Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Local APIC handling, local APIC timers
4 *
5 * (c) 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
6 *
7 * Fixes
8 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
9 * thanks to Eric Gilmore
10 * and Rolf G. Tews
11 * for testing these extensively.
12 * Maciej W. Rozycki : Various updates and fixes.
13 * Mikael Pettersson : Power Management for UP-APIC.
14 * Pavel Machek and
15 * Mikael Pettersson : PM converted to driver model.
16 */
17
18#include <linux/perf_event.h>
19#include <linux/kernel_stat.h>
20#include <linux/mc146818rtc.h>
21#include <linux/acpi_pmtmr.h>
22#include <linux/clockchips.h>
23#include <linux/interrupt.h>
24#include <linux/memblock.h>
25#include <linux/ftrace.h>
26#include <linux/ioport.h>
27#include <linux/export.h>
28#include <linux/syscore_ops.h>
29#include <linux/delay.h>
30#include <linux/timex.h>
31#include <linux/i8253.h>
32#include <linux/dmar.h>
33#include <linux/init.h>
34#include <linux/cpu.h>
35#include <linux/dmi.h>
36#include <linux/smp.h>
37#include <linux/mm.h>
38
39#include <asm/trace/irq_vectors.h>
40#include <asm/irq_remapping.h>
41#include <asm/perf_event.h>
42#include <asm/x86_init.h>
43#include <asm/pgalloc.h>
44#include <linux/atomic.h>
45#include <asm/mpspec.h>
46#include <asm/i8259.h>
47#include <asm/proto.h>
48#include <asm/traps.h>
49#include <asm/apic.h>
50#include <asm/io_apic.h>
51#include <asm/desc.h>
52#include <asm/hpet.h>
53#include <asm/mtrr.h>
54#include <asm/time.h>
55#include <asm/smp.h>
56#include <asm/mce.h>
57#include <asm/tsc.h>
58#include <asm/hypervisor.h>
59#include <asm/cpu_device_id.h>
60#include <asm/intel-family.h>
61#include <asm/irq_regs.h>
62
63unsigned int num_processors;
64
65unsigned disabled_cpus;
66
67/* Processor that is doing the boot up */
68unsigned int boot_cpu_physical_apicid __ro_after_init = -1U;
69EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
70
71u8 boot_cpu_apic_version __ro_after_init;
72
73/*
74 * The highest APIC ID seen during enumeration.
75 */
76static unsigned int max_physical_apicid;
77
78/*
79 * Bitmask of physically existing CPUs:
80 */
81physid_mask_t phys_cpu_present_map;
82
83/*
84 * Processor to be disabled specified by kernel parameter
85 * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
86 * avoid undefined behaviour caused by sending INIT from AP to BSP.
87 */
88static unsigned int disabled_cpu_apicid __ro_after_init = BAD_APICID;
89
90/*
91 * This variable controls which CPUs receive external NMIs. By default,
92 * external NMIs are delivered only to the BSP.
93 */
94static int apic_extnmi __ro_after_init = APIC_EXTNMI_BSP;
95
96/*
97 * Map cpu index to physical APIC ID
98 */
99DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
100DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
101DEFINE_EARLY_PER_CPU_READ_MOSTLY(u32, x86_cpu_to_acpiid, U32_MAX);
102EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
103EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
104EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_acpiid);
105
106#ifdef CONFIG_X86_32
107
108/*
109 * On x86_32, the mapping between cpu and logical apicid may vary
110 * depending on apic in use. The following early percpu variable is
111 * used for the mapping. This is where the behaviors of x86_64 and 32
112 * actually diverge. Let's keep it ugly for now.
113 */
114DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
115
116/* Local APIC was disabled by the BIOS and enabled by the kernel */
117static int enabled_via_apicbase __ro_after_init;
118
119/*
120 * Handle interrupt mode configuration register (IMCR).
121 * This register controls whether the interrupt signals
122 * that reach the BSP come from the master PIC or from the
123 * local APIC. Before entering Symmetric I/O Mode, either
124 * the BIOS or the operating system must switch out of
125 * PIC Mode by changing the IMCR.
126 */
127static inline void imcr_pic_to_apic(void)
128{
129 /* select IMCR register */
130 outb(0x70, 0x22);
131 /* NMI and 8259 INTR go through APIC */
132 outb(0x01, 0x23);
133}
134
135static inline void imcr_apic_to_pic(void)
136{
137 /* select IMCR register */
138 outb(0x70, 0x22);
139 /* NMI and 8259 INTR go directly to BSP */
140 outb(0x00, 0x23);
141}
142#endif
143
144/*
145 * Knob to control our willingness to enable the local APIC.
146 *
147 * +1=force-enable
148 */
149static int force_enable_local_apic __initdata;
150
151/*
152 * APIC command line parameters
153 */
154static int __init parse_lapic(char *arg)
155{
156 if (IS_ENABLED(CONFIG_X86_32) && !arg)
157 force_enable_local_apic = 1;
158 else if (arg && !strncmp(arg, "notscdeadline", 13))
159 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
160 return 0;
161}
162early_param("lapic", parse_lapic);
163
164#ifdef CONFIG_X86_64
165static int apic_calibrate_pmtmr __initdata;
166static __init int setup_apicpmtimer(char *s)
167{
168 apic_calibrate_pmtmr = 1;
169 notsc_setup(NULL);
170 return 0;
171}
172__setup("apicpmtimer", setup_apicpmtimer);
173#endif
174
175unsigned long mp_lapic_addr __ro_after_init;
176int disable_apic __ro_after_init;
177/* Disable local APIC timer from the kernel commandline or via dmi quirk */
178static int disable_apic_timer __initdata;
179/* Local APIC timer works in C2 */
180int local_apic_timer_c2_ok __ro_after_init;
181EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
182
183/*
184 * Debug level, exported for io_apic.c
185 */
186int apic_verbosity __ro_after_init;
187
188int pic_mode __ro_after_init;
189
190/* Have we found an MP table */
191int smp_found_config __ro_after_init;
192
193static struct resource lapic_resource = {
194 .name = "Local APIC",
195 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
196};
197
198unsigned int lapic_timer_period = 0;
199
200static void apic_pm_activate(void);
201
202static unsigned long apic_phys __ro_after_init;
203
204/*
205 * Get the LAPIC version
206 */
207static inline int lapic_get_version(void)
208{
209 return GET_APIC_VERSION(apic_read(APIC_LVR));
210}
211
212/*
213 * Check, if the APIC is integrated or a separate chip
214 */
215static inline int lapic_is_integrated(void)
216{
217 return APIC_INTEGRATED(lapic_get_version());
218}
219
220/*
221 * Check, whether this is a modern or a first generation APIC
222 */
223static int modern_apic(void)
224{
225 /* AMD systems use old APIC versions, so check the CPU */
226 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
227 boot_cpu_data.x86 >= 0xf)
228 return 1;
229
230 /* Hygon systems use modern APIC */
231 if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
232 return 1;
233
234 return lapic_get_version() >= 0x14;
235}
236
237/*
238 * right after this call apic become NOOP driven
239 * so apic->write/read doesn't do anything
240 */
241static void __init apic_disable(void)
242{
243 pr_info("APIC: switched to apic NOOP\n");
244 apic = &apic_noop;
245}
246
247void native_apic_wait_icr_idle(void)
248{
249 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
250 cpu_relax();
251}
252
253u32 native_safe_apic_wait_icr_idle(void)
254{
255 u32 send_status;
256 int timeout;
257
258 timeout = 0;
259 do {
260 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
261 if (!send_status)
262 break;
263 inc_irq_stat(icr_read_retry_count);
264 udelay(100);
265 } while (timeout++ < 1000);
266
267 return send_status;
268}
269
270void native_apic_icr_write(u32 low, u32 id)
271{
272 unsigned long flags;
273
274 local_irq_save(flags);
275 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
276 apic_write(APIC_ICR, low);
277 local_irq_restore(flags);
278}
279
280u64 native_apic_icr_read(void)
281{
282 u32 icr1, icr2;
283
284 icr2 = apic_read(APIC_ICR2);
285 icr1 = apic_read(APIC_ICR);
286
287 return icr1 | ((u64)icr2 << 32);
288}
289
290#ifdef CONFIG_X86_32
291/**
292 * get_physical_broadcast - Get number of physical broadcast IDs
293 */
294int get_physical_broadcast(void)
295{
296 return modern_apic() ? 0xff : 0xf;
297}
298#endif
299
300/**
301 * lapic_get_maxlvt - get the maximum number of local vector table entries
302 */
303int lapic_get_maxlvt(void)
304{
305 /*
306 * - we always have APIC integrated on 64bit mode
307 * - 82489DXs do not report # of LVT entries
308 */
309 return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
310}
311
312/*
313 * Local APIC timer
314 */
315
316/* Clock divisor */
317#define APIC_DIVISOR 16
318#define TSC_DIVISOR 8
319
320/*
321 * This function sets up the local APIC timer, with a timeout of
322 * 'clocks' APIC bus clock. During calibration we actually call
323 * this function twice on the boot CPU, once with a bogus timeout
324 * value, second time for real. The other (noncalibrating) CPUs
325 * call this function only once, with the real, calibrated value.
326 *
327 * We do reads before writes even if unnecessary, to get around the
328 * P5 APIC double write bug.
329 */
330static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
331{
332 unsigned int lvtt_value, tmp_value;
333
334 lvtt_value = LOCAL_TIMER_VECTOR;
335 if (!oneshot)
336 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
337 else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
338 lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
339
340 if (!lapic_is_integrated())
341 lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
342
343 if (!irqen)
344 lvtt_value |= APIC_LVT_MASKED;
345
346 apic_write(APIC_LVTT, lvtt_value);
347
348 if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
349 /*
350 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
351 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
352 * According to Intel, MFENCE can do the serialization here.
353 */
354 asm volatile("mfence" : : : "memory");
355
356 printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
357 return;
358 }
359
360 /*
361 * Divide PICLK by 16
362 */
363 tmp_value = apic_read(APIC_TDCR);
364 apic_write(APIC_TDCR,
365 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
366 APIC_TDR_DIV_16);
367
368 if (!oneshot)
369 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
370}
371
372/*
373 * Setup extended LVT, AMD specific
374 *
375 * Software should use the LVT offsets the BIOS provides. The offsets
376 * are determined by the subsystems using it like those for MCE
377 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts
378 * are supported. Beginning with family 10h at least 4 offsets are
379 * available.
380 *
381 * Since the offsets must be consistent for all cores, we keep track
382 * of the LVT offsets in software and reserve the offset for the same
383 * vector also to be used on other cores. An offset is freed by
384 * setting the entry to APIC_EILVT_MASKED.
385 *
386 * If the BIOS is right, there should be no conflicts. Otherwise a
387 * "[Firmware Bug]: ..." error message is generated. However, if
388 * software does not properly determines the offsets, it is not
389 * necessarily a BIOS bug.
390 */
391
392static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
393
394static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
395{
396 return (old & APIC_EILVT_MASKED)
397 || (new == APIC_EILVT_MASKED)
398 || ((new & ~APIC_EILVT_MASKED) == old);
399}
400
401static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
402{
403 unsigned int rsvd, vector;
404
405 if (offset >= APIC_EILVT_NR_MAX)
406 return ~0;
407
408 rsvd = atomic_read(&eilvt_offsets[offset]);
409 do {
410 vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */
411 if (vector && !eilvt_entry_is_changeable(vector, new))
412 /* may not change if vectors are different */
413 return rsvd;
414 rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
415 } while (rsvd != new);
416
417 rsvd &= ~APIC_EILVT_MASKED;
418 if (rsvd && rsvd != vector)
419 pr_info("LVT offset %d assigned for vector 0x%02x\n",
420 offset, rsvd);
421
422 return new;
423}
424
425/*
426 * If mask=1, the LVT entry does not generate interrupts while mask=0
427 * enables the vector. See also the BKDGs. Must be called with
428 * preemption disabled.
429 */
430
431int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
432{
433 unsigned long reg = APIC_EILVTn(offset);
434 unsigned int new, old, reserved;
435
436 new = (mask << 16) | (msg_type << 8) | vector;
437 old = apic_read(reg);
438 reserved = reserve_eilvt_offset(offset, new);
439
440 if (reserved != 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 another cpu\n",
444 smp_processor_id(), reg, offset, new, reserved);
445 return -EINVAL;
446 }
447
448 if (!eilvt_entry_is_changeable(old, new)) {
449 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
450 "vector 0x%x, but the register is already in use for "
451 "vector 0x%x on this cpu\n",
452 smp_processor_id(), reg, offset, new, old);
453 return -EBUSY;
454 }
455
456 apic_write(reg, new);
457
458 return 0;
459}
460EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
461
462/*
463 * Program the next event, relative to now
464 */
465static int lapic_next_event(unsigned long delta,
466 struct clock_event_device *evt)
467{
468 apic_write(APIC_TMICT, delta);
469 return 0;
470}
471
472static int lapic_next_deadline(unsigned long delta,
473 struct clock_event_device *evt)
474{
475 u64 tsc;
476
477 tsc = rdtsc();
478 wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
479 return 0;
480}
481
482static int lapic_timer_shutdown(struct clock_event_device *evt)
483{
484 unsigned int v;
485
486 /* Lapic used as dummy for broadcast ? */
487 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
488 return 0;
489
490 v = apic_read(APIC_LVTT);
491 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
492 apic_write(APIC_LVTT, v);
493 apic_write(APIC_TMICT, 0);
494 return 0;
495}
496
497static inline int
498lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
499{
500 /* Lapic used as dummy for broadcast ? */
501 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
502 return 0;
503
504 __setup_APIC_LVTT(lapic_timer_period, oneshot, 1);
505 return 0;
506}
507
508static int lapic_timer_set_periodic(struct clock_event_device *evt)
509{
510 return lapic_timer_set_periodic_oneshot(evt, false);
511}
512
513static int lapic_timer_set_oneshot(struct clock_event_device *evt)
514{
515 return lapic_timer_set_periodic_oneshot(evt, true);
516}
517
518/*
519 * Local APIC timer broadcast function
520 */
521static void lapic_timer_broadcast(const struct cpumask *mask)
522{
523#ifdef CONFIG_SMP
524 apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
525#endif
526}
527
528
529/*
530 * The local apic timer can be used for any function which is CPU local.
531 */
532static struct clock_event_device lapic_clockevent = {
533 .name = "lapic",
534 .features = CLOCK_EVT_FEAT_PERIODIC |
535 CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
536 | CLOCK_EVT_FEAT_DUMMY,
537 .shift = 32,
538 .set_state_shutdown = lapic_timer_shutdown,
539 .set_state_periodic = lapic_timer_set_periodic,
540 .set_state_oneshot = lapic_timer_set_oneshot,
541 .set_state_oneshot_stopped = lapic_timer_shutdown,
542 .set_next_event = lapic_next_event,
543 .broadcast = lapic_timer_broadcast,
544 .rating = 100,
545 .irq = -1,
546};
547static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
548
549#define DEADLINE_MODEL_MATCH_FUNC(model, func) \
550 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&func }
551
552#define DEADLINE_MODEL_MATCH_REV(model, rev) \
553 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)rev }
554
555static u32 hsx_deadline_rev(void)
556{
557 switch (boot_cpu_data.x86_stepping) {
558 case 0x02: return 0x3a; /* EP */
559 case 0x04: return 0x0f; /* EX */
560 }
561
562 return ~0U;
563}
564
565static u32 bdx_deadline_rev(void)
566{
567 switch (boot_cpu_data.x86_stepping) {
568 case 0x02: return 0x00000011;
569 case 0x03: return 0x0700000e;
570 case 0x04: return 0x0f00000c;
571 case 0x05: return 0x0e000003;
572 }
573
574 return ~0U;
575}
576
577static u32 skx_deadline_rev(void)
578{
579 switch (boot_cpu_data.x86_stepping) {
580 case 0x03: return 0x01000136;
581 case 0x04: return 0x02000014;
582 }
583
584 if (boot_cpu_data.x86_stepping > 4)
585 return 0;
586
587 return ~0U;
588}
589
590static const struct x86_cpu_id deadline_match[] = {
591 DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_HASWELL_X, hsx_deadline_rev),
592 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_X, 0x0b000020),
593 DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_BROADWELL_D, bdx_deadline_rev),
594 DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_SKYLAKE_X, skx_deadline_rev),
595
596 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL, 0x22),
597 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_L, 0x20),
598 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_G, 0x17),
599
600 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL, 0x25),
601 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_G, 0x17),
602
603 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_L, 0xb2),
604 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE, 0xb2),
605
606 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_L, 0x52),
607 DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE, 0x52),
608
609 {},
610};
611
612static void apic_check_deadline_errata(void)
613{
614 const struct x86_cpu_id *m;
615 u32 rev;
616
617 if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER) ||
618 boot_cpu_has(X86_FEATURE_HYPERVISOR))
619 return;
620
621 m = x86_match_cpu(deadline_match);
622 if (!m)
623 return;
624
625 /*
626 * Function pointers will have the MSB set due to address layout,
627 * immediate revisions will not.
628 */
629 if ((long)m->driver_data < 0)
630 rev = ((u32 (*)(void))(m->driver_data))();
631 else
632 rev = (u32)m->driver_data;
633
634 if (boot_cpu_data.microcode >= rev)
635 return;
636
637 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
638 pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
639 "please update microcode to version: 0x%x (or later)\n", rev);
640}
641
642/*
643 * Setup the local APIC timer for this CPU. Copy the initialized values
644 * of the boot CPU and register the clock event in the framework.
645 */
646static void setup_APIC_timer(void)
647{
648 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
649
650 if (this_cpu_has(X86_FEATURE_ARAT)) {
651 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
652 /* Make LAPIC timer preferrable over percpu HPET */
653 lapic_clockevent.rating = 150;
654 }
655
656 memcpy(levt, &lapic_clockevent, sizeof(*levt));
657 levt->cpumask = cpumask_of(smp_processor_id());
658
659 if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
660 levt->name = "lapic-deadline";
661 levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
662 CLOCK_EVT_FEAT_DUMMY);
663 levt->set_next_event = lapic_next_deadline;
664 clockevents_config_and_register(levt,
665 tsc_khz * (1000 / TSC_DIVISOR),
666 0xF, ~0UL);
667 } else
668 clockevents_register_device(levt);
669}
670
671/*
672 * Install the updated TSC frequency from recalibration at the TSC
673 * deadline clockevent devices.
674 */
675static void __lapic_update_tsc_freq(void *info)
676{
677 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
678
679 if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
680 return;
681
682 clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
683}
684
685void lapic_update_tsc_freq(void)
686{
687 /*
688 * The clockevent device's ->mult and ->shift can both be
689 * changed. In order to avoid races, schedule the frequency
690 * update code on each CPU.
691 */
692 on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
693}
694
695/*
696 * In this functions we calibrate APIC bus clocks to the external timer.
697 *
698 * We want to do the calibration only once since we want to have local timer
699 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
700 * frequency.
701 *
702 * This was previously done by reading the PIT/HPET and waiting for a wrap
703 * around to find out, that a tick has elapsed. I have a box, where the PIT
704 * readout is broken, so it never gets out of the wait loop again. This was
705 * also reported by others.
706 *
707 * Monitoring the jiffies value is inaccurate and the clockevents
708 * infrastructure allows us to do a simple substitution of the interrupt
709 * handler.
710 *
711 * The calibration routine also uses the pm_timer when possible, as the PIT
712 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
713 * back to normal later in the boot process).
714 */
715
716#define LAPIC_CAL_LOOPS (HZ/10)
717
718static __initdata int lapic_cal_loops = -1;
719static __initdata long lapic_cal_t1, lapic_cal_t2;
720static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
721static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
722static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
723
724/*
725 * Temporary interrupt handler and polled calibration function.
726 */
727static void __init lapic_cal_handler(struct clock_event_device *dev)
728{
729 unsigned long long tsc = 0;
730 long tapic = apic_read(APIC_TMCCT);
731 unsigned long pm = acpi_pm_read_early();
732
733 if (boot_cpu_has(X86_FEATURE_TSC))
734 tsc = rdtsc();
735
736 switch (lapic_cal_loops++) {
737 case 0:
738 lapic_cal_t1 = tapic;
739 lapic_cal_tsc1 = tsc;
740 lapic_cal_pm1 = pm;
741 lapic_cal_j1 = jiffies;
742 break;
743
744 case LAPIC_CAL_LOOPS:
745 lapic_cal_t2 = tapic;
746 lapic_cal_tsc2 = tsc;
747 if (pm < lapic_cal_pm1)
748 pm += ACPI_PM_OVRRUN;
749 lapic_cal_pm2 = pm;
750 lapic_cal_j2 = jiffies;
751 break;
752 }
753}
754
755static int __init
756calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
757{
758 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
759 const long pm_thresh = pm_100ms / 100;
760 unsigned long mult;
761 u64 res;
762
763#ifndef CONFIG_X86_PM_TIMER
764 return -1;
765#endif
766
767 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
768
769 /* Check, if the PM timer is available */
770 if (!deltapm)
771 return -1;
772
773 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
774
775 if (deltapm > (pm_100ms - pm_thresh) &&
776 deltapm < (pm_100ms + pm_thresh)) {
777 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
778 return 0;
779 }
780
781 res = (((u64)deltapm) * mult) >> 22;
782 do_div(res, 1000000);
783 pr_warning("APIC calibration not consistent "
784 "with PM-Timer: %ldms instead of 100ms\n",(long)res);
785
786 /* Correct the lapic counter value */
787 res = (((u64)(*delta)) * pm_100ms);
788 do_div(res, deltapm);
789 pr_info("APIC delta adjusted to PM-Timer: "
790 "%lu (%ld)\n", (unsigned long)res, *delta);
791 *delta = (long)res;
792
793 /* Correct the tsc counter value */
794 if (boot_cpu_has(X86_FEATURE_TSC)) {
795 res = (((u64)(*deltatsc)) * pm_100ms);
796 do_div(res, deltapm);
797 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
798 "PM-Timer: %lu (%ld)\n",
799 (unsigned long)res, *deltatsc);
800 *deltatsc = (long)res;
801 }
802
803 return 0;
804}
805
806static int __init lapic_init_clockevent(void)
807{
808 if (!lapic_timer_period)
809 return -1;
810
811 /* Calculate the scaled math multiplication factor */
812 lapic_clockevent.mult = div_sc(lapic_timer_period/APIC_DIVISOR,
813 TICK_NSEC, lapic_clockevent.shift);
814 lapic_clockevent.max_delta_ns =
815 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
816 lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
817 lapic_clockevent.min_delta_ns =
818 clockevent_delta2ns(0xF, &lapic_clockevent);
819 lapic_clockevent.min_delta_ticks = 0xF;
820
821 return 0;
822}
823
824bool __init apic_needs_pit(void)
825{
826 /*
827 * If the frequencies are not known, PIT is required for both TSC
828 * and apic timer calibration.
829 */
830 if (!tsc_khz || !cpu_khz)
831 return true;
832
833 /* Is there an APIC at all? */
834 if (!boot_cpu_has(X86_FEATURE_APIC))
835 return true;
836
837 /* Virt guests may lack ARAT, but still have DEADLINE */
838 if (!boot_cpu_has(X86_FEATURE_ARAT))
839 return true;
840
841 /* Deadline timer is based on TSC so no further PIT action required */
842 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
843 return false;
844
845 /* APIC timer disabled? */
846 if (disable_apic_timer)
847 return true;
848 /*
849 * The APIC timer frequency is known already, no PIT calibration
850 * required. If unknown, let the PIT be initialized.
851 */
852 return lapic_timer_period == 0;
853}
854
855static int __init calibrate_APIC_clock(void)
856{
857 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
858 u64 tsc_perj = 0, tsc_start = 0;
859 unsigned long jif_start;
860 unsigned long deltaj;
861 long delta, deltatsc;
862 int pm_referenced = 0;
863
864 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
865 return 0;
866
867 /*
868 * Check if lapic timer has already been calibrated by platform
869 * specific routine, such as tsc calibration code. If so just fill
870 * in the clockevent structure and return.
871 */
872 if (!lapic_init_clockevent()) {
873 apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
874 lapic_timer_period);
875 /*
876 * Direct calibration methods must have an always running
877 * local APIC timer, no need for broadcast timer.
878 */
879 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
880 return 0;
881 }
882
883 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
884 "calibrating APIC timer ...\n");
885
886 /*
887 * There are platforms w/o global clockevent devices. Instead of
888 * making the calibration conditional on that, use a polling based
889 * approach everywhere.
890 */
891 local_irq_disable();
892
893 /*
894 * Setup the APIC counter to maximum. There is no way the lapic
895 * can underflow in the 100ms detection time frame
896 */
897 __setup_APIC_LVTT(0xffffffff, 0, 0);
898
899 /*
900 * Methods to terminate the calibration loop:
901 * 1) Global clockevent if available (jiffies)
902 * 2) TSC if available and frequency is known
903 */
904 jif_start = READ_ONCE(jiffies);
905
906 if (tsc_khz) {
907 tsc_start = rdtsc();
908 tsc_perj = div_u64((u64)tsc_khz * 1000, HZ);
909 }
910
911 /*
912 * Enable interrupts so the tick can fire, if a global
913 * clockevent device is available
914 */
915 local_irq_enable();
916
917 while (lapic_cal_loops <= LAPIC_CAL_LOOPS) {
918 /* Wait for a tick to elapse */
919 while (1) {
920 if (tsc_khz) {
921 u64 tsc_now = rdtsc();
922 if ((tsc_now - tsc_start) >= tsc_perj) {
923 tsc_start += tsc_perj;
924 break;
925 }
926 } else {
927 unsigned long jif_now = READ_ONCE(jiffies);
928
929 if (time_after(jif_now, jif_start)) {
930 jif_start = jif_now;
931 break;
932 }
933 }
934 cpu_relax();
935 }
936
937 /* Invoke the calibration routine */
938 local_irq_disable();
939 lapic_cal_handler(NULL);
940 local_irq_enable();
941 }
942
943 local_irq_disable();
944
945 /* Build delta t1-t2 as apic timer counts down */
946 delta = lapic_cal_t1 - lapic_cal_t2;
947 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
948
949 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
950
951 /* we trust the PM based calibration if possible */
952 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
953 &delta, &deltatsc);
954
955 lapic_timer_period = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
956 lapic_init_clockevent();
957
958 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
959 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
960 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
961 lapic_timer_period);
962
963 if (boot_cpu_has(X86_FEATURE_TSC)) {
964 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
965 "%ld.%04ld MHz.\n",
966 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
967 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
968 }
969
970 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
971 "%u.%04u MHz.\n",
972 lapic_timer_period / (1000000 / HZ),
973 lapic_timer_period % (1000000 / HZ));
974
975 /*
976 * Do a sanity check on the APIC calibration result
977 */
978 if (lapic_timer_period < (1000000 / HZ)) {
979 local_irq_enable();
980 pr_warning("APIC frequency too slow, disabling apic timer\n");
981 return -1;
982 }
983
984 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
985
986 /*
987 * PM timer calibration failed or not turned on so lets try APIC
988 * timer based calibration, if a global clockevent device is
989 * available.
990 */
991 if (!pm_referenced && global_clock_event) {
992 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
993
994 /*
995 * Setup the apic timer manually
996 */
997 levt->event_handler = lapic_cal_handler;
998 lapic_timer_set_periodic(levt);
999 lapic_cal_loops = -1;
1000
1001 /* Let the interrupts run */
1002 local_irq_enable();
1003
1004 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
1005 cpu_relax();
1006
1007 /* Stop the lapic timer */
1008 local_irq_disable();
1009 lapic_timer_shutdown(levt);
1010
1011 /* Jiffies delta */
1012 deltaj = lapic_cal_j2 - lapic_cal_j1;
1013 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
1014
1015 /* Check, if the jiffies result is consistent */
1016 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
1017 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
1018 else
1019 levt->features |= CLOCK_EVT_FEAT_DUMMY;
1020 }
1021 local_irq_enable();
1022
1023 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
1024 pr_warning("APIC timer disabled due to verification failure\n");
1025 return -1;
1026 }
1027
1028 return 0;
1029}
1030
1031/*
1032 * Setup the boot APIC
1033 *
1034 * Calibrate and verify the result.
1035 */
1036void __init setup_boot_APIC_clock(void)
1037{
1038 /*
1039 * The local apic timer can be disabled via the kernel
1040 * commandline or from the CPU detection code. Register the lapic
1041 * timer as a dummy clock event source on SMP systems, so the
1042 * broadcast mechanism is used. On UP systems simply ignore it.
1043 */
1044 if (disable_apic_timer) {
1045 pr_info("Disabling APIC timer\n");
1046 /* No broadcast on UP ! */
1047 if (num_possible_cpus() > 1) {
1048 lapic_clockevent.mult = 1;
1049 setup_APIC_timer();
1050 }
1051 return;
1052 }
1053
1054 if (calibrate_APIC_clock()) {
1055 /* No broadcast on UP ! */
1056 if (num_possible_cpus() > 1)
1057 setup_APIC_timer();
1058 return;
1059 }
1060
1061 /*
1062 * If nmi_watchdog is set to IO_APIC, we need the
1063 * PIT/HPET going. Otherwise register lapic as a dummy
1064 * device.
1065 */
1066 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
1067
1068 /* Setup the lapic or request the broadcast */
1069 setup_APIC_timer();
1070 amd_e400_c1e_apic_setup();
1071}
1072
1073void setup_secondary_APIC_clock(void)
1074{
1075 setup_APIC_timer();
1076 amd_e400_c1e_apic_setup();
1077}
1078
1079/*
1080 * The guts of the apic timer interrupt
1081 */
1082static void local_apic_timer_interrupt(void)
1083{
1084 struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1085
1086 /*
1087 * Normally we should not be here till LAPIC has been initialized but
1088 * in some cases like kdump, its possible that there is a pending LAPIC
1089 * timer interrupt from previous kernel's context and is delivered in
1090 * new kernel the moment interrupts are enabled.
1091 *
1092 * Interrupts are enabled early and LAPIC is setup much later, hence
1093 * its possible that when we get here evt->event_handler is NULL.
1094 * Check for event_handler being NULL and discard the interrupt as
1095 * spurious.
1096 */
1097 if (!evt->event_handler) {
1098 pr_warning("Spurious LAPIC timer interrupt on cpu %d\n",
1099 smp_processor_id());
1100 /* Switch it off */
1101 lapic_timer_shutdown(evt);
1102 return;
1103 }
1104
1105 /*
1106 * the NMI deadlock-detector uses this.
1107 */
1108 inc_irq_stat(apic_timer_irqs);
1109
1110 evt->event_handler(evt);
1111}
1112
1113/*
1114 * Local APIC timer interrupt. This is the most natural way for doing
1115 * local interrupts, but local timer interrupts can be emulated by
1116 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1117 *
1118 * [ if a single-CPU system runs an SMP kernel then we call the local
1119 * interrupt as well. Thus we cannot inline the local irq ... ]
1120 */
1121__visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
1122{
1123 struct pt_regs *old_regs = set_irq_regs(regs);
1124
1125 /*
1126 * NOTE! We'd better ACK the irq immediately,
1127 * because timer handling can be slow.
1128 *
1129 * update_process_times() expects us to have done irq_enter().
1130 * Besides, if we don't timer interrupts ignore the global
1131 * interrupt lock, which is the WrongThing (tm) to do.
1132 */
1133 entering_ack_irq();
1134 trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1135 local_apic_timer_interrupt();
1136 trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1137 exiting_irq();
1138
1139 set_irq_regs(old_regs);
1140}
1141
1142int setup_profiling_timer(unsigned int multiplier)
1143{
1144 return -EINVAL;
1145}
1146
1147/*
1148 * Local APIC start and shutdown
1149 */
1150
1151/**
1152 * clear_local_APIC - shutdown the local APIC
1153 *
1154 * This is called, when a CPU is disabled and before rebooting, so the state of
1155 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1156 * leftovers during boot.
1157 */
1158void clear_local_APIC(void)
1159{
1160 int maxlvt;
1161 u32 v;
1162
1163 /* APIC hasn't been mapped yet */
1164 if (!x2apic_mode && !apic_phys)
1165 return;
1166
1167 maxlvt = lapic_get_maxlvt();
1168 /*
1169 * Masking an LVT entry can trigger a local APIC error
1170 * if the vector is zero. Mask LVTERR first to prevent this.
1171 */
1172 if (maxlvt >= 3) {
1173 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1174 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1175 }
1176 /*
1177 * Careful: we have to set masks only first to deassert
1178 * any level-triggered sources.
1179 */
1180 v = apic_read(APIC_LVTT);
1181 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1182 v = apic_read(APIC_LVT0);
1183 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1184 v = apic_read(APIC_LVT1);
1185 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1186 if (maxlvt >= 4) {
1187 v = apic_read(APIC_LVTPC);
1188 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1189 }
1190
1191 /* lets not touch this if we didn't frob it */
1192#ifdef CONFIG_X86_THERMAL_VECTOR
1193 if (maxlvt >= 5) {
1194 v = apic_read(APIC_LVTTHMR);
1195 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1196 }
1197#endif
1198#ifdef CONFIG_X86_MCE_INTEL
1199 if (maxlvt >= 6) {
1200 v = apic_read(APIC_LVTCMCI);
1201 if (!(v & APIC_LVT_MASKED))
1202 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1203 }
1204#endif
1205
1206 /*
1207 * Clean APIC state for other OSs:
1208 */
1209 apic_write(APIC_LVTT, APIC_LVT_MASKED);
1210 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1211 apic_write(APIC_LVT1, APIC_LVT_MASKED);
1212 if (maxlvt >= 3)
1213 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1214 if (maxlvt >= 4)
1215 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1216
1217 /* Integrated APIC (!82489DX) ? */
1218 if (lapic_is_integrated()) {
1219 if (maxlvt > 3)
1220 /* Clear ESR due to Pentium errata 3AP and 11AP */
1221 apic_write(APIC_ESR, 0);
1222 apic_read(APIC_ESR);
1223 }
1224}
1225
1226/**
1227 * apic_soft_disable - Clears and software disables the local APIC on hotplug
1228 *
1229 * Contrary to disable_local_APIC() this does not touch the enable bit in
1230 * MSR_IA32_APICBASE. Clearing that bit on systems based on the 3 wire APIC
1231 * bus would require a hardware reset as the APIC would lose track of bus
1232 * arbitration. On systems with FSB delivery APICBASE could be disabled,
1233 * but it has to be guaranteed that no interrupt is sent to the APIC while
1234 * in that state and it's not clear from the SDM whether it still responds
1235 * to INIT/SIPI messages. Stay on the safe side and use software disable.
1236 */
1237void apic_soft_disable(void)
1238{
1239 u32 value;
1240
1241 clear_local_APIC();
1242
1243 /* Soft disable APIC (implies clearing of registers for 82489DX!). */
1244 value = apic_read(APIC_SPIV);
1245 value &= ~APIC_SPIV_APIC_ENABLED;
1246 apic_write(APIC_SPIV, value);
1247}
1248
1249/**
1250 * disable_local_APIC - clear and disable the local APIC
1251 */
1252void disable_local_APIC(void)
1253{
1254 /* APIC hasn't been mapped yet */
1255 if (!x2apic_mode && !apic_phys)
1256 return;
1257
1258 apic_soft_disable();
1259
1260#ifdef CONFIG_X86_32
1261 /*
1262 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1263 * restore the disabled state.
1264 */
1265 if (enabled_via_apicbase) {
1266 unsigned int l, h;
1267
1268 rdmsr(MSR_IA32_APICBASE, l, h);
1269 l &= ~MSR_IA32_APICBASE_ENABLE;
1270 wrmsr(MSR_IA32_APICBASE, l, h);
1271 }
1272#endif
1273}
1274
1275/*
1276 * If Linux enabled the LAPIC against the BIOS default disable it down before
1277 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1278 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1279 * for the case where Linux didn't enable the LAPIC.
1280 */
1281void lapic_shutdown(void)
1282{
1283 unsigned long flags;
1284
1285 if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1286 return;
1287
1288 local_irq_save(flags);
1289
1290#ifdef CONFIG_X86_32
1291 if (!enabled_via_apicbase)
1292 clear_local_APIC();
1293 else
1294#endif
1295 disable_local_APIC();
1296
1297
1298 local_irq_restore(flags);
1299}
1300
1301/**
1302 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1303 */
1304void __init sync_Arb_IDs(void)
1305{
1306 /*
1307 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1308 * needed on AMD.
1309 */
1310 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1311 return;
1312
1313 /*
1314 * Wait for idle.
1315 */
1316 apic_wait_icr_idle();
1317
1318 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1319 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1320 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1321}
1322
1323enum apic_intr_mode_id apic_intr_mode __ro_after_init;
1324
1325static int __init apic_intr_mode_select(void)
1326{
1327 /* Check kernel option */
1328 if (disable_apic) {
1329 pr_info("APIC disabled via kernel command line\n");
1330 return APIC_PIC;
1331 }
1332
1333 /* Check BIOS */
1334#ifdef CONFIG_X86_64
1335 /* On 64-bit, the APIC must be integrated, Check local APIC only */
1336 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1337 disable_apic = 1;
1338 pr_info("APIC disabled by BIOS\n");
1339 return APIC_PIC;
1340 }
1341#else
1342 /* On 32-bit, the APIC may be integrated APIC or 82489DX */
1343
1344 /* Neither 82489DX nor integrated APIC ? */
1345 if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1346 disable_apic = 1;
1347 return APIC_PIC;
1348 }
1349
1350 /* If the BIOS pretends there is an integrated APIC ? */
1351 if (!boot_cpu_has(X86_FEATURE_APIC) &&
1352 APIC_INTEGRATED(boot_cpu_apic_version)) {
1353 disable_apic = 1;
1354 pr_err(FW_BUG "Local APIC %d not detected, force emulation\n",
1355 boot_cpu_physical_apicid);
1356 return APIC_PIC;
1357 }
1358#endif
1359
1360 /* Check MP table or ACPI MADT configuration */
1361 if (!smp_found_config) {
1362 disable_ioapic_support();
1363 if (!acpi_lapic) {
1364 pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1365 return APIC_VIRTUAL_WIRE_NO_CONFIG;
1366 }
1367 return APIC_VIRTUAL_WIRE;
1368 }
1369
1370#ifdef CONFIG_SMP
1371 /* If SMP should be disabled, then really disable it! */
1372 if (!setup_max_cpus) {
1373 pr_info("APIC: SMP mode deactivated\n");
1374 return APIC_SYMMETRIC_IO_NO_ROUTING;
1375 }
1376
1377 if (read_apic_id() != boot_cpu_physical_apicid) {
1378 panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1379 read_apic_id(), boot_cpu_physical_apicid);
1380 /* Or can we switch back to PIC here? */
1381 }
1382#endif
1383
1384 return APIC_SYMMETRIC_IO;
1385}
1386
1387/*
1388 * An initial setup of the virtual wire mode.
1389 */
1390void __init init_bsp_APIC(void)
1391{
1392 unsigned int value;
1393
1394 /*
1395 * Don't do the setup now if we have a SMP BIOS as the
1396 * through-I/O-APIC virtual wire mode might be active.
1397 */
1398 if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1399 return;
1400
1401 /*
1402 * Do not trust the local APIC being empty at bootup.
1403 */
1404 clear_local_APIC();
1405
1406 /*
1407 * Enable APIC.
1408 */
1409 value = apic_read(APIC_SPIV);
1410 value &= ~APIC_VECTOR_MASK;
1411 value |= APIC_SPIV_APIC_ENABLED;
1412
1413#ifdef CONFIG_X86_32
1414 /* This bit is reserved on P4/Xeon and should be cleared */
1415 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1416 (boot_cpu_data.x86 == 15))
1417 value &= ~APIC_SPIV_FOCUS_DISABLED;
1418 else
1419#endif
1420 value |= APIC_SPIV_FOCUS_DISABLED;
1421 value |= SPURIOUS_APIC_VECTOR;
1422 apic_write(APIC_SPIV, value);
1423
1424 /*
1425 * Set up the virtual wire mode.
1426 */
1427 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1428 value = APIC_DM_NMI;
1429 if (!lapic_is_integrated()) /* 82489DX */
1430 value |= APIC_LVT_LEVEL_TRIGGER;
1431 if (apic_extnmi == APIC_EXTNMI_NONE)
1432 value |= APIC_LVT_MASKED;
1433 apic_write(APIC_LVT1, value);
1434}
1435
1436static void __init apic_bsp_setup(bool upmode);
1437
1438/* Init the interrupt delivery mode for the BSP */
1439void __init apic_intr_mode_init(void)
1440{
1441 bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1442
1443 apic_intr_mode = apic_intr_mode_select();
1444
1445 switch (apic_intr_mode) {
1446 case APIC_PIC:
1447 pr_info("APIC: Keep in PIC mode(8259)\n");
1448 return;
1449 case APIC_VIRTUAL_WIRE:
1450 pr_info("APIC: Switch to virtual wire mode setup\n");
1451 default_setup_apic_routing();
1452 break;
1453 case APIC_VIRTUAL_WIRE_NO_CONFIG:
1454 pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1455 upmode = true;
1456 default_setup_apic_routing();
1457 break;
1458 case APIC_SYMMETRIC_IO:
1459 pr_info("APIC: Switch to symmetric I/O mode setup\n");
1460 default_setup_apic_routing();
1461 break;
1462 case APIC_SYMMETRIC_IO_NO_ROUTING:
1463 pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1464 break;
1465 }
1466
1467 apic_bsp_setup(upmode);
1468}
1469
1470static void lapic_setup_esr(void)
1471{
1472 unsigned int oldvalue, value, maxlvt;
1473
1474 if (!lapic_is_integrated()) {
1475 pr_info("No ESR for 82489DX.\n");
1476 return;
1477 }
1478
1479 if (apic->disable_esr) {
1480 /*
1481 * Something untraceable is creating bad interrupts on
1482 * secondary quads ... for the moment, just leave the
1483 * ESR disabled - we can't do anything useful with the
1484 * errors anyway - mbligh
1485 */
1486 pr_info("Leaving ESR disabled.\n");
1487 return;
1488 }
1489
1490 maxlvt = lapic_get_maxlvt();
1491 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1492 apic_write(APIC_ESR, 0);
1493 oldvalue = apic_read(APIC_ESR);
1494
1495 /* enables sending errors */
1496 value = ERROR_APIC_VECTOR;
1497 apic_write(APIC_LVTERR, value);
1498
1499 /*
1500 * spec says clear errors after enabling vector.
1501 */
1502 if (maxlvt > 3)
1503 apic_write(APIC_ESR, 0);
1504 value = apic_read(APIC_ESR);
1505 if (value != oldvalue)
1506 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1507 "vector: 0x%08x after: 0x%08x\n",
1508 oldvalue, value);
1509}
1510
1511#define APIC_IR_REGS APIC_ISR_NR
1512#define APIC_IR_BITS (APIC_IR_REGS * 32)
1513#define APIC_IR_MAPSIZE (APIC_IR_BITS / BITS_PER_LONG)
1514
1515union apic_ir {
1516 unsigned long map[APIC_IR_MAPSIZE];
1517 u32 regs[APIC_IR_REGS];
1518};
1519
1520static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr)
1521{
1522 int i, bit;
1523
1524 /* Read the IRRs */
1525 for (i = 0; i < APIC_IR_REGS; i++)
1526 irr->regs[i] = apic_read(APIC_IRR + i * 0x10);
1527
1528 /* Read the ISRs */
1529 for (i = 0; i < APIC_IR_REGS; i++)
1530 isr->regs[i] = apic_read(APIC_ISR + i * 0x10);
1531
1532 /*
1533 * If the ISR map is not empty. ACK the APIC and run another round
1534 * to verify whether a pending IRR has been unblocked and turned
1535 * into a ISR.
1536 */
1537 if (!bitmap_empty(isr->map, APIC_IR_BITS)) {
1538 /*
1539 * There can be multiple ISR bits set when a high priority
1540 * interrupt preempted a lower priority one. Issue an ACK
1541 * per set bit.
1542 */
1543 for_each_set_bit(bit, isr->map, APIC_IR_BITS)
1544 ack_APIC_irq();
1545 return true;
1546 }
1547
1548 return !bitmap_empty(irr->map, APIC_IR_BITS);
1549}
1550
1551/*
1552 * After a crash, we no longer service the interrupts and a pending
1553 * interrupt from previous kernel might still have ISR bit set.
1554 *
1555 * Most probably by now the CPU has serviced that pending interrupt and it
1556 * might not have done the ack_APIC_irq() because it thought, interrupt
1557 * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear
1558 * the ISR bit and cpu thinks it has already serivced the interrupt. Hence
1559 * a vector might get locked. It was noticed for timer irq (vector
1560 * 0x31). Issue an extra EOI to clear ISR.
1561 *
1562 * If there are pending IRR bits they turn into ISR bits after a higher
1563 * priority ISR bit has been acked.
1564 */
1565static void apic_pending_intr_clear(void)
1566{
1567 union apic_ir irr, isr;
1568 unsigned int i;
1569
1570 /* 512 loops are way oversized and give the APIC a chance to obey. */
1571 for (i = 0; i < 512; i++) {
1572 if (!apic_check_and_ack(&irr, &isr))
1573 return;
1574 }
1575 /* Dump the IRR/ISR content if that failed */
1576 pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map);
1577}
1578
1579/**
1580 * setup_local_APIC - setup the local APIC
1581 *
1582 * Used to setup local APIC while initializing BSP or bringing up APs.
1583 * Always called with preemption disabled.
1584 */
1585static void setup_local_APIC(void)
1586{
1587 int cpu = smp_processor_id();
1588 unsigned int value;
1589
1590 if (disable_apic) {
1591 disable_ioapic_support();
1592 return;
1593 }
1594
1595 /*
1596 * If this comes from kexec/kcrash the APIC might be enabled in
1597 * SPIV. Soft disable it before doing further initialization.
1598 */
1599 value = apic_read(APIC_SPIV);
1600 value &= ~APIC_SPIV_APIC_ENABLED;
1601 apic_write(APIC_SPIV, value);
1602
1603#ifdef CONFIG_X86_32
1604 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1605 if (lapic_is_integrated() && apic->disable_esr) {
1606 apic_write(APIC_ESR, 0);
1607 apic_write(APIC_ESR, 0);
1608 apic_write(APIC_ESR, 0);
1609 apic_write(APIC_ESR, 0);
1610 }
1611#endif
1612 /*
1613 * Double-check whether this APIC is really registered.
1614 * This is meaningless in clustered apic mode, so we skip it.
1615 */
1616 BUG_ON(!apic->apic_id_registered());
1617
1618 /*
1619 * Intel recommends to set DFR, LDR and TPR before enabling
1620 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1621 * document number 292116). So here it goes...
1622 */
1623 apic->init_apic_ldr();
1624
1625#ifdef CONFIG_X86_32
1626 if (apic->dest_logical) {
1627 int logical_apicid, ldr_apicid;
1628
1629 /*
1630 * APIC LDR is initialized. If logical_apicid mapping was
1631 * initialized during get_smp_config(), make sure it matches
1632 * the actual value.
1633 */
1634 logical_apicid = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1635 ldr_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1636 if (logical_apicid != BAD_APICID)
1637 WARN_ON(logical_apicid != ldr_apicid);
1638 /* Always use the value from LDR. */
1639 early_per_cpu(x86_cpu_to_logical_apicid, cpu) = ldr_apicid;
1640 }
1641#endif
1642
1643 /*
1644 * Set Task Priority to 'accept all except vectors 0-31'. An APIC
1645 * vector in the 16-31 range could be delivered if TPR == 0, but we
1646 * would think it's an exception and terrible things will happen. We
1647 * never change this later on.
1648 */
1649 value = apic_read(APIC_TASKPRI);
1650 value &= ~APIC_TPRI_MASK;
1651 value |= 0x10;
1652 apic_write(APIC_TASKPRI, value);
1653
1654 /* Clear eventually stale ISR/IRR bits */
1655 apic_pending_intr_clear();
1656
1657 /*
1658 * Now that we are all set up, enable the APIC
1659 */
1660 value = apic_read(APIC_SPIV);
1661 value &= ~APIC_VECTOR_MASK;
1662 /*
1663 * Enable APIC
1664 */
1665 value |= APIC_SPIV_APIC_ENABLED;
1666
1667#ifdef CONFIG_X86_32
1668 /*
1669 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1670 * certain networking cards. If high frequency interrupts are
1671 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1672 * entry is masked/unmasked at a high rate as well then sooner or
1673 * later IOAPIC line gets 'stuck', no more interrupts are received
1674 * from the device. If focus CPU is disabled then the hang goes
1675 * away, oh well :-(
1676 *
1677 * [ This bug can be reproduced easily with a level-triggered
1678 * PCI Ne2000 networking cards and PII/PIII processors, dual
1679 * BX chipset. ]
1680 */
1681 /*
1682 * Actually disabling the focus CPU check just makes the hang less
1683 * frequent as it makes the interrupt distributon model be more
1684 * like LRU than MRU (the short-term load is more even across CPUs).
1685 */
1686
1687 /*
1688 * - enable focus processor (bit==0)
1689 * - 64bit mode always use processor focus
1690 * so no need to set it
1691 */
1692 value &= ~APIC_SPIV_FOCUS_DISABLED;
1693#endif
1694
1695 /*
1696 * Set spurious IRQ vector
1697 */
1698 value |= SPURIOUS_APIC_VECTOR;
1699 apic_write(APIC_SPIV, value);
1700
1701 perf_events_lapic_init();
1702
1703 /*
1704 * Set up LVT0, LVT1:
1705 *
1706 * set up through-local-APIC on the boot CPU's LINT0. This is not
1707 * strictly necessary in pure symmetric-IO mode, but sometimes
1708 * we delegate interrupts to the 8259A.
1709 */
1710 /*
1711 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1712 */
1713 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1714 if (!cpu && (pic_mode || !value || skip_ioapic_setup)) {
1715 value = APIC_DM_EXTINT;
1716 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1717 } else {
1718 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1719 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1720 }
1721 apic_write(APIC_LVT0, value);
1722
1723 /*
1724 * Only the BSP sees the LINT1 NMI signal by default. This can be
1725 * modified by apic_extnmi= boot option.
1726 */
1727 if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1728 apic_extnmi == APIC_EXTNMI_ALL)
1729 value = APIC_DM_NMI;
1730 else
1731 value = APIC_DM_NMI | APIC_LVT_MASKED;
1732
1733 /* Is 82489DX ? */
1734 if (!lapic_is_integrated())
1735 value |= APIC_LVT_LEVEL_TRIGGER;
1736 apic_write(APIC_LVT1, value);
1737
1738#ifdef CONFIG_X86_MCE_INTEL
1739 /* Recheck CMCI information after local APIC is up on CPU #0 */
1740 if (!cpu)
1741 cmci_recheck();
1742#endif
1743}
1744
1745static void end_local_APIC_setup(void)
1746{
1747 lapic_setup_esr();
1748
1749#ifdef CONFIG_X86_32
1750 {
1751 unsigned int value;
1752 /* Disable the local apic timer */
1753 value = apic_read(APIC_LVTT);
1754 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1755 apic_write(APIC_LVTT, value);
1756 }
1757#endif
1758
1759 apic_pm_activate();
1760}
1761
1762/*
1763 * APIC setup function for application processors. Called from smpboot.c
1764 */
1765void apic_ap_setup(void)
1766{
1767 setup_local_APIC();
1768 end_local_APIC_setup();
1769}
1770
1771#ifdef CONFIG_X86_X2APIC
1772int x2apic_mode;
1773
1774enum {
1775 X2APIC_OFF,
1776 X2APIC_ON,
1777 X2APIC_DISABLED,
1778};
1779static int x2apic_state;
1780
1781static void __x2apic_disable(void)
1782{
1783 u64 msr;
1784
1785 if (!boot_cpu_has(X86_FEATURE_APIC))
1786 return;
1787
1788 rdmsrl(MSR_IA32_APICBASE, msr);
1789 if (!(msr & X2APIC_ENABLE))
1790 return;
1791 /* Disable xapic and x2apic first and then reenable xapic mode */
1792 wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1793 wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1794 printk_once(KERN_INFO "x2apic disabled\n");
1795}
1796
1797static void __x2apic_enable(void)
1798{
1799 u64 msr;
1800
1801 rdmsrl(MSR_IA32_APICBASE, msr);
1802 if (msr & X2APIC_ENABLE)
1803 return;
1804 wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1805 printk_once(KERN_INFO "x2apic enabled\n");
1806}
1807
1808static int __init setup_nox2apic(char *str)
1809{
1810 if (x2apic_enabled()) {
1811 int apicid = native_apic_msr_read(APIC_ID);
1812
1813 if (apicid >= 255) {
1814 pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
1815 apicid);
1816 return 0;
1817 }
1818 pr_warning("x2apic already enabled.\n");
1819 __x2apic_disable();
1820 }
1821 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1822 x2apic_state = X2APIC_DISABLED;
1823 x2apic_mode = 0;
1824 return 0;
1825}
1826early_param("nox2apic", setup_nox2apic);
1827
1828/* Called from cpu_init() to enable x2apic on (secondary) cpus */
1829void x2apic_setup(void)
1830{
1831 /*
1832 * If x2apic is not in ON state, disable it if already enabled
1833 * from BIOS.
1834 */
1835 if (x2apic_state != X2APIC_ON) {
1836 __x2apic_disable();
1837 return;
1838 }
1839 __x2apic_enable();
1840}
1841
1842static __init void x2apic_disable(void)
1843{
1844 u32 x2apic_id, state = x2apic_state;
1845
1846 x2apic_mode = 0;
1847 x2apic_state = X2APIC_DISABLED;
1848
1849 if (state != X2APIC_ON)
1850 return;
1851
1852 x2apic_id = read_apic_id();
1853 if (x2apic_id >= 255)
1854 panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1855
1856 __x2apic_disable();
1857 register_lapic_address(mp_lapic_addr);
1858}
1859
1860static __init void x2apic_enable(void)
1861{
1862 if (x2apic_state != X2APIC_OFF)
1863 return;
1864
1865 x2apic_mode = 1;
1866 x2apic_state = X2APIC_ON;
1867 __x2apic_enable();
1868}
1869
1870static __init void try_to_enable_x2apic(int remap_mode)
1871{
1872 if (x2apic_state == X2APIC_DISABLED)
1873 return;
1874
1875 if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1876 /* IR is required if there is APIC ID > 255 even when running
1877 * under KVM
1878 */
1879 if (max_physical_apicid > 255 ||
1880 !x86_init.hyper.x2apic_available()) {
1881 pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1882 x2apic_disable();
1883 return;
1884 }
1885
1886 /*
1887 * without IR all CPUs can be addressed by IOAPIC/MSI
1888 * only in physical mode
1889 */
1890 x2apic_phys = 1;
1891 }
1892 x2apic_enable();
1893}
1894
1895void __init check_x2apic(void)
1896{
1897 if (x2apic_enabled()) {
1898 pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1899 x2apic_mode = 1;
1900 x2apic_state = X2APIC_ON;
1901 } else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1902 x2apic_state = X2APIC_DISABLED;
1903 }
1904}
1905#else /* CONFIG_X86_X2APIC */
1906static int __init validate_x2apic(void)
1907{
1908 if (!apic_is_x2apic_enabled())
1909 return 0;
1910 /*
1911 * Checkme: Can we simply turn off x2apic here instead of panic?
1912 */
1913 panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n");
1914}
1915early_initcall(validate_x2apic);
1916
1917static inline void try_to_enable_x2apic(int remap_mode) { }
1918static inline void __x2apic_enable(void) { }
1919#endif /* !CONFIG_X86_X2APIC */
1920
1921void __init enable_IR_x2apic(void)
1922{
1923 unsigned long flags;
1924 int ret, ir_stat;
1925
1926 if (skip_ioapic_setup) {
1927 pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1928 return;
1929 }
1930
1931 ir_stat = irq_remapping_prepare();
1932 if (ir_stat < 0 && !x2apic_supported())
1933 return;
1934
1935 ret = save_ioapic_entries();
1936 if (ret) {
1937 pr_info("Saving IO-APIC state failed: %d\n", ret);
1938 return;
1939 }
1940
1941 local_irq_save(flags);
1942 legacy_pic->mask_all();
1943 mask_ioapic_entries();
1944
1945 /* If irq_remapping_prepare() succeeded, try to enable it */
1946 if (ir_stat >= 0)
1947 ir_stat = irq_remapping_enable();
1948 /* ir_stat contains the remap mode or an error code */
1949 try_to_enable_x2apic(ir_stat);
1950
1951 if (ir_stat < 0)
1952 restore_ioapic_entries();
1953 legacy_pic->restore_mask();
1954 local_irq_restore(flags);
1955}
1956
1957#ifdef CONFIG_X86_64
1958/*
1959 * Detect and enable local APICs on non-SMP boards.
1960 * Original code written by Keir Fraser.
1961 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1962 * not correctly set up (usually the APIC timer won't work etc.)
1963 */
1964static int __init detect_init_APIC(void)
1965{
1966 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1967 pr_info("No local APIC present\n");
1968 return -1;
1969 }
1970
1971 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1972 return 0;
1973}
1974#else
1975
1976static int __init apic_verify(void)
1977{
1978 u32 features, h, l;
1979
1980 /*
1981 * The APIC feature bit should now be enabled
1982 * in `cpuid'
1983 */
1984 features = cpuid_edx(1);
1985 if (!(features & (1 << X86_FEATURE_APIC))) {
1986 pr_warning("Could not enable APIC!\n");
1987 return -1;
1988 }
1989 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1990 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1991
1992 /* The BIOS may have set up the APIC at some other address */
1993 if (boot_cpu_data.x86 >= 6) {
1994 rdmsr(MSR_IA32_APICBASE, l, h);
1995 if (l & MSR_IA32_APICBASE_ENABLE)
1996 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1997 }
1998
1999 pr_info("Found and enabled local APIC!\n");
2000 return 0;
2001}
2002
2003int __init apic_force_enable(unsigned long addr)
2004{
2005 u32 h, l;
2006
2007 if (disable_apic)
2008 return -1;
2009
2010 /*
2011 * Some BIOSes disable the local APIC in the APIC_BASE
2012 * MSR. This can only be done in software for Intel P6 or later
2013 * and AMD K7 (Model > 1) or later.
2014 */
2015 if (boot_cpu_data.x86 >= 6) {
2016 rdmsr(MSR_IA32_APICBASE, l, h);
2017 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
2018 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
2019 l &= ~MSR_IA32_APICBASE_BASE;
2020 l |= MSR_IA32_APICBASE_ENABLE | addr;
2021 wrmsr(MSR_IA32_APICBASE, l, h);
2022 enabled_via_apicbase = 1;
2023 }
2024 }
2025 return apic_verify();
2026}
2027
2028/*
2029 * Detect and initialize APIC
2030 */
2031static int __init detect_init_APIC(void)
2032{
2033 /* Disabled by kernel option? */
2034 if (disable_apic)
2035 return -1;
2036
2037 switch (boot_cpu_data.x86_vendor) {
2038 case X86_VENDOR_AMD:
2039 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
2040 (boot_cpu_data.x86 >= 15))
2041 break;
2042 goto no_apic;
2043 case X86_VENDOR_HYGON:
2044 break;
2045 case X86_VENDOR_INTEL:
2046 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
2047 (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
2048 break;
2049 goto no_apic;
2050 default:
2051 goto no_apic;
2052 }
2053
2054 if (!boot_cpu_has(X86_FEATURE_APIC)) {
2055 /*
2056 * Over-ride BIOS and try to enable the local APIC only if
2057 * "lapic" specified.
2058 */
2059 if (!force_enable_local_apic) {
2060 pr_info("Local APIC disabled by BIOS -- "
2061 "you can enable it with \"lapic\"\n");
2062 return -1;
2063 }
2064 if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
2065 return -1;
2066 } else {
2067 if (apic_verify())
2068 return -1;
2069 }
2070
2071 apic_pm_activate();
2072
2073 return 0;
2074
2075no_apic:
2076 pr_info("No local APIC present or hardware disabled\n");
2077 return -1;
2078}
2079#endif
2080
2081/**
2082 * init_apic_mappings - initialize APIC mappings
2083 */
2084void __init init_apic_mappings(void)
2085{
2086 unsigned int new_apicid;
2087
2088 apic_check_deadline_errata();
2089
2090 if (x2apic_mode) {
2091 boot_cpu_physical_apicid = read_apic_id();
2092 return;
2093 }
2094
2095 /* If no local APIC can be found return early */
2096 if (!smp_found_config && detect_init_APIC()) {
2097 /* lets NOP'ify apic operations */
2098 pr_info("APIC: disable apic facility\n");
2099 apic_disable();
2100 } else {
2101 apic_phys = mp_lapic_addr;
2102
2103 /*
2104 * If the system has ACPI MADT tables or MP info, the LAPIC
2105 * address is already registered.
2106 */
2107 if (!acpi_lapic && !smp_found_config)
2108 register_lapic_address(apic_phys);
2109 }
2110
2111 /*
2112 * Fetch the APIC ID of the BSP in case we have a
2113 * default configuration (or the MP table is broken).
2114 */
2115 new_apicid = read_apic_id();
2116 if (boot_cpu_physical_apicid != new_apicid) {
2117 boot_cpu_physical_apicid = new_apicid;
2118 /*
2119 * yeah -- we lie about apic_version
2120 * in case if apic was disabled via boot option
2121 * but it's not a problem for SMP compiled kernel
2122 * since apic_intr_mode_select is prepared for such
2123 * a case and disable smp mode
2124 */
2125 boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2126 }
2127}
2128
2129void __init register_lapic_address(unsigned long address)
2130{
2131 mp_lapic_addr = address;
2132
2133 if (!x2apic_mode) {
2134 set_fixmap_nocache(FIX_APIC_BASE, address);
2135 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
2136 APIC_BASE, address);
2137 }
2138 if (boot_cpu_physical_apicid == -1U) {
2139 boot_cpu_physical_apicid = read_apic_id();
2140 boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2141 }
2142}
2143
2144/*
2145 * Local APIC interrupts
2146 */
2147
2148/*
2149 * This interrupt should _never_ happen with our APIC/SMP architecture
2150 */
2151__visible void __irq_entry smp_spurious_interrupt(struct pt_regs *regs)
2152{
2153 u8 vector = ~regs->orig_ax;
2154 u32 v;
2155
2156 entering_irq();
2157 trace_spurious_apic_entry(vector);
2158
2159 inc_irq_stat(irq_spurious_count);
2160
2161 /*
2162 * If this is a spurious interrupt then do not acknowledge
2163 */
2164 if (vector == SPURIOUS_APIC_VECTOR) {
2165 /* See SDM vol 3 */
2166 pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n",
2167 smp_processor_id());
2168 goto out;
2169 }
2170
2171 /*
2172 * If it is a vectored one, verify it's set in the ISR. If set,
2173 * acknowledge it.
2174 */
2175 v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2176 if (v & (1 << (vector & 0x1f))) {
2177 pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n",
2178 vector, smp_processor_id());
2179 ack_APIC_irq();
2180 } else {
2181 pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n",
2182 vector, smp_processor_id());
2183 }
2184out:
2185 trace_spurious_apic_exit(vector);
2186 exiting_irq();
2187}
2188
2189/*
2190 * This interrupt should never happen with our APIC/SMP architecture
2191 */
2192__visible void __irq_entry smp_error_interrupt(struct pt_regs *regs)
2193{
2194 static const char * const error_interrupt_reason[] = {
2195 "Send CS error", /* APIC Error Bit 0 */
2196 "Receive CS error", /* APIC Error Bit 1 */
2197 "Send accept error", /* APIC Error Bit 2 */
2198 "Receive accept error", /* APIC Error Bit 3 */
2199 "Redirectable IPI", /* APIC Error Bit 4 */
2200 "Send illegal vector", /* APIC Error Bit 5 */
2201 "Received illegal vector", /* APIC Error Bit 6 */
2202 "Illegal register address", /* APIC Error Bit 7 */
2203 };
2204 u32 v, i = 0;
2205
2206 entering_irq();
2207 trace_error_apic_entry(ERROR_APIC_VECTOR);
2208
2209 /* First tickle the hardware, only then report what went on. -- REW */
2210 if (lapic_get_maxlvt() > 3) /* Due to the Pentium erratum 3AP. */
2211 apic_write(APIC_ESR, 0);
2212 v = apic_read(APIC_ESR);
2213 ack_APIC_irq();
2214 atomic_inc(&irq_err_count);
2215
2216 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2217 smp_processor_id(), v);
2218
2219 v &= 0xff;
2220 while (v) {
2221 if (v & 0x1)
2222 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2223 i++;
2224 v >>= 1;
2225 }
2226
2227 apic_printk(APIC_DEBUG, KERN_CONT "\n");
2228
2229 trace_error_apic_exit(ERROR_APIC_VECTOR);
2230 exiting_irq();
2231}
2232
2233/**
2234 * connect_bsp_APIC - attach the APIC to the interrupt system
2235 */
2236static void __init connect_bsp_APIC(void)
2237{
2238#ifdef CONFIG_X86_32
2239 if (pic_mode) {
2240 /*
2241 * Do not trust the local APIC being empty at bootup.
2242 */
2243 clear_local_APIC();
2244 /*
2245 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
2246 * local APIC to INT and NMI lines.
2247 */
2248 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2249 "enabling APIC mode.\n");
2250 imcr_pic_to_apic();
2251 }
2252#endif
2253}
2254
2255/**
2256 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2257 * @virt_wire_setup: indicates, whether virtual wire mode is selected
2258 *
2259 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2260 * APIC is disabled.
2261 */
2262void disconnect_bsp_APIC(int virt_wire_setup)
2263{
2264 unsigned int value;
2265
2266#ifdef CONFIG_X86_32
2267 if (pic_mode) {
2268 /*
2269 * Put the board back into PIC mode (has an effect only on
2270 * certain older boards). Note that APIC interrupts, including
2271 * IPIs, won't work beyond this point! The only exception are
2272 * INIT IPIs.
2273 */
2274 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2275 "entering PIC mode.\n");
2276 imcr_apic_to_pic();
2277 return;
2278 }
2279#endif
2280
2281 /* Go back to Virtual Wire compatibility mode */
2282
2283 /* For the spurious interrupt use vector F, and enable it */
2284 value = apic_read(APIC_SPIV);
2285 value &= ~APIC_VECTOR_MASK;
2286 value |= APIC_SPIV_APIC_ENABLED;
2287 value |= 0xf;
2288 apic_write(APIC_SPIV, value);
2289
2290 if (!virt_wire_setup) {
2291 /*
2292 * For LVT0 make it edge triggered, active high,
2293 * external and enabled
2294 */
2295 value = apic_read(APIC_LVT0);
2296 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2297 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2298 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2299 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2300 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2301 apic_write(APIC_LVT0, value);
2302 } else {
2303 /* Disable LVT0 */
2304 apic_write(APIC_LVT0, APIC_LVT_MASKED);
2305 }
2306
2307 /*
2308 * For LVT1 make it edge triggered, active high,
2309 * nmi and enabled
2310 */
2311 value = apic_read(APIC_LVT1);
2312 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2313 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2314 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2315 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2316 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2317 apic_write(APIC_LVT1, value);
2318}
2319
2320/*
2321 * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2322 * contiguously, it equals to current allocated max logical CPU ID plus 1.
2323 * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2324 * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2325 *
2326 * NOTE: Reserve 0 for BSP.
2327 */
2328static int nr_logical_cpuids = 1;
2329
2330/*
2331 * Used to store mapping between logical CPU IDs and APIC IDs.
2332 */
2333static int cpuid_to_apicid[] = {
2334 [0 ... NR_CPUS - 1] = -1,
2335};
2336
2337#ifdef CONFIG_SMP
2338/**
2339 * apic_id_is_primary_thread - Check whether APIC ID belongs to a primary thread
2340 * @id: APIC ID to check
2341 */
2342bool apic_id_is_primary_thread(unsigned int apicid)
2343{
2344 u32 mask;
2345
2346 if (smp_num_siblings == 1)
2347 return true;
2348 /* Isolate the SMT bit(s) in the APICID and check for 0 */
2349 mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
2350 return !(apicid & mask);
2351}
2352#endif
2353
2354/*
2355 * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2356 * and cpuid_to_apicid[] synchronized.
2357 */
2358static int allocate_logical_cpuid(int apicid)
2359{
2360 int i;
2361
2362 /*
2363 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2364 * check if the kernel has allocated a cpuid for it.
2365 */
2366 for (i = 0; i < nr_logical_cpuids; i++) {
2367 if (cpuid_to_apicid[i] == apicid)
2368 return i;
2369 }
2370
2371 /* Allocate a new cpuid. */
2372 if (nr_logical_cpuids >= nr_cpu_ids) {
2373 WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %u reached. "
2374 "Processor %d/0x%x and the rest are ignored.\n",
2375 nr_cpu_ids, nr_logical_cpuids, apicid);
2376 return -EINVAL;
2377 }
2378
2379 cpuid_to_apicid[nr_logical_cpuids] = apicid;
2380 return nr_logical_cpuids++;
2381}
2382
2383int generic_processor_info(int apicid, int version)
2384{
2385 int cpu, max = nr_cpu_ids;
2386 bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2387 phys_cpu_present_map);
2388
2389 /*
2390 * boot_cpu_physical_apicid is designed to have the apicid
2391 * returned by read_apic_id(), i.e, the apicid of the
2392 * currently booting-up processor. However, on some platforms,
2393 * it is temporarily modified by the apicid reported as BSP
2394 * through MP table. Concretely:
2395 *
2396 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2397 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2398 *
2399 * This function is executed with the modified
2400 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2401 * parameter doesn't work to disable APs on kdump 2nd kernel.
2402 *
2403 * Since fixing handling of boot_cpu_physical_apicid requires
2404 * another discussion and tests on each platform, we leave it
2405 * for now and here we use read_apic_id() directly in this
2406 * function, generic_processor_info().
2407 */
2408 if (disabled_cpu_apicid != BAD_APICID &&
2409 disabled_cpu_apicid != read_apic_id() &&
2410 disabled_cpu_apicid == apicid) {
2411 int thiscpu = num_processors + disabled_cpus;
2412
2413 pr_warning("APIC: Disabling requested cpu."
2414 " Processor %d/0x%x ignored.\n",
2415 thiscpu, apicid);
2416
2417 disabled_cpus++;
2418 return -ENODEV;
2419 }
2420
2421 /*
2422 * If boot cpu has not been detected yet, then only allow upto
2423 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2424 */
2425 if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2426 apicid != boot_cpu_physical_apicid) {
2427 int thiscpu = max + disabled_cpus - 1;
2428
2429 pr_warning(
2430 "APIC: NR_CPUS/possible_cpus limit of %i almost"
2431 " reached. Keeping one slot for boot cpu."
2432 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2433
2434 disabled_cpus++;
2435 return -ENODEV;
2436 }
2437
2438 if (num_processors >= nr_cpu_ids) {
2439 int thiscpu = max + disabled_cpus;
2440
2441 pr_warning("APIC: NR_CPUS/possible_cpus limit of %i "
2442 "reached. Processor %d/0x%x ignored.\n",
2443 max, thiscpu, apicid);
2444
2445 disabled_cpus++;
2446 return -EINVAL;
2447 }
2448
2449 if (apicid == boot_cpu_physical_apicid) {
2450 /*
2451 * x86_bios_cpu_apicid is required to have processors listed
2452 * in same order as logical cpu numbers. Hence the first
2453 * entry is BSP, and so on.
2454 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2455 * for BSP.
2456 */
2457 cpu = 0;
2458
2459 /* Logical cpuid 0 is reserved for BSP. */
2460 cpuid_to_apicid[0] = apicid;
2461 } else {
2462 cpu = allocate_logical_cpuid(apicid);
2463 if (cpu < 0) {
2464 disabled_cpus++;
2465 return -EINVAL;
2466 }
2467 }
2468
2469 /*
2470 * Validate version
2471 */
2472 if (version == 0x0) {
2473 pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2474 cpu, apicid);
2475 version = 0x10;
2476 }
2477
2478 if (version != boot_cpu_apic_version) {
2479 pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2480 boot_cpu_apic_version, cpu, version);
2481 }
2482
2483 if (apicid > max_physical_apicid)
2484 max_physical_apicid = apicid;
2485
2486#if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2487 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2488 early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2489#endif
2490#ifdef CONFIG_X86_32
2491 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2492 apic->x86_32_early_logical_apicid(cpu);
2493#endif
2494 set_cpu_possible(cpu, true);
2495 physid_set(apicid, phys_cpu_present_map);
2496 set_cpu_present(cpu, true);
2497 num_processors++;
2498
2499 return cpu;
2500}
2501
2502int hard_smp_processor_id(void)
2503{
2504 return read_apic_id();
2505}
2506
2507/*
2508 * Override the generic EOI implementation with an optimized version.
2509 * Only called during early boot when only one CPU is active and with
2510 * interrupts disabled, so we know this does not race with actual APIC driver
2511 * use.
2512 */
2513void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2514{
2515 struct apic **drv;
2516
2517 for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2518 /* Should happen once for each apic */
2519 WARN_ON((*drv)->eoi_write == eoi_write);
2520 (*drv)->native_eoi_write = (*drv)->eoi_write;
2521 (*drv)->eoi_write = eoi_write;
2522 }
2523}
2524
2525static void __init apic_bsp_up_setup(void)
2526{
2527#ifdef CONFIG_X86_64
2528 apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2529#else
2530 /*
2531 * Hack: In case of kdump, after a crash, kernel might be booting
2532 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2533 * might be zero if read from MP tables. Get it from LAPIC.
2534 */
2535# ifdef CONFIG_CRASH_DUMP
2536 boot_cpu_physical_apicid = read_apic_id();
2537# endif
2538#endif
2539 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2540}
2541
2542/**
2543 * apic_bsp_setup - Setup function for local apic and io-apic
2544 * @upmode: Force UP mode (for APIC_init_uniprocessor)
2545 */
2546static void __init apic_bsp_setup(bool upmode)
2547{
2548 connect_bsp_APIC();
2549 if (upmode)
2550 apic_bsp_up_setup();
2551 setup_local_APIC();
2552
2553 enable_IO_APIC();
2554 end_local_APIC_setup();
2555 irq_remap_enable_fault_handling();
2556 setup_IO_APIC();
2557}
2558
2559#ifdef CONFIG_UP_LATE_INIT
2560void __init up_late_init(void)
2561{
2562 if (apic_intr_mode == APIC_PIC)
2563 return;
2564
2565 /* Setup local timer */
2566 x86_init.timers.setup_percpu_clockev();
2567}
2568#endif
2569
2570/*
2571 * Power management
2572 */
2573#ifdef CONFIG_PM
2574
2575static struct {
2576 /*
2577 * 'active' is true if the local APIC was enabled by us and
2578 * not the BIOS; this signifies that we are also responsible
2579 * for disabling it before entering apm/acpi suspend
2580 */
2581 int active;
2582 /* r/w apic fields */
2583 unsigned int apic_id;
2584 unsigned int apic_taskpri;
2585 unsigned int apic_ldr;
2586 unsigned int apic_dfr;
2587 unsigned int apic_spiv;
2588 unsigned int apic_lvtt;
2589 unsigned int apic_lvtpc;
2590 unsigned int apic_lvt0;
2591 unsigned int apic_lvt1;
2592 unsigned int apic_lvterr;
2593 unsigned int apic_tmict;
2594 unsigned int apic_tdcr;
2595 unsigned int apic_thmr;
2596 unsigned int apic_cmci;
2597} apic_pm_state;
2598
2599static int lapic_suspend(void)
2600{
2601 unsigned long flags;
2602 int maxlvt;
2603
2604 if (!apic_pm_state.active)
2605 return 0;
2606
2607 maxlvt = lapic_get_maxlvt();
2608
2609 apic_pm_state.apic_id = apic_read(APIC_ID);
2610 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2611 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2612 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2613 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2614 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2615 if (maxlvt >= 4)
2616 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2617 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2618 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2619 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2620 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2621 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2622#ifdef CONFIG_X86_THERMAL_VECTOR
2623 if (maxlvt >= 5)
2624 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2625#endif
2626#ifdef CONFIG_X86_MCE_INTEL
2627 if (maxlvt >= 6)
2628 apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2629#endif
2630
2631 local_irq_save(flags);
2632 disable_local_APIC();
2633
2634 irq_remapping_disable();
2635
2636 local_irq_restore(flags);
2637 return 0;
2638}
2639
2640static void lapic_resume(void)
2641{
2642 unsigned int l, h;
2643 unsigned long flags;
2644 int maxlvt;
2645
2646 if (!apic_pm_state.active)
2647 return;
2648
2649 local_irq_save(flags);
2650
2651 /*
2652 * IO-APIC and PIC have their own resume routines.
2653 * We just mask them here to make sure the interrupt
2654 * subsystem is completely quiet while we enable x2apic
2655 * and interrupt-remapping.
2656 */
2657 mask_ioapic_entries();
2658 legacy_pic->mask_all();
2659
2660 if (x2apic_mode) {
2661 __x2apic_enable();
2662 } else {
2663 /*
2664 * Make sure the APICBASE points to the right address
2665 *
2666 * FIXME! This will be wrong if we ever support suspend on
2667 * SMP! We'll need to do this as part of the CPU restore!
2668 */
2669 if (boot_cpu_data.x86 >= 6) {
2670 rdmsr(MSR_IA32_APICBASE, l, h);
2671 l &= ~MSR_IA32_APICBASE_BASE;
2672 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2673 wrmsr(MSR_IA32_APICBASE, l, h);
2674 }
2675 }
2676
2677 maxlvt = lapic_get_maxlvt();
2678 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2679 apic_write(APIC_ID, apic_pm_state.apic_id);
2680 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2681 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2682 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2683 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2684 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2685 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2686#ifdef CONFIG_X86_THERMAL_VECTOR
2687 if (maxlvt >= 5)
2688 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2689#endif
2690#ifdef CONFIG_X86_MCE_INTEL
2691 if (maxlvt >= 6)
2692 apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2693#endif
2694 if (maxlvt >= 4)
2695 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2696 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2697 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2698 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2699 apic_write(APIC_ESR, 0);
2700 apic_read(APIC_ESR);
2701 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2702 apic_write(APIC_ESR, 0);
2703 apic_read(APIC_ESR);
2704
2705 irq_remapping_reenable(x2apic_mode);
2706
2707 local_irq_restore(flags);
2708}
2709
2710/*
2711 * This device has no shutdown method - fully functioning local APICs
2712 * are needed on every CPU up until machine_halt/restart/poweroff.
2713 */
2714
2715static struct syscore_ops lapic_syscore_ops = {
2716 .resume = lapic_resume,
2717 .suspend = lapic_suspend,
2718};
2719
2720static void apic_pm_activate(void)
2721{
2722 apic_pm_state.active = 1;
2723}
2724
2725static int __init init_lapic_sysfs(void)
2726{
2727 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2728 if (boot_cpu_has(X86_FEATURE_APIC))
2729 register_syscore_ops(&lapic_syscore_ops);
2730
2731 return 0;
2732}
2733
2734/* local apic needs to resume before other devices access its registers. */
2735core_initcall(init_lapic_sysfs);
2736
2737#else /* CONFIG_PM */
2738
2739static void apic_pm_activate(void) { }
2740
2741#endif /* CONFIG_PM */
2742
2743#ifdef CONFIG_X86_64
2744
2745static int multi_checked;
2746static int multi;
2747
2748static int set_multi(const struct dmi_system_id *d)
2749{
2750 if (multi)
2751 return 0;
2752 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2753 multi = 1;
2754 return 0;
2755}
2756
2757static const struct dmi_system_id multi_dmi_table[] = {
2758 {
2759 .callback = set_multi,
2760 .ident = "IBM System Summit2",
2761 .matches = {
2762 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2763 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2764 },
2765 },
2766 {}
2767};
2768
2769static void dmi_check_multi(void)
2770{
2771 if (multi_checked)
2772 return;
2773
2774 dmi_check_system(multi_dmi_table);
2775 multi_checked = 1;
2776}
2777
2778/*
2779 * apic_is_clustered_box() -- Check if we can expect good TSC
2780 *
2781 * Thus far, the major user of this is IBM's Summit2 series:
2782 * Clustered boxes may have unsynced TSC problems if they are
2783 * multi-chassis.
2784 * Use DMI to check them
2785 */
2786int apic_is_clustered_box(void)
2787{
2788 dmi_check_multi();
2789 return multi;
2790}
2791#endif
2792
2793/*
2794 * APIC command line parameters
2795 */
2796static int __init setup_disableapic(char *arg)
2797{
2798 disable_apic = 1;
2799 setup_clear_cpu_cap(X86_FEATURE_APIC);
2800 return 0;
2801}
2802early_param("disableapic", setup_disableapic);
2803
2804/* same as disableapic, for compatibility */
2805static int __init setup_nolapic(char *arg)
2806{
2807 return setup_disableapic(arg);
2808}
2809early_param("nolapic", setup_nolapic);
2810
2811static int __init parse_lapic_timer_c2_ok(char *arg)
2812{
2813 local_apic_timer_c2_ok = 1;
2814 return 0;
2815}
2816early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2817
2818static int __init parse_disable_apic_timer(char *arg)
2819{
2820 disable_apic_timer = 1;
2821 return 0;
2822}
2823early_param("noapictimer", parse_disable_apic_timer);
2824
2825static int __init parse_nolapic_timer(char *arg)
2826{
2827 disable_apic_timer = 1;
2828 return 0;
2829}
2830early_param("nolapic_timer", parse_nolapic_timer);
2831
2832static int __init apic_set_verbosity(char *arg)
2833{
2834 if (!arg) {
2835#ifdef CONFIG_X86_64
2836 skip_ioapic_setup = 0;
2837 return 0;
2838#endif
2839 return -EINVAL;
2840 }
2841
2842 if (strcmp("debug", arg) == 0)
2843 apic_verbosity = APIC_DEBUG;
2844 else if (strcmp("verbose", arg) == 0)
2845 apic_verbosity = APIC_VERBOSE;
2846#ifdef CONFIG_X86_64
2847 else {
2848 pr_warning("APIC Verbosity level %s not recognised"
2849 " use apic=verbose or apic=debug\n", arg);
2850 return -EINVAL;
2851 }
2852#endif
2853
2854 return 0;
2855}
2856early_param("apic", apic_set_verbosity);
2857
2858static int __init lapic_insert_resource(void)
2859{
2860 if (!apic_phys)
2861 return -1;
2862
2863 /* Put local APIC into the resource map. */
2864 lapic_resource.start = apic_phys;
2865 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2866 insert_resource(&iomem_resource, &lapic_resource);
2867
2868 return 0;
2869}
2870
2871/*
2872 * need call insert after e820__reserve_resources()
2873 * that is using request_resource
2874 */
2875late_initcall(lapic_insert_resource);
2876
2877static int __init apic_set_disabled_cpu_apicid(char *arg)
2878{
2879 if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2880 return -EINVAL;
2881
2882 return 0;
2883}
2884early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2885
2886static int __init apic_set_extnmi(char *arg)
2887{
2888 if (!arg)
2889 return -EINVAL;
2890
2891 if (!strncmp("all", arg, 3))
2892 apic_extnmi = APIC_EXTNMI_ALL;
2893 else if (!strncmp("none", arg, 4))
2894 apic_extnmi = APIC_EXTNMI_NONE;
2895 else if (!strncmp("bsp", arg, 3))
2896 apic_extnmi = APIC_EXTNMI_BSP;
2897 else {
2898 pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2899 return -EINVAL;
2900 }
2901
2902 return 0;
2903}
2904early_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 <asm/trace/irq_vectors.h>
40#include <asm/irq_remapping.h>
41#include <asm/pc-conf-reg.h>
42#include <asm/perf_event.h>
43#include <asm/x86_init.h>
44#include <linux/atomic.h>
45#include <asm/barrier.h>
46#include <asm/mpspec.h>
47#include <asm/i8259.h>
48#include <asm/proto.h>
49#include <asm/traps.h>
50#include <asm/apic.h>
51#include <asm/acpi.h>
52#include <asm/io_apic.h>
53#include <asm/desc.h>
54#include <asm/hpet.h>
55#include <asm/mtrr.h>
56#include <asm/time.h>
57#include <asm/smp.h>
58#include <asm/mce.h>
59#include <asm/tsc.h>
60#include <asm/hypervisor.h>
61#include <asm/cpu_device_id.h>
62#include <asm/intel-family.h>
63#include <asm/irq_regs.h>
64#include <asm/cpu.h>
65
66unsigned int num_processors;
67
68unsigned disabled_cpus;
69
70/* Processor that is doing the boot up */
71unsigned int boot_cpu_physical_apicid __ro_after_init = -1U;
72EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
73
74u8 boot_cpu_apic_version __ro_after_init;
75
76/*
77 * The highest APIC ID seen during enumeration.
78 */
79static unsigned int max_physical_apicid;
80
81/*
82 * Bitmask of physically existing CPUs:
83 */
84physid_mask_t phys_cpu_present_map;
85
86/*
87 * Processor to be disabled specified by kernel parameter
88 * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
89 * avoid undefined behaviour caused by sending INIT from AP to BSP.
90 */
91static unsigned int disabled_cpu_apicid __ro_after_init = BAD_APICID;
92
93/*
94 * This variable controls which CPUs receive external NMIs. By default,
95 * external NMIs are delivered only to the BSP.
96 */
97static int apic_extnmi __ro_after_init = APIC_EXTNMI_BSP;
98
99/*
100 * Hypervisor supports 15 bits of APIC ID in MSI Extended Destination ID
101 */
102static bool virt_ext_dest_id __ro_after_init;
103
104/*
105 * Map cpu index to physical APIC ID
106 */
107DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
108DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
109DEFINE_EARLY_PER_CPU_READ_MOSTLY(u32, x86_cpu_to_acpiid, U32_MAX);
110EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
111EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
112EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_acpiid);
113
114#ifdef CONFIG_X86_32
115
116/*
117 * On x86_32, the mapping between cpu and logical apicid may vary
118 * depending on apic in use. The following early percpu variable is
119 * used for the mapping. This is where the behaviors of x86_64 and 32
120 * actually diverge. Let's keep it ugly for now.
121 */
122DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
123
124/* Local APIC was disabled by the BIOS and enabled by the kernel */
125static int enabled_via_apicbase __ro_after_init;
126
127/*
128 * Handle interrupt mode configuration register (IMCR).
129 * This register controls whether the interrupt signals
130 * that reach the BSP come from the master PIC or from the
131 * local APIC. Before entering Symmetric I/O Mode, either
132 * the BIOS or the operating system must switch out of
133 * PIC Mode by changing the IMCR.
134 */
135static inline void imcr_pic_to_apic(void)
136{
137 /* NMI and 8259 INTR go through APIC */
138 pc_conf_set(PC_CONF_MPS_IMCR, 0x01);
139}
140
141static inline void imcr_apic_to_pic(void)
142{
143 /* NMI and 8259 INTR go directly to BSP */
144 pc_conf_set(PC_CONF_MPS_IMCR, 0x00);
145}
146#endif
147
148/*
149 * Knob to control our willingness to enable the local APIC.
150 *
151 * +1=force-enable
152 */
153static int force_enable_local_apic __initdata;
154
155/*
156 * APIC command line parameters
157 */
158static int __init parse_lapic(char *arg)
159{
160 if (IS_ENABLED(CONFIG_X86_32) && !arg)
161 force_enable_local_apic = 1;
162 else if (arg && !strncmp(arg, "notscdeadline", 13))
163 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
164 return 0;
165}
166early_param("lapic", parse_lapic);
167
168#ifdef CONFIG_X86_64
169static int apic_calibrate_pmtmr __initdata;
170static __init int setup_apicpmtimer(char *s)
171{
172 apic_calibrate_pmtmr = 1;
173 notsc_setup(NULL);
174 return 1;
175}
176__setup("apicpmtimer", setup_apicpmtimer);
177#endif
178
179unsigned long mp_lapic_addr __ro_after_init;
180int disable_apic __ro_after_init;
181/* Disable local APIC timer from the kernel commandline or via dmi quirk */
182static int disable_apic_timer __initdata;
183/* Local APIC timer works in C2 */
184int local_apic_timer_c2_ok __ro_after_init;
185EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
186
187/*
188 * Debug level, exported for io_apic.c
189 */
190int apic_verbosity __ro_after_init;
191
192int pic_mode __ro_after_init;
193
194/* Have we found an MP table */
195int smp_found_config __ro_after_init;
196
197static struct resource lapic_resource = {
198 .name = "Local APIC",
199 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
200};
201
202unsigned int lapic_timer_period = 0;
203
204static void apic_pm_activate(void);
205
206static unsigned long apic_phys __ro_after_init;
207
208/*
209 * Get the LAPIC version
210 */
211static inline int lapic_get_version(void)
212{
213 return GET_APIC_VERSION(apic_read(APIC_LVR));
214}
215
216/*
217 * Check, if the APIC is integrated or a separate chip
218 */
219static inline int lapic_is_integrated(void)
220{
221 return APIC_INTEGRATED(lapic_get_version());
222}
223
224/*
225 * Check, whether this is a modern or a first generation APIC
226 */
227static int modern_apic(void)
228{
229 /* AMD systems use old APIC versions, so check the CPU */
230 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
231 boot_cpu_data.x86 >= 0xf)
232 return 1;
233
234 /* Hygon systems use modern APIC */
235 if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
236 return 1;
237
238 return lapic_get_version() >= 0x14;
239}
240
241/*
242 * right after this call apic become NOOP driven
243 * so apic->write/read doesn't do anything
244 */
245static void __init apic_disable(void)
246{
247 pr_info("APIC: switched to apic NOOP\n");
248 apic = &apic_noop;
249}
250
251void native_apic_wait_icr_idle(void)
252{
253 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
254 cpu_relax();
255}
256
257u32 native_safe_apic_wait_icr_idle(void)
258{
259 u32 send_status;
260 int timeout;
261
262 timeout = 0;
263 do {
264 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
265 if (!send_status)
266 break;
267 inc_irq_stat(icr_read_retry_count);
268 udelay(100);
269 } while (timeout++ < 1000);
270
271 return send_status;
272}
273
274void native_apic_icr_write(u32 low, u32 id)
275{
276 unsigned long flags;
277
278 local_irq_save(flags);
279 apic_write(APIC_ICR2, SET_XAPIC_DEST_FIELD(id));
280 apic_write(APIC_ICR, low);
281 local_irq_restore(flags);
282}
283
284u64 native_apic_icr_read(void)
285{
286 u32 icr1, icr2;
287
288 icr2 = apic_read(APIC_ICR2);
289 icr1 = apic_read(APIC_ICR);
290
291 return icr1 | ((u64)icr2 << 32);
292}
293
294#ifdef CONFIG_X86_32
295/**
296 * get_physical_broadcast - Get number of physical broadcast IDs
297 */
298int get_physical_broadcast(void)
299{
300 return modern_apic() ? 0xff : 0xf;
301}
302#endif
303
304/**
305 * lapic_get_maxlvt - get the maximum number of local vector table entries
306 */
307int lapic_get_maxlvt(void)
308{
309 /*
310 * - we always have APIC integrated on 64bit mode
311 * - 82489DXs do not report # of LVT entries
312 */
313 return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
314}
315
316/*
317 * Local APIC timer
318 */
319
320/* Clock divisor */
321#define APIC_DIVISOR 16
322#define TSC_DIVISOR 8
323
324/* i82489DX specific */
325#define I82489DX_BASE_DIVIDER (((0x2) << 18))
326
327/*
328 * This function sets up the local APIC timer, with a timeout of
329 * 'clocks' APIC bus clock. During calibration we actually call
330 * this function twice on the boot CPU, once with a bogus timeout
331 * value, second time for real. The other (noncalibrating) CPUs
332 * call this function only once, with the real, calibrated value.
333 *
334 * We do reads before writes even if unnecessary, to get around the
335 * P5 APIC double write bug.
336 */
337static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
338{
339 unsigned int lvtt_value, tmp_value;
340
341 lvtt_value = LOCAL_TIMER_VECTOR;
342 if (!oneshot)
343 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
344 else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
345 lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
346
347 /*
348 * The i82489DX APIC uses bit 18 and 19 for the base divider. This
349 * overlaps with bit 18 on integrated APICs, but is not documented
350 * in the SDM. No problem though. i82489DX equipped systems do not
351 * have TSC deadline timer.
352 */
353 if (!lapic_is_integrated())
354 lvtt_value |= I82489DX_BASE_DIVIDER;
355
356 if (!irqen)
357 lvtt_value |= APIC_LVT_MASKED;
358
359 apic_write(APIC_LVTT, lvtt_value);
360
361 if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
362 /*
363 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
364 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
365 * According to Intel, MFENCE can do the serialization here.
366 */
367 asm volatile("mfence" : : : "memory");
368 return;
369 }
370
371 /*
372 * Divide PICLK by 16
373 */
374 tmp_value = apic_read(APIC_TDCR);
375 apic_write(APIC_TDCR,
376 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
377 APIC_TDR_DIV_16);
378
379 if (!oneshot)
380 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
381}
382
383/*
384 * Setup extended LVT, AMD specific
385 *
386 * Software should use the LVT offsets the BIOS provides. The offsets
387 * are determined by the subsystems using it like those for MCE
388 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts
389 * are supported. Beginning with family 10h at least 4 offsets are
390 * available.
391 *
392 * Since the offsets must be consistent for all cores, we keep track
393 * of the LVT offsets in software and reserve the offset for the same
394 * vector also to be used on other cores. An offset is freed by
395 * setting the entry to APIC_EILVT_MASKED.
396 *
397 * If the BIOS is right, there should be no conflicts. Otherwise a
398 * "[Firmware Bug]: ..." error message is generated. However, if
399 * software does not properly determines the offsets, it is not
400 * necessarily a BIOS bug.
401 */
402
403static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
404
405static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
406{
407 return (old & APIC_EILVT_MASKED)
408 || (new == APIC_EILVT_MASKED)
409 || ((new & ~APIC_EILVT_MASKED) == old);
410}
411
412static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
413{
414 unsigned int rsvd, vector;
415
416 if (offset >= APIC_EILVT_NR_MAX)
417 return ~0;
418
419 rsvd = atomic_read(&eilvt_offsets[offset]);
420 do {
421 vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */
422 if (vector && !eilvt_entry_is_changeable(vector, new))
423 /* may not change if vectors are different */
424 return rsvd;
425 rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
426 } while (rsvd != new);
427
428 rsvd &= ~APIC_EILVT_MASKED;
429 if (rsvd && rsvd != vector)
430 pr_info("LVT offset %d assigned for vector 0x%02x\n",
431 offset, rsvd);
432
433 return new;
434}
435
436/*
437 * If mask=1, the LVT entry does not generate interrupts while mask=0
438 * enables the vector. See also the BKDGs. Must be called with
439 * preemption disabled.
440 */
441
442int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
443{
444 unsigned long reg = APIC_EILVTn(offset);
445 unsigned int new, old, reserved;
446
447 new = (mask << 16) | (msg_type << 8) | vector;
448 old = apic_read(reg);
449 reserved = reserve_eilvt_offset(offset, new);
450
451 if (reserved != new) {
452 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
453 "vector 0x%x, but the register is already in use for "
454 "vector 0x%x on another cpu\n",
455 smp_processor_id(), reg, offset, new, reserved);
456 return -EINVAL;
457 }
458
459 if (!eilvt_entry_is_changeable(old, new)) {
460 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
461 "vector 0x%x, but the register is already in use for "
462 "vector 0x%x on this cpu\n",
463 smp_processor_id(), reg, offset, new, old);
464 return -EBUSY;
465 }
466
467 apic_write(reg, new);
468
469 return 0;
470}
471EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
472
473/*
474 * Program the next event, relative to now
475 */
476static int lapic_next_event(unsigned long delta,
477 struct clock_event_device *evt)
478{
479 apic_write(APIC_TMICT, delta);
480 return 0;
481}
482
483static int lapic_next_deadline(unsigned long delta,
484 struct clock_event_device *evt)
485{
486 u64 tsc;
487
488 /* This MSR is special and need a special fence: */
489 weak_wrmsr_fence();
490
491 tsc = rdtsc();
492 wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
493 return 0;
494}
495
496static int lapic_timer_shutdown(struct clock_event_device *evt)
497{
498 unsigned int v;
499
500 /* Lapic used as dummy for broadcast ? */
501 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
502 return 0;
503
504 v = apic_read(APIC_LVTT);
505 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
506 apic_write(APIC_LVTT, v);
507 apic_write(APIC_TMICT, 0);
508 return 0;
509}
510
511static inline int
512lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
513{
514 /* Lapic used as dummy for broadcast ? */
515 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
516 return 0;
517
518 __setup_APIC_LVTT(lapic_timer_period, oneshot, 1);
519 return 0;
520}
521
522static int lapic_timer_set_periodic(struct clock_event_device *evt)
523{
524 return lapic_timer_set_periodic_oneshot(evt, false);
525}
526
527static int lapic_timer_set_oneshot(struct clock_event_device *evt)
528{
529 return lapic_timer_set_periodic_oneshot(evt, true);
530}
531
532/*
533 * Local APIC timer broadcast function
534 */
535static void lapic_timer_broadcast(const struct cpumask *mask)
536{
537#ifdef CONFIG_SMP
538 apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
539#endif
540}
541
542
543/*
544 * The local apic timer can be used for any function which is CPU local.
545 */
546static struct clock_event_device lapic_clockevent = {
547 .name = "lapic",
548 .features = CLOCK_EVT_FEAT_PERIODIC |
549 CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
550 | CLOCK_EVT_FEAT_DUMMY,
551 .shift = 32,
552 .set_state_shutdown = lapic_timer_shutdown,
553 .set_state_periodic = lapic_timer_set_periodic,
554 .set_state_oneshot = lapic_timer_set_oneshot,
555 .set_state_oneshot_stopped = lapic_timer_shutdown,
556 .set_next_event = lapic_next_event,
557 .broadcast = lapic_timer_broadcast,
558 .rating = 100,
559 .irq = -1,
560};
561static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
562
563static const struct x86_cpu_id deadline_match[] __initconst = {
564 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x2, 0x2), 0x3a), /* EP */
565 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x4, 0x4), 0x0f), /* EX */
566
567 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_X, 0x0b000020),
568
569 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x2, 0x2), 0x00000011),
570 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x3, 0x3), 0x0700000e),
571 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x4, 0x4), 0x0f00000c),
572 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x5, 0x5), 0x0e000003),
573
574 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x3, 0x3), 0x01000136),
575 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x4, 0x4), 0x02000014),
576 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x5, 0xf), 0),
577
578 X86_MATCH_INTEL_FAM6_MODEL( HASWELL, 0x22),
579 X86_MATCH_INTEL_FAM6_MODEL( HASWELL_L, 0x20),
580 X86_MATCH_INTEL_FAM6_MODEL( HASWELL_G, 0x17),
581
582 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL, 0x25),
583 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_G, 0x17),
584
585 X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE_L, 0xb2),
586 X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE, 0xb2),
587
588 X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE_L, 0x52),
589 X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE, 0x52),
590
591 {},
592};
593
594static __init bool apic_validate_deadline_timer(void)
595{
596 const struct x86_cpu_id *m;
597 u32 rev;
598
599 if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
600 return false;
601 if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
602 return true;
603
604 m = x86_match_cpu(deadline_match);
605 if (!m)
606 return true;
607
608 rev = (u32)m->driver_data;
609
610 if (boot_cpu_data.microcode >= rev)
611 return true;
612
613 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
614 pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
615 "please update microcode to version: 0x%x (or later)\n", rev);
616 return false;
617}
618
619/*
620 * Setup the local APIC timer for this CPU. Copy the initialized values
621 * of the boot CPU and register the clock event in the framework.
622 */
623static void setup_APIC_timer(void)
624{
625 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
626
627 if (this_cpu_has(X86_FEATURE_ARAT)) {
628 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
629 /* Make LAPIC timer preferable over percpu HPET */
630 lapic_clockevent.rating = 150;
631 }
632
633 memcpy(levt, &lapic_clockevent, sizeof(*levt));
634 levt->cpumask = cpumask_of(smp_processor_id());
635
636 if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
637 levt->name = "lapic-deadline";
638 levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
639 CLOCK_EVT_FEAT_DUMMY);
640 levt->set_next_event = lapic_next_deadline;
641 clockevents_config_and_register(levt,
642 tsc_khz * (1000 / TSC_DIVISOR),
643 0xF, ~0UL);
644 } else
645 clockevents_register_device(levt);
646}
647
648/*
649 * Install the updated TSC frequency from recalibration at the TSC
650 * deadline clockevent devices.
651 */
652static void __lapic_update_tsc_freq(void *info)
653{
654 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
655
656 if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
657 return;
658
659 clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
660}
661
662void lapic_update_tsc_freq(void)
663{
664 /*
665 * The clockevent device's ->mult and ->shift can both be
666 * changed. In order to avoid races, schedule the frequency
667 * update code on each CPU.
668 */
669 on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
670}
671
672/*
673 * In this functions we calibrate APIC bus clocks to the external timer.
674 *
675 * We want to do the calibration only once since we want to have local timer
676 * irqs synchronous. CPUs connected by the same APIC bus have the very same bus
677 * frequency.
678 *
679 * This was previously done by reading the PIT/HPET and waiting for a wrap
680 * around to find out, that a tick has elapsed. I have a box, where the PIT
681 * readout is broken, so it never gets out of the wait loop again. This was
682 * also reported by others.
683 *
684 * Monitoring the jiffies value is inaccurate and the clockevents
685 * infrastructure allows us to do a simple substitution of the interrupt
686 * handler.
687 *
688 * The calibration routine also uses the pm_timer when possible, as the PIT
689 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
690 * back to normal later in the boot process).
691 */
692
693#define LAPIC_CAL_LOOPS (HZ/10)
694
695static __initdata int lapic_cal_loops = -1;
696static __initdata long lapic_cal_t1, lapic_cal_t2;
697static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
698static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
699static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
700
701/*
702 * Temporary interrupt handler and polled calibration function.
703 */
704static void __init lapic_cal_handler(struct clock_event_device *dev)
705{
706 unsigned long long tsc = 0;
707 long tapic = apic_read(APIC_TMCCT);
708 unsigned long pm = acpi_pm_read_early();
709
710 if (boot_cpu_has(X86_FEATURE_TSC))
711 tsc = rdtsc();
712
713 switch (lapic_cal_loops++) {
714 case 0:
715 lapic_cal_t1 = tapic;
716 lapic_cal_tsc1 = tsc;
717 lapic_cal_pm1 = pm;
718 lapic_cal_j1 = jiffies;
719 break;
720
721 case LAPIC_CAL_LOOPS:
722 lapic_cal_t2 = tapic;
723 lapic_cal_tsc2 = tsc;
724 if (pm < lapic_cal_pm1)
725 pm += ACPI_PM_OVRRUN;
726 lapic_cal_pm2 = pm;
727 lapic_cal_j2 = jiffies;
728 break;
729 }
730}
731
732static int __init
733calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
734{
735 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
736 const long pm_thresh = pm_100ms / 100;
737 unsigned long mult;
738 u64 res;
739
740#ifndef CONFIG_X86_PM_TIMER
741 return -1;
742#endif
743
744 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
745
746 /* Check, if the PM timer is available */
747 if (!deltapm)
748 return -1;
749
750 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
751
752 if (deltapm > (pm_100ms - pm_thresh) &&
753 deltapm < (pm_100ms + pm_thresh)) {
754 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
755 return 0;
756 }
757
758 res = (((u64)deltapm) * mult) >> 22;
759 do_div(res, 1000000);
760 pr_warn("APIC calibration not consistent "
761 "with PM-Timer: %ldms instead of 100ms\n", (long)res);
762
763 /* Correct the lapic counter value */
764 res = (((u64)(*delta)) * pm_100ms);
765 do_div(res, deltapm);
766 pr_info("APIC delta adjusted to PM-Timer: "
767 "%lu (%ld)\n", (unsigned long)res, *delta);
768 *delta = (long)res;
769
770 /* Correct the tsc counter value */
771 if (boot_cpu_has(X86_FEATURE_TSC)) {
772 res = (((u64)(*deltatsc)) * pm_100ms);
773 do_div(res, deltapm);
774 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
775 "PM-Timer: %lu (%ld)\n",
776 (unsigned long)res, *deltatsc);
777 *deltatsc = (long)res;
778 }
779
780 return 0;
781}
782
783static int __init lapic_init_clockevent(void)
784{
785 if (!lapic_timer_period)
786 return -1;
787
788 /* Calculate the scaled math multiplication factor */
789 lapic_clockevent.mult = div_sc(lapic_timer_period/APIC_DIVISOR,
790 TICK_NSEC, lapic_clockevent.shift);
791 lapic_clockevent.max_delta_ns =
792 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
793 lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
794 lapic_clockevent.min_delta_ns =
795 clockevent_delta2ns(0xF, &lapic_clockevent);
796 lapic_clockevent.min_delta_ticks = 0xF;
797
798 return 0;
799}
800
801bool __init apic_needs_pit(void)
802{
803 /*
804 * If the frequencies are not known, PIT is required for both TSC
805 * and apic timer calibration.
806 */
807 if (!tsc_khz || !cpu_khz)
808 return true;
809
810 /* Is there an APIC at all or is it disabled? */
811 if (!boot_cpu_has(X86_FEATURE_APIC) || disable_apic)
812 return true;
813
814 /*
815 * If interrupt delivery mode is legacy PIC or virtual wire without
816 * configuration, the local APIC timer wont be set up. Make sure
817 * that the PIT is initialized.
818 */
819 if (apic_intr_mode == APIC_PIC ||
820 apic_intr_mode == APIC_VIRTUAL_WIRE_NO_CONFIG)
821 return true;
822
823 /* Virt guests may lack ARAT, but still have DEADLINE */
824 if (!boot_cpu_has(X86_FEATURE_ARAT))
825 return true;
826
827 /* Deadline timer is based on TSC so no further PIT action required */
828 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
829 return false;
830
831 /* APIC timer disabled? */
832 if (disable_apic_timer)
833 return true;
834 /*
835 * The APIC timer frequency is known already, no PIT calibration
836 * required. If unknown, let the PIT be initialized.
837 */
838 return lapic_timer_period == 0;
839}
840
841static int __init calibrate_APIC_clock(void)
842{
843 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
844 u64 tsc_perj = 0, tsc_start = 0;
845 unsigned long jif_start;
846 unsigned long deltaj;
847 long delta, deltatsc;
848 int pm_referenced = 0;
849
850 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
851 return 0;
852
853 /*
854 * Check if lapic timer has already been calibrated by platform
855 * specific routine, such as tsc calibration code. If so just fill
856 * in the clockevent structure and return.
857 */
858 if (!lapic_init_clockevent()) {
859 apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
860 lapic_timer_period);
861 /*
862 * Direct calibration methods must have an always running
863 * local APIC timer, no need for broadcast timer.
864 */
865 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
866 return 0;
867 }
868
869 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
870 "calibrating APIC timer ...\n");
871
872 /*
873 * There are platforms w/o global clockevent devices. Instead of
874 * making the calibration conditional on that, use a polling based
875 * approach everywhere.
876 */
877 local_irq_disable();
878
879 /*
880 * Setup the APIC counter to maximum. There is no way the lapic
881 * can underflow in the 100ms detection time frame
882 */
883 __setup_APIC_LVTT(0xffffffff, 0, 0);
884
885 /*
886 * Methods to terminate the calibration loop:
887 * 1) Global clockevent if available (jiffies)
888 * 2) TSC if available and frequency is known
889 */
890 jif_start = READ_ONCE(jiffies);
891
892 if (tsc_khz) {
893 tsc_start = rdtsc();
894 tsc_perj = div_u64((u64)tsc_khz * 1000, HZ);
895 }
896
897 /*
898 * Enable interrupts so the tick can fire, if a global
899 * clockevent device is available
900 */
901 local_irq_enable();
902
903 while (lapic_cal_loops <= LAPIC_CAL_LOOPS) {
904 /* Wait for a tick to elapse */
905 while (1) {
906 if (tsc_khz) {
907 u64 tsc_now = rdtsc();
908 if ((tsc_now - tsc_start) >= tsc_perj) {
909 tsc_start += tsc_perj;
910 break;
911 }
912 } else {
913 unsigned long jif_now = READ_ONCE(jiffies);
914
915 if (time_after(jif_now, jif_start)) {
916 jif_start = jif_now;
917 break;
918 }
919 }
920 cpu_relax();
921 }
922
923 /* Invoke the calibration routine */
924 local_irq_disable();
925 lapic_cal_handler(NULL);
926 local_irq_enable();
927 }
928
929 local_irq_disable();
930
931 /* Build delta t1-t2 as apic timer counts down */
932 delta = lapic_cal_t1 - lapic_cal_t2;
933 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
934
935 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
936
937 /* we trust the PM based calibration if possible */
938 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
939 &delta, &deltatsc);
940
941 lapic_timer_period = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
942 lapic_init_clockevent();
943
944 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
945 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
946 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
947 lapic_timer_period);
948
949 if (boot_cpu_has(X86_FEATURE_TSC)) {
950 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
951 "%ld.%04ld MHz.\n",
952 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
953 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
954 }
955
956 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
957 "%u.%04u MHz.\n",
958 lapic_timer_period / (1000000 / HZ),
959 lapic_timer_period % (1000000 / HZ));
960
961 /*
962 * Do a sanity check on the APIC calibration result
963 */
964 if (lapic_timer_period < (1000000 / HZ)) {
965 local_irq_enable();
966 pr_warn("APIC frequency too slow, disabling apic timer\n");
967 return -1;
968 }
969
970 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
971
972 /*
973 * PM timer calibration failed or not turned on so lets try APIC
974 * timer based calibration, if a global clockevent device is
975 * available.
976 */
977 if (!pm_referenced && global_clock_event) {
978 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
979
980 /*
981 * Setup the apic timer manually
982 */
983 levt->event_handler = lapic_cal_handler;
984 lapic_timer_set_periodic(levt);
985 lapic_cal_loops = -1;
986
987 /* Let the interrupts run */
988 local_irq_enable();
989
990 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
991 cpu_relax();
992
993 /* Stop the lapic timer */
994 local_irq_disable();
995 lapic_timer_shutdown(levt);
996
997 /* Jiffies delta */
998 deltaj = lapic_cal_j2 - lapic_cal_j1;
999 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
1000
1001 /* Check, if the jiffies result is consistent */
1002 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
1003 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
1004 else
1005 levt->features |= CLOCK_EVT_FEAT_DUMMY;
1006 }
1007 local_irq_enable();
1008
1009 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
1010 pr_warn("APIC timer disabled due to verification failure\n");
1011 return -1;
1012 }
1013
1014 return 0;
1015}
1016
1017/*
1018 * Setup the boot APIC
1019 *
1020 * Calibrate and verify the result.
1021 */
1022void __init setup_boot_APIC_clock(void)
1023{
1024 /*
1025 * The local apic timer can be disabled via the kernel
1026 * commandline or from the CPU detection code. Register the lapic
1027 * timer as a dummy clock event source on SMP systems, so the
1028 * broadcast mechanism is used. On UP systems simply ignore it.
1029 */
1030 if (disable_apic_timer) {
1031 pr_info("Disabling APIC timer\n");
1032 /* No broadcast on UP ! */
1033 if (num_possible_cpus() > 1) {
1034 lapic_clockevent.mult = 1;
1035 setup_APIC_timer();
1036 }
1037 return;
1038 }
1039
1040 if (calibrate_APIC_clock()) {
1041 /* No broadcast on UP ! */
1042 if (num_possible_cpus() > 1)
1043 setup_APIC_timer();
1044 return;
1045 }
1046
1047 /*
1048 * If nmi_watchdog is set to IO_APIC, we need the
1049 * PIT/HPET going. Otherwise register lapic as a dummy
1050 * device.
1051 */
1052 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
1053
1054 /* Setup the lapic or request the broadcast */
1055 setup_APIC_timer();
1056 amd_e400_c1e_apic_setup();
1057}
1058
1059void setup_secondary_APIC_clock(void)
1060{
1061 setup_APIC_timer();
1062 amd_e400_c1e_apic_setup();
1063}
1064
1065/*
1066 * The guts of the apic timer interrupt
1067 */
1068static void local_apic_timer_interrupt(void)
1069{
1070 struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1071
1072 /*
1073 * Normally we should not be here till LAPIC has been initialized but
1074 * in some cases like kdump, its possible that there is a pending LAPIC
1075 * timer interrupt from previous kernel's context and is delivered in
1076 * new kernel the moment interrupts are enabled.
1077 *
1078 * Interrupts are enabled early and LAPIC is setup much later, hence
1079 * its possible that when we get here evt->event_handler is NULL.
1080 * Check for event_handler being NULL and discard the interrupt as
1081 * spurious.
1082 */
1083 if (!evt->event_handler) {
1084 pr_warn("Spurious LAPIC timer interrupt on cpu %d\n",
1085 smp_processor_id());
1086 /* Switch it off */
1087 lapic_timer_shutdown(evt);
1088 return;
1089 }
1090
1091 /*
1092 * the NMI deadlock-detector uses this.
1093 */
1094 inc_irq_stat(apic_timer_irqs);
1095
1096 evt->event_handler(evt);
1097}
1098
1099/*
1100 * Local APIC timer interrupt. This is the most natural way for doing
1101 * local interrupts, but local timer interrupts can be emulated by
1102 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1103 *
1104 * [ if a single-CPU system runs an SMP kernel then we call the local
1105 * interrupt as well. Thus we cannot inline the local irq ... ]
1106 */
1107DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt)
1108{
1109 struct pt_regs *old_regs = set_irq_regs(regs);
1110
1111 ack_APIC_irq();
1112 trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1113 local_apic_timer_interrupt();
1114 trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1115
1116 set_irq_regs(old_regs);
1117}
1118
1119/*
1120 * Local APIC start and shutdown
1121 */
1122
1123/**
1124 * clear_local_APIC - shutdown the local APIC
1125 *
1126 * This is called, when a CPU is disabled and before rebooting, so the state of
1127 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1128 * leftovers during boot.
1129 */
1130void clear_local_APIC(void)
1131{
1132 int maxlvt;
1133 u32 v;
1134
1135 /* APIC hasn't been mapped yet */
1136 if (!x2apic_mode && !apic_phys)
1137 return;
1138
1139 maxlvt = lapic_get_maxlvt();
1140 /*
1141 * Masking an LVT entry can trigger a local APIC error
1142 * if the vector is zero. Mask LVTERR first to prevent this.
1143 */
1144 if (maxlvt >= 3) {
1145 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1146 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1147 }
1148 /*
1149 * Careful: we have to set masks only first to deassert
1150 * any level-triggered sources.
1151 */
1152 v = apic_read(APIC_LVTT);
1153 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1154 v = apic_read(APIC_LVT0);
1155 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1156 v = apic_read(APIC_LVT1);
1157 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1158 if (maxlvt >= 4) {
1159 v = apic_read(APIC_LVTPC);
1160 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1161 }
1162
1163 /* lets not touch this if we didn't frob it */
1164#ifdef CONFIG_X86_THERMAL_VECTOR
1165 if (maxlvt >= 5) {
1166 v = apic_read(APIC_LVTTHMR);
1167 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1168 }
1169#endif
1170#ifdef CONFIG_X86_MCE_INTEL
1171 if (maxlvt >= 6) {
1172 v = apic_read(APIC_LVTCMCI);
1173 if (!(v & APIC_LVT_MASKED))
1174 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1175 }
1176#endif
1177
1178 /*
1179 * Clean APIC state for other OSs:
1180 */
1181 apic_write(APIC_LVTT, APIC_LVT_MASKED);
1182 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1183 apic_write(APIC_LVT1, APIC_LVT_MASKED);
1184 if (maxlvt >= 3)
1185 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1186 if (maxlvt >= 4)
1187 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1188
1189 /* Integrated APIC (!82489DX) ? */
1190 if (lapic_is_integrated()) {
1191 if (maxlvt > 3)
1192 /* Clear ESR due to Pentium errata 3AP and 11AP */
1193 apic_write(APIC_ESR, 0);
1194 apic_read(APIC_ESR);
1195 }
1196}
1197
1198/**
1199 * apic_soft_disable - Clears and software disables the local APIC on hotplug
1200 *
1201 * Contrary to disable_local_APIC() this does not touch the enable bit in
1202 * MSR_IA32_APICBASE. Clearing that bit on systems based on the 3 wire APIC
1203 * bus would require a hardware reset as the APIC would lose track of bus
1204 * arbitration. On systems with FSB delivery APICBASE could be disabled,
1205 * but it has to be guaranteed that no interrupt is sent to the APIC while
1206 * in that state and it's not clear from the SDM whether it still responds
1207 * to INIT/SIPI messages. Stay on the safe side and use software disable.
1208 */
1209void apic_soft_disable(void)
1210{
1211 u32 value;
1212
1213 clear_local_APIC();
1214
1215 /* Soft disable APIC (implies clearing of registers for 82489DX!). */
1216 value = apic_read(APIC_SPIV);
1217 value &= ~APIC_SPIV_APIC_ENABLED;
1218 apic_write(APIC_SPIV, value);
1219}
1220
1221/**
1222 * disable_local_APIC - clear and disable the local APIC
1223 */
1224void disable_local_APIC(void)
1225{
1226 /* APIC hasn't been mapped yet */
1227 if (!x2apic_mode && !apic_phys)
1228 return;
1229
1230 apic_soft_disable();
1231
1232#ifdef CONFIG_X86_32
1233 /*
1234 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1235 * restore the disabled state.
1236 */
1237 if (enabled_via_apicbase) {
1238 unsigned int l, h;
1239
1240 rdmsr(MSR_IA32_APICBASE, l, h);
1241 l &= ~MSR_IA32_APICBASE_ENABLE;
1242 wrmsr(MSR_IA32_APICBASE, l, h);
1243 }
1244#endif
1245}
1246
1247/*
1248 * If Linux enabled the LAPIC against the BIOS default disable it down before
1249 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1250 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1251 * for the case where Linux didn't enable the LAPIC.
1252 */
1253void lapic_shutdown(void)
1254{
1255 unsigned long flags;
1256
1257 if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1258 return;
1259
1260 local_irq_save(flags);
1261
1262#ifdef CONFIG_X86_32
1263 if (!enabled_via_apicbase)
1264 clear_local_APIC();
1265 else
1266#endif
1267 disable_local_APIC();
1268
1269
1270 local_irq_restore(flags);
1271}
1272
1273/**
1274 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1275 */
1276void __init sync_Arb_IDs(void)
1277{
1278 /*
1279 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1280 * needed on AMD.
1281 */
1282 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1283 return;
1284
1285 /*
1286 * Wait for idle.
1287 */
1288 apic_wait_icr_idle();
1289
1290 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1291 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1292 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1293}
1294
1295enum apic_intr_mode_id apic_intr_mode __ro_after_init;
1296
1297static int __init __apic_intr_mode_select(void)
1298{
1299 /* Check kernel option */
1300 if (disable_apic) {
1301 pr_info("APIC disabled via kernel command line\n");
1302 return APIC_PIC;
1303 }
1304
1305 /* Check BIOS */
1306#ifdef CONFIG_X86_64
1307 /* On 64-bit, the APIC must be integrated, Check local APIC only */
1308 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1309 disable_apic = 1;
1310 pr_info("APIC disabled by BIOS\n");
1311 return APIC_PIC;
1312 }
1313#else
1314 /* On 32-bit, the APIC may be integrated APIC or 82489DX */
1315
1316 /* Neither 82489DX nor integrated APIC ? */
1317 if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1318 disable_apic = 1;
1319 return APIC_PIC;
1320 }
1321
1322 /* If the BIOS pretends there is an integrated APIC ? */
1323 if (!boot_cpu_has(X86_FEATURE_APIC) &&
1324 APIC_INTEGRATED(boot_cpu_apic_version)) {
1325 disable_apic = 1;
1326 pr_err(FW_BUG "Local APIC %d not detected, force emulation\n",
1327 boot_cpu_physical_apicid);
1328 return APIC_PIC;
1329 }
1330#endif
1331
1332 /* Check MP table or ACPI MADT configuration */
1333 if (!smp_found_config) {
1334 disable_ioapic_support();
1335 if (!acpi_lapic) {
1336 pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1337 return APIC_VIRTUAL_WIRE_NO_CONFIG;
1338 }
1339 return APIC_VIRTUAL_WIRE;
1340 }
1341
1342#ifdef CONFIG_SMP
1343 /* If SMP should be disabled, then really disable it! */
1344 if (!setup_max_cpus) {
1345 pr_info("APIC: SMP mode deactivated\n");
1346 return APIC_SYMMETRIC_IO_NO_ROUTING;
1347 }
1348
1349 if (read_apic_id() != boot_cpu_physical_apicid) {
1350 panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1351 read_apic_id(), boot_cpu_physical_apicid);
1352 /* Or can we switch back to PIC here? */
1353 }
1354#endif
1355
1356 return APIC_SYMMETRIC_IO;
1357}
1358
1359/* Select the interrupt delivery mode for the BSP */
1360void __init apic_intr_mode_select(void)
1361{
1362 apic_intr_mode = __apic_intr_mode_select();
1363}
1364
1365/*
1366 * An initial setup of the virtual wire mode.
1367 */
1368void __init init_bsp_APIC(void)
1369{
1370 unsigned int value;
1371
1372 /*
1373 * Don't do the setup now if we have a SMP BIOS as the
1374 * through-I/O-APIC virtual wire mode might be active.
1375 */
1376 if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1377 return;
1378
1379 /*
1380 * Do not trust the local APIC being empty at bootup.
1381 */
1382 clear_local_APIC();
1383
1384 /*
1385 * Enable APIC.
1386 */
1387 value = apic_read(APIC_SPIV);
1388 value &= ~APIC_VECTOR_MASK;
1389 value |= APIC_SPIV_APIC_ENABLED;
1390
1391#ifdef CONFIG_X86_32
1392 /* This bit is reserved on P4/Xeon and should be cleared */
1393 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1394 (boot_cpu_data.x86 == 15))
1395 value &= ~APIC_SPIV_FOCUS_DISABLED;
1396 else
1397#endif
1398 value |= APIC_SPIV_FOCUS_DISABLED;
1399 value |= SPURIOUS_APIC_VECTOR;
1400 apic_write(APIC_SPIV, value);
1401
1402 /*
1403 * Set up the virtual wire mode.
1404 */
1405 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1406 value = APIC_DM_NMI;
1407 if (!lapic_is_integrated()) /* 82489DX */
1408 value |= APIC_LVT_LEVEL_TRIGGER;
1409 if (apic_extnmi == APIC_EXTNMI_NONE)
1410 value |= APIC_LVT_MASKED;
1411 apic_write(APIC_LVT1, value);
1412}
1413
1414static void __init apic_bsp_setup(bool upmode);
1415
1416/* Init the interrupt delivery mode for the BSP */
1417void __init apic_intr_mode_init(void)
1418{
1419 bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1420
1421 switch (apic_intr_mode) {
1422 case APIC_PIC:
1423 pr_info("APIC: Keep in PIC mode(8259)\n");
1424 return;
1425 case APIC_VIRTUAL_WIRE:
1426 pr_info("APIC: Switch to virtual wire mode setup\n");
1427 break;
1428 case APIC_VIRTUAL_WIRE_NO_CONFIG:
1429 pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1430 upmode = true;
1431 break;
1432 case APIC_SYMMETRIC_IO:
1433 pr_info("APIC: Switch to symmetric I/O mode setup\n");
1434 break;
1435 case APIC_SYMMETRIC_IO_NO_ROUTING:
1436 pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1437 break;
1438 }
1439
1440 default_setup_apic_routing();
1441
1442 if (x86_platform.apic_post_init)
1443 x86_platform.apic_post_init();
1444
1445 apic_bsp_setup(upmode);
1446}
1447
1448static void lapic_setup_esr(void)
1449{
1450 unsigned int oldvalue, value, maxlvt;
1451
1452 if (!lapic_is_integrated()) {
1453 pr_info("No ESR for 82489DX.\n");
1454 return;
1455 }
1456
1457 if (apic->disable_esr) {
1458 /*
1459 * Something untraceable is creating bad interrupts on
1460 * secondary quads ... for the moment, just leave the
1461 * ESR disabled - we can't do anything useful with the
1462 * errors anyway - mbligh
1463 */
1464 pr_info("Leaving ESR disabled.\n");
1465 return;
1466 }
1467
1468 maxlvt = lapic_get_maxlvt();
1469 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1470 apic_write(APIC_ESR, 0);
1471 oldvalue = apic_read(APIC_ESR);
1472
1473 /* enables sending errors */
1474 value = ERROR_APIC_VECTOR;
1475 apic_write(APIC_LVTERR, value);
1476
1477 /*
1478 * spec says clear errors after enabling vector.
1479 */
1480 if (maxlvt > 3)
1481 apic_write(APIC_ESR, 0);
1482 value = apic_read(APIC_ESR);
1483 if (value != oldvalue)
1484 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1485 "vector: 0x%08x after: 0x%08x\n",
1486 oldvalue, value);
1487}
1488
1489#define APIC_IR_REGS APIC_ISR_NR
1490#define APIC_IR_BITS (APIC_IR_REGS * 32)
1491#define APIC_IR_MAPSIZE (APIC_IR_BITS / BITS_PER_LONG)
1492
1493union apic_ir {
1494 unsigned long map[APIC_IR_MAPSIZE];
1495 u32 regs[APIC_IR_REGS];
1496};
1497
1498static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr)
1499{
1500 int i, bit;
1501
1502 /* Read the IRRs */
1503 for (i = 0; i < APIC_IR_REGS; i++)
1504 irr->regs[i] = apic_read(APIC_IRR + i * 0x10);
1505
1506 /* Read the ISRs */
1507 for (i = 0; i < APIC_IR_REGS; i++)
1508 isr->regs[i] = apic_read(APIC_ISR + i * 0x10);
1509
1510 /*
1511 * If the ISR map is not empty. ACK the APIC and run another round
1512 * to verify whether a pending IRR has been unblocked and turned
1513 * into a ISR.
1514 */
1515 if (!bitmap_empty(isr->map, APIC_IR_BITS)) {
1516 /*
1517 * There can be multiple ISR bits set when a high priority
1518 * interrupt preempted a lower priority one. Issue an ACK
1519 * per set bit.
1520 */
1521 for_each_set_bit(bit, isr->map, APIC_IR_BITS)
1522 ack_APIC_irq();
1523 return true;
1524 }
1525
1526 return !bitmap_empty(irr->map, APIC_IR_BITS);
1527}
1528
1529/*
1530 * After a crash, we no longer service the interrupts and a pending
1531 * interrupt from previous kernel might still have ISR bit set.
1532 *
1533 * Most probably by now the CPU has serviced that pending interrupt and it
1534 * might not have done the ack_APIC_irq() because it thought, interrupt
1535 * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear
1536 * the ISR bit and cpu thinks it has already serviced the interrupt. Hence
1537 * a vector might get locked. It was noticed for timer irq (vector
1538 * 0x31). Issue an extra EOI to clear ISR.
1539 *
1540 * If there are pending IRR bits they turn into ISR bits after a higher
1541 * priority ISR bit has been acked.
1542 */
1543static void apic_pending_intr_clear(void)
1544{
1545 union apic_ir irr, isr;
1546 unsigned int i;
1547
1548 /* 512 loops are way oversized and give the APIC a chance to obey. */
1549 for (i = 0; i < 512; i++) {
1550 if (!apic_check_and_ack(&irr, &isr))
1551 return;
1552 }
1553 /* Dump the IRR/ISR content if that failed */
1554 pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map);
1555}
1556
1557/**
1558 * setup_local_APIC - setup the local APIC
1559 *
1560 * Used to setup local APIC while initializing BSP or bringing up APs.
1561 * Always called with preemption disabled.
1562 */
1563static void setup_local_APIC(void)
1564{
1565 int cpu = smp_processor_id();
1566 unsigned int value;
1567
1568 if (disable_apic) {
1569 disable_ioapic_support();
1570 return;
1571 }
1572
1573 /*
1574 * If this comes from kexec/kcrash the APIC might be enabled in
1575 * SPIV. Soft disable it before doing further initialization.
1576 */
1577 value = apic_read(APIC_SPIV);
1578 value &= ~APIC_SPIV_APIC_ENABLED;
1579 apic_write(APIC_SPIV, value);
1580
1581#ifdef CONFIG_X86_32
1582 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1583 if (lapic_is_integrated() && apic->disable_esr) {
1584 apic_write(APIC_ESR, 0);
1585 apic_write(APIC_ESR, 0);
1586 apic_write(APIC_ESR, 0);
1587 apic_write(APIC_ESR, 0);
1588 }
1589#endif
1590 /*
1591 * Double-check whether this APIC is really registered.
1592 * This is meaningless in clustered apic mode, so we skip it.
1593 */
1594 BUG_ON(!apic->apic_id_registered());
1595
1596 /*
1597 * Intel recommends to set DFR, LDR and TPR before enabling
1598 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1599 * document number 292116). So here it goes...
1600 */
1601 apic->init_apic_ldr();
1602
1603#ifdef CONFIG_X86_32
1604 if (apic->dest_mode_logical) {
1605 int logical_apicid, ldr_apicid;
1606
1607 /*
1608 * APIC LDR is initialized. If logical_apicid mapping was
1609 * initialized during get_smp_config(), make sure it matches
1610 * the actual value.
1611 */
1612 logical_apicid = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1613 ldr_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1614 if (logical_apicid != BAD_APICID)
1615 WARN_ON(logical_apicid != ldr_apicid);
1616 /* Always use the value from LDR. */
1617 early_per_cpu(x86_cpu_to_logical_apicid, cpu) = ldr_apicid;
1618 }
1619#endif
1620
1621 /*
1622 * Set Task Priority to 'accept all except vectors 0-31'. An APIC
1623 * vector in the 16-31 range could be delivered if TPR == 0, but we
1624 * would think it's an exception and terrible things will happen. We
1625 * never change this later on.
1626 */
1627 value = apic_read(APIC_TASKPRI);
1628 value &= ~APIC_TPRI_MASK;
1629 value |= 0x10;
1630 apic_write(APIC_TASKPRI, value);
1631
1632 /* Clear eventually stale ISR/IRR bits */
1633 apic_pending_intr_clear();
1634
1635 /*
1636 * Now that we are all set up, enable the APIC
1637 */
1638 value = apic_read(APIC_SPIV);
1639 value &= ~APIC_VECTOR_MASK;
1640 /*
1641 * Enable APIC
1642 */
1643 value |= APIC_SPIV_APIC_ENABLED;
1644
1645#ifdef CONFIG_X86_32
1646 /*
1647 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1648 * certain networking cards. If high frequency interrupts are
1649 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1650 * entry is masked/unmasked at a high rate as well then sooner or
1651 * later IOAPIC line gets 'stuck', no more interrupts are received
1652 * from the device. If focus CPU is disabled then the hang goes
1653 * away, oh well :-(
1654 *
1655 * [ This bug can be reproduced easily with a level-triggered
1656 * PCI Ne2000 networking cards and PII/PIII processors, dual
1657 * BX chipset. ]
1658 */
1659 /*
1660 * Actually disabling the focus CPU check just makes the hang less
1661 * frequent as it makes the interrupt distribution model be more
1662 * like LRU than MRU (the short-term load is more even across CPUs).
1663 */
1664
1665 /*
1666 * - enable focus processor (bit==0)
1667 * - 64bit mode always use processor focus
1668 * so no need to set it
1669 */
1670 value &= ~APIC_SPIV_FOCUS_DISABLED;
1671#endif
1672
1673 /*
1674 * Set spurious IRQ vector
1675 */
1676 value |= SPURIOUS_APIC_VECTOR;
1677 apic_write(APIC_SPIV, value);
1678
1679 perf_events_lapic_init();
1680
1681 /*
1682 * Set up LVT0, LVT1:
1683 *
1684 * set up through-local-APIC on the boot CPU's LINT0. This is not
1685 * strictly necessary in pure symmetric-IO mode, but sometimes
1686 * we delegate interrupts to the 8259A.
1687 */
1688 /*
1689 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1690 */
1691 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1692 if (!cpu && (pic_mode || !value || skip_ioapic_setup)) {
1693 value = APIC_DM_EXTINT;
1694 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1695 } else {
1696 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1697 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1698 }
1699 apic_write(APIC_LVT0, value);
1700
1701 /*
1702 * Only the BSP sees the LINT1 NMI signal by default. This can be
1703 * modified by apic_extnmi= boot option.
1704 */
1705 if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1706 apic_extnmi == APIC_EXTNMI_ALL)
1707 value = APIC_DM_NMI;
1708 else
1709 value = APIC_DM_NMI | APIC_LVT_MASKED;
1710
1711 /* Is 82489DX ? */
1712 if (!lapic_is_integrated())
1713 value |= APIC_LVT_LEVEL_TRIGGER;
1714 apic_write(APIC_LVT1, value);
1715
1716#ifdef CONFIG_X86_MCE_INTEL
1717 /* Recheck CMCI information after local APIC is up on CPU #0 */
1718 if (!cpu)
1719 cmci_recheck();
1720#endif
1721}
1722
1723static void end_local_APIC_setup(void)
1724{
1725 lapic_setup_esr();
1726
1727#ifdef CONFIG_X86_32
1728 {
1729 unsigned int value;
1730 /* Disable the local apic timer */
1731 value = apic_read(APIC_LVTT);
1732 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1733 apic_write(APIC_LVTT, value);
1734 }
1735#endif
1736
1737 apic_pm_activate();
1738}
1739
1740/*
1741 * APIC setup function for application processors. Called from smpboot.c
1742 */
1743void apic_ap_setup(void)
1744{
1745 setup_local_APIC();
1746 end_local_APIC_setup();
1747}
1748
1749#ifdef CONFIG_X86_X2APIC
1750int x2apic_mode;
1751EXPORT_SYMBOL_GPL(x2apic_mode);
1752
1753enum {
1754 X2APIC_OFF,
1755 X2APIC_DISABLED,
1756 /* All states below here have X2APIC enabled */
1757 X2APIC_ON,
1758 X2APIC_ON_LOCKED
1759};
1760static int x2apic_state;
1761
1762static bool x2apic_hw_locked(void)
1763{
1764 u64 ia32_cap;
1765 u64 msr;
1766
1767 ia32_cap = x86_read_arch_cap_msr();
1768 if (ia32_cap & ARCH_CAP_XAPIC_DISABLE) {
1769 rdmsrl(MSR_IA32_XAPIC_DISABLE_STATUS, msr);
1770 return (msr & LEGACY_XAPIC_DISABLED);
1771 }
1772 return false;
1773}
1774
1775static void __x2apic_disable(void)
1776{
1777 u64 msr;
1778
1779 if (!boot_cpu_has(X86_FEATURE_APIC))
1780 return;
1781
1782 rdmsrl(MSR_IA32_APICBASE, msr);
1783 if (!(msr & X2APIC_ENABLE))
1784 return;
1785 /* Disable xapic and x2apic first and then reenable xapic mode */
1786 wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1787 wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1788 printk_once(KERN_INFO "x2apic disabled\n");
1789}
1790
1791static void __x2apic_enable(void)
1792{
1793 u64 msr;
1794
1795 rdmsrl(MSR_IA32_APICBASE, msr);
1796 if (msr & X2APIC_ENABLE)
1797 return;
1798 wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1799 printk_once(KERN_INFO "x2apic enabled\n");
1800}
1801
1802static int __init setup_nox2apic(char *str)
1803{
1804 if (x2apic_enabled()) {
1805 int apicid = native_apic_msr_read(APIC_ID);
1806
1807 if (apicid >= 255) {
1808 pr_warn("Apicid: %08x, cannot enforce nox2apic\n",
1809 apicid);
1810 return 0;
1811 }
1812 if (x2apic_hw_locked()) {
1813 pr_warn("APIC locked in x2apic mode, can't disable\n");
1814 return 0;
1815 }
1816 pr_warn("x2apic already enabled.\n");
1817 __x2apic_disable();
1818 }
1819 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1820 x2apic_state = X2APIC_DISABLED;
1821 x2apic_mode = 0;
1822 return 0;
1823}
1824early_param("nox2apic", setup_nox2apic);
1825
1826/* Called from cpu_init() to enable x2apic on (secondary) cpus */
1827void x2apic_setup(void)
1828{
1829 /*
1830 * Try to make the AP's APIC state match that of the BSP, but if the
1831 * BSP is unlocked and the AP is locked then there is a state mismatch.
1832 * Warn about the mismatch in case a GP fault occurs due to a locked AP
1833 * trying to be turned off.
1834 */
1835 if (x2apic_state != X2APIC_ON_LOCKED && x2apic_hw_locked())
1836 pr_warn("x2apic lock mismatch between BSP and AP.\n");
1837 /*
1838 * If x2apic is not in ON or LOCKED state, disable it if already enabled
1839 * from BIOS.
1840 */
1841 if (x2apic_state < X2APIC_ON) {
1842 __x2apic_disable();
1843 return;
1844 }
1845 __x2apic_enable();
1846}
1847
1848static __init void x2apic_disable(void)
1849{
1850 u32 x2apic_id, state = x2apic_state;
1851
1852 x2apic_mode = 0;
1853 x2apic_state = X2APIC_DISABLED;
1854
1855 if (state != X2APIC_ON)
1856 return;
1857
1858 x2apic_id = read_apic_id();
1859 if (x2apic_id >= 255)
1860 panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1861
1862 if (x2apic_hw_locked()) {
1863 pr_warn("Cannot disable locked x2apic, id: %08x\n", x2apic_id);
1864 return;
1865 }
1866
1867 __x2apic_disable();
1868 register_lapic_address(mp_lapic_addr);
1869}
1870
1871static __init void x2apic_enable(void)
1872{
1873 if (x2apic_state != X2APIC_OFF)
1874 return;
1875
1876 x2apic_mode = 1;
1877 x2apic_state = X2APIC_ON;
1878 __x2apic_enable();
1879}
1880
1881static __init void try_to_enable_x2apic(int remap_mode)
1882{
1883 if (x2apic_state == X2APIC_DISABLED)
1884 return;
1885
1886 if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1887 u32 apic_limit = 255;
1888
1889 /*
1890 * Using X2APIC without IR is not architecturally supported
1891 * on bare metal but may be supported in guests.
1892 */
1893 if (!x86_init.hyper.x2apic_available()) {
1894 pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1895 x2apic_disable();
1896 return;
1897 }
1898
1899 /*
1900 * If the hypervisor supports extended destination ID in
1901 * MSI, that increases the maximum APIC ID that can be
1902 * used for non-remapped IRQ domains.
1903 */
1904 if (x86_init.hyper.msi_ext_dest_id()) {
1905 virt_ext_dest_id = 1;
1906 apic_limit = 32767;
1907 }
1908
1909 /*
1910 * Without IR, all CPUs can be addressed by IOAPIC/MSI only
1911 * in physical mode, and CPUs with an APIC ID that cannot
1912 * be addressed must not be brought online.
1913 */
1914 x2apic_set_max_apicid(apic_limit);
1915 x2apic_phys = 1;
1916 }
1917 x2apic_enable();
1918}
1919
1920void __init check_x2apic(void)
1921{
1922 if (x2apic_enabled()) {
1923 pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1924 x2apic_mode = 1;
1925 if (x2apic_hw_locked())
1926 x2apic_state = X2APIC_ON_LOCKED;
1927 else
1928 x2apic_state = X2APIC_ON;
1929 } else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1930 x2apic_state = X2APIC_DISABLED;
1931 }
1932}
1933#else /* CONFIG_X86_X2APIC */
1934void __init check_x2apic(void)
1935{
1936 if (!apic_is_x2apic_enabled())
1937 return;
1938 /*
1939 * Checkme: Can we simply turn off x2APIC here instead of disabling the APIC?
1940 */
1941 pr_err("Kernel does not support x2APIC, please recompile with CONFIG_X86_X2APIC.\n");
1942 pr_err("Disabling APIC, expect reduced performance and functionality.\n");
1943
1944 disable_apic = 1;
1945 setup_clear_cpu_cap(X86_FEATURE_APIC);
1946}
1947
1948static inline void try_to_enable_x2apic(int remap_mode) { }
1949static inline void __x2apic_enable(void) { }
1950#endif /* !CONFIG_X86_X2APIC */
1951
1952void __init enable_IR_x2apic(void)
1953{
1954 unsigned long flags;
1955 int ret, ir_stat;
1956
1957 if (skip_ioapic_setup) {
1958 pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1959 return;
1960 }
1961
1962 ir_stat = irq_remapping_prepare();
1963 if (ir_stat < 0 && !x2apic_supported())
1964 return;
1965
1966 ret = save_ioapic_entries();
1967 if (ret) {
1968 pr_info("Saving IO-APIC state failed: %d\n", ret);
1969 return;
1970 }
1971
1972 local_irq_save(flags);
1973 legacy_pic->mask_all();
1974 mask_ioapic_entries();
1975
1976 /* If irq_remapping_prepare() succeeded, try to enable it */
1977 if (ir_stat >= 0)
1978 ir_stat = irq_remapping_enable();
1979 /* ir_stat contains the remap mode or an error code */
1980 try_to_enable_x2apic(ir_stat);
1981
1982 if (ir_stat < 0)
1983 restore_ioapic_entries();
1984 legacy_pic->restore_mask();
1985 local_irq_restore(flags);
1986}
1987
1988#ifdef CONFIG_X86_64
1989/*
1990 * Detect and enable local APICs on non-SMP boards.
1991 * Original code written by Keir Fraser.
1992 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1993 * not correctly set up (usually the APIC timer won't work etc.)
1994 */
1995static int __init detect_init_APIC(void)
1996{
1997 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1998 pr_info("No local APIC present\n");
1999 return -1;
2000 }
2001
2002 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
2003 return 0;
2004}
2005#else
2006
2007static int __init apic_verify(void)
2008{
2009 u32 features, h, l;
2010
2011 /*
2012 * The APIC feature bit should now be enabled
2013 * in `cpuid'
2014 */
2015 features = cpuid_edx(1);
2016 if (!(features & (1 << X86_FEATURE_APIC))) {
2017 pr_warn("Could not enable APIC!\n");
2018 return -1;
2019 }
2020 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
2021 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
2022
2023 /* The BIOS may have set up the APIC at some other address */
2024 if (boot_cpu_data.x86 >= 6) {
2025 rdmsr(MSR_IA32_APICBASE, l, h);
2026 if (l & MSR_IA32_APICBASE_ENABLE)
2027 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
2028 }
2029
2030 pr_info("Found and enabled local APIC!\n");
2031 return 0;
2032}
2033
2034int __init apic_force_enable(unsigned long addr)
2035{
2036 u32 h, l;
2037
2038 if (disable_apic)
2039 return -1;
2040
2041 /*
2042 * Some BIOSes disable the local APIC in the APIC_BASE
2043 * MSR. This can only be done in software for Intel P6 or later
2044 * and AMD K7 (Model > 1) or later.
2045 */
2046 if (boot_cpu_data.x86 >= 6) {
2047 rdmsr(MSR_IA32_APICBASE, l, h);
2048 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
2049 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
2050 l &= ~MSR_IA32_APICBASE_BASE;
2051 l |= MSR_IA32_APICBASE_ENABLE | addr;
2052 wrmsr(MSR_IA32_APICBASE, l, h);
2053 enabled_via_apicbase = 1;
2054 }
2055 }
2056 return apic_verify();
2057}
2058
2059/*
2060 * Detect and initialize APIC
2061 */
2062static int __init detect_init_APIC(void)
2063{
2064 /* Disabled by kernel option? */
2065 if (disable_apic)
2066 return -1;
2067
2068 switch (boot_cpu_data.x86_vendor) {
2069 case X86_VENDOR_AMD:
2070 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
2071 (boot_cpu_data.x86 >= 15))
2072 break;
2073 goto no_apic;
2074 case X86_VENDOR_HYGON:
2075 break;
2076 case X86_VENDOR_INTEL:
2077 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
2078 (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
2079 break;
2080 goto no_apic;
2081 default:
2082 goto no_apic;
2083 }
2084
2085 if (!boot_cpu_has(X86_FEATURE_APIC)) {
2086 /*
2087 * Over-ride BIOS and try to enable the local APIC only if
2088 * "lapic" specified.
2089 */
2090 if (!force_enable_local_apic) {
2091 pr_info("Local APIC disabled by BIOS -- "
2092 "you can enable it with \"lapic\"\n");
2093 return -1;
2094 }
2095 if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
2096 return -1;
2097 } else {
2098 if (apic_verify())
2099 return -1;
2100 }
2101
2102 apic_pm_activate();
2103
2104 return 0;
2105
2106no_apic:
2107 pr_info("No local APIC present or hardware disabled\n");
2108 return -1;
2109}
2110#endif
2111
2112/**
2113 * init_apic_mappings - initialize APIC mappings
2114 */
2115void __init init_apic_mappings(void)
2116{
2117 unsigned int new_apicid;
2118
2119 if (apic_validate_deadline_timer())
2120 pr_info("TSC deadline timer available\n");
2121
2122 if (x2apic_mode) {
2123 boot_cpu_physical_apicid = read_apic_id();
2124 return;
2125 }
2126
2127 /* If no local APIC can be found return early */
2128 if (!smp_found_config && detect_init_APIC()) {
2129 /* lets NOP'ify apic operations */
2130 pr_info("APIC: disable apic facility\n");
2131 apic_disable();
2132 } else {
2133 apic_phys = mp_lapic_addr;
2134
2135 /*
2136 * If the system has ACPI MADT tables or MP info, the LAPIC
2137 * address is already registered.
2138 */
2139 if (!acpi_lapic && !smp_found_config)
2140 register_lapic_address(apic_phys);
2141 }
2142
2143 /*
2144 * Fetch the APIC ID of the BSP in case we have a
2145 * default configuration (or the MP table is broken).
2146 */
2147 new_apicid = read_apic_id();
2148 if (boot_cpu_physical_apicid != new_apicid) {
2149 boot_cpu_physical_apicid = new_apicid;
2150 /*
2151 * yeah -- we lie about apic_version
2152 * in case if apic was disabled via boot option
2153 * but it's not a problem for SMP compiled kernel
2154 * since apic_intr_mode_select is prepared for such
2155 * a case and disable smp mode
2156 */
2157 boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2158 }
2159}
2160
2161void __init register_lapic_address(unsigned long address)
2162{
2163 mp_lapic_addr = address;
2164
2165 if (!x2apic_mode) {
2166 set_fixmap_nocache(FIX_APIC_BASE, address);
2167 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
2168 APIC_BASE, address);
2169 }
2170 if (boot_cpu_physical_apicid == -1U) {
2171 boot_cpu_physical_apicid = read_apic_id();
2172 boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2173 }
2174}
2175
2176/*
2177 * Local APIC interrupts
2178 */
2179
2180/*
2181 * Common handling code for spurious_interrupt and spurious_vector entry
2182 * points below. No point in allowing the compiler to inline it twice.
2183 */
2184static noinline void handle_spurious_interrupt(u8 vector)
2185{
2186 u32 v;
2187
2188 trace_spurious_apic_entry(vector);
2189
2190 inc_irq_stat(irq_spurious_count);
2191
2192 /*
2193 * If this is a spurious interrupt then do not acknowledge
2194 */
2195 if (vector == SPURIOUS_APIC_VECTOR) {
2196 /* See SDM vol 3 */
2197 pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n",
2198 smp_processor_id());
2199 goto out;
2200 }
2201
2202 /*
2203 * If it is a vectored one, verify it's set in the ISR. If set,
2204 * acknowledge it.
2205 */
2206 v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2207 if (v & (1 << (vector & 0x1f))) {
2208 pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n",
2209 vector, smp_processor_id());
2210 ack_APIC_irq();
2211 } else {
2212 pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n",
2213 vector, smp_processor_id());
2214 }
2215out:
2216 trace_spurious_apic_exit(vector);
2217}
2218
2219/**
2220 * spurious_interrupt - Catch all for interrupts raised on unused vectors
2221 * @regs: Pointer to pt_regs on stack
2222 * @vector: The vector number
2223 *
2224 * This is invoked from ASM entry code to catch all interrupts which
2225 * trigger on an entry which is routed to the common_spurious idtentry
2226 * point.
2227 */
2228DEFINE_IDTENTRY_IRQ(spurious_interrupt)
2229{
2230 handle_spurious_interrupt(vector);
2231}
2232
2233DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt)
2234{
2235 handle_spurious_interrupt(SPURIOUS_APIC_VECTOR);
2236}
2237
2238/*
2239 * This interrupt should never happen with our APIC/SMP architecture
2240 */
2241DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt)
2242{
2243 static const char * const error_interrupt_reason[] = {
2244 "Send CS error", /* APIC Error Bit 0 */
2245 "Receive CS error", /* APIC Error Bit 1 */
2246 "Send accept error", /* APIC Error Bit 2 */
2247 "Receive accept error", /* APIC Error Bit 3 */
2248 "Redirectable IPI", /* APIC Error Bit 4 */
2249 "Send illegal vector", /* APIC Error Bit 5 */
2250 "Received illegal vector", /* APIC Error Bit 6 */
2251 "Illegal register address", /* APIC Error Bit 7 */
2252 };
2253 u32 v, i = 0;
2254
2255 trace_error_apic_entry(ERROR_APIC_VECTOR);
2256
2257 /* First tickle the hardware, only then report what went on. -- REW */
2258 if (lapic_get_maxlvt() > 3) /* Due to the Pentium erratum 3AP. */
2259 apic_write(APIC_ESR, 0);
2260 v = apic_read(APIC_ESR);
2261 ack_APIC_irq();
2262 atomic_inc(&irq_err_count);
2263
2264 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2265 smp_processor_id(), v);
2266
2267 v &= 0xff;
2268 while (v) {
2269 if (v & 0x1)
2270 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2271 i++;
2272 v >>= 1;
2273 }
2274
2275 apic_printk(APIC_DEBUG, KERN_CONT "\n");
2276
2277 trace_error_apic_exit(ERROR_APIC_VECTOR);
2278}
2279
2280/**
2281 * connect_bsp_APIC - attach the APIC to the interrupt system
2282 */
2283static void __init connect_bsp_APIC(void)
2284{
2285#ifdef CONFIG_X86_32
2286 if (pic_mode) {
2287 /*
2288 * Do not trust the local APIC being empty at bootup.
2289 */
2290 clear_local_APIC();
2291 /*
2292 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
2293 * local APIC to INT and NMI lines.
2294 */
2295 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2296 "enabling APIC mode.\n");
2297 imcr_pic_to_apic();
2298 }
2299#endif
2300}
2301
2302/**
2303 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2304 * @virt_wire_setup: indicates, whether virtual wire mode is selected
2305 *
2306 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2307 * APIC is disabled.
2308 */
2309void disconnect_bsp_APIC(int virt_wire_setup)
2310{
2311 unsigned int value;
2312
2313#ifdef CONFIG_X86_32
2314 if (pic_mode) {
2315 /*
2316 * Put the board back into PIC mode (has an effect only on
2317 * certain older boards). Note that APIC interrupts, including
2318 * IPIs, won't work beyond this point! The only exception are
2319 * INIT IPIs.
2320 */
2321 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2322 "entering PIC mode.\n");
2323 imcr_apic_to_pic();
2324 return;
2325 }
2326#endif
2327
2328 /* Go back to Virtual Wire compatibility mode */
2329
2330 /* For the spurious interrupt use vector F, and enable it */
2331 value = apic_read(APIC_SPIV);
2332 value &= ~APIC_VECTOR_MASK;
2333 value |= APIC_SPIV_APIC_ENABLED;
2334 value |= 0xf;
2335 apic_write(APIC_SPIV, value);
2336
2337 if (!virt_wire_setup) {
2338 /*
2339 * For LVT0 make it edge triggered, active high,
2340 * external and enabled
2341 */
2342 value = apic_read(APIC_LVT0);
2343 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2344 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2345 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2346 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2347 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2348 apic_write(APIC_LVT0, value);
2349 } else {
2350 /* Disable LVT0 */
2351 apic_write(APIC_LVT0, APIC_LVT_MASKED);
2352 }
2353
2354 /*
2355 * For LVT1 make it edge triggered, active high,
2356 * nmi and enabled
2357 */
2358 value = apic_read(APIC_LVT1);
2359 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2360 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2361 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2362 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2363 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2364 apic_write(APIC_LVT1, value);
2365}
2366
2367/*
2368 * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2369 * contiguously, it equals to current allocated max logical CPU ID plus 1.
2370 * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2371 * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2372 *
2373 * NOTE: Reserve 0 for BSP.
2374 */
2375static int nr_logical_cpuids = 1;
2376
2377/*
2378 * Used to store mapping between logical CPU IDs and APIC IDs.
2379 */
2380static int cpuid_to_apicid[] = {
2381 [0 ... NR_CPUS - 1] = -1,
2382};
2383
2384bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
2385{
2386 return phys_id == cpuid_to_apicid[cpu];
2387}
2388
2389#ifdef CONFIG_SMP
2390/**
2391 * apic_id_is_primary_thread - Check whether APIC ID belongs to a primary thread
2392 * @apicid: APIC ID to check
2393 */
2394bool apic_id_is_primary_thread(unsigned int apicid)
2395{
2396 u32 mask;
2397
2398 if (smp_num_siblings == 1)
2399 return true;
2400 /* Isolate the SMT bit(s) in the APICID and check for 0 */
2401 mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
2402 return !(apicid & mask);
2403}
2404#endif
2405
2406/*
2407 * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2408 * and cpuid_to_apicid[] synchronized.
2409 */
2410static int allocate_logical_cpuid(int apicid)
2411{
2412 int i;
2413
2414 /*
2415 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2416 * check if the kernel has allocated a cpuid for it.
2417 */
2418 for (i = 0; i < nr_logical_cpuids; i++) {
2419 if (cpuid_to_apicid[i] == apicid)
2420 return i;
2421 }
2422
2423 /* Allocate a new cpuid. */
2424 if (nr_logical_cpuids >= nr_cpu_ids) {
2425 WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %u reached. "
2426 "Processor %d/0x%x and the rest are ignored.\n",
2427 nr_cpu_ids, nr_logical_cpuids, apicid);
2428 return -EINVAL;
2429 }
2430
2431 cpuid_to_apicid[nr_logical_cpuids] = apicid;
2432 return nr_logical_cpuids++;
2433}
2434
2435int generic_processor_info(int apicid, int version)
2436{
2437 int cpu, max = nr_cpu_ids;
2438 bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2439 phys_cpu_present_map);
2440
2441 /*
2442 * boot_cpu_physical_apicid is designed to have the apicid
2443 * returned by read_apic_id(), i.e, the apicid of the
2444 * currently booting-up processor. However, on some platforms,
2445 * it is temporarily modified by the apicid reported as BSP
2446 * through MP table. Concretely:
2447 *
2448 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2449 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2450 *
2451 * This function is executed with the modified
2452 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2453 * parameter doesn't work to disable APs on kdump 2nd kernel.
2454 *
2455 * Since fixing handling of boot_cpu_physical_apicid requires
2456 * another discussion and tests on each platform, we leave it
2457 * for now and here we use read_apic_id() directly in this
2458 * function, generic_processor_info().
2459 */
2460 if (disabled_cpu_apicid != BAD_APICID &&
2461 disabled_cpu_apicid != read_apic_id() &&
2462 disabled_cpu_apicid == apicid) {
2463 int thiscpu = num_processors + disabled_cpus;
2464
2465 pr_warn("APIC: Disabling requested cpu."
2466 " Processor %d/0x%x ignored.\n", thiscpu, apicid);
2467
2468 disabled_cpus++;
2469 return -ENODEV;
2470 }
2471
2472 /*
2473 * If boot cpu has not been detected yet, then only allow upto
2474 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2475 */
2476 if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2477 apicid != boot_cpu_physical_apicid) {
2478 int thiscpu = max + disabled_cpus - 1;
2479
2480 pr_warn("APIC: NR_CPUS/possible_cpus limit of %i almost"
2481 " reached. Keeping one slot for boot cpu."
2482 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2483
2484 disabled_cpus++;
2485 return -ENODEV;
2486 }
2487
2488 if (num_processors >= nr_cpu_ids) {
2489 int thiscpu = max + disabled_cpus;
2490
2491 pr_warn("APIC: NR_CPUS/possible_cpus limit of %i reached. "
2492 "Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2493
2494 disabled_cpus++;
2495 return -EINVAL;
2496 }
2497
2498 if (apicid == boot_cpu_physical_apicid) {
2499 /*
2500 * x86_bios_cpu_apicid is required to have processors listed
2501 * in same order as logical cpu numbers. Hence the first
2502 * entry is BSP, and so on.
2503 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2504 * for BSP.
2505 */
2506 cpu = 0;
2507
2508 /* Logical cpuid 0 is reserved for BSP. */
2509 cpuid_to_apicid[0] = apicid;
2510 } else {
2511 cpu = allocate_logical_cpuid(apicid);
2512 if (cpu < 0) {
2513 disabled_cpus++;
2514 return -EINVAL;
2515 }
2516 }
2517
2518 /*
2519 * Validate version
2520 */
2521 if (version == 0x0) {
2522 pr_warn("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2523 cpu, apicid);
2524 version = 0x10;
2525 }
2526
2527 if (version != boot_cpu_apic_version) {
2528 pr_warn("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2529 boot_cpu_apic_version, cpu, version);
2530 }
2531
2532 if (apicid > max_physical_apicid)
2533 max_physical_apicid = apicid;
2534
2535#if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2536 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2537 early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2538#endif
2539#ifdef CONFIG_X86_32
2540 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2541 apic->x86_32_early_logical_apicid(cpu);
2542#endif
2543 set_cpu_possible(cpu, true);
2544 physid_set(apicid, phys_cpu_present_map);
2545 set_cpu_present(cpu, true);
2546 num_processors++;
2547
2548 return cpu;
2549}
2550
2551int hard_smp_processor_id(void)
2552{
2553 return read_apic_id();
2554}
2555
2556void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,
2557 bool dmar)
2558{
2559 memset(msg, 0, sizeof(*msg));
2560
2561 msg->arch_addr_lo.base_address = X86_MSI_BASE_ADDRESS_LOW;
2562 msg->arch_addr_lo.dest_mode_logical = apic->dest_mode_logical;
2563 msg->arch_addr_lo.destid_0_7 = cfg->dest_apicid & 0xFF;
2564
2565 msg->arch_data.delivery_mode = APIC_DELIVERY_MODE_FIXED;
2566 msg->arch_data.vector = cfg->vector;
2567
2568 msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH;
2569 /*
2570 * Only the IOMMU itself can use the trick of putting destination
2571 * APIC ID into the high bits of the address. Anything else would
2572 * just be writing to memory if it tried that, and needs IR to
2573 * address APICs which can't be addressed in the normal 32-bit
2574 * address range at 0xFFExxxxx. That is typically just 8 bits, but
2575 * some hypervisors allow the extended destination ID field in bits
2576 * 5-11 to be used, giving support for 15 bits of APIC IDs in total.
2577 */
2578 if (dmar)
2579 msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >> 8;
2580 else if (virt_ext_dest_id && cfg->dest_apicid < 0x8000)
2581 msg->arch_addr_lo.virt_destid_8_14 = cfg->dest_apicid >> 8;
2582 else
2583 WARN_ON_ONCE(cfg->dest_apicid > 0xFF);
2584}
2585
2586u32 x86_msi_msg_get_destid(struct msi_msg *msg, bool extid)
2587{
2588 u32 dest = msg->arch_addr_lo.destid_0_7;
2589
2590 if (extid)
2591 dest |= msg->arch_addr_hi.destid_8_31 << 8;
2592 return dest;
2593}
2594EXPORT_SYMBOL_GPL(x86_msi_msg_get_destid);
2595
2596#ifdef CONFIG_X86_64
2597void __init acpi_wake_cpu_handler_update(wakeup_cpu_handler handler)
2598{
2599 struct apic **drv;
2600
2601 for (drv = __apicdrivers; drv < __apicdrivers_end; drv++)
2602 (*drv)->wakeup_secondary_cpu_64 = handler;
2603}
2604#endif
2605
2606/*
2607 * Override the generic EOI implementation with an optimized version.
2608 * Only called during early boot when only one CPU is active and with
2609 * interrupts disabled, so we know this does not race with actual APIC driver
2610 * use.
2611 */
2612void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2613{
2614 struct apic **drv;
2615
2616 for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2617 /* Should happen once for each apic */
2618 WARN_ON((*drv)->eoi_write == eoi_write);
2619 (*drv)->native_eoi_write = (*drv)->eoi_write;
2620 (*drv)->eoi_write = eoi_write;
2621 }
2622}
2623
2624static void __init apic_bsp_up_setup(void)
2625{
2626#ifdef CONFIG_X86_64
2627 apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2628#else
2629 /*
2630 * Hack: In case of kdump, after a crash, kernel might be booting
2631 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2632 * might be zero if read from MP tables. Get it from LAPIC.
2633 */
2634# ifdef CONFIG_CRASH_DUMP
2635 boot_cpu_physical_apicid = read_apic_id();
2636# endif
2637#endif
2638 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2639}
2640
2641/**
2642 * apic_bsp_setup - Setup function for local apic and io-apic
2643 * @upmode: Force UP mode (for APIC_init_uniprocessor)
2644 */
2645static void __init apic_bsp_setup(bool upmode)
2646{
2647 connect_bsp_APIC();
2648 if (upmode)
2649 apic_bsp_up_setup();
2650 setup_local_APIC();
2651
2652 enable_IO_APIC();
2653 end_local_APIC_setup();
2654 irq_remap_enable_fault_handling();
2655 setup_IO_APIC();
2656 lapic_update_legacy_vectors();
2657}
2658
2659#ifdef CONFIG_UP_LATE_INIT
2660void __init up_late_init(void)
2661{
2662 if (apic_intr_mode == APIC_PIC)
2663 return;
2664
2665 /* Setup local timer */
2666 x86_init.timers.setup_percpu_clockev();
2667}
2668#endif
2669
2670/*
2671 * Power management
2672 */
2673#ifdef CONFIG_PM
2674
2675static struct {
2676 /*
2677 * 'active' is true if the local APIC was enabled by us and
2678 * not the BIOS; this signifies that we are also responsible
2679 * for disabling it before entering apm/acpi suspend
2680 */
2681 int active;
2682 /* r/w apic fields */
2683 unsigned int apic_id;
2684 unsigned int apic_taskpri;
2685 unsigned int apic_ldr;
2686 unsigned int apic_dfr;
2687 unsigned int apic_spiv;
2688 unsigned int apic_lvtt;
2689 unsigned int apic_lvtpc;
2690 unsigned int apic_lvt0;
2691 unsigned int apic_lvt1;
2692 unsigned int apic_lvterr;
2693 unsigned int apic_tmict;
2694 unsigned int apic_tdcr;
2695 unsigned int apic_thmr;
2696 unsigned int apic_cmci;
2697} apic_pm_state;
2698
2699static int lapic_suspend(void)
2700{
2701 unsigned long flags;
2702 int maxlvt;
2703
2704 if (!apic_pm_state.active)
2705 return 0;
2706
2707 maxlvt = lapic_get_maxlvt();
2708
2709 apic_pm_state.apic_id = apic_read(APIC_ID);
2710 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2711 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2712 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2713 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2714 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2715 if (maxlvt >= 4)
2716 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2717 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2718 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2719 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2720 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2721 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2722#ifdef CONFIG_X86_THERMAL_VECTOR
2723 if (maxlvt >= 5)
2724 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2725#endif
2726#ifdef CONFIG_X86_MCE_INTEL
2727 if (maxlvt >= 6)
2728 apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2729#endif
2730
2731 local_irq_save(flags);
2732
2733 /*
2734 * Mask IOAPIC before disabling the local APIC to prevent stale IRR
2735 * entries on some implementations.
2736 */
2737 mask_ioapic_entries();
2738
2739 disable_local_APIC();
2740
2741 irq_remapping_disable();
2742
2743 local_irq_restore(flags);
2744 return 0;
2745}
2746
2747static void lapic_resume(void)
2748{
2749 unsigned int l, h;
2750 unsigned long flags;
2751 int maxlvt;
2752
2753 if (!apic_pm_state.active)
2754 return;
2755
2756 local_irq_save(flags);
2757
2758 /*
2759 * IO-APIC and PIC have their own resume routines.
2760 * We just mask them here to make sure the interrupt
2761 * subsystem is completely quiet while we enable x2apic
2762 * and interrupt-remapping.
2763 */
2764 mask_ioapic_entries();
2765 legacy_pic->mask_all();
2766
2767 if (x2apic_mode) {
2768 __x2apic_enable();
2769 } else {
2770 /*
2771 * Make sure the APICBASE points to the right address
2772 *
2773 * FIXME! This will be wrong if we ever support suspend on
2774 * SMP! We'll need to do this as part of the CPU restore!
2775 */
2776 if (boot_cpu_data.x86 >= 6) {
2777 rdmsr(MSR_IA32_APICBASE, l, h);
2778 l &= ~MSR_IA32_APICBASE_BASE;
2779 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2780 wrmsr(MSR_IA32_APICBASE, l, h);
2781 }
2782 }
2783
2784 maxlvt = lapic_get_maxlvt();
2785 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2786 apic_write(APIC_ID, apic_pm_state.apic_id);
2787 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2788 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2789 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2790 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2791 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2792 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2793#ifdef CONFIG_X86_THERMAL_VECTOR
2794 if (maxlvt >= 5)
2795 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2796#endif
2797#ifdef CONFIG_X86_MCE_INTEL
2798 if (maxlvt >= 6)
2799 apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2800#endif
2801 if (maxlvt >= 4)
2802 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2803 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2804 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2805 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2806 apic_write(APIC_ESR, 0);
2807 apic_read(APIC_ESR);
2808 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2809 apic_write(APIC_ESR, 0);
2810 apic_read(APIC_ESR);
2811
2812 irq_remapping_reenable(x2apic_mode);
2813
2814 local_irq_restore(flags);
2815}
2816
2817/*
2818 * This device has no shutdown method - fully functioning local APICs
2819 * are needed on every CPU up until machine_halt/restart/poweroff.
2820 */
2821
2822static struct syscore_ops lapic_syscore_ops = {
2823 .resume = lapic_resume,
2824 .suspend = lapic_suspend,
2825};
2826
2827static void apic_pm_activate(void)
2828{
2829 apic_pm_state.active = 1;
2830}
2831
2832static int __init init_lapic_sysfs(void)
2833{
2834 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2835 if (boot_cpu_has(X86_FEATURE_APIC))
2836 register_syscore_ops(&lapic_syscore_ops);
2837
2838 return 0;
2839}
2840
2841/* local apic needs to resume before other devices access its registers. */
2842core_initcall(init_lapic_sysfs);
2843
2844#else /* CONFIG_PM */
2845
2846static void apic_pm_activate(void) { }
2847
2848#endif /* CONFIG_PM */
2849
2850#ifdef CONFIG_X86_64
2851
2852static int multi_checked;
2853static int multi;
2854
2855static int set_multi(const struct dmi_system_id *d)
2856{
2857 if (multi)
2858 return 0;
2859 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2860 multi = 1;
2861 return 0;
2862}
2863
2864static const struct dmi_system_id multi_dmi_table[] = {
2865 {
2866 .callback = set_multi,
2867 .ident = "IBM System Summit2",
2868 .matches = {
2869 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2870 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2871 },
2872 },
2873 {}
2874};
2875
2876static void dmi_check_multi(void)
2877{
2878 if (multi_checked)
2879 return;
2880
2881 dmi_check_system(multi_dmi_table);
2882 multi_checked = 1;
2883}
2884
2885/*
2886 * apic_is_clustered_box() -- Check if we can expect good TSC
2887 *
2888 * Thus far, the major user of this is IBM's Summit2 series:
2889 * Clustered boxes may have unsynced TSC problems if they are
2890 * multi-chassis.
2891 * Use DMI to check them
2892 */
2893int apic_is_clustered_box(void)
2894{
2895 dmi_check_multi();
2896 return multi;
2897}
2898#endif
2899
2900/*
2901 * APIC command line parameters
2902 */
2903static int __init setup_disableapic(char *arg)
2904{
2905 disable_apic = 1;
2906 setup_clear_cpu_cap(X86_FEATURE_APIC);
2907 return 0;
2908}
2909early_param("disableapic", setup_disableapic);
2910
2911/* same as disableapic, for compatibility */
2912static int __init setup_nolapic(char *arg)
2913{
2914 return setup_disableapic(arg);
2915}
2916early_param("nolapic", setup_nolapic);
2917
2918static int __init parse_lapic_timer_c2_ok(char *arg)
2919{
2920 local_apic_timer_c2_ok = 1;
2921 return 0;
2922}
2923early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2924
2925static int __init parse_disable_apic_timer(char *arg)
2926{
2927 disable_apic_timer = 1;
2928 return 0;
2929}
2930early_param("noapictimer", parse_disable_apic_timer);
2931
2932static int __init parse_nolapic_timer(char *arg)
2933{
2934 disable_apic_timer = 1;
2935 return 0;
2936}
2937early_param("nolapic_timer", parse_nolapic_timer);
2938
2939static int __init apic_set_verbosity(char *arg)
2940{
2941 if (!arg) {
2942#ifdef CONFIG_X86_64
2943 skip_ioapic_setup = 0;
2944 return 0;
2945#endif
2946 return -EINVAL;
2947 }
2948
2949 if (strcmp("debug", arg) == 0)
2950 apic_verbosity = APIC_DEBUG;
2951 else if (strcmp("verbose", arg) == 0)
2952 apic_verbosity = APIC_VERBOSE;
2953#ifdef CONFIG_X86_64
2954 else {
2955 pr_warn("APIC Verbosity level %s not recognised"
2956 " use apic=verbose or apic=debug\n", arg);
2957 return -EINVAL;
2958 }
2959#endif
2960
2961 return 0;
2962}
2963early_param("apic", apic_set_verbosity);
2964
2965static int __init lapic_insert_resource(void)
2966{
2967 if (!apic_phys)
2968 return -1;
2969
2970 /* Put local APIC into the resource map. */
2971 lapic_resource.start = apic_phys;
2972 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2973 insert_resource(&iomem_resource, &lapic_resource);
2974
2975 return 0;
2976}
2977
2978/*
2979 * need call insert after e820__reserve_resources()
2980 * that is using request_resource
2981 */
2982late_initcall(lapic_insert_resource);
2983
2984static int __init apic_set_disabled_cpu_apicid(char *arg)
2985{
2986 if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2987 return -EINVAL;
2988
2989 return 0;
2990}
2991early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2992
2993static int __init apic_set_extnmi(char *arg)
2994{
2995 if (!arg)
2996 return -EINVAL;
2997
2998 if (!strncmp("all", arg, 3))
2999 apic_extnmi = APIC_EXTNMI_ALL;
3000 else if (!strncmp("none", arg, 4))
3001 apic_extnmi = APIC_EXTNMI_NONE;
3002 else if (!strncmp("bsp", arg, 3))
3003 apic_extnmi = APIC_EXTNMI_BSP;
3004 else {
3005 pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
3006 return -EINVAL;
3007 }
3008
3009 return 0;
3010}
3011early_param("apic_extnmi", apic_set_extnmi);