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/bitmap.h>
23#include <linux/clockchips.h>
24#include <linux/interrupt.h>
25#include <linux/memblock.h>
26#include <linux/ftrace.h>
27#include <linux/ioport.h>
28#include <linux/export.h>
29#include <linux/syscore_ops.h>
30#include <linux/delay.h>
31#include <linux/timex.h>
32#include <linux/i8253.h>
33#include <linux/dmar.h>
34#include <linux/init.h>
35#include <linux/cpu.h>
36#include <linux/dmi.h>
37#include <linux/smp.h>
38#include <linux/mm.h>
39
40#include <xen/xen.h>
41
42#include <asm/trace/irq_vectors.h>
43#include <asm/irq_remapping.h>
44#include <asm/pc-conf-reg.h>
45#include <asm/perf_event.h>
46#include <asm/x86_init.h>
47#include <linux/atomic.h>
48#include <asm/barrier.h>
49#include <asm/mpspec.h>
50#include <asm/i8259.h>
51#include <asm/proto.h>
52#include <asm/traps.h>
53#include <asm/apic.h>
54#include <asm/acpi.h>
55#include <asm/io_apic.h>
56#include <asm/desc.h>
57#include <asm/hpet.h>
58#include <asm/mtrr.h>
59#include <asm/time.h>
60#include <asm/smp.h>
61#include <asm/mce.h>
62#include <asm/tsc.h>
63#include <asm/hypervisor.h>
64#include <asm/cpu_device_id.h>
65#include <asm/intel-family.h>
66#include <asm/irq_regs.h>
67#include <asm/cpu.h>
68
69#include "local.h"
70
71/* Processor that is doing the boot up */
72u32 boot_cpu_physical_apicid __ro_after_init = BAD_APICID;
73EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
74
75u8 boot_cpu_apic_version __ro_after_init;
76
77/*
78 * This variable controls which CPUs receive external NMIs. By default,
79 * external NMIs are delivered only to the BSP.
80 */
81static int apic_extnmi __ro_after_init = APIC_EXTNMI_BSP;
82
83/*
84 * Hypervisor supports 15 bits of APIC ID in MSI Extended Destination ID
85 */
86static bool virt_ext_dest_id __ro_after_init;
87
88/* For parallel bootup. */
89unsigned long apic_mmio_base __ro_after_init;
90
91static inline bool apic_accessible(void)
92{
93 return x2apic_mode || apic_mmio_base;
94}
95
96#ifdef CONFIG_X86_32
97/* Local APIC was disabled by the BIOS and enabled by the kernel */
98static int enabled_via_apicbase __ro_after_init;
99
100/*
101 * Handle interrupt mode configuration register (IMCR).
102 * This register controls whether the interrupt signals
103 * that reach the BSP come from the master PIC or from the
104 * local APIC. Before entering Symmetric I/O Mode, either
105 * the BIOS or the operating system must switch out of
106 * PIC Mode by changing the IMCR.
107 */
108static inline void imcr_pic_to_apic(void)
109{
110 /* NMI and 8259 INTR go through APIC */
111 pc_conf_set(PC_CONF_MPS_IMCR, 0x01);
112}
113
114static inline void imcr_apic_to_pic(void)
115{
116 /* NMI and 8259 INTR go directly to BSP */
117 pc_conf_set(PC_CONF_MPS_IMCR, 0x00);
118}
119#endif
120
121/*
122 * Knob to control our willingness to enable the local APIC.
123 *
124 * +1=force-enable
125 */
126static int force_enable_local_apic __initdata;
127
128/*
129 * APIC command line parameters
130 */
131static int __init parse_lapic(char *arg)
132{
133 if (IS_ENABLED(CONFIG_X86_32) && !arg)
134 force_enable_local_apic = 1;
135 else if (arg && !strncmp(arg, "notscdeadline", 13))
136 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
137 return 0;
138}
139early_param("lapic", parse_lapic);
140
141#ifdef CONFIG_X86_64
142static int apic_calibrate_pmtmr __initdata;
143static __init int setup_apicpmtimer(char *s)
144{
145 apic_calibrate_pmtmr = 1;
146 notsc_setup(NULL);
147 return 1;
148}
149__setup("apicpmtimer", setup_apicpmtimer);
150#endif
151
152static unsigned long mp_lapic_addr __ro_after_init;
153bool apic_is_disabled __ro_after_init;
154/* Disable local APIC timer from the kernel commandline or via dmi quirk */
155static int disable_apic_timer __initdata;
156/* Local APIC timer works in C2 */
157int local_apic_timer_c2_ok __ro_after_init;
158EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
159
160/*
161 * Debug level, exported for io_apic.c
162 */
163int apic_verbosity __ro_after_init;
164
165int pic_mode __ro_after_init;
166
167/* Have we found an MP table */
168int smp_found_config __ro_after_init;
169
170static struct resource lapic_resource = {
171 .name = "Local APIC",
172 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
173};
174
175unsigned int lapic_timer_period = 0;
176
177static void apic_pm_activate(void);
178
179/*
180 * Get the LAPIC version
181 */
182static inline int lapic_get_version(void)
183{
184 return GET_APIC_VERSION(apic_read(APIC_LVR));
185}
186
187/*
188 * Check, if the APIC is integrated or a separate chip
189 */
190static inline int lapic_is_integrated(void)
191{
192 return APIC_INTEGRATED(lapic_get_version());
193}
194
195/*
196 * Check, whether this is a modern or a first generation APIC
197 */
198static int modern_apic(void)
199{
200 /* AMD systems use old APIC versions, so check the CPU */
201 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
202 boot_cpu_data.x86 >= 0xf)
203 return 1;
204
205 /* Hygon systems use modern APIC */
206 if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
207 return 1;
208
209 return lapic_get_version() >= 0x14;
210}
211
212/*
213 * right after this call apic become NOOP driven
214 * so apic->write/read doesn't do anything
215 */
216static void __init apic_disable(void)
217{
218 apic_install_driver(driver: &apic_noop);
219}
220
221void native_apic_icr_write(u32 low, u32 id)
222{
223 unsigned long flags;
224
225 local_irq_save(flags);
226 apic_write(APIC_ICR2, SET_XAPIC_DEST_FIELD(id));
227 apic_write(APIC_ICR, val: low);
228 local_irq_restore(flags);
229}
230
231u64 native_apic_icr_read(void)
232{
233 u32 icr1, icr2;
234
235 icr2 = apic_read(APIC_ICR2);
236 icr1 = apic_read(APIC_ICR);
237
238 return icr1 | ((u64)icr2 << 32);
239}
240
241/**
242 * lapic_get_maxlvt - get the maximum number of local vector table entries
243 */
244int lapic_get_maxlvt(void)
245{
246 /*
247 * - we always have APIC integrated on 64bit mode
248 * - 82489DXs do not report # of LVT entries
249 */
250 return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
251}
252
253/*
254 * Local APIC timer
255 */
256
257/* Clock divisor */
258#define APIC_DIVISOR 16
259#define TSC_DIVISOR 8
260
261/* i82489DX specific */
262#define I82489DX_BASE_DIVIDER (((0x2) << 18))
263
264/*
265 * This function sets up the local APIC timer, with a timeout of
266 * 'clocks' APIC bus clock. During calibration we actually call
267 * this function twice on the boot CPU, once with a bogus timeout
268 * value, second time for real. The other (noncalibrating) CPUs
269 * call this function only once, with the real, calibrated value.
270 *
271 * We do reads before writes even if unnecessary, to get around the
272 * P5 APIC double write bug.
273 */
274static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
275{
276 unsigned int lvtt_value, tmp_value;
277
278 lvtt_value = LOCAL_TIMER_VECTOR;
279 if (!oneshot)
280 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
281 else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
282 lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
283
284 /*
285 * The i82489DX APIC uses bit 18 and 19 for the base divider. This
286 * overlaps with bit 18 on integrated APICs, but is not documented
287 * in the SDM. No problem though. i82489DX equipped systems do not
288 * have TSC deadline timer.
289 */
290 if (!lapic_is_integrated())
291 lvtt_value |= I82489DX_BASE_DIVIDER;
292
293 if (!irqen)
294 lvtt_value |= APIC_LVT_MASKED;
295
296 apic_write(APIC_LVTT, val: lvtt_value);
297
298 if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
299 /*
300 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
301 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
302 * According to Intel, MFENCE can do the serialization here.
303 */
304 asm volatile("mfence" : : : "memory");
305 return;
306 }
307
308 /*
309 * Divide PICLK by 16
310 */
311 tmp_value = apic_read(APIC_TDCR);
312 apic_write(APIC_TDCR,
313 val: (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
314 APIC_TDR_DIV_16);
315
316 if (!oneshot)
317 apic_write(APIC_TMICT, val: clocks / APIC_DIVISOR);
318}
319
320/*
321 * Setup extended LVT, AMD specific
322 *
323 * Software should use the LVT offsets the BIOS provides. The offsets
324 * are determined by the subsystems using it like those for MCE
325 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts
326 * are supported. Beginning with family 10h at least 4 offsets are
327 * available.
328 *
329 * Since the offsets must be consistent for all cores, we keep track
330 * of the LVT offsets in software and reserve the offset for the same
331 * vector also to be used on other cores. An offset is freed by
332 * setting the entry to APIC_EILVT_MASKED.
333 *
334 * If the BIOS is right, there should be no conflicts. Otherwise a
335 * "[Firmware Bug]: ..." error message is generated. However, if
336 * software does not properly determines the offsets, it is not
337 * necessarily a BIOS bug.
338 */
339
340static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
341
342static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
343{
344 return (old & APIC_EILVT_MASKED)
345 || (new == APIC_EILVT_MASKED)
346 || ((new & ~APIC_EILVT_MASKED) == old);
347}
348
349static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
350{
351 unsigned int rsvd, vector;
352
353 if (offset >= APIC_EILVT_NR_MAX)
354 return ~0;
355
356 rsvd = atomic_read(v: &eilvt_offsets[offset]);
357 do {
358 vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */
359 if (vector && !eilvt_entry_is_changeable(old: vector, new))
360 /* may not change if vectors are different */
361 return rsvd;
362 } while (!atomic_try_cmpxchg(v: &eilvt_offsets[offset], old: &rsvd, new));
363
364 rsvd = new & ~APIC_EILVT_MASKED;
365 if (rsvd && rsvd != vector)
366 pr_info("LVT offset %d assigned for vector 0x%02x\n",
367 offset, rsvd);
368
369 return new;
370}
371
372/*
373 * If mask=1, the LVT entry does not generate interrupts while mask=0
374 * enables the vector. See also the BKDGs. Must be called with
375 * preemption disabled.
376 */
377
378int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
379{
380 unsigned long reg = APIC_EILVTn(offset);
381 unsigned int new, old, reserved;
382
383 new = (mask << 16) | (msg_type << 8) | vector;
384 old = apic_read(reg);
385 reserved = reserve_eilvt_offset(offset, new);
386
387 if (reserved != new) {
388 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
389 "vector 0x%x, but the register is already in use for "
390 "vector 0x%x on another cpu\n",
391 smp_processor_id(), reg, offset, new, reserved);
392 return -EINVAL;
393 }
394
395 if (!eilvt_entry_is_changeable(old, new)) {
396 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
397 "vector 0x%x, but the register is already in use for "
398 "vector 0x%x on this cpu\n",
399 smp_processor_id(), reg, offset, new, old);
400 return -EBUSY;
401 }
402
403 apic_write(reg, val: new);
404
405 return 0;
406}
407EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
408
409/*
410 * Program the next event, relative to now
411 */
412static int lapic_next_event(unsigned long delta,
413 struct clock_event_device *evt)
414{
415 apic_write(APIC_TMICT, val: delta);
416 return 0;
417}
418
419static int lapic_next_deadline(unsigned long delta,
420 struct clock_event_device *evt)
421{
422 u64 tsc;
423
424 /* This MSR is special and need a special fence: */
425 weak_wrmsr_fence();
426
427 tsc = rdtsc();
428 wrmsrl(MSR_IA32_TSC_DEADLINE, val: tsc + (((u64) delta) * TSC_DIVISOR));
429 return 0;
430}
431
432static int lapic_timer_shutdown(struct clock_event_device *evt)
433{
434 unsigned int v;
435
436 /* Lapic used as dummy for broadcast ? */
437 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
438 return 0;
439
440 v = apic_read(APIC_LVTT);
441 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
442 apic_write(APIC_LVTT, val: v);
443 apic_write(APIC_TMICT, val: 0);
444 return 0;
445}
446
447static inline int
448lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
449{
450 /* Lapic used as dummy for broadcast ? */
451 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
452 return 0;
453
454 __setup_APIC_LVTT(clocks: lapic_timer_period, oneshot, irqen: 1);
455 return 0;
456}
457
458static int lapic_timer_set_periodic(struct clock_event_device *evt)
459{
460 return lapic_timer_set_periodic_oneshot(evt, oneshot: false);
461}
462
463static int lapic_timer_set_oneshot(struct clock_event_device *evt)
464{
465 return lapic_timer_set_periodic_oneshot(evt, oneshot: true);
466}
467
468/*
469 * Local APIC timer broadcast function
470 */
471static void lapic_timer_broadcast(const struct cpumask *mask)
472{
473#ifdef CONFIG_SMP
474 __apic_send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
475#endif
476}
477
478
479/*
480 * The local apic timer can be used for any function which is CPU local.
481 */
482static struct clock_event_device lapic_clockevent = {
483 .name = "lapic",
484 .features = CLOCK_EVT_FEAT_PERIODIC |
485 CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
486 | CLOCK_EVT_FEAT_DUMMY,
487 .shift = 32,
488 .set_state_shutdown = lapic_timer_shutdown,
489 .set_state_periodic = lapic_timer_set_periodic,
490 .set_state_oneshot = lapic_timer_set_oneshot,
491 .set_state_oneshot_stopped = lapic_timer_shutdown,
492 .set_next_event = lapic_next_event,
493 .broadcast = lapic_timer_broadcast,
494 .rating = 100,
495 .irq = -1,
496};
497static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
498
499static const struct x86_cpu_id deadline_match[] __initconst = {
500 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x2, 0x2), 0x3a), /* EP */
501 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x4, 0x4), 0x0f), /* EX */
502
503 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_X, 0x0b000020),
504
505 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x2, 0x2), 0x00000011),
506 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x3, 0x3), 0x0700000e),
507 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x4, 0x4), 0x0f00000c),
508 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x5, 0x5), 0x0e000003),
509
510 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x3, 0x3), 0x01000136),
511 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x4, 0x4), 0x02000014),
512 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x5, 0xf), 0),
513
514 X86_MATCH_INTEL_FAM6_MODEL( HASWELL, 0x22),
515 X86_MATCH_INTEL_FAM6_MODEL( HASWELL_L, 0x20),
516 X86_MATCH_INTEL_FAM6_MODEL( HASWELL_G, 0x17),
517
518 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL, 0x25),
519 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_G, 0x17),
520
521 X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE_L, 0xb2),
522 X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE, 0xb2),
523
524 X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE_L, 0x52),
525 X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE, 0x52),
526
527 {},
528};
529
530static __init bool apic_validate_deadline_timer(void)
531{
532 const struct x86_cpu_id *m;
533 u32 rev;
534
535 if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
536 return false;
537 if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
538 return true;
539
540 m = x86_match_cpu(match: deadline_match);
541 if (!m)
542 return true;
543
544 rev = (u32)m->driver_data;
545
546 if (boot_cpu_data.microcode >= rev)
547 return true;
548
549 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
550 pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
551 "please update microcode to version: 0x%x (or later)\n", rev);
552 return false;
553}
554
555/*
556 * Setup the local APIC timer for this CPU. Copy the initialized values
557 * of the boot CPU and register the clock event in the framework.
558 */
559static void setup_APIC_timer(void)
560{
561 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
562
563 if (this_cpu_has(X86_FEATURE_ARAT)) {
564 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
565 /* Make LAPIC timer preferable over percpu HPET */
566 lapic_clockevent.rating = 150;
567 }
568
569 memcpy(levt, &lapic_clockevent, sizeof(*levt));
570 levt->cpumask = cpumask_of(smp_processor_id());
571
572 if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
573 levt->name = "lapic-deadline";
574 levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
575 CLOCK_EVT_FEAT_DUMMY);
576 levt->set_next_event = lapic_next_deadline;
577 clockevents_config_and_register(dev: levt,
578 freq: tsc_khz * (1000 / TSC_DIVISOR),
579 min_delta: 0xF, max_delta: ~0UL);
580 } else
581 clockevents_register_device(dev: levt);
582}
583
584/*
585 * Install the updated TSC frequency from recalibration at the TSC
586 * deadline clockevent devices.
587 */
588static void __lapic_update_tsc_freq(void *info)
589{
590 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
591
592 if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
593 return;
594
595 clockevents_update_freq(ce: levt, freq: tsc_khz * (1000 / TSC_DIVISOR));
596}
597
598void lapic_update_tsc_freq(void)
599{
600 /*
601 * The clockevent device's ->mult and ->shift can both be
602 * changed. In order to avoid races, schedule the frequency
603 * update code on each CPU.
604 */
605 on_each_cpu(func: __lapic_update_tsc_freq, NULL, wait: 0);
606}
607
608/*
609 * In this functions we calibrate APIC bus clocks to the external timer.
610 *
611 * We want to do the calibration only once since we want to have local timer
612 * irqs synchronous. CPUs connected by the same APIC bus have the very same bus
613 * frequency.
614 *
615 * This was previously done by reading the PIT/HPET and waiting for a wrap
616 * around to find out, that a tick has elapsed. I have a box, where the PIT
617 * readout is broken, so it never gets out of the wait loop again. This was
618 * also reported by others.
619 *
620 * Monitoring the jiffies value is inaccurate and the clockevents
621 * infrastructure allows us to do a simple substitution of the interrupt
622 * handler.
623 *
624 * The calibration routine also uses the pm_timer when possible, as the PIT
625 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
626 * back to normal later in the boot process).
627 */
628
629#define LAPIC_CAL_LOOPS (HZ/10)
630
631static __initdata int lapic_cal_loops = -1;
632static __initdata long lapic_cal_t1, lapic_cal_t2;
633static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
634static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
635static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
636
637/*
638 * Temporary interrupt handler and polled calibration function.
639 */
640static void __init lapic_cal_handler(struct clock_event_device *dev)
641{
642 unsigned long long tsc = 0;
643 long tapic = apic_read(APIC_TMCCT);
644 unsigned long pm = acpi_pm_read_early();
645
646 if (boot_cpu_has(X86_FEATURE_TSC))
647 tsc = rdtsc();
648
649 switch (lapic_cal_loops++) {
650 case 0:
651 lapic_cal_t1 = tapic;
652 lapic_cal_tsc1 = tsc;
653 lapic_cal_pm1 = pm;
654 lapic_cal_j1 = jiffies;
655 break;
656
657 case LAPIC_CAL_LOOPS:
658 lapic_cal_t2 = tapic;
659 lapic_cal_tsc2 = tsc;
660 if (pm < lapic_cal_pm1)
661 pm += ACPI_PM_OVRRUN;
662 lapic_cal_pm2 = pm;
663 lapic_cal_j2 = jiffies;
664 break;
665 }
666}
667
668static int __init
669calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
670{
671 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
672 const long pm_thresh = pm_100ms / 100;
673 unsigned long mult;
674 u64 res;
675
676#ifndef CONFIG_X86_PM_TIMER
677 return -1;
678#endif
679
680 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
681
682 /* Check, if the PM timer is available */
683 if (!deltapm)
684 return -1;
685
686 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, shift_constant: 22);
687
688 if (deltapm > (pm_100ms - pm_thresh) &&
689 deltapm < (pm_100ms + pm_thresh)) {
690 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
691 return 0;
692 }
693
694 res = (((u64)deltapm) * mult) >> 22;
695 do_div(res, 1000000);
696 pr_warn("APIC calibration not consistent "
697 "with PM-Timer: %ldms instead of 100ms\n", (long)res);
698
699 /* Correct the lapic counter value */
700 res = (((u64)(*delta)) * pm_100ms);
701 do_div(res, deltapm);
702 pr_info("APIC delta adjusted to PM-Timer: "
703 "%lu (%ld)\n", (unsigned long)res, *delta);
704 *delta = (long)res;
705
706 /* Correct the tsc counter value */
707 if (boot_cpu_has(X86_FEATURE_TSC)) {
708 res = (((u64)(*deltatsc)) * pm_100ms);
709 do_div(res, deltapm);
710 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
711 "PM-Timer: %lu (%ld)\n",
712 (unsigned long)res, *deltatsc);
713 *deltatsc = (long)res;
714 }
715
716 return 0;
717}
718
719static int __init lapic_init_clockevent(void)
720{
721 if (!lapic_timer_period)
722 return -1;
723
724 /* Calculate the scaled math multiplication factor */
725 lapic_clockevent.mult = div_sc(ticks: lapic_timer_period/APIC_DIVISOR,
726 TICK_NSEC, shift: lapic_clockevent.shift);
727 lapic_clockevent.max_delta_ns =
728 clockevent_delta2ns(latch: 0x7FFFFFFF, evt: &lapic_clockevent);
729 lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
730 lapic_clockevent.min_delta_ns =
731 clockevent_delta2ns(latch: 0xF, evt: &lapic_clockevent);
732 lapic_clockevent.min_delta_ticks = 0xF;
733
734 return 0;
735}
736
737bool __init apic_needs_pit(void)
738{
739 /*
740 * If the frequencies are not known, PIT is required for both TSC
741 * and apic timer calibration.
742 */
743 if (!tsc_khz || !cpu_khz)
744 return true;
745
746 /* Is there an APIC at all or is it disabled? */
747 if (!boot_cpu_has(X86_FEATURE_APIC) || apic_is_disabled)
748 return true;
749
750 /*
751 * If interrupt delivery mode is legacy PIC or virtual wire without
752 * configuration, the local APIC timer won't be set up. Make sure
753 * that the PIT is initialized.
754 */
755 if (apic_intr_mode == APIC_PIC ||
756 apic_intr_mode == APIC_VIRTUAL_WIRE_NO_CONFIG)
757 return true;
758
759 /* Virt guests may lack ARAT, but still have DEADLINE */
760 if (!boot_cpu_has(X86_FEATURE_ARAT))
761 return true;
762
763 /* Deadline timer is based on TSC so no further PIT action required */
764 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
765 return false;
766
767 /* APIC timer disabled? */
768 if (disable_apic_timer)
769 return true;
770 /*
771 * The APIC timer frequency is known already, no PIT calibration
772 * required. If unknown, let the PIT be initialized.
773 */
774 return lapic_timer_period == 0;
775}
776
777static int __init calibrate_APIC_clock(void)
778{
779 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
780 u64 tsc_perj = 0, tsc_start = 0;
781 unsigned long jif_start;
782 unsigned long deltaj;
783 long delta, deltatsc;
784 int pm_referenced = 0;
785
786 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
787 return 0;
788
789 /*
790 * Check if lapic timer has already been calibrated by platform
791 * specific routine, such as tsc calibration code. If so just fill
792 * in the clockevent structure and return.
793 */
794 if (!lapic_init_clockevent()) {
795 apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
796 lapic_timer_period);
797 /*
798 * Direct calibration methods must have an always running
799 * local APIC timer, no need for broadcast timer.
800 */
801 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
802 return 0;
803 }
804
805 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
806 "calibrating APIC timer ...\n");
807
808 /*
809 * There are platforms w/o global clockevent devices. Instead of
810 * making the calibration conditional on that, use a polling based
811 * approach everywhere.
812 */
813 local_irq_disable();
814
815 /*
816 * Setup the APIC counter to maximum. There is no way the lapic
817 * can underflow in the 100ms detection time frame
818 */
819 __setup_APIC_LVTT(clocks: 0xffffffff, oneshot: 0, irqen: 0);
820
821 /*
822 * Methods to terminate the calibration loop:
823 * 1) Global clockevent if available (jiffies)
824 * 2) TSC if available and frequency is known
825 */
826 jif_start = READ_ONCE(jiffies);
827
828 if (tsc_khz) {
829 tsc_start = rdtsc();
830 tsc_perj = div_u64(dividend: (u64)tsc_khz * 1000, HZ);
831 }
832
833 /*
834 * Enable interrupts so the tick can fire, if a global
835 * clockevent device is available
836 */
837 local_irq_enable();
838
839 while (lapic_cal_loops <= LAPIC_CAL_LOOPS) {
840 /* Wait for a tick to elapse */
841 while (1) {
842 if (tsc_khz) {
843 u64 tsc_now = rdtsc();
844 if ((tsc_now - tsc_start) >= tsc_perj) {
845 tsc_start += tsc_perj;
846 break;
847 }
848 } else {
849 unsigned long jif_now = READ_ONCE(jiffies);
850
851 if (time_after(jif_now, jif_start)) {
852 jif_start = jif_now;
853 break;
854 }
855 }
856 cpu_relax();
857 }
858
859 /* Invoke the calibration routine */
860 local_irq_disable();
861 lapic_cal_handler(NULL);
862 local_irq_enable();
863 }
864
865 local_irq_disable();
866
867 /* Build delta t1-t2 as apic timer counts down */
868 delta = lapic_cal_t1 - lapic_cal_t2;
869 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
870
871 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
872
873 /* we trust the PM based calibration if possible */
874 pm_referenced = !calibrate_by_pmtimer(deltapm: lapic_cal_pm2 - lapic_cal_pm1,
875 delta: &delta, deltatsc: &deltatsc);
876
877 lapic_timer_period = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
878 lapic_init_clockevent();
879
880 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
881 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
882 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
883 lapic_timer_period);
884
885 if (boot_cpu_has(X86_FEATURE_TSC)) {
886 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
887 "%ld.%04ld MHz.\n",
888 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
889 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
890 }
891
892 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
893 "%u.%04u MHz.\n",
894 lapic_timer_period / (1000000 / HZ),
895 lapic_timer_period % (1000000 / HZ));
896
897 /*
898 * Do a sanity check on the APIC calibration result
899 */
900 if (lapic_timer_period < (1000000 / HZ)) {
901 local_irq_enable();
902 pr_warn("APIC frequency too slow, disabling apic timer\n");
903 return -1;
904 }
905
906 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
907
908 /*
909 * PM timer calibration failed or not turned on so lets try APIC
910 * timer based calibration, if a global clockevent device is
911 * available.
912 */
913 if (!pm_referenced && global_clock_event) {
914 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
915
916 /*
917 * Setup the apic timer manually
918 */
919 levt->event_handler = lapic_cal_handler;
920 lapic_timer_set_periodic(evt: levt);
921 lapic_cal_loops = -1;
922
923 /* Let the interrupts run */
924 local_irq_enable();
925
926 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
927 cpu_relax();
928
929 /* Stop the lapic timer */
930 local_irq_disable();
931 lapic_timer_shutdown(evt: levt);
932
933 /* Jiffies delta */
934 deltaj = lapic_cal_j2 - lapic_cal_j1;
935 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
936
937 /* Check, if the jiffies result is consistent */
938 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
939 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
940 else
941 levt->features |= CLOCK_EVT_FEAT_DUMMY;
942 }
943 local_irq_enable();
944
945 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
946 pr_warn("APIC timer disabled due to verification failure\n");
947 return -1;
948 }
949
950 return 0;
951}
952
953/*
954 * Setup the boot APIC
955 *
956 * Calibrate and verify the result.
957 */
958void __init setup_boot_APIC_clock(void)
959{
960 /*
961 * The local apic timer can be disabled via the kernel
962 * commandline or from the CPU detection code. Register the lapic
963 * timer as a dummy clock event source on SMP systems, so the
964 * broadcast mechanism is used. On UP systems simply ignore it.
965 */
966 if (disable_apic_timer) {
967 pr_info("Disabling APIC timer\n");
968 /* No broadcast on UP ! */
969 if (num_possible_cpus() > 1) {
970 lapic_clockevent.mult = 1;
971 setup_APIC_timer();
972 }
973 return;
974 }
975
976 if (calibrate_APIC_clock()) {
977 /* No broadcast on UP ! */
978 if (num_possible_cpus() > 1)
979 setup_APIC_timer();
980 return;
981 }
982
983 /*
984 * If nmi_watchdog is set to IO_APIC, we need the
985 * PIT/HPET going. Otherwise register lapic as a dummy
986 * device.
987 */
988 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
989
990 /* Setup the lapic or request the broadcast */
991 setup_APIC_timer();
992 amd_e400_c1e_apic_setup();
993}
994
995void setup_secondary_APIC_clock(void)
996{
997 setup_APIC_timer();
998 amd_e400_c1e_apic_setup();
999}
1000
1001/*
1002 * The guts of the apic timer interrupt
1003 */
1004static void local_apic_timer_interrupt(void)
1005{
1006 struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1007
1008 /*
1009 * Normally we should not be here till LAPIC has been initialized but
1010 * in some cases like kdump, its possible that there is a pending LAPIC
1011 * timer interrupt from previous kernel's context and is delivered in
1012 * new kernel the moment interrupts are enabled.
1013 *
1014 * Interrupts are enabled early and LAPIC is setup much later, hence
1015 * its possible that when we get here evt->event_handler is NULL.
1016 * Check for event_handler being NULL and discard the interrupt as
1017 * spurious.
1018 */
1019 if (!evt->event_handler) {
1020 pr_warn("Spurious LAPIC timer interrupt on cpu %d\n",
1021 smp_processor_id());
1022 /* Switch it off */
1023 lapic_timer_shutdown(evt);
1024 return;
1025 }
1026
1027 /*
1028 * the NMI deadlock-detector uses this.
1029 */
1030 inc_irq_stat(apic_timer_irqs);
1031
1032 evt->event_handler(evt);
1033}
1034
1035/*
1036 * Local APIC timer interrupt. This is the most natural way for doing
1037 * local interrupts, but local timer interrupts can be emulated by
1038 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1039 *
1040 * [ if a single-CPU system runs an SMP kernel then we call the local
1041 * interrupt as well. Thus we cannot inline the local irq ... ]
1042 */
1043DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt)
1044{
1045 struct pt_regs *old_regs = set_irq_regs(regs);
1046
1047 apic_eoi();
1048 trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1049 local_apic_timer_interrupt();
1050 trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1051
1052 set_irq_regs(old_regs);
1053}
1054
1055/*
1056 * Local APIC start and shutdown
1057 */
1058
1059/**
1060 * clear_local_APIC - shutdown the local APIC
1061 *
1062 * This is called, when a CPU is disabled and before rebooting, so the state of
1063 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1064 * leftovers during boot.
1065 */
1066void clear_local_APIC(void)
1067{
1068 int maxlvt;
1069 u32 v;
1070
1071 if (!apic_accessible())
1072 return;
1073
1074 maxlvt = lapic_get_maxlvt();
1075 /*
1076 * Masking an LVT entry can trigger a local APIC error
1077 * if the vector is zero. Mask LVTERR first to prevent this.
1078 */
1079 if (maxlvt >= 3) {
1080 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1081 apic_write(APIC_LVTERR, val: v | APIC_LVT_MASKED);
1082 }
1083 /*
1084 * Careful: we have to set masks only first to deassert
1085 * any level-triggered sources.
1086 */
1087 v = apic_read(APIC_LVTT);
1088 apic_write(APIC_LVTT, val: v | APIC_LVT_MASKED);
1089 v = apic_read(APIC_LVT0);
1090 apic_write(APIC_LVT0, val: v | APIC_LVT_MASKED);
1091 v = apic_read(APIC_LVT1);
1092 apic_write(APIC_LVT1, val: v | APIC_LVT_MASKED);
1093 if (maxlvt >= 4) {
1094 v = apic_read(APIC_LVTPC);
1095 apic_write(APIC_LVTPC, val: v | APIC_LVT_MASKED);
1096 }
1097
1098 /* lets not touch this if we didn't frob it */
1099#ifdef CONFIG_X86_THERMAL_VECTOR
1100 if (maxlvt >= 5) {
1101 v = apic_read(APIC_LVTTHMR);
1102 apic_write(APIC_LVTTHMR, val: v | APIC_LVT_MASKED);
1103 }
1104#endif
1105#ifdef CONFIG_X86_MCE_INTEL
1106 if (maxlvt >= 6) {
1107 v = apic_read(APIC_LVTCMCI);
1108 if (!(v & APIC_LVT_MASKED))
1109 apic_write(APIC_LVTCMCI, val: v | APIC_LVT_MASKED);
1110 }
1111#endif
1112
1113 /*
1114 * Clean APIC state for other OSs:
1115 */
1116 apic_write(APIC_LVTT, APIC_LVT_MASKED);
1117 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1118 apic_write(APIC_LVT1, APIC_LVT_MASKED);
1119 if (maxlvt >= 3)
1120 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1121 if (maxlvt >= 4)
1122 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1123
1124 /* Integrated APIC (!82489DX) ? */
1125 if (lapic_is_integrated()) {
1126 if (maxlvt > 3)
1127 /* Clear ESR due to Pentium errata 3AP and 11AP */
1128 apic_write(APIC_ESR, val: 0);
1129 apic_read(APIC_ESR);
1130 }
1131}
1132
1133/**
1134 * apic_soft_disable - Clears and software disables the local APIC on hotplug
1135 *
1136 * Contrary to disable_local_APIC() this does not touch the enable bit in
1137 * MSR_IA32_APICBASE. Clearing that bit on systems based on the 3 wire APIC
1138 * bus would require a hardware reset as the APIC would lose track of bus
1139 * arbitration. On systems with FSB delivery APICBASE could be disabled,
1140 * but it has to be guaranteed that no interrupt is sent to the APIC while
1141 * in that state and it's not clear from the SDM whether it still responds
1142 * to INIT/SIPI messages. Stay on the safe side and use software disable.
1143 */
1144void apic_soft_disable(void)
1145{
1146 u32 value;
1147
1148 clear_local_APIC();
1149
1150 /* Soft disable APIC (implies clearing of registers for 82489DX!). */
1151 value = apic_read(APIC_SPIV);
1152 value &= ~APIC_SPIV_APIC_ENABLED;
1153 apic_write(APIC_SPIV, val: value);
1154}
1155
1156/**
1157 * disable_local_APIC - clear and disable the local APIC
1158 */
1159void disable_local_APIC(void)
1160{
1161 if (!apic_accessible())
1162 return;
1163
1164 apic_soft_disable();
1165
1166#ifdef CONFIG_X86_32
1167 /*
1168 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1169 * restore the disabled state.
1170 */
1171 if (enabled_via_apicbase) {
1172 unsigned int l, h;
1173
1174 rdmsr(MSR_IA32_APICBASE, l, h);
1175 l &= ~MSR_IA32_APICBASE_ENABLE;
1176 wrmsr(MSR_IA32_APICBASE, l, h);
1177 }
1178#endif
1179}
1180
1181/*
1182 * If Linux enabled the LAPIC against the BIOS default disable it down before
1183 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1184 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1185 * for the case where Linux didn't enable the LAPIC.
1186 */
1187void lapic_shutdown(void)
1188{
1189 unsigned long flags;
1190
1191 if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1192 return;
1193
1194 local_irq_save(flags);
1195
1196#ifdef CONFIG_X86_32
1197 if (!enabled_via_apicbase)
1198 clear_local_APIC();
1199 else
1200#endif
1201 disable_local_APIC();
1202
1203
1204 local_irq_restore(flags);
1205}
1206
1207/**
1208 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1209 */
1210void __init sync_Arb_IDs(void)
1211{
1212 /*
1213 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1214 * needed on AMD.
1215 */
1216 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1217 return;
1218
1219 /*
1220 * Wait for idle.
1221 */
1222 apic_wait_icr_idle();
1223
1224 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1225 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1226 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1227}
1228
1229enum apic_intr_mode_id apic_intr_mode __ro_after_init;
1230
1231static int __init __apic_intr_mode_select(void)
1232{
1233 /* Check kernel option */
1234 if (apic_is_disabled) {
1235 pr_info("APIC disabled via kernel command line\n");
1236 return APIC_PIC;
1237 }
1238
1239 /* Check BIOS */
1240#ifdef CONFIG_X86_64
1241 /* On 64-bit, the APIC must be integrated, Check local APIC only */
1242 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1243 apic_is_disabled = true;
1244 pr_info("APIC disabled by BIOS\n");
1245 return APIC_PIC;
1246 }
1247#else
1248 /* On 32-bit, the APIC may be integrated APIC or 82489DX */
1249
1250 /* Neither 82489DX nor integrated APIC ? */
1251 if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1252 apic_is_disabled = true;
1253 return APIC_PIC;
1254 }
1255
1256 /* If the BIOS pretends there is an integrated APIC ? */
1257 if (!boot_cpu_has(X86_FEATURE_APIC) &&
1258 APIC_INTEGRATED(boot_cpu_apic_version)) {
1259 apic_is_disabled = true;
1260 pr_err(FW_BUG "Local APIC not detected, force emulation\n");
1261 return APIC_PIC;
1262 }
1263#endif
1264
1265 /* Check MP table or ACPI MADT configuration */
1266 if (!smp_found_config) {
1267 disable_ioapic_support();
1268 if (!acpi_lapic) {
1269 pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1270 return APIC_VIRTUAL_WIRE_NO_CONFIG;
1271 }
1272 return APIC_VIRTUAL_WIRE;
1273 }
1274
1275#ifdef CONFIG_SMP
1276 /* If SMP should be disabled, then really disable it! */
1277 if (!setup_max_cpus) {
1278 pr_info("APIC: SMP mode deactivated\n");
1279 return APIC_SYMMETRIC_IO_NO_ROUTING;
1280 }
1281#endif
1282
1283 return APIC_SYMMETRIC_IO;
1284}
1285
1286/* Select the interrupt delivery mode for the BSP */
1287void __init apic_intr_mode_select(void)
1288{
1289 apic_intr_mode = __apic_intr_mode_select();
1290}
1291
1292/*
1293 * An initial setup of the virtual wire mode.
1294 */
1295void __init init_bsp_APIC(void)
1296{
1297 unsigned int value;
1298
1299 /*
1300 * Don't do the setup now if we have a SMP BIOS as the
1301 * through-I/O-APIC virtual wire mode might be active.
1302 */
1303 if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1304 return;
1305
1306 /*
1307 * Do not trust the local APIC being empty at bootup.
1308 */
1309 clear_local_APIC();
1310
1311 /*
1312 * Enable APIC.
1313 */
1314 value = apic_read(APIC_SPIV);
1315 value &= ~APIC_VECTOR_MASK;
1316 value |= APIC_SPIV_APIC_ENABLED;
1317
1318#ifdef CONFIG_X86_32
1319 /* This bit is reserved on P4/Xeon and should be cleared */
1320 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1321 (boot_cpu_data.x86 == 15))
1322 value &= ~APIC_SPIV_FOCUS_DISABLED;
1323 else
1324#endif
1325 value |= APIC_SPIV_FOCUS_DISABLED;
1326 value |= SPURIOUS_APIC_VECTOR;
1327 apic_write(APIC_SPIV, val: value);
1328
1329 /*
1330 * Set up the virtual wire mode.
1331 */
1332 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1333 value = APIC_DM_NMI;
1334 if (!lapic_is_integrated()) /* 82489DX */
1335 value |= APIC_LVT_LEVEL_TRIGGER;
1336 if (apic_extnmi == APIC_EXTNMI_NONE)
1337 value |= APIC_LVT_MASKED;
1338 apic_write(APIC_LVT1, val: value);
1339}
1340
1341static void __init apic_bsp_setup(bool upmode);
1342
1343/* Init the interrupt delivery mode for the BSP */
1344void __init apic_intr_mode_init(void)
1345{
1346 bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1347
1348 switch (apic_intr_mode) {
1349 case APIC_PIC:
1350 pr_info("APIC: Keep in PIC mode(8259)\n");
1351 return;
1352 case APIC_VIRTUAL_WIRE:
1353 pr_info("APIC: Switch to virtual wire mode setup\n");
1354 break;
1355 case APIC_VIRTUAL_WIRE_NO_CONFIG:
1356 pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1357 upmode = true;
1358 break;
1359 case APIC_SYMMETRIC_IO:
1360 pr_info("APIC: Switch to symmetric I/O mode setup\n");
1361 break;
1362 case APIC_SYMMETRIC_IO_NO_ROUTING:
1363 pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1364 break;
1365 }
1366
1367 x86_64_probe_apic();
1368
1369 x86_32_install_bigsmp();
1370
1371 if (x86_platform.apic_post_init)
1372 x86_platform.apic_post_init();
1373
1374 apic_bsp_setup(upmode);
1375}
1376
1377static void lapic_setup_esr(void)
1378{
1379 unsigned int oldvalue, value, maxlvt;
1380
1381 if (!lapic_is_integrated()) {
1382 pr_info("No ESR for 82489DX.\n");
1383 return;
1384 }
1385
1386 if (apic->disable_esr) {
1387 /*
1388 * Something untraceable is creating bad interrupts on
1389 * secondary quads ... for the moment, just leave the
1390 * ESR disabled - we can't do anything useful with the
1391 * errors anyway - mbligh
1392 */
1393 pr_info("Leaving ESR disabled.\n");
1394 return;
1395 }
1396
1397 maxlvt = lapic_get_maxlvt();
1398 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1399 apic_write(APIC_ESR, val: 0);
1400 oldvalue = apic_read(APIC_ESR);
1401
1402 /* enables sending errors */
1403 value = ERROR_APIC_VECTOR;
1404 apic_write(APIC_LVTERR, val: value);
1405
1406 /*
1407 * spec says clear errors after enabling vector.
1408 */
1409 if (maxlvt > 3)
1410 apic_write(APIC_ESR, val: 0);
1411 value = apic_read(APIC_ESR);
1412 if (value != oldvalue)
1413 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1414 "vector: 0x%08x after: 0x%08x\n",
1415 oldvalue, value);
1416}
1417
1418#define APIC_IR_REGS APIC_ISR_NR
1419#define APIC_IR_BITS (APIC_IR_REGS * 32)
1420#define APIC_IR_MAPSIZE (APIC_IR_BITS / BITS_PER_LONG)
1421
1422union apic_ir {
1423 unsigned long map[APIC_IR_MAPSIZE];
1424 u32 regs[APIC_IR_REGS];
1425};
1426
1427static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr)
1428{
1429 int i, bit;
1430
1431 /* Read the IRRs */
1432 for (i = 0; i < APIC_IR_REGS; i++)
1433 irr->regs[i] = apic_read(APIC_IRR + i * 0x10);
1434
1435 /* Read the ISRs */
1436 for (i = 0; i < APIC_IR_REGS; i++)
1437 isr->regs[i] = apic_read(APIC_ISR + i * 0x10);
1438
1439 /*
1440 * If the ISR map is not empty. ACK the APIC and run another round
1441 * to verify whether a pending IRR has been unblocked and turned
1442 * into a ISR.
1443 */
1444 if (!bitmap_empty(src: isr->map, APIC_IR_BITS)) {
1445 /*
1446 * There can be multiple ISR bits set when a high priority
1447 * interrupt preempted a lower priority one. Issue an ACK
1448 * per set bit.
1449 */
1450 for_each_set_bit(bit, isr->map, APIC_IR_BITS)
1451 apic_eoi();
1452 return true;
1453 }
1454
1455 return !bitmap_empty(src: irr->map, APIC_IR_BITS);
1456}
1457
1458/*
1459 * After a crash, we no longer service the interrupts and a pending
1460 * interrupt from previous kernel might still have ISR bit set.
1461 *
1462 * Most probably by now the CPU has serviced that pending interrupt and it
1463 * might not have done the apic_eoi() because it thought, interrupt
1464 * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear
1465 * the ISR bit and cpu thinks it has already serviced the interrupt. Hence
1466 * a vector might get locked. It was noticed for timer irq (vector
1467 * 0x31). Issue an extra EOI to clear ISR.
1468 *
1469 * If there are pending IRR bits they turn into ISR bits after a higher
1470 * priority ISR bit has been acked.
1471 */
1472static void apic_pending_intr_clear(void)
1473{
1474 union apic_ir irr, isr;
1475 unsigned int i;
1476
1477 /* 512 loops are way oversized and give the APIC a chance to obey. */
1478 for (i = 0; i < 512; i++) {
1479 if (!apic_check_and_ack(irr: &irr, isr: &isr))
1480 return;
1481 }
1482 /* Dump the IRR/ISR content if that failed */
1483 pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map);
1484}
1485
1486/**
1487 * setup_local_APIC - setup the local APIC
1488 *
1489 * Used to setup local APIC while initializing BSP or bringing up APs.
1490 * Always called with preemption disabled.
1491 */
1492static void setup_local_APIC(void)
1493{
1494 int cpu = smp_processor_id();
1495 unsigned int value;
1496
1497 if (apic_is_disabled) {
1498 disable_ioapic_support();
1499 return;
1500 }
1501
1502 /*
1503 * If this comes from kexec/kcrash the APIC might be enabled in
1504 * SPIV. Soft disable it before doing further initialization.
1505 */
1506 value = apic_read(APIC_SPIV);
1507 value &= ~APIC_SPIV_APIC_ENABLED;
1508 apic_write(APIC_SPIV, val: value);
1509
1510#ifdef CONFIG_X86_32
1511 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1512 if (lapic_is_integrated() && apic->disable_esr) {
1513 apic_write(APIC_ESR, 0);
1514 apic_write(APIC_ESR, 0);
1515 apic_write(APIC_ESR, 0);
1516 apic_write(APIC_ESR, 0);
1517 }
1518#endif
1519 /*
1520 * Intel recommends to set DFR, LDR and TPR before enabling
1521 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1522 * document number 292116).
1523 *
1524 * Except for APICs which operate in physical destination mode.
1525 */
1526 if (apic->init_apic_ldr)
1527 apic->init_apic_ldr();
1528
1529 /*
1530 * Set Task Priority to 'accept all except vectors 0-31'. An APIC
1531 * vector in the 16-31 range could be delivered if TPR == 0, but we
1532 * would think it's an exception and terrible things will happen. We
1533 * never change this later on.
1534 */
1535 value = apic_read(APIC_TASKPRI);
1536 value &= ~APIC_TPRI_MASK;
1537 value |= 0x10;
1538 apic_write(APIC_TASKPRI, val: value);
1539
1540 /* Clear eventually stale ISR/IRR bits */
1541 apic_pending_intr_clear();
1542
1543 /*
1544 * Now that we are all set up, enable the APIC
1545 */
1546 value = apic_read(APIC_SPIV);
1547 value &= ~APIC_VECTOR_MASK;
1548 /*
1549 * Enable APIC
1550 */
1551 value |= APIC_SPIV_APIC_ENABLED;
1552
1553#ifdef CONFIG_X86_32
1554 /*
1555 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1556 * certain networking cards. If high frequency interrupts are
1557 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1558 * entry is masked/unmasked at a high rate as well then sooner or
1559 * later IOAPIC line gets 'stuck', no more interrupts are received
1560 * from the device. If focus CPU is disabled then the hang goes
1561 * away, oh well :-(
1562 *
1563 * [ This bug can be reproduced easily with a level-triggered
1564 * PCI Ne2000 networking cards and PII/PIII processors, dual
1565 * BX chipset. ]
1566 */
1567 /*
1568 * Actually disabling the focus CPU check just makes the hang less
1569 * frequent as it makes the interrupt distribution model be more
1570 * like LRU than MRU (the short-term load is more even across CPUs).
1571 */
1572
1573 /*
1574 * - enable focus processor (bit==0)
1575 * - 64bit mode always use processor focus
1576 * so no need to set it
1577 */
1578 value &= ~APIC_SPIV_FOCUS_DISABLED;
1579#endif
1580
1581 /*
1582 * Set spurious IRQ vector
1583 */
1584 value |= SPURIOUS_APIC_VECTOR;
1585 apic_write(APIC_SPIV, val: value);
1586
1587 perf_events_lapic_init();
1588
1589 /*
1590 * Set up LVT0, LVT1:
1591 *
1592 * set up through-local-APIC on the boot CPU's LINT0. This is not
1593 * strictly necessary in pure symmetric-IO mode, but sometimes
1594 * we delegate interrupts to the 8259A.
1595 */
1596 /*
1597 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1598 */
1599 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1600 if (!cpu && (pic_mode || !value || ioapic_is_disabled)) {
1601 value = APIC_DM_EXTINT;
1602 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1603 } else {
1604 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1605 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1606 }
1607 apic_write(APIC_LVT0, val: value);
1608
1609 /*
1610 * Only the BSP sees the LINT1 NMI signal by default. This can be
1611 * modified by apic_extnmi= boot option.
1612 */
1613 if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1614 apic_extnmi == APIC_EXTNMI_ALL)
1615 value = APIC_DM_NMI;
1616 else
1617 value = APIC_DM_NMI | APIC_LVT_MASKED;
1618
1619 /* Is 82489DX ? */
1620 if (!lapic_is_integrated())
1621 value |= APIC_LVT_LEVEL_TRIGGER;
1622 apic_write(APIC_LVT1, val: value);
1623
1624#ifdef CONFIG_X86_MCE_INTEL
1625 /* Recheck CMCI information after local APIC is up on CPU #0 */
1626 if (!cpu)
1627 cmci_recheck();
1628#endif
1629}
1630
1631static void end_local_APIC_setup(void)
1632{
1633 lapic_setup_esr();
1634
1635#ifdef CONFIG_X86_32
1636 {
1637 unsigned int value;
1638 /* Disable the local apic timer */
1639 value = apic_read(APIC_LVTT);
1640 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1641 apic_write(APIC_LVTT, value);
1642 }
1643#endif
1644
1645 apic_pm_activate();
1646}
1647
1648/*
1649 * APIC setup function for application processors. Called from smpboot.c
1650 */
1651void apic_ap_setup(void)
1652{
1653 setup_local_APIC();
1654 end_local_APIC_setup();
1655}
1656
1657static __init void apic_read_boot_cpu_id(bool x2apic)
1658{
1659 /*
1660 * This can be invoked from check_x2apic() before the APIC has been
1661 * selected. But that code knows for sure that the BIOS enabled
1662 * X2APIC.
1663 */
1664 if (x2apic) {
1665 boot_cpu_physical_apicid = native_apic_msr_read(APIC_ID);
1666 boot_cpu_apic_version = GET_APIC_VERSION(native_apic_msr_read(APIC_LVR));
1667 } else {
1668 boot_cpu_physical_apicid = read_apic_id();
1669 boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
1670 }
1671 topology_register_boot_apic(apic_id: boot_cpu_physical_apicid);
1672 x86_32_probe_bigsmp_early();
1673}
1674
1675#ifdef CONFIG_X86_X2APIC
1676int x2apic_mode;
1677EXPORT_SYMBOL_GPL(x2apic_mode);
1678
1679enum {
1680 X2APIC_OFF,
1681 X2APIC_DISABLED,
1682 /* All states below here have X2APIC enabled */
1683 X2APIC_ON,
1684 X2APIC_ON_LOCKED
1685};
1686static int x2apic_state;
1687
1688static bool x2apic_hw_locked(void)
1689{
1690 u64 x86_arch_cap_msr;
1691 u64 msr;
1692
1693 x86_arch_cap_msr = x86_read_arch_cap_msr();
1694 if (x86_arch_cap_msr & ARCH_CAP_XAPIC_DISABLE) {
1695 rdmsrl(MSR_IA32_XAPIC_DISABLE_STATUS, msr);
1696 return (msr & LEGACY_XAPIC_DISABLED);
1697 }
1698 return false;
1699}
1700
1701static void __x2apic_disable(void)
1702{
1703 u64 msr;
1704
1705 if (!boot_cpu_has(X86_FEATURE_APIC))
1706 return;
1707
1708 rdmsrl(MSR_IA32_APICBASE, msr);
1709 if (!(msr & X2APIC_ENABLE))
1710 return;
1711 /* Disable xapic and x2apic first and then reenable xapic mode */
1712 wrmsrl(MSR_IA32_APICBASE, val: msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1713 wrmsrl(MSR_IA32_APICBASE, val: msr & ~X2APIC_ENABLE);
1714 printk_once(KERN_INFO "x2apic disabled\n");
1715}
1716
1717static void __x2apic_enable(void)
1718{
1719 u64 msr;
1720
1721 rdmsrl(MSR_IA32_APICBASE, msr);
1722 if (msr & X2APIC_ENABLE)
1723 return;
1724 wrmsrl(MSR_IA32_APICBASE, val: msr | X2APIC_ENABLE);
1725 printk_once(KERN_INFO "x2apic enabled\n");
1726}
1727
1728static int __init setup_nox2apic(char *str)
1729{
1730 if (x2apic_enabled()) {
1731 u32 apicid = native_apic_msr_read(APIC_ID);
1732
1733 if (apicid >= 255) {
1734 pr_warn("Apicid: %08x, cannot enforce nox2apic\n",
1735 apicid);
1736 return 0;
1737 }
1738 if (x2apic_hw_locked()) {
1739 pr_warn("APIC locked in x2apic mode, can't disable\n");
1740 return 0;
1741 }
1742 pr_warn("x2apic already enabled.\n");
1743 __x2apic_disable();
1744 }
1745 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1746 x2apic_state = X2APIC_DISABLED;
1747 x2apic_mode = 0;
1748 return 0;
1749}
1750early_param("nox2apic", setup_nox2apic);
1751
1752/* Called from cpu_init() to enable x2apic on (secondary) cpus */
1753void x2apic_setup(void)
1754{
1755 /*
1756 * Try to make the AP's APIC state match that of the BSP, but if the
1757 * BSP is unlocked and the AP is locked then there is a state mismatch.
1758 * Warn about the mismatch in case a GP fault occurs due to a locked AP
1759 * trying to be turned off.
1760 */
1761 if (x2apic_state != X2APIC_ON_LOCKED && x2apic_hw_locked())
1762 pr_warn("x2apic lock mismatch between BSP and AP.\n");
1763 /*
1764 * If x2apic is not in ON or LOCKED state, disable it if already enabled
1765 * from BIOS.
1766 */
1767 if (x2apic_state < X2APIC_ON) {
1768 __x2apic_disable();
1769 return;
1770 }
1771 __x2apic_enable();
1772}
1773
1774static __init void apic_set_fixmap(void);
1775
1776static __init void x2apic_disable(void)
1777{
1778 u32 x2apic_id, state = x2apic_state;
1779
1780 x2apic_mode = 0;
1781 x2apic_state = X2APIC_DISABLED;
1782
1783 if (state != X2APIC_ON)
1784 return;
1785
1786 x2apic_id = read_apic_id();
1787 if (x2apic_id >= 255)
1788 panic(fmt: "Cannot disable x2apic, id: %08x\n", x2apic_id);
1789
1790 if (x2apic_hw_locked()) {
1791 pr_warn("Cannot disable locked x2apic, id: %08x\n", x2apic_id);
1792 return;
1793 }
1794
1795 __x2apic_disable();
1796 apic_set_fixmap();
1797}
1798
1799static __init void x2apic_enable(void)
1800{
1801 if (x2apic_state != X2APIC_OFF)
1802 return;
1803
1804 x2apic_mode = 1;
1805 x2apic_state = X2APIC_ON;
1806 __x2apic_enable();
1807}
1808
1809static __init void try_to_enable_x2apic(int remap_mode)
1810{
1811 if (x2apic_state == X2APIC_DISABLED)
1812 return;
1813
1814 if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1815 u32 apic_limit = 255;
1816
1817 /*
1818 * Using X2APIC without IR is not architecturally supported
1819 * on bare metal but may be supported in guests.
1820 */
1821 if (!x86_init.hyper.x2apic_available()) {
1822 pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1823 x2apic_disable();
1824 return;
1825 }
1826
1827 /*
1828 * If the hypervisor supports extended destination ID in
1829 * MSI, that increases the maximum APIC ID that can be
1830 * used for non-remapped IRQ domains.
1831 */
1832 if (x86_init.hyper.msi_ext_dest_id()) {
1833 virt_ext_dest_id = 1;
1834 apic_limit = 32767;
1835 }
1836
1837 /*
1838 * Without IR, all CPUs can be addressed by IOAPIC/MSI only
1839 * in physical mode, and CPUs with an APIC ID that cannot
1840 * be addressed must not be brought online.
1841 */
1842 x2apic_set_max_apicid(apicid: apic_limit);
1843 x2apic_phys = 1;
1844 }
1845 x2apic_enable();
1846}
1847
1848void __init check_x2apic(void)
1849{
1850 if (x2apic_enabled()) {
1851 pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1852 x2apic_mode = 1;
1853 if (x2apic_hw_locked())
1854 x2apic_state = X2APIC_ON_LOCKED;
1855 else
1856 x2apic_state = X2APIC_ON;
1857 apic_read_boot_cpu_id(x2apic: true);
1858 } else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1859 x2apic_state = X2APIC_DISABLED;
1860 }
1861}
1862#else /* CONFIG_X86_X2APIC */
1863void __init check_x2apic(void)
1864{
1865 if (!apic_is_x2apic_enabled())
1866 return;
1867 /*
1868 * Checkme: Can we simply turn off x2APIC here instead of disabling the APIC?
1869 */
1870 pr_err("Kernel does not support x2APIC, please recompile with CONFIG_X86_X2APIC.\n");
1871 pr_err("Disabling APIC, expect reduced performance and functionality.\n");
1872
1873 apic_is_disabled = true;
1874 setup_clear_cpu_cap(X86_FEATURE_APIC);
1875}
1876
1877static inline void try_to_enable_x2apic(int remap_mode) { }
1878static inline void __x2apic_enable(void) { }
1879#endif /* !CONFIG_X86_X2APIC */
1880
1881void __init enable_IR_x2apic(void)
1882{
1883 unsigned long flags;
1884 int ret, ir_stat;
1885
1886 if (ioapic_is_disabled) {
1887 pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1888 return;
1889 }
1890
1891 ir_stat = irq_remapping_prepare();
1892 if (ir_stat < 0 && !x2apic_supported())
1893 return;
1894
1895 ret = save_ioapic_entries();
1896 if (ret) {
1897 pr_info("Saving IO-APIC state failed: %d\n", ret);
1898 return;
1899 }
1900
1901 local_irq_save(flags);
1902 legacy_pic->mask_all();
1903 mask_ioapic_entries();
1904
1905 /* If irq_remapping_prepare() succeeded, try to enable it */
1906 if (ir_stat >= 0)
1907 ir_stat = irq_remapping_enable();
1908 /* ir_stat contains the remap mode or an error code */
1909 try_to_enable_x2apic(remap_mode: ir_stat);
1910
1911 if (ir_stat < 0)
1912 restore_ioapic_entries();
1913 legacy_pic->restore_mask();
1914 local_irq_restore(flags);
1915}
1916
1917#ifdef CONFIG_X86_64
1918/*
1919 * Detect and enable local APICs on non-SMP boards.
1920 * Original code written by Keir Fraser.
1921 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1922 * not correctly set up (usually the APIC timer won't work etc.)
1923 */
1924static bool __init detect_init_APIC(void)
1925{
1926 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1927 pr_info("No local APIC present\n");
1928 return false;
1929 }
1930
1931 register_lapic_address(APIC_DEFAULT_PHYS_BASE);
1932 return true;
1933}
1934#else
1935
1936static bool __init apic_verify(unsigned long addr)
1937{
1938 u32 features, h, l;
1939
1940 /*
1941 * The APIC feature bit should now be enabled
1942 * in `cpuid'
1943 */
1944 features = cpuid_edx(1);
1945 if (!(features & (1 << X86_FEATURE_APIC))) {
1946 pr_warn("Could not enable APIC!\n");
1947 return false;
1948 }
1949 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1950
1951 /* The BIOS may have set up the APIC at some other address */
1952 if (boot_cpu_data.x86 >= 6) {
1953 rdmsr(MSR_IA32_APICBASE, l, h);
1954 if (l & MSR_IA32_APICBASE_ENABLE)
1955 addr = l & MSR_IA32_APICBASE_BASE;
1956 }
1957
1958 register_lapic_address(addr);
1959 pr_info("Found and enabled local APIC!\n");
1960 return true;
1961}
1962
1963bool __init apic_force_enable(unsigned long addr)
1964{
1965 u32 h, l;
1966
1967 if (apic_is_disabled)
1968 return false;
1969
1970 /*
1971 * Some BIOSes disable the local APIC in the APIC_BASE
1972 * MSR. This can only be done in software for Intel P6 or later
1973 * and AMD K7 (Model > 1) or later.
1974 */
1975 if (boot_cpu_data.x86 >= 6) {
1976 rdmsr(MSR_IA32_APICBASE, l, h);
1977 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1978 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1979 l &= ~MSR_IA32_APICBASE_BASE;
1980 l |= MSR_IA32_APICBASE_ENABLE | addr;
1981 wrmsr(MSR_IA32_APICBASE, l, h);
1982 enabled_via_apicbase = 1;
1983 }
1984 }
1985 return apic_verify(addr);
1986}
1987
1988/*
1989 * Detect and initialize APIC
1990 */
1991static bool __init detect_init_APIC(void)
1992{
1993 /* Disabled by kernel option? */
1994 if (apic_is_disabled)
1995 return false;
1996
1997 switch (boot_cpu_data.x86_vendor) {
1998 case X86_VENDOR_AMD:
1999 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
2000 (boot_cpu_data.x86 >= 15))
2001 break;
2002 goto no_apic;
2003 case X86_VENDOR_HYGON:
2004 break;
2005 case X86_VENDOR_INTEL:
2006 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
2007 (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
2008 break;
2009 goto no_apic;
2010 default:
2011 goto no_apic;
2012 }
2013
2014 if (!boot_cpu_has(X86_FEATURE_APIC)) {
2015 /*
2016 * Over-ride BIOS and try to enable the local APIC only if
2017 * "lapic" specified.
2018 */
2019 if (!force_enable_local_apic) {
2020 pr_info("Local APIC disabled by BIOS -- "
2021 "you can enable it with \"lapic\"\n");
2022 return false;
2023 }
2024 if (!apic_force_enable(APIC_DEFAULT_PHYS_BASE))
2025 return false;
2026 } else {
2027 if (!apic_verify(APIC_DEFAULT_PHYS_BASE))
2028 return false;
2029 }
2030
2031 apic_pm_activate();
2032
2033 return true;
2034
2035no_apic:
2036 pr_info("No local APIC present or hardware disabled\n");
2037 return false;
2038}
2039#endif
2040
2041/**
2042 * init_apic_mappings - initialize APIC mappings
2043 */
2044void __init init_apic_mappings(void)
2045{
2046 if (apic_validate_deadline_timer())
2047 pr_info("TSC deadline timer available\n");
2048
2049 if (x2apic_mode)
2050 return;
2051
2052 if (!smp_found_config) {
2053 if (!detect_init_APIC()) {
2054 pr_info("APIC: disable apic facility\n");
2055 apic_disable();
2056 }
2057 }
2058}
2059
2060static __init void apic_set_fixmap(void)
2061{
2062 set_fixmap_nocache(FIX_APIC_BASE, mp_lapic_addr);
2063 apic_mmio_base = APIC_BASE;
2064 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
2065 apic_mmio_base, mp_lapic_addr);
2066 apic_read_boot_cpu_id(x2apic: false);
2067}
2068
2069void __init register_lapic_address(unsigned long address)
2070{
2071 /* This should only happen once */
2072 WARN_ON_ONCE(mp_lapic_addr);
2073 mp_lapic_addr = address;
2074
2075 if (!x2apic_mode)
2076 apic_set_fixmap();
2077}
2078
2079/*
2080 * Local APIC interrupts
2081 */
2082
2083/*
2084 * Common handling code for spurious_interrupt and spurious_vector entry
2085 * points below. No point in allowing the compiler to inline it twice.
2086 */
2087static noinline void handle_spurious_interrupt(u8 vector)
2088{
2089 u32 v;
2090
2091 trace_spurious_apic_entry(vector);
2092
2093 inc_irq_stat(irq_spurious_count);
2094
2095 /*
2096 * If this is a spurious interrupt then do not acknowledge
2097 */
2098 if (vector == SPURIOUS_APIC_VECTOR) {
2099 /* See SDM vol 3 */
2100 pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n",
2101 smp_processor_id());
2102 goto out;
2103 }
2104
2105 /*
2106 * If it is a vectored one, verify it's set in the ISR. If set,
2107 * acknowledge it.
2108 */
2109 v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2110 if (v & (1 << (vector & 0x1f))) {
2111 pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n",
2112 vector, smp_processor_id());
2113 apic_eoi();
2114 } else {
2115 pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n",
2116 vector, smp_processor_id());
2117 }
2118out:
2119 trace_spurious_apic_exit(vector);
2120}
2121
2122/**
2123 * spurious_interrupt - Catch all for interrupts raised on unused vectors
2124 * @regs: Pointer to pt_regs on stack
2125 * @vector: The vector number
2126 *
2127 * This is invoked from ASM entry code to catch all interrupts which
2128 * trigger on an entry which is routed to the common_spurious idtentry
2129 * point.
2130 */
2131DEFINE_IDTENTRY_IRQ(spurious_interrupt)
2132{
2133 handle_spurious_interrupt(vector);
2134}
2135
2136DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt)
2137{
2138 handle_spurious_interrupt(SPURIOUS_APIC_VECTOR);
2139}
2140
2141/*
2142 * This interrupt should never happen with our APIC/SMP architecture
2143 */
2144DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt)
2145{
2146 static const char * const error_interrupt_reason[] = {
2147 "Send CS error", /* APIC Error Bit 0 */
2148 "Receive CS error", /* APIC Error Bit 1 */
2149 "Send accept error", /* APIC Error Bit 2 */
2150 "Receive accept error", /* APIC Error Bit 3 */
2151 "Redirectable IPI", /* APIC Error Bit 4 */
2152 "Send illegal vector", /* APIC Error Bit 5 */
2153 "Received illegal vector", /* APIC Error Bit 6 */
2154 "Illegal register address", /* APIC Error Bit 7 */
2155 };
2156 u32 v, i = 0;
2157
2158 trace_error_apic_entry(ERROR_APIC_VECTOR);
2159
2160 /* First tickle the hardware, only then report what went on. -- REW */
2161 if (lapic_get_maxlvt() > 3) /* Due to the Pentium erratum 3AP. */
2162 apic_write(APIC_ESR, val: 0);
2163 v = apic_read(APIC_ESR);
2164 apic_eoi();
2165 atomic_inc(v: &irq_err_count);
2166
2167 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2168 smp_processor_id(), v);
2169
2170 v &= 0xff;
2171 while (v) {
2172 if (v & 0x1)
2173 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2174 i++;
2175 v >>= 1;
2176 }
2177
2178 apic_printk(APIC_DEBUG, KERN_CONT "\n");
2179
2180 trace_error_apic_exit(ERROR_APIC_VECTOR);
2181}
2182
2183/**
2184 * connect_bsp_APIC - attach the APIC to the interrupt system
2185 */
2186static void __init connect_bsp_APIC(void)
2187{
2188#ifdef CONFIG_X86_32
2189 if (pic_mode) {
2190 /*
2191 * Do not trust the local APIC being empty at bootup.
2192 */
2193 clear_local_APIC();
2194 /*
2195 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
2196 * local APIC to INT and NMI lines.
2197 */
2198 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2199 "enabling APIC mode.\n");
2200 imcr_pic_to_apic();
2201 }
2202#endif
2203}
2204
2205/**
2206 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2207 * @virt_wire_setup: indicates, whether virtual wire mode is selected
2208 *
2209 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2210 * APIC is disabled.
2211 */
2212void disconnect_bsp_APIC(int virt_wire_setup)
2213{
2214 unsigned int value;
2215
2216#ifdef CONFIG_X86_32
2217 if (pic_mode) {
2218 /*
2219 * Put the board back into PIC mode (has an effect only on
2220 * certain older boards). Note that APIC interrupts, including
2221 * IPIs, won't work beyond this point! The only exception are
2222 * INIT IPIs.
2223 */
2224 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2225 "entering PIC mode.\n");
2226 imcr_apic_to_pic();
2227 return;
2228 }
2229#endif
2230
2231 /* Go back to Virtual Wire compatibility mode */
2232
2233 /* For the spurious interrupt use vector F, and enable it */
2234 value = apic_read(APIC_SPIV);
2235 value &= ~APIC_VECTOR_MASK;
2236 value |= APIC_SPIV_APIC_ENABLED;
2237 value |= 0xf;
2238 apic_write(APIC_SPIV, val: value);
2239
2240 if (!virt_wire_setup) {
2241 /*
2242 * For LVT0 make it edge triggered, active high,
2243 * external and enabled
2244 */
2245 value = apic_read(APIC_LVT0);
2246 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2247 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2248 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2249 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2250 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2251 apic_write(APIC_LVT0, val: value);
2252 } else {
2253 /* Disable LVT0 */
2254 apic_write(APIC_LVT0, APIC_LVT_MASKED);
2255 }
2256
2257 /*
2258 * For LVT1 make it edge triggered, active high,
2259 * nmi and enabled
2260 */
2261 value = apic_read(APIC_LVT1);
2262 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2263 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2264 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2265 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2266 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2267 apic_write(APIC_LVT1, val: value);
2268}
2269
2270void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,
2271 bool dmar)
2272{
2273 memset(msg, 0, sizeof(*msg));
2274
2275 msg->arch_addr_lo.base_address = X86_MSI_BASE_ADDRESS_LOW;
2276 msg->arch_addr_lo.dest_mode_logical = apic->dest_mode_logical;
2277 msg->arch_addr_lo.destid_0_7 = cfg->dest_apicid & 0xFF;
2278
2279 msg->arch_data.delivery_mode = APIC_DELIVERY_MODE_FIXED;
2280 msg->arch_data.vector = cfg->vector;
2281
2282 msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH;
2283 /*
2284 * Only the IOMMU itself can use the trick of putting destination
2285 * APIC ID into the high bits of the address. Anything else would
2286 * just be writing to memory if it tried that, and needs IR to
2287 * address APICs which can't be addressed in the normal 32-bit
2288 * address range at 0xFFExxxxx. That is typically just 8 bits, but
2289 * some hypervisors allow the extended destination ID field in bits
2290 * 5-11 to be used, giving support for 15 bits of APIC IDs in total.
2291 */
2292 if (dmar)
2293 msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >> 8;
2294 else if (virt_ext_dest_id && cfg->dest_apicid < 0x8000)
2295 msg->arch_addr_lo.virt_destid_8_14 = cfg->dest_apicid >> 8;
2296 else
2297 WARN_ON_ONCE(cfg->dest_apicid > 0xFF);
2298}
2299
2300u32 x86_msi_msg_get_destid(struct msi_msg *msg, bool extid)
2301{
2302 u32 dest = msg->arch_addr_lo.destid_0_7;
2303
2304 if (extid)
2305 dest |= msg->arch_addr_hi.destid_8_31 << 8;
2306 return dest;
2307}
2308EXPORT_SYMBOL_GPL(x86_msi_msg_get_destid);
2309
2310static void __init apic_bsp_up_setup(void)
2311{
2312 reset_phys_cpu_present_map(apicid: boot_cpu_physical_apicid);
2313}
2314
2315/**
2316 * apic_bsp_setup - Setup function for local apic and io-apic
2317 * @upmode: Force UP mode (for APIC_init_uniprocessor)
2318 */
2319static void __init apic_bsp_setup(bool upmode)
2320{
2321 connect_bsp_APIC();
2322 if (upmode)
2323 apic_bsp_up_setup();
2324 setup_local_APIC();
2325
2326 enable_IO_APIC();
2327 end_local_APIC_setup();
2328 irq_remap_enable_fault_handling();
2329 setup_IO_APIC();
2330 lapic_update_legacy_vectors();
2331}
2332
2333#ifdef CONFIG_UP_LATE_INIT
2334void __init up_late_init(void)
2335{
2336 if (apic_intr_mode == APIC_PIC)
2337 return;
2338
2339 /* Setup local timer */
2340 x86_init.timers.setup_percpu_clockev();
2341}
2342#endif
2343
2344/*
2345 * Power management
2346 */
2347#ifdef CONFIG_PM
2348
2349static struct {
2350 /*
2351 * 'active' is true if the local APIC was enabled by us and
2352 * not the BIOS; this signifies that we are also responsible
2353 * for disabling it before entering apm/acpi suspend
2354 */
2355 int active;
2356 /* r/w apic fields */
2357 u32 apic_id;
2358 unsigned int apic_taskpri;
2359 unsigned int apic_ldr;
2360 unsigned int apic_dfr;
2361 unsigned int apic_spiv;
2362 unsigned int apic_lvtt;
2363 unsigned int apic_lvtpc;
2364 unsigned int apic_lvt0;
2365 unsigned int apic_lvt1;
2366 unsigned int apic_lvterr;
2367 unsigned int apic_tmict;
2368 unsigned int apic_tdcr;
2369 unsigned int apic_thmr;
2370 unsigned int apic_cmci;
2371} apic_pm_state;
2372
2373static int lapic_suspend(void)
2374{
2375 unsigned long flags;
2376 int maxlvt;
2377
2378 if (!apic_pm_state.active)
2379 return 0;
2380
2381 maxlvt = lapic_get_maxlvt();
2382
2383 apic_pm_state.apic_id = apic_read(APIC_ID);
2384 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2385 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2386 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2387 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2388 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2389 if (maxlvt >= 4)
2390 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2391 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2392 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2393 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2394 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2395 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2396#ifdef CONFIG_X86_THERMAL_VECTOR
2397 if (maxlvt >= 5)
2398 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2399#endif
2400#ifdef CONFIG_X86_MCE_INTEL
2401 if (maxlvt >= 6)
2402 apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2403#endif
2404
2405 local_irq_save(flags);
2406
2407 /*
2408 * Mask IOAPIC before disabling the local APIC to prevent stale IRR
2409 * entries on some implementations.
2410 */
2411 mask_ioapic_entries();
2412
2413 disable_local_APIC();
2414
2415 irq_remapping_disable();
2416
2417 local_irq_restore(flags);
2418 return 0;
2419}
2420
2421static void lapic_resume(void)
2422{
2423 unsigned int l, h;
2424 unsigned long flags;
2425 int maxlvt;
2426
2427 if (!apic_pm_state.active)
2428 return;
2429
2430 local_irq_save(flags);
2431
2432 /*
2433 * IO-APIC and PIC have their own resume routines.
2434 * We just mask them here to make sure the interrupt
2435 * subsystem is completely quiet while we enable x2apic
2436 * and interrupt-remapping.
2437 */
2438 mask_ioapic_entries();
2439 legacy_pic->mask_all();
2440
2441 if (x2apic_mode) {
2442 __x2apic_enable();
2443 } else {
2444 /*
2445 * Make sure the APICBASE points to the right address
2446 *
2447 * FIXME! This will be wrong if we ever support suspend on
2448 * SMP! We'll need to do this as part of the CPU restore!
2449 */
2450 if (boot_cpu_data.x86 >= 6) {
2451 rdmsr(MSR_IA32_APICBASE, l, h);
2452 l &= ~MSR_IA32_APICBASE_BASE;
2453 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2454 wrmsr(MSR_IA32_APICBASE, l, h);
2455 }
2456 }
2457
2458 maxlvt = lapic_get_maxlvt();
2459 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2460 apic_write(APIC_ID, val: apic_pm_state.apic_id);
2461 apic_write(APIC_DFR, val: apic_pm_state.apic_dfr);
2462 apic_write(APIC_LDR, val: apic_pm_state.apic_ldr);
2463 apic_write(APIC_TASKPRI, val: apic_pm_state.apic_taskpri);
2464 apic_write(APIC_SPIV, val: apic_pm_state.apic_spiv);
2465 apic_write(APIC_LVT0, val: apic_pm_state.apic_lvt0);
2466 apic_write(APIC_LVT1, val: apic_pm_state.apic_lvt1);
2467#ifdef CONFIG_X86_THERMAL_VECTOR
2468 if (maxlvt >= 5)
2469 apic_write(APIC_LVTTHMR, val: apic_pm_state.apic_thmr);
2470#endif
2471#ifdef CONFIG_X86_MCE_INTEL
2472 if (maxlvt >= 6)
2473 apic_write(APIC_LVTCMCI, val: apic_pm_state.apic_cmci);
2474#endif
2475 if (maxlvt >= 4)
2476 apic_write(APIC_LVTPC, val: apic_pm_state.apic_lvtpc);
2477 apic_write(APIC_LVTT, val: apic_pm_state.apic_lvtt);
2478 apic_write(APIC_TDCR, val: apic_pm_state.apic_tdcr);
2479 apic_write(APIC_TMICT, val: apic_pm_state.apic_tmict);
2480 apic_write(APIC_ESR, val: 0);
2481 apic_read(APIC_ESR);
2482 apic_write(APIC_LVTERR, val: apic_pm_state.apic_lvterr);
2483 apic_write(APIC_ESR, val: 0);
2484 apic_read(APIC_ESR);
2485
2486 irq_remapping_reenable(x2apic_mode);
2487
2488 local_irq_restore(flags);
2489}
2490
2491/*
2492 * This device has no shutdown method - fully functioning local APICs
2493 * are needed on every CPU up until machine_halt/restart/poweroff.
2494 */
2495
2496static struct syscore_ops lapic_syscore_ops = {
2497 .resume = lapic_resume,
2498 .suspend = lapic_suspend,
2499};
2500
2501static void apic_pm_activate(void)
2502{
2503 apic_pm_state.active = 1;
2504}
2505
2506static int __init init_lapic_sysfs(void)
2507{
2508 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2509 if (boot_cpu_has(X86_FEATURE_APIC))
2510 register_syscore_ops(ops: &lapic_syscore_ops);
2511
2512 return 0;
2513}
2514
2515/* local apic needs to resume before other devices access its registers. */
2516core_initcall(init_lapic_sysfs);
2517
2518#else /* CONFIG_PM */
2519
2520static void apic_pm_activate(void) { }
2521
2522#endif /* CONFIG_PM */
2523
2524#ifdef CONFIG_X86_64
2525
2526static int multi_checked;
2527static int multi;
2528
2529static int set_multi(const struct dmi_system_id *d)
2530{
2531 if (multi)
2532 return 0;
2533 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2534 multi = 1;
2535 return 0;
2536}
2537
2538static const struct dmi_system_id multi_dmi_table[] = {
2539 {
2540 .callback = set_multi,
2541 .ident = "IBM System Summit2",
2542 .matches = {
2543 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2544 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2545 },
2546 },
2547 {}
2548};
2549
2550static void dmi_check_multi(void)
2551{
2552 if (multi_checked)
2553 return;
2554
2555 dmi_check_system(list: multi_dmi_table);
2556 multi_checked = 1;
2557}
2558
2559/*
2560 * apic_is_clustered_box() -- Check if we can expect good TSC
2561 *
2562 * Thus far, the major user of this is IBM's Summit2 series:
2563 * Clustered boxes may have unsynced TSC problems if they are
2564 * multi-chassis.
2565 * Use DMI to check them
2566 */
2567int apic_is_clustered_box(void)
2568{
2569 dmi_check_multi();
2570 return multi;
2571}
2572#endif
2573
2574/*
2575 * APIC command line parameters
2576 */
2577static int __init setup_disableapic(char *arg)
2578{
2579 apic_is_disabled = true;
2580 setup_clear_cpu_cap(X86_FEATURE_APIC);
2581 return 0;
2582}
2583early_param("disableapic", setup_disableapic);
2584
2585/* same as disableapic, for compatibility */
2586static int __init setup_nolapic(char *arg)
2587{
2588 return setup_disableapic(arg);
2589}
2590early_param("nolapic", setup_nolapic);
2591
2592static int __init parse_lapic_timer_c2_ok(char *arg)
2593{
2594 local_apic_timer_c2_ok = 1;
2595 return 0;
2596}
2597early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2598
2599static int __init parse_disable_apic_timer(char *arg)
2600{
2601 disable_apic_timer = 1;
2602 return 0;
2603}
2604early_param("noapictimer", parse_disable_apic_timer);
2605
2606static int __init parse_nolapic_timer(char *arg)
2607{
2608 disable_apic_timer = 1;
2609 return 0;
2610}
2611early_param("nolapic_timer", parse_nolapic_timer);
2612
2613static int __init apic_set_verbosity(char *arg)
2614{
2615 if (!arg) {
2616 if (IS_ENABLED(CONFIG_X86_32))
2617 return -EINVAL;
2618
2619 ioapic_is_disabled = false;
2620 return 0;
2621 }
2622
2623 if (strcmp("debug", arg) == 0)
2624 apic_verbosity = APIC_DEBUG;
2625 else if (strcmp("verbose", arg) == 0)
2626 apic_verbosity = APIC_VERBOSE;
2627#ifdef CONFIG_X86_64
2628 else {
2629 pr_warn("APIC Verbosity level %s not recognised"
2630 " use apic=verbose or apic=debug\n", arg);
2631 return -EINVAL;
2632 }
2633#endif
2634
2635 return 0;
2636}
2637early_param("apic", apic_set_verbosity);
2638
2639static int __init lapic_insert_resource(void)
2640{
2641 if (!apic_mmio_base)
2642 return -1;
2643
2644 /* Put local APIC into the resource map. */
2645 lapic_resource.start = apic_mmio_base;
2646 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2647 insert_resource(parent: &iomem_resource, new: &lapic_resource);
2648
2649 return 0;
2650}
2651
2652/*
2653 * need call insert after e820__reserve_resources()
2654 * that is using request_resource
2655 */
2656late_initcall(lapic_insert_resource);
2657
2658static int __init apic_set_extnmi(char *arg)
2659{
2660 if (!arg)
2661 return -EINVAL;
2662
2663 if (!strncmp("all", arg, 3))
2664 apic_extnmi = APIC_EXTNMI_ALL;
2665 else if (!strncmp("none", arg, 4))
2666 apic_extnmi = APIC_EXTNMI_NONE;
2667 else if (!strncmp("bsp", arg, 3))
2668 apic_extnmi = APIC_EXTNMI_BSP;
2669 else {
2670 pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2671 return -EINVAL;
2672 }
2673
2674 return 0;
2675}
2676early_param("apic_extnmi", apic_set_extnmi);
2677

source code of linux/arch/x86/kernel/apic/apic.c