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