable();
1883}
1884
1885void __init check_x2apic(void)
1886{
1887 if (x2apic_enabled()) {
1888 pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1889 x2apic_mode = 1;
1890 if (x2apic_hw_locked())
1891 x2apic_state = X2APIC_ON_LOCKED;
1892 else
1893 x2apic_state = X2APIC_ON;
1894 apic_read_boot_cpu_id(true);
1895 } else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1896 x2apic_state = X2APIC_DISABLED;
1897 }
1898}
1899#else /* CONFIG_X86_X2APIC */
1900void __init check_x2apic(void)
1901{
1902 if (!apic_is_x2apic_enabled())
1903 return;
1904 /*
1905 * Checkme: Can we simply turn off x2APIC here instead of disabling the APIC?
1906 */
1907 pr_err("Kernel does not support x2APIC, please recompile with CONFIG_X86_X2APIC.\n");
1908 pr_err("Disabling APIC, expect reduced performance and functionality.\n");
1909
1910 apic_is_disabled = true;
1911 setup_clear_cpu_cap(X86_FEATURE_APIC);
1912}
1913
1914static inline void try_to_enable_x2apic(int remap_mode) { }
1915static inline void __x2apic_enable(void) { }
1916#endif /* !CONFIG_X86_X2APIC */
1917
1918void __init enable_IR_x2apic(void)
1919{
1920 unsigned long flags;
1921 int ret, ir_stat;
1922
1923 if (ioapic_is_disabled) {
1924 pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1925 return;
1926 }
1927
1928 ir_stat = irq_remapping_prepare();
1929 if (ir_stat < 0 && !x2apic_supported())
1930 return;
1931
1932 ret = save_ioapic_entries();
1933 if (ret) {
1934 pr_info("Saving IO-APIC state failed: %d\n", ret);
1935 return;
1936 }
1937
1938 local_irq_save(flags);
1939 legacy_pic->mask_all();
1940 mask_ioapic_entries();
1941
1942 /* If irq_remapping_prepare() succeeded, try to enable it */
1943 if (ir_stat >= 0)
1944 ir_stat = irq_remapping_enable();
1945 /* ir_stat contains the remap mode or an error code */
1946 try_to_enable_x2apic(ir_stat);
1947
1948 if (ir_stat < 0)
1949 restore_ioapic_entries();
1950 legacy_pic->restore_mask();
1951 local_irq_restore(flags);
1952}
1953
1954#ifdef CONFIG_X86_64
1955/*
1956 * Detect and enable local APICs on non-SMP boards.
1957 * Original code written by Keir Fraser.
1958 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1959 * not correctly set up (usually the APIC timer won't work etc.)
1960 */
1961static bool __init detect_init_APIC(void)
1962{
1963 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1964 pr_info("No local APIC present\n");
1965 return false;
1966 }
1967
1968 register_lapic_address(APIC_DEFAULT_PHYS_BASE);
1969 return true;
1970}
1971#else
1972
1973static bool __init apic_verify(unsigned long addr)
1974{
1975 u32 features, h, l;
1976
1977 /*
1978 * The APIC feature bit should now be enabled
1979 * in `cpuid'
1980 */
1981 features = cpuid_edx(1);
1982 if (!(features & (1 << X86_FEATURE_APIC))) {
1983 pr_warn("Could not enable APIC!\n");
1984 return false;
1985 }
1986 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1987
1988 /* The BIOS may have set up the APIC at some other address */
1989 if (boot_cpu_data.x86 >= 6) {
1990 rdmsr(MSR_IA32_APICBASE, l, h);
1991 if (l & MSR_IA32_APICBASE_ENABLE)
1992 addr = l & MSR_IA32_APICBASE_BASE;
1993 }
1994
1995 register_lapic_address(addr);
1996 pr_info("Found and enabled local APIC!\n");
1997 return true;
1998}
1999
2000bool __init apic_force_enable(unsigned long addr)
2001{
2002 u32 h, l;
2003
2004 if (apic_is_disabled)
2005 return false;
2006
2007 /*
2008 * Some BIOSes disable the local APIC in the APIC_BASE
2009 * MSR. This can only be done in software for Intel P6 or later
2010 * and AMD K7 (Model > 1) or later.
2011 */
2012 if (boot_cpu_data.x86 >= 6) {
2013 rdmsr(MSR_IA32_APICBASE, l, h);
2014 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
2015 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
2016 l &= ~MSR_IA32_APICBASE_BASE;
2017 l |= MSR_IA32_APICBASE_ENABLE | addr;
2018 wrmsr(MSR_IA32_APICBASE, l, h);
2019 enabled_via_apicbase = 1;
2020 }
2021 }
2022 return apic_verify(addr);
2023}
2024
2025/*
2026 * Detect and initialize APIC
2027 */
2028static bool __init detect_init_APIC(void)
2029{
2030 /* Disabled by kernel option? */
2031 if (apic_is_disabled)
2032 return false;
2033
2034 switch (boot_cpu_data.x86_vendor) {
2035 case X86_VENDOR_AMD:
2036 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
2037 (boot_cpu_data.x86 >= 15))
2038 break;
2039 goto no_apic;
2040 case X86_VENDOR_HYGON:
2041 break;
2042 case X86_VENDOR_INTEL:
2043 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
2044 (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
2045 break;
2046 goto no_apic;
2047 default:
2048 goto no_apic;
2049 }
2050
2051 if (!boot_cpu_has(X86_FEATURE_APIC)) {
2052 /*
2053 * Over-ride BIOS and try to enable the local APIC only if
2054 * "lapic" specified.
2055 */
2056 if (!force_enable_local_apic) {
2057 pr_info("Local APIC disabled by BIOS -- "
2058 "you can enable it with \"lapic\"\n");
2059 return false;
2060 }
2061 if (!apic_force_enable(APIC_DEFAULT_PHYS_BASE))
2062 return false;
2063 } else {
2064 if (!apic_verify(APIC_DEFAULT_PHYS_BASE))
2065 return false;
2066 }
2067
2068 apic_pm_activate();
2069
2070 return true;
2071
2072no_apic:
2073 pr_info("No local APIC present or hardware disabled\n");
2074 return false;
2075}
2076#endif
2077
2078/**
2079 * init_apic_mappings - initialize APIC mappings
2080 */
2081void __init init_apic_mappings(void)
2082{
2083 if (apic_validate_deadline_timer())
2084 pr_info("TSC deadline timer available\n");
2085
2086 if (x2apic_mode)
2087 return;
2088
2089 if (!smp_found_config) {
2090 if (!detect_init_APIC()) {
2091 pr_info("APIC: disable apic facility\n");
2092 apic_disable();
2093 }
2094 num_processors = 1;
2095 }
2096}
2097
2098static __init void apic_set_fixmap(void)
2099{
2100 set_fixmap_nocache(FIX_APIC_BASE, mp_lapic_addr);
2101 apic_mmio_base = APIC_BASE;
2102 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
2103 apic_mmio_base, mp_lapic_addr);
2104 apic_read_boot_cpu_id(false);
2105}
2106
2107void __init register_lapic_address(unsigned long address)
2108{
2109 /* This should only happen once */
2110 WARN_ON_ONCE(mp_lapic_addr);
2111 mp_lapic_addr = address;
2112
2113 if (!x2apic_mode)
2114 apic_set_fixmap();
2115}
2116
2117/*
2118 * Local APIC interrupts
2119 */
2120
2121/*
2122 * Common handling code for spurious_interrupt and spurious_vector entry
2123 * points below. No point in allowing the compiler to inline it twice.
2124 */
2125static noinline void handle_spurious_interrupt(u8 vector)
2126{
2127 u32 v;
2128
2129 trace_spurious_apic_entry(vector);
2130
2131 inc_irq_stat(irq_spurious_count);
2132
2133 /*
2134 * If this is a spurious interrupt then do not acknowledge
2135 */
2136 if (vector == SPURIOUS_APIC_VECTOR) {
2137 /* See SDM vol 3 */
2138 pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n",
2139 smp_processor_id());
2140 goto out;
2141 }
2142
2143 /*
2144 * If it is a vectored one, verify it's set in the ISR. If set,
2145 * acknowledge it.
2146 */
2147 v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2148 if (v & (1 << (vector & 0x1f))) {
2149 pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n",
2150 vector, smp_processor_id());
2151 apic_eoi();
2152 } else {
2153 pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n",
2154 vector, smp_processor_id());
2155 }
2156out:
2157 trace_spurious_apic_exit(vector);
2158}
2159
2160/**
2161 * spurious_interrupt - Catch all for interrupts raised on unused vectors
2162 * @regs: Pointer to pt_regs on stack
2163 * @vector: The vector number
2164 *
2165 * This is invoked from ASM entry code to catch all interrupts which
2166 * trigger on an entry which is routed to the common_spurious idtentry
2167 * point.
2168 */
2169DEFINE_IDTENTRY_IRQ(spurious_interrupt)
2170{
2171 handle_spurious_interrupt(vector);
2172}
2173
2174DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt)
2175{
2176 handle_spurious_interrupt(SPURIOUS_APIC_VECTOR);
2177}
2178
2179/*
2180 * This interrupt should never happen with our APIC/SMP architecture
2181 */
2182DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt)
2183{
2184 static const char * const error_interrupt_reason[] = {
2185 "Send CS error", /* APIC Error Bit 0 */
2186 "Receive CS error", /* APIC Error Bit 1 */
2187 "Send accept error", /* APIC Error Bit 2 */
2188 "Receive accept error", /* APIC Error Bit 3 */
2189 "Redirectable IPI", /* APIC Error Bit 4 */
2190 "Send illegal vector", /* APIC Error Bit 5 */
2191 "Received illegal vector", /* APIC Error Bit 6 */
2192 "Illegal register address", /* APIC Error Bit 7 */
2193 };
2194 u32 v, i = 0;
2195
2196 trace_error_apic_entry(ERROR_APIC_VECTOR);
2197
2198 /* First tickle the hardware, only then report what went on. -- REW */
2199 if (lapic_get_maxlvt() > 3) /* Due to the Pentium erratum 3AP. */
2200 apic_write(APIC_ESR, 0);
2201 v = apic_read(APIC_ESR);
2202 apic_eoi();
2203 atomic_inc(&irq_err_count);
2204
2205 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2206 smp_processor_id(), v);
2207
2208 v &= 0xff;
2209 while (v) {
2210 if (v & 0x1)
2211 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2212 i++;
2213 v >>= 1;
2214 }
2215
2216 apic_printk(APIC_DEBUG, KERN_CONT "\n");
2217
2218 trace_error_apic_exit(ERROR_APIC_VECTOR);
2219}
2220
2221/**
2222 * connect_bsp_APIC - attach the APIC to the interrupt system
2223 */
2224static void __init connect_bsp_APIC(void)
2225{
2226#ifdef CONFIG_X86_32
2227 if (pic_mode) {
2228 /*
2229 * Do not trust the local APIC being empty at bootup.
2230 */
2231 clear_local_APIC();
2232 /*
2233 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
2234 * local APIC to INT and NMI lines.
2235 */
2236 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2237 "enabling APIC mode.\n");
2238 imcr_pic_to_apic();
2239 }
2240#endif
2241}
2242
2243/**
2244 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2245 * @virt_wire_setup: indicates, whether virtual wire mode is selected
2246 *
2247 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2248 * APIC is disabled.
2249 */
2250void disconnect_bsp_APIC(int virt_wire_setup)
2251{
2252 unsigned int value;
2253
2254#ifdef CONFIG_X86_32
2255 if (pic_mode) {
2256 /*
2257 * Put the board back into PIC mode (has an effect only on
2258 * certain older boards). Note that APIC interrupts, including
2259 * IPIs, won't work beyond this point! The only exception are
2260 * INIT IPIs.
2261 */
2262 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2263 "entering PIC mode.\n");
2264 imcr_apic_to_pic();
2265 return;
2266 }
2267#endif
2268
2269 /* Go back to Virtual Wire compatibility mode */
2270
2271 /* For the spurious interrupt use vector F, and enable it */
2272 value = apic_read(APIC_SPIV);
2273 value &= ~APIC_VECTOR_MASK;
2274 value |= APIC_SPIV_APIC_ENABLED;
2275 value |= 0xf;
2276 apic_write(APIC_SPIV, value);
2277
2278 if (!virt_wire_setup) {
2279 /*
2280 * For LVT0 make it edge triggered, active high,
2281 * external and enabled
2282 */
2283 value = apic_read(APIC_LVT0);
2284 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2285 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2286 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2287 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2288 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2289 apic_write(APIC_LVT0, value);
2290 } else {
2291 /* Disable LVT0 */
2292 apic_write(APIC_LVT0, APIC_LVT_MASKED);
2293 }
2294
2295 /*
2296 * For LVT1 make it edge triggered, active high,
2297 * nmi and enabled
2298 */
2299 value = apic_read(APIC_LVT1);
2300 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2301 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2302 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2303 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2304 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2305 apic_write(APIC_LVT1, value);
2306}
2307
2308/*
2309 * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2310 * contiguously, it equals to current allocated max logical CPU ID plus 1.
2311 * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2312 * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2313 *
2314 * NOTE: Reserve 0 for BSP.
2315 */
2316static int nr_logical_cpuids = 1;
2317
2318/*
2319 * Used to store mapping between logical CPU IDs and APIC IDs.
2320 */
2321u32 cpuid_to_apicid[] = { [0 ... NR_CPUS - 1] = BAD_APICID, };
2322
2323bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
2324{
2325 return phys_id == (u64)cpuid_to_apicid[cpu];
2326}
2327
2328#ifdef CONFIG_SMP
2329static void cpu_mark_primary_thread(unsigned int cpu, unsigned int apicid)
2330{
2331 /* Isolate the SMT bit(s) in the APICID and check for 0 */
2332 u32 mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
2333
2334 if (smp_num_siblings == 1 || !(apicid & mask))
2335 cpumask_set_cpu(cpu, &__cpu_primary_thread_mask);
2336}
2337
2338/*
2339 * Due to the utter mess of CPUID evaluation smp_num_siblings is not valid
2340 * during early boot. Initialize the primary thread mask before SMP
2341 * bringup.
2342 */
2343static int __init smp_init_primary_thread_mask(void)
2344{
2345 unsigned int cpu;
2346
2347 /*
2348 * XEN/PV provides either none or useless topology information.
2349 * Pretend that all vCPUs are primary threads.
2350 */
2351 if (xen_pv_domain()) {
2352 cpumask_copy(&__cpu_primary_thread_mask, cpu_possible_mask);
2353 return 0;
2354 }
2355
2356 for (cpu = 0; cpu < nr_logical_cpuids; cpu++)
2357 cpu_mark_primary_thread(cpu, cpuid_to_apicid[cpu]);
2358 return 0;
2359}
2360early_initcall(smp_init_primary_thread_mask);
2361#else
2362static inline void cpu_mark_primary_thread(unsigned int cpu, unsigned int apicid) { }
2363#endif
2364
2365/*
2366 * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2367 * and cpuid_to_apicid[] synchronized.
2368 */
2369static int allocate_logical_cpuid(int apicid)
2370{
2371 int i;
2372
2373 /*
2374 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2375 * check if the kernel has allocated a cpuid for it.
2376 */
2377 for (i = 0; i < nr_logical_cpuids; i++) {
2378 if (cpuid_to_apicid[i] == apicid)
2379 return i;
2380 }
2381
2382 /* Allocate a new cpuid. */
2383 if (nr_logical_cpuids >= nr_cpu_ids) {
2384 WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %u reached. "
2385 "Processor %d/0x%x and the rest are ignored.\n",
2386 nr_cpu_ids, nr_logical_cpuids, apicid);
2387 return -EINVAL;
2388 }
2389
2390 cpuid_to_apicid[nr_logical_cpuids] = apicid;
2391 return nr_logical_cpuids++;
2392}
2393
2394static void cpu_update_apic(int cpu, u32 apicid)
2395{
2396#if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2397 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2398#endif
2399 set_cpu_possible(cpu, true);
2400 physid_set(apicid, phys_cpu_present_map);
2401 set_cpu_present(cpu, true);
2402 num_processors++;
2403
2404 if (system_state != SYSTEM_BOOTING)
2405 cpu_mark_primary_thread(cpu, apicid);
2406}
2407
2408static __init void cpu_set_boot_apic(void)
2409{
2410 cpuid_to_apicid[0] = boot_cpu_physical_apicid;
2411 cpu_update_apic(0, boot_cpu_physical_apicid);
2412 x86_32_probe_bigsmp_early();
2413}
2414
2415int generic_processor_info(int apicid)
2416{
2417 int cpu, max = nr_cpu_ids;
2418
2419 /* The boot CPU must be set before MADT/MPTABLE parsing happens */
2420 if (cpuid_to_apicid[0] == BAD_APICID)
2421 panic("Boot CPU APIC not registered yet\n");
2422
2423 if (apicid == boot_cpu_physical_apicid)
2424 return 0;
2425
2426 if (disabled_cpu_apicid == apicid) {
2427 int thiscpu = num_processors + disabled_cpus;
2428
2429 pr_warn("APIC: Disabling requested cpu. Processor %d/0x%x ignored.\n",
2430 thiscpu, apicid);
2431
2432 disabled_cpus++;
2433 return -ENODEV;
2434 }
2435
2436 if (num_processors >= nr_cpu_ids) {
2437 int thiscpu = max + disabled_cpus;
2438
2439 pr_warn("APIC: NR_CPUS/possible_cpus limit of %i reached. "
2440 "Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2441
2442 disabled_cpus++;
2443 return -EINVAL;
2444 }
2445
2446 cpu = allocate_logical_cpuid(apicid);
2447 if (cpu < 0) {
2448 disabled_cpus++;
2449 return -EINVAL;
2450 }
2451
2452 cpu_update_apic(cpu, apicid);
2453 return cpu;
2454}
2455
2456
2457void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,
2458 bool dmar)
2459{
2460 memset(msg, 0, sizeof(*msg));
2461
2462 msg->arch_addr_lo.base_address = X86_MSI_BASE_ADDRESS_LOW;
2463 msg->arch_addr_lo.dest_mode_logical = apic->dest_mode_logical;
2464 msg->arch_addr_lo.destid_0_7 = cfg->dest_apicid & 0xFF;
2465
2466 msg->arch_data.delivery_mode = APIC_DELIVERY_MODE_FIXED;
2467 msg->arch_data.vector = cfg->vector;
2468
2469 msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH;
2470 /*
2471 * Only the IOMMU itself can use the trick of putting destination
2472 * APIC ID into the high bits of the address. Anything else would
2473 * just be writing to memory if it tried that, and needs IR to
2474 * address APICs which can't be addressed in the normal 32-bit
2475 * address range at 0xFFExxxxx. That is typically just 8 bits, but
2476 * some hypervisors allow the extended destination ID field in bits
2477 * 5-11 to be used, giving support for 15 bits of APIC IDs in total.
2478 */
2479 if (dmar)
2480 msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >> 8;
2481 else if (virt_ext_dest_id && cfg->dest_apicid < 0x8000)
2482 msg->arch_addr_lo.virt_destid_8_14 = cfg->dest_apicid >> 8;
2483 else
2484 WARN_ON_ONCE(cfg->dest_apicid > 0xFF);
2485}
2486
2487u32 x86_msi_msg_get_destid(struct msi_msg *msg, bool extid)
2488{
2489 u32 dest = msg->arch_addr_lo.destid_0_7;
2490
2491 if (extid)
2492 dest |= msg->arch_addr_hi.destid_8_31 << 8;
2493 return dest;
2494}
2495EXPORT_SYMBOL_GPL(x86_msi_msg_get_destid);
2496
2497static void __init apic_bsp_up_setup(void)
2498{
2499#ifdef CONFIG_X86_64
2500 apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2501#endif
2502 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2503}
2504
2505/**
2506 * apic_bsp_setup - Setup function for local apic and io-apic
2507 * @upmode: Force UP mode (for APIC_init_uniprocessor)
2508 */
2509static void __init apic_bsp_setup(bool upmode)
2510{
2511 connect_bsp_APIC();
2512 if (upmode)
2513 apic_bsp_up_setup();
2514 setup_local_APIC();
2515
2516 enable_IO_APIC();
2517 end_local_APIC_setup();
2518 irq_remap_enable_fault_handling();
2519 setup_IO_APIC();
2520 lapic_update_legacy_vectors();
2521}
2522
2523#ifdef CONFIG_UP_LATE_INIT
2524void __init up_late_init(void)
2525{
2526 if (apic_intr_mode == APIC_PIC)
2527 return;
2528
2529 /* Setup local timer */
2530 x86_init.timers.setup_percpu_clockev();
2531}
2532#endif
2533
2534/*
2535 * Power management
2536 */
2537#ifdef CONFIG_PM
2538
2539static struct {
2540 /*
2541 * 'active' is true if the local APIC was enabled by us and
2542 * not the BIOS; this signifies that we are also responsible
2543 * for disabling it before entering apm/acpi suspend
2544 */
2545 int active;
2546 /* r/w apic fields */
2547 u32 apic_id;
2548 unsigned int apic_taskpri;
2549 unsigned int apic_ldr;
2550 unsigned int apic_dfr;
2551 unsigned int apic_spiv;
2552 unsigned int apic_lvtt;
2553 unsigned int apic_lvtpc;
2554 unsigned int apic_lvt0;
2555 unsigned int apic_lvt1;
2556 unsigned int apic_lvterr;
2557 unsigned int apic_tmict;
2558 unsigned int apic_tdcr;
2559 unsigned int apic_thmr;
2560 unsigned int apic_cmci;
2561} apic_pm_state;
2562
2563static int lapic_suspend(void)
2564{
2565 unsigned long flags;
2566 int maxlvt;
2567
2568 if (!apic_pm_state.active)
2569 return 0;
2570
2571 maxlvt = lapic_get_maxlvt();
2572
2573 apic_pm_state.apic_id = apic_read(APIC_ID);
2574 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2575 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2576 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2577 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2578 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2579 if (maxlvt >= 4)
2580 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2581 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2582 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2583 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2584 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2585 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2586#ifdef CONFIG_X86_THERMAL_VECTOR
2587 if (maxlvt >= 5)
2588 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2589#endif
2590#ifdef CONFIG_X86_MCE_INTEL
2591 if (maxlvt >= 6)
2592 apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2593#endif
2594
2595 local_irq_save(flags);
2596
2597 /*
2598 * Mask IOAPIC before disabling the local APIC to prevent stale IRR
2599 * entries on some implementations.
2600 */
2601 mask_ioapic_entries();
2602
2603 disable_local_APIC();
2604
2605 irq_remapping_disable();
2606
2607 local_irq_restore(flags);
2608 return 0;
2609}
2610
2611static void lapic_resume(void)
2612{
2613 unsigned int l, h;
2614 unsigned long flags;
2615 int maxlvt;
2616
2617 if (!apic_pm_state.active)
2618 return;
2619
2620 local_irq_save(flags);
2621
2622 /*
2623 * IO-APIC and PIC have their own resume routines.
2624 * We just mask them here to make sure the interrupt
2625 * subsystem is completely quiet while we enable x2apic
2626 * and interrupt-remapping.
2627 */
2628 mask_ioapic_entries();
2629 legacy_pic->mask_all();
2630
2631 if (x2apic_mode) {
2632 __x2apic_enable();
2633 } else {
2634 /*
2635 * Make sure the APICBASE points to the right address
2636 *
2637 * FIXME! This will be wrong if we ever support suspend on
2638 * SMP! We'll need to do this as part of the CPU restore!
2639 */
2640 if (boot_cpu_data.x86 >= 6) {
2641 rdmsr(MSR_IA32_APICBASE, l, h);
2642 l &= ~MSR_IA32_APICBASE_BASE;
2643 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2644 wrmsr(MSR_IA32_APICBASE, l, h);
2645 }
2646 }
2647
2648 maxlvt = lapic_get_maxlvt();
2649 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2650 apic_write(APIC_ID, apic_pm_state.apic_id);
2651 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2652 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2653 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2654 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2655 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2656 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2657#ifdef CONFIG_X86_THERMAL_VECTOR
2658 if (maxlvt >= 5)
2659 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2660#endif
2661#ifdef CONFIG_X86_MCE_INTEL
2662 if (maxlvt >= 6)
2663 apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2664#endif
2665 if (maxlvt >= 4)
2666 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2667 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2668 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2669 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2670 apic_write(APIC_ESR, 0);
2671 apic_read(APIC_ESR);
2672 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2673 apic_write(APIC_ESR, 0);
2674 apic_read(APIC_ESR);
2675
2676 irq_remapping_reenable(x2apic_mode);
2677
2678 local_irq_restore(flags);
2679}
2680
2681/*
2682 * This device has no shutdown method - fully functioning local APICs
2683 * are needed on every CPU up until machine_halt/restart/poweroff.
2684 */
2685
2686static struct syscore_ops lapic_syscore_ops = {
2687 .resume = lapic_resume,
2688 .suspend = lapic_suspend,
2689};
2690
2691static void apic_pm_activate(void)
2692{
2693 apic_pm_state.active = 1;
2694}
2695
2696static int __init init_lapic_sysfs(void)
2697{
2698 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2699 if (boot_cpu_has(X86_FEATURE_APIC))
2700 register_syscore_ops(&lapic_syscore_ops);
2701
2702 return 0;
2703}
2704
2705/* local apic needs to resume before other devices access its registers. */
2706core_initcall(init_lapic_sysfs);
2707
2708#else /* CONFIG_PM */
2709
2710static void apic_pm_activate(void) { }
2711
2712#endif /* CONFIG_PM */
2713
2714#ifdef CONFIG_X86_64
2715
2716static int multi_checked;
2717static int multi;
2718
2719static int set_multi(const struct dmi_system_id *d)
2720{
2721 if (multi)
2722 return 0;
2723 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2724 multi = 1;
2725 return 0;
2726}
2727
2728static const struct dmi_system_id multi_dmi_table[] = {
2729 {
2730 .callback = set_multi,
2731 .ident = "IBM System Summit2",
2732 .matches = {
2733 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2734 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2735 },
2736 },
2737 {}
2738};
2739
2740static void dmi_check_multi(void)
2741{
2742 if (multi_checked)
2743 return;
2744
2745 dmi_check_system(multi_dmi_table);
2746 multi_checked = 1;
2747}
2748
2749/*
2750 * apic_is_clustered_box() -- Check if we can expect good TSC
2751 *
2752 * Thus far, the major user of this is IBM's Summit2 series:
2753 * Clustered boxes may have unsynced TSC problems if they are
2754 * multi-chassis.
2755 * Use DMI to check them
2756 */
2757int apic_is_clustered_box(void)
2758{
2759 dmi_check_multi();
2760 return multi;
2761}
2762#endif
2763
2764/*
2765 * APIC command line parameters
2766 */
2767static int __init setup_disableapic(char *arg)
2768{
2769 apic_is_disabled = true;
2770 setup_clear_cpu_cap(X86_FEATURE_APIC);
2771 return 0;
2772}
2773early_param("disableapic", setup_disableapic);
2774
2775/* same as disableapic, for compatibility */
2776static int __init setup_nolapic(char *arg)
2777{
2778 return setup_disableapic(arg);
2779}
2780early_param("nolapic", setup_nolapic);
2781
2782static int __init parse_lapic_timer_c2_ok(char *arg)
2783{
2784 local_apic_timer_c2_ok = 1;
2785 return 0;
2786}
2787early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2788
2789static int __init parse_disable_apic_timer(char *arg)
2790{
2791 disable_apic_timer = 1;
2792 return 0;
2793}
2794early_param("noapictimer", parse_disable_apic_timer);
2795
2796static int __init parse_nolapic_timer(char *arg)
2797{
2798 disable_apic_timer = 1;
2799 return 0;
2800}
2801early_param("nolapic_timer", parse_nolapic_timer);
2802
2803static int __init apic_set_verbosity(char *arg)
2804{
2805 if (!arg) {
2806 if (IS_ENABLED(CONFIG_X86_32))
2807 return -EINVAL;
2808
2809 ioapic_is_disabled = false;
2810 return 0;
2811 }
2812
2813 if (strcmp("debug", arg) == 0)
2814 apic_verbosity = APIC_DEBUG;
2815 else if (strcmp("verbose", arg) == 0)
2816 apic_verbosity = APIC_VERBOSE;
2817#ifdef CONFIG_X86_64
2818 else {
2819 pr_warn("APIC Verbosity level %s not recognised"
2820 " use apic=verbose or apic=debug\n", arg);
2821 return -EINVAL;
2822 }
2823#endif
2824
2825 return 0;
2826}
2827early_param("apic", apic_set_verbosity);
2828
2829static int __init lapic_insert_resource(void)
2830{
2831 if (!apic_mmio_base)
2832 return -1;
2833
2834 /* Put local APIC into the resource map. */
2835 lapic_resource.start = apic_mmio_base;
2836 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2837 insert_resource(&iomem_resource, &lapic_resource);
2838
2839 return 0;
2840}
2841
2842/*
2843 * need call insert after e820__reserve_resources()
2844 * that is using request_resource
2845 */
2846late_initcall(lapic_insert_resource);
2847
2848static int __init apic_set_disabled_cpu_apicid(char *arg)
2849{
2850 if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2851 return -EINVAL;
2852
2853 return 0;
2854}
2855early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2856
2857static int __init apic_set_extnmi(char *arg)
2858{
2859 if (!arg)
2860 return -EINVAL;
2861
2862 if (!strncmp("all", arg, 3))
2863 apic_extnmi = APIC_EXTNMI_ALL;
2864 else if (!strncmp("none", arg, 4))
2865 apic_extnmi = APIC_EXTNMI_NONE;
2866 else if (!strncmp("bsp", arg, 3))
2867 apic_extnmi = APIC_EXTNMI_BSP;
2868 else {
2869 pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2870 return -EINVAL;
2871 }
2872
2873 return 0;
2874}
2875early_param("apic_extnmi", apic_set_extnmi);