2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1996, by Steve Passe
6 * Copyright (c) 2003 John Baldwin <jhb@FreeBSD.org>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. The name of the developer may NOT be used to endorse or promote products
14 * derived from this software without specific prior written permission.
15 * 3. Neither the name of the author nor the names of any co-contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * Local APIC support on Pentium and later processors.
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
39 #include "opt_atpic.h"
40 #include "opt_hwpmc_hooks.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
47 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/mutex.h>
53 #include <sys/sched.h>
55 #include <sys/sysctl.h>
56 #include <sys/timeet.h>
61 #include <x86/apicreg.h>
62 #include <machine/clock.h>
63 #include <machine/cpufunc.h>
64 #include <machine/cputypes.h>
65 #include <machine/frame.h>
66 #include <machine/intr_machdep.h>
67 #include <x86/apicvar.h>
69 #include <machine/md_var.h>
70 #include <machine/smp.h>
71 #include <machine/specialreg.h>
75 #include <sys/interrupt.h>
80 #define SDT_APIC SDT_SYSIGT
83 #define SDT_APIC SDT_SYS386IGT
84 #define GSEL_APIC GSEL(GCODE_SEL, SEL_KPL)
87 static MALLOC_DEFINE(M_LAPIC, "local_apic", "Local APIC items");
89 /* Sanity checks on IDT vectors. */
90 CTASSERT(APIC_IO_INTS + APIC_NUM_IOINTS == APIC_TIMER_INT);
91 CTASSERT(APIC_TIMER_INT < APIC_LOCAL_INTS);
92 CTASSERT(APIC_LOCAL_INTS == 240);
93 CTASSERT(IPI_STOP < APIC_SPURIOUS_INT);
96 * I/O interrupts use non-negative IRQ values. These values are used
97 * to mark unused IDT entries or IDT entries reserved for a non-I/O
102 #define IRQ_SYSCALL -3
103 #define IRQ_DTRACE_RET -4
104 #define IRQ_EVTCHN -5
106 enum lat_timer_mode {
108 LAT_MODE_PERIODIC = 1,
109 LAT_MODE_ONESHOT = 2,
110 LAT_MODE_DEADLINE = 3,
114 * Support for local APICs. Local APICs manage interrupts on each
115 * individual processor as opposed to I/O APICs which receive interrupts
116 * from I/O devices and then forward them on to the local APICs.
118 * Local APICs can also send interrupts to each other thus providing the
119 * mechanism for IPIs.
123 u_int lvt_edgetrigger:1;
124 u_int lvt_activehi:1;
132 struct lvt la_lvts[APIC_LVT_MAX + 1];
133 struct lvt la_elvts[APIC_ELVT_MAX + 1];
136 u_int la_cluster_id:2;
138 u_long *la_timer_count;
139 uint64_t la_timer_period;
140 enum lat_timer_mode la_timer_mode;
141 uint32_t lvt_timer_base;
142 uint32_t lvt_timer_last;
143 /* Include IDT_SYSCALL to make indexing easier. */
144 int la_ioint_irqs[APIC_NUM_IOINTS + 1];
147 /* Global defaults for local APIC LVT entries. */
148 static struct lvt lvts[APIC_LVT_MAX + 1] = {
149 { 1, 1, 1, 1, APIC_LVT_DM_EXTINT, 0 }, /* LINT0: masked ExtINT */
150 { 1, 1, 0, 1, APIC_LVT_DM_NMI, 0 }, /* LINT1: NMI */
151 { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_TIMER_INT }, /* Timer */
152 { 1, 1, 0, 1, APIC_LVT_DM_FIXED, APIC_ERROR_INT }, /* Error */
153 { 1, 1, 1, 1, APIC_LVT_DM_NMI, 0 }, /* PMC */
154 { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_THERMAL_INT }, /* Thermal */
155 { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_CMC_INT }, /* CMCI */
158 /* Global defaults for AMD local APIC ELVT entries. */
159 static struct lvt elvts[APIC_ELVT_MAX + 1] = {
160 { 1, 1, 1, 0, APIC_LVT_DM_FIXED, 0 },
161 { 1, 1, 1, 0, APIC_LVT_DM_FIXED, APIC_CMC_INT },
162 { 1, 1, 1, 0, APIC_LVT_DM_FIXED, 0 },
163 { 1, 1, 1, 0, APIC_LVT_DM_FIXED, 0 },
166 static inthand_t *ioint_handlers[] = {
168 IDTVEC(apic_isr1), /* 32 - 63 */
169 IDTVEC(apic_isr2), /* 64 - 95 */
170 IDTVEC(apic_isr3), /* 96 - 127 */
171 IDTVEC(apic_isr4), /* 128 - 159 */
172 IDTVEC(apic_isr5), /* 160 - 191 */
173 IDTVEC(apic_isr6), /* 192 - 223 */
174 IDTVEC(apic_isr7), /* 224 - 255 */
177 static inthand_t *ioint_pti_handlers[] = {
179 IDTVEC(apic_isr1_pti), /* 32 - 63 */
180 IDTVEC(apic_isr2_pti), /* 64 - 95 */
181 IDTVEC(apic_isr3_pti), /* 96 - 127 */
182 IDTVEC(apic_isr4_pti), /* 128 - 159 */
183 IDTVEC(apic_isr5_pti), /* 160 - 191 */
184 IDTVEC(apic_isr6_pti), /* 192 - 223 */
185 IDTVEC(apic_isr7_pti), /* 224 - 255 */
188 static u_int32_t lapic_timer_divisors[] = {
189 APIC_TDCR_1, APIC_TDCR_2, APIC_TDCR_4, APIC_TDCR_8, APIC_TDCR_16,
190 APIC_TDCR_32, APIC_TDCR_64, APIC_TDCR_128
193 extern inthand_t IDTVEC(rsvd_pti), IDTVEC(rsvd);
195 volatile char *lapic_map;
196 vm_paddr_t lapic_paddr;
198 int lapic_eoi_suppression;
199 static int lapic_timer_tsc_deadline;
200 static u_long lapic_timer_divisor, count_freq;
201 static struct eventtimer lapic_et;
203 static uint64_t lapic_ipi_wait_mult;
205 unsigned int max_apic_id;
207 SYSCTL_NODE(_hw, OID_AUTO, apic, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
209 SYSCTL_INT(_hw_apic, OID_AUTO, x2apic_mode, CTLFLAG_RD, &x2apic_mode, 0, "");
210 SYSCTL_INT(_hw_apic, OID_AUTO, eoi_suppression, CTLFLAG_RD,
211 &lapic_eoi_suppression, 0, "");
212 SYSCTL_INT(_hw_apic, OID_AUTO, timer_tsc_deadline, CTLFLAG_RD,
213 &lapic_timer_tsc_deadline, 0, "");
215 static void lapic_calibrate_initcount(struct lapic *la);
216 static void lapic_calibrate_deadline(struct lapic *la);
219 * Use __nosanitizethread to exempt the LAPIC I/O accessors from KCSan
220 * instrumentation. Otherwise, if x2APIC is not available, use of the global
221 * lapic_map will generate a KCSan false positive. While the mapping is
222 * shared among all CPUs, the physical access will always take place on the
223 * local CPU's APIC, so there isn't in fact a race here. Furthermore, the
224 * KCSan warning printf can cause a panic if issued during LAPIC access,
225 * due to attempted recursive use of event timer resources.
228 static uint32_t __nosanitizethread
229 lapic_read32(enum LAPIC_REGISTERS reg)
234 res = rdmsr32(MSR_APIC_000 + reg);
236 res = *(volatile uint32_t *)(lapic_map + reg * LAPIC_MEM_MUL);
241 static void __nosanitizethread
242 lapic_write32(enum LAPIC_REGISTERS reg, uint32_t val)
248 wrmsr(MSR_APIC_000 + reg, val);
250 *(volatile uint32_t *)(lapic_map + reg * LAPIC_MEM_MUL) = val;
254 static void __nosanitizethread
255 lapic_write32_nofence(enum LAPIC_REGISTERS reg, uint32_t val)
259 wrmsr(MSR_APIC_000 + reg, val);
261 *(volatile uint32_t *)(lapic_map + reg * LAPIC_MEM_MUL) = val;
267 lapic_read_icr_lo(void)
270 return (lapic_read32(LAPIC_ICR_LO));
274 lapic_write_icr(uint32_t vhi, uint32_t vlo)
280 v = ((uint64_t)vhi << 32) | vlo;
282 wrmsr(MSR_APIC_000 + LAPIC_ICR_LO, v);
284 saveintr = intr_disable();
285 lapic_write32(LAPIC_ICR_HI, vhi);
286 lapic_write32(LAPIC_ICR_LO, vlo);
287 intr_restore(saveintr);
292 lapic_write_icr_lo(uint32_t vlo)
297 wrmsr(MSR_APIC_000 + LAPIC_ICR_LO, vlo);
299 lapic_write32(LAPIC_ICR_LO, vlo);
304 lapic_write_self_ipi(uint32_t vector)
307 KASSERT(x2apic_mode, ("SELF IPI write in xAPIC mode"));
308 wrmsr(MSR_APIC_000 + LAPIC_SELF_IPI, vector);
313 native_lapic_enable_x2apic(void)
317 apic_base = rdmsr(MSR_APICBASE);
318 apic_base |= APICBASE_X2APIC | APICBASE_ENABLED;
319 wrmsr(MSR_APICBASE, apic_base);
323 native_lapic_is_x2apic(void)
327 apic_base = rdmsr(MSR_APICBASE);
328 return ((apic_base & (APICBASE_X2APIC | APICBASE_ENABLED)) ==
329 (APICBASE_X2APIC | APICBASE_ENABLED));
332 static void lapic_enable(void);
333 static void lapic_resume(struct pic *pic, bool suspend_cancelled);
334 static void lapic_timer_oneshot(struct lapic *);
335 static void lapic_timer_oneshot_nointr(struct lapic *, uint32_t);
336 static void lapic_timer_periodic(struct lapic *);
337 static void lapic_timer_deadline(struct lapic *);
338 static void lapic_timer_stop(struct lapic *);
339 static void lapic_timer_set_divisor(u_int divisor);
340 static uint32_t lvt_mode(struct lapic *la, u_int pin, uint32_t value);
341 static int lapic_et_start(struct eventtimer *et,
342 sbintime_t first, sbintime_t period);
343 static int lapic_et_stop(struct eventtimer *et);
344 static u_int apic_idt_to_irq(u_int apic_id, u_int vector);
345 static void lapic_set_tpr(u_int vector);
347 struct pic lapic_pic = { .pic_resume = lapic_resume };
349 /* Forward declarations for apic_ops */
350 static void native_lapic_create(u_int apic_id, int boot_cpu);
351 static void native_lapic_init(vm_paddr_t addr);
352 static void native_lapic_xapic_mode(void);
353 static void native_lapic_setup(int boot);
354 static void native_lapic_dump(const char *str);
355 static void native_lapic_disable(void);
356 static void native_lapic_eoi(void);
357 static int native_lapic_id(void);
358 static int native_lapic_intr_pending(u_int vector);
359 static u_int native_apic_cpuid(u_int apic_id);
360 static u_int native_apic_alloc_vector(u_int apic_id, u_int irq);
361 static u_int native_apic_alloc_vectors(u_int apic_id, u_int *irqs,
362 u_int count, u_int align);
363 static void native_apic_disable_vector(u_int apic_id, u_int vector);
364 static void native_apic_enable_vector(u_int apic_id, u_int vector);
365 static void native_apic_free_vector(u_int apic_id, u_int vector, u_int irq);
366 static void native_lapic_set_logical_id(u_int apic_id, u_int cluster,
368 static int native_lapic_enable_pmc(void);
369 static void native_lapic_disable_pmc(void);
370 static void native_lapic_reenable_pmc(void);
371 static void native_lapic_enable_cmc(void);
372 static int native_lapic_enable_mca_elvt(void);
373 static int native_lapic_set_lvt_mask(u_int apic_id, u_int lvt,
375 static int native_lapic_set_lvt_mode(u_int apic_id, u_int lvt,
377 static int native_lapic_set_lvt_polarity(u_int apic_id, u_int lvt,
378 enum intr_polarity pol);
379 static int native_lapic_set_lvt_triggermode(u_int apic_id, u_int lvt,
380 enum intr_trigger trigger);
382 static void native_lapic_ipi_raw(register_t icrlo, u_int dest);
383 static void native_lapic_ipi_vectored(u_int vector, int dest);
384 static int native_lapic_ipi_wait(int delay);
386 static int native_lapic_ipi_alloc(inthand_t *ipifunc);
387 static void native_lapic_ipi_free(int vector);
389 struct apic_ops apic_ops = {
390 .create = native_lapic_create,
391 .init = native_lapic_init,
392 .xapic_mode = native_lapic_xapic_mode,
393 .is_x2apic = native_lapic_is_x2apic,
394 .setup = native_lapic_setup,
395 .dump = native_lapic_dump,
396 .disable = native_lapic_disable,
397 .eoi = native_lapic_eoi,
398 .id = native_lapic_id,
399 .intr_pending = native_lapic_intr_pending,
400 .set_logical_id = native_lapic_set_logical_id,
401 .cpuid = native_apic_cpuid,
402 .alloc_vector = native_apic_alloc_vector,
403 .alloc_vectors = native_apic_alloc_vectors,
404 .enable_vector = native_apic_enable_vector,
405 .disable_vector = native_apic_disable_vector,
406 .free_vector = native_apic_free_vector,
407 .enable_pmc = native_lapic_enable_pmc,
408 .disable_pmc = native_lapic_disable_pmc,
409 .reenable_pmc = native_lapic_reenable_pmc,
410 .enable_cmc = native_lapic_enable_cmc,
411 .enable_mca_elvt = native_lapic_enable_mca_elvt,
413 .ipi_raw = native_lapic_ipi_raw,
414 .ipi_vectored = native_lapic_ipi_vectored,
415 .ipi_wait = native_lapic_ipi_wait,
417 .ipi_alloc = native_lapic_ipi_alloc,
418 .ipi_free = native_lapic_ipi_free,
419 .set_lvt_mask = native_lapic_set_lvt_mask,
420 .set_lvt_mode = native_lapic_set_lvt_mode,
421 .set_lvt_polarity = native_lapic_set_lvt_polarity,
422 .set_lvt_triggermode = native_lapic_set_lvt_triggermode,
426 lvt_mode_impl(struct lapic *la, struct lvt *lvt, u_int pin, uint32_t value)
429 value &= ~(APIC_LVT_M | APIC_LVT_TM | APIC_LVT_IIPP | APIC_LVT_DM |
431 if (lvt->lvt_edgetrigger == 0)
432 value |= APIC_LVT_TM;
433 if (lvt->lvt_activehi == 0)
434 value |= APIC_LVT_IIPP_INTALO;
437 value |= lvt->lvt_mode;
438 switch (lvt->lvt_mode) {
439 case APIC_LVT_DM_NMI:
440 case APIC_LVT_DM_SMI:
441 case APIC_LVT_DM_INIT:
442 case APIC_LVT_DM_EXTINT:
443 if (!lvt->lvt_edgetrigger && bootverbose) {
444 printf("lapic%u: Forcing LINT%u to edge trigger\n",
446 value &= ~APIC_LVT_TM;
448 /* Use a vector of 0. */
450 case APIC_LVT_DM_FIXED:
451 value |= lvt->lvt_vector;
454 panic("bad APIC LVT delivery mode: %#x\n", value);
460 lvt_mode(struct lapic *la, u_int pin, uint32_t value)
464 KASSERT(pin <= APIC_LVT_MAX,
465 ("%s: pin %u out of range", __func__, pin));
466 if (la->la_lvts[pin].lvt_active)
467 lvt = &la->la_lvts[pin];
471 return (lvt_mode_impl(la, lvt, pin, value));
475 elvt_mode(struct lapic *la, u_int idx, uint32_t value)
479 KASSERT(idx <= APIC_ELVT_MAX,
480 ("%s: idx %u out of range", __func__, idx));
482 elvt = &la->la_elvts[idx];
483 KASSERT(elvt->lvt_active, ("%s: ELVT%u is not active", __func__, idx));
484 KASSERT(elvt->lvt_edgetrigger,
485 ("%s: ELVT%u is not edge triggered", __func__, idx));
486 KASSERT(elvt->lvt_activehi,
487 ("%s: ELVT%u is not active high", __func__, idx));
488 return (lvt_mode_impl(la, elvt, idx, value));
492 * Map the local APIC and setup necessary interrupt vectors.
495 native_lapic_init(vm_paddr_t addr)
498 uint64_t r, r1, r2, rx;
505 * Enable x2APIC mode if possible. Map the local APIC
508 * Keep the LAPIC registers page mapped uncached for x2APIC
509 * mode too, to have direct map page attribute set to
510 * uncached. This is needed to work around CPU errata present
511 * on all Intel processors.
513 KASSERT(trunc_page(addr) == addr,
514 ("local APIC not aligned on a page boundary"));
516 lapic_map = pmap_mapdev(addr, PAGE_SIZE);
518 native_lapic_enable_x2apic();
522 /* Setup the spurious interrupt handler. */
523 setidt(APIC_SPURIOUS_INT, IDTVEC(spuriousint), SDT_APIC, SEL_KPL,
526 /* Perform basic initialization of the BSP's local APIC. */
529 /* Set BSP's per-CPU local APIC ID. */
530 PCPU_SET(apic_id, lapic_id());
532 /* Local APIC timer interrupt. */
533 setidt(APIC_TIMER_INT, pti ? IDTVEC(timerint_pti) : IDTVEC(timerint),
534 SDT_APIC, SEL_KPL, GSEL_APIC);
536 /* Local APIC error interrupt. */
537 setidt(APIC_ERROR_INT, pti ? IDTVEC(errorint_pti) : IDTVEC(errorint),
538 SDT_APIC, SEL_KPL, GSEL_APIC);
540 /* XXX: Thermal interrupt */
542 /* Local APIC CMCI. */
543 setidt(APIC_CMC_INT, pti ? IDTVEC(cmcint_pti) : IDTVEC(cmcint),
544 SDT_APIC, SEL_KPL, GSEL_APIC);
546 if ((resource_int_value("apic", 0, "clock", &i) != 0 || i != 0)) {
547 /* Set if APIC timer runs in C3. */
548 arat = (cpu_power_eax & CPUTPM1_ARAT);
550 bzero(&lapic_et, sizeof(lapic_et));
551 lapic_et.et_name = "LAPIC";
552 lapic_et.et_flags = ET_FLAGS_PERIODIC | ET_FLAGS_ONESHOT |
554 lapic_et.et_quality = 600;
556 lapic_et.et_flags |= ET_FLAGS_C3STOP;
557 lapic_et.et_quality = 100;
559 if ((cpu_feature & CPUID_TSC) != 0 &&
560 (cpu_feature2 & CPUID2_TSCDLT) != 0 &&
561 tsc_is_invariant && tsc_freq != 0) {
562 lapic_timer_tsc_deadline = 1;
563 TUNABLE_INT_FETCH("hw.lapic_tsc_deadline",
564 &lapic_timer_tsc_deadline);
567 lapic_et.et_frequency = 0;
568 /* We don't know frequency yet, so trying to guess. */
569 lapic_et.et_min_period = 0x00001000LL;
570 lapic_et.et_max_period = SBT_1S;
571 lapic_et.et_start = lapic_et_start;
572 lapic_et.et_stop = lapic_et_stop;
573 lapic_et.et_priv = NULL;
574 et_register(&lapic_et);
578 * Set lapic_eoi_suppression after lapic_enable(), to not
579 * enable suppression in the hardware prematurely. Note that
580 * we by default enable suppression even when system only has
581 * one IO-APIC, since EOI is broadcasted to all APIC agents,
582 * including CPUs, otherwise.
584 * It seems that at least some KVM versions report
585 * EOI_SUPPRESSION bit, but auto-EOI does not work.
587 ver = lapic_read32(LAPIC_VERSION);
588 if ((ver & APIC_VER_EOI_SUPPRESSION) != 0) {
589 lapic_eoi_suppression = 1;
590 if (vm_guest == VM_GUEST_KVM) {
593 "KVM -- disabling lapic eoi suppression\n");
594 lapic_eoi_suppression = 0;
596 TUNABLE_INT_FETCH("hw.lapic_eoi_suppression",
597 &lapic_eoi_suppression);
603 * Calibrate the busy loop waiting for IPI ack in xAPIC mode.
604 * lapic_ipi_wait_mult contains the number of iterations which
605 * approximately delay execution for 1 microsecond (the
606 * argument to native_lapic_ipi_wait() is in microseconds).
608 * We assume that TSC is present and already measured.
609 * Possible TSC frequency jumps are irrelevant to the
610 * calibration loop below, the CPU clock management code is
611 * not yet started, and we do not enter sleep states.
613 KASSERT((cpu_feature & CPUID_TSC) != 0 && tsc_freq != 0,
614 ("TSC not initialized"));
617 for (rx = 0; rx < LOOPS; rx++) {
618 (void)lapic_read_icr_lo();
622 r1 = tsc_freq * LOOPS;
624 lapic_ipi_wait_mult = r1 >= r2 ? r1 / r2 : 1;
626 printf("LAPIC: ipi_wait() us multiplier %ju (r %ju "
627 "tsc %ju)\n", (uintmax_t)lapic_ipi_wait_mult,
628 (uintmax_t)r, (uintmax_t)tsc_freq);
636 * Create a local APIC instance.
639 native_lapic_create(u_int apic_id, int boot_cpu)
643 if (apic_id > max_apic_id) {
644 printf("APIC: Ignoring local APIC with ID %d\n", apic_id);
646 panic("Can't ignore BSP");
649 KASSERT(!lapics[apic_id].la_present, ("duplicate local APIC %u",
653 * Assume no local LVT overrides and a cluster of 0 and
654 * intra-cluster ID of 0.
656 lapics[apic_id].la_present = 1;
657 lapics[apic_id].la_id = apic_id;
658 for (i = 0; i <= APIC_LVT_MAX; i++) {
659 lapics[apic_id].la_lvts[i] = lvts[i];
660 lapics[apic_id].la_lvts[i].lvt_active = 0;
662 for (i = 0; i <= APIC_ELVT_MAX; i++) {
663 lapics[apic_id].la_elvts[i] = elvts[i];
664 lapics[apic_id].la_elvts[i].lvt_active = 0;
666 for (i = 0; i <= APIC_NUM_IOINTS; i++)
667 lapics[apic_id].la_ioint_irqs[i] = IRQ_FREE;
668 lapics[apic_id].la_ioint_irqs[IDT_SYSCALL - APIC_IO_INTS] = IRQ_SYSCALL;
669 lapics[apic_id].la_ioint_irqs[APIC_TIMER_INT - APIC_IO_INTS] =
672 lapics[apic_id].la_ioint_irqs[IDT_DTRACE_RET - APIC_IO_INTS] =
676 lapics[apic_id].la_ioint_irqs[IDT_EVTCHN - APIC_IO_INTS] = IRQ_EVTCHN;
680 cpu_add(apic_id, boot_cpu);
684 static inline uint32_t
685 amd_read_ext_features(void)
689 if (cpu_vendor_id != CPU_VENDOR_AMD &&
690 cpu_vendor_id != CPU_VENDOR_HYGON)
692 version = lapic_read32(LAPIC_VERSION);
693 if ((version & APIC_VER_AMD_EXT_SPACE) != 0)
694 return (lapic_read32(LAPIC_EXT_FEATURES));
699 static inline uint32_t
700 amd_read_elvt_count(void)
705 extf = amd_read_ext_features();
706 count = (extf & APIC_EXTF_ELVT_MASK) >> APIC_EXTF_ELVT_SHIFT;
707 count = min(count, APIC_ELVT_MAX + 1);
712 * Dump contents of local APIC registers
715 native_lapic_dump(const char* str)
723 version = lapic_read32(LAPIC_VERSION);
724 maxlvt = (version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
725 printf("cpu%d %s:\n", PCPU_GET(cpuid), str);
726 printf(" ID: 0x%08x VER: 0x%08x LDR: 0x%08x DFR: 0x%08x",
727 lapic_read32(LAPIC_ID), version,
728 lapic_read32(LAPIC_LDR), x2apic_mode ? 0 : lapic_read32(LAPIC_DFR));
729 if ((cpu_feature2 & CPUID2_X2APIC) != 0)
730 printf(" x2APIC: %d", x2apic_mode);
731 printf("\n lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
732 lapic_read32(LAPIC_LVT_LINT0), lapic_read32(LAPIC_LVT_LINT1),
733 lapic_read32(LAPIC_TPR), lapic_read32(LAPIC_SVR));
734 printf(" timer: 0x%08x therm: 0x%08x err: 0x%08x",
735 lapic_read32(LAPIC_LVT_TIMER), lapic_read32(LAPIC_LVT_THERMAL),
736 lapic_read32(LAPIC_LVT_ERROR));
737 if (maxlvt >= APIC_LVT_PMC)
738 printf(" pmc: 0x%08x", lapic_read32(LAPIC_LVT_PCINT));
740 if (maxlvt >= APIC_LVT_CMCI)
741 printf(" cmci: 0x%08x\n", lapic_read32(LAPIC_LVT_CMCI));
742 extf = amd_read_ext_features();
744 printf(" AMD ext features: 0x%08x\n", extf);
745 elvt_count = amd_read_elvt_count();
746 for (i = 0; i < elvt_count; i++)
747 printf(" AMD elvt%d: 0x%08x\n", i,
748 lapic_read32(LAPIC_EXT_LVT0 + i));
753 native_lapic_xapic_mode(void)
757 saveintr = intr_disable();
759 native_lapic_enable_x2apic();
760 intr_restore(saveintr);
764 native_lapic_setup(int boot)
773 saveintr = intr_disable();
775 la = &lapics[lapic_id()];
776 KASSERT(la->la_present, ("missing APIC structure"));
777 version = lapic_read32(LAPIC_VERSION);
778 maxlvt = (version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
780 /* Initialize the TPR to allow all interrupts. */
783 /* Setup spurious vector and enable the local APIC. */
786 /* Program LINT[01] LVT entries. */
787 lapic_write32(LAPIC_LVT_LINT0, lvt_mode(la, APIC_LVT_LINT0,
788 lapic_read32(LAPIC_LVT_LINT0)));
789 lapic_write32(LAPIC_LVT_LINT1, lvt_mode(la, APIC_LVT_LINT1,
790 lapic_read32(LAPIC_LVT_LINT1)));
792 /* Program the PMC LVT entry if present. */
793 if (maxlvt >= APIC_LVT_PMC) {
794 lapic_write32(LAPIC_LVT_PCINT, lvt_mode(la, APIC_LVT_PMC,
798 /* Program timer LVT. */
799 la->lvt_timer_base = lvt_mode(la, APIC_LVT_TIMER,
800 lapic_read32(LAPIC_LVT_TIMER));
801 la->lvt_timer_last = la->lvt_timer_base;
802 lapic_write32(LAPIC_LVT_TIMER, la->lvt_timer_base);
804 /* Calibrate the timer parameters using BSP. */
805 if (boot && IS_BSP()) {
806 lapic_calibrate_initcount(la);
807 if (lapic_timer_tsc_deadline)
808 lapic_calibrate_deadline(la);
811 /* Setup the timer if configured. */
812 if (la->la_timer_mode != LAT_MODE_UNDEF) {
813 KASSERT(la->la_timer_period != 0, ("lapic%u: zero divisor",
815 switch (la->la_timer_mode) {
816 case LAT_MODE_PERIODIC:
817 lapic_timer_set_divisor(lapic_timer_divisor);
818 lapic_timer_periodic(la);
820 case LAT_MODE_ONESHOT:
821 lapic_timer_set_divisor(lapic_timer_divisor);
822 lapic_timer_oneshot(la);
824 case LAT_MODE_DEADLINE:
825 lapic_timer_deadline(la);
828 panic("corrupted la_timer_mode %p %d", la,
833 /* Program error LVT and clear any existing errors. */
834 lapic_write32(LAPIC_LVT_ERROR, lvt_mode(la, APIC_LVT_ERROR,
835 lapic_read32(LAPIC_LVT_ERROR)));
836 lapic_write32(LAPIC_ESR, 0);
838 /* XXX: Thermal LVT */
840 /* Program the CMCI LVT entry if present. */
841 if (maxlvt >= APIC_LVT_CMCI) {
842 lapic_write32(LAPIC_LVT_CMCI, lvt_mode(la, APIC_LVT_CMCI,
843 lapic_read32(LAPIC_LVT_CMCI)));
846 elvt_count = amd_read_elvt_count();
847 for (i = 0; i < elvt_count; i++) {
848 if (la->la_elvts[i].lvt_active)
849 lapic_write32(LAPIC_EXT_LVT0 + i,
850 elvt_mode(la, i, lapic_read32(LAPIC_EXT_LVT0 + i)));
853 intr_restore(saveintr);
857 native_lapic_intrcnt(void *dummy __unused)
861 char buf[MAXCOMLEN + 1];
863 /* If there are no APICs, skip this function. */
867 STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
868 la = &lapics[pc->pc_apic_id];
872 snprintf(buf, sizeof(buf), "cpu%d:timer", pc->pc_cpuid);
873 intrcnt_add(buf, &la->la_timer_count);
876 SYSINIT(native_lapic_intrcnt, SI_SUB_INTR, SI_ORDER_MIDDLE, native_lapic_intrcnt,
880 native_lapic_reenable_pmc(void)
885 value = lapic_read32(LAPIC_LVT_PCINT);
886 value &= ~APIC_LVT_M;
887 lapic_write32(LAPIC_LVT_PCINT, value);
893 lapic_update_pmc(void *dummy)
897 la = &lapics[lapic_id()];
898 lapic_write32(LAPIC_LVT_PCINT, lvt_mode(la, APIC_LVT_PMC,
899 lapic_read32(LAPIC_LVT_PCINT)));
904 native_lapic_enable_pmc(void)
909 /* Fail if the local APIC is not present. */
910 if (!x2apic_mode && lapic_map == NULL)
913 /* Fail if the PMC LVT is not present. */
914 maxlvt = (lapic_read32(LAPIC_VERSION) & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
915 if (maxlvt < APIC_LVT_PMC)
918 lvts[APIC_LVT_PMC].lvt_masked = 0;
920 #ifdef EARLY_AP_STARTUP
921 MPASS(mp_ncpus == 1 || smp_started);
922 smp_rendezvous(NULL, lapic_update_pmc, NULL, NULL);
926 * If hwpmc was loaded at boot time then the APs may not be
927 * started yet. In that case, don't forward the request to
928 * them as they will program the lvt when they start.
931 smp_rendezvous(NULL, lapic_update_pmc, NULL, NULL);
934 lapic_update_pmc(NULL);
943 native_lapic_disable_pmc(void)
948 /* Fail if the local APIC is not present. */
949 if (!x2apic_mode && lapic_map == NULL)
952 /* Fail if the PMC LVT is not present. */
953 maxlvt = (lapic_read32(LAPIC_VERSION) & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
954 if (maxlvt < APIC_LVT_PMC)
957 lvts[APIC_LVT_PMC].lvt_masked = 1;
960 /* The APs should always be started when hwpmc is unloaded. */
961 KASSERT(mp_ncpus == 1 || smp_started, ("hwpmc unloaded too early"));
963 smp_rendezvous(NULL, lapic_update_pmc, NULL, NULL);
968 lapic_calibrate_initcount(struct lapic *la)
972 /* Start off with a divisor of 2 (power on reset default). */
973 lapic_timer_divisor = 2;
974 /* Try to calibrate the local APIC timer. */
976 lapic_timer_set_divisor(lapic_timer_divisor);
977 lapic_timer_oneshot_nointr(la, APIC_TIMER_MAX_COUNT);
979 value = APIC_TIMER_MAX_COUNT - lapic_read32(LAPIC_CCR_TIMER);
980 if (value != APIC_TIMER_MAX_COUNT)
982 lapic_timer_divisor <<= 1;
983 } while (lapic_timer_divisor <= 128);
984 if (lapic_timer_divisor > 128)
985 panic("lapic: Divisor too big");
987 printf("lapic: Divisor %lu, Frequency %lu Hz\n",
988 lapic_timer_divisor, value);
994 lapic_calibrate_deadline(struct lapic *la __unused)
998 printf("lapic: deadline tsc mode, Frequency %ju Hz\n",
999 (uintmax_t)tsc_freq);
1004 lapic_change_mode(struct eventtimer *et, struct lapic *la,
1005 enum lat_timer_mode newmode)
1008 if (la->la_timer_mode == newmode)
1011 case LAT_MODE_PERIODIC:
1012 lapic_timer_set_divisor(lapic_timer_divisor);
1013 et->et_frequency = count_freq;
1015 case LAT_MODE_DEADLINE:
1016 et->et_frequency = tsc_freq;
1018 case LAT_MODE_ONESHOT:
1019 lapic_timer_set_divisor(lapic_timer_divisor);
1020 et->et_frequency = count_freq;
1023 panic("lapic_change_mode %d", newmode);
1025 la->la_timer_mode = newmode;
1026 et->et_min_period = (0x00000002LLU << 32) / et->et_frequency;
1027 et->et_max_period = (0xfffffffeLLU << 32) / et->et_frequency;
1031 lapic_et_start(struct eventtimer *et, sbintime_t first, sbintime_t period)
1035 la = &lapics[PCPU_GET(apic_id)];
1037 lapic_change_mode(et, la, LAT_MODE_PERIODIC);
1038 la->la_timer_period = ((uint32_t)et->et_frequency * period) >>
1040 lapic_timer_periodic(la);
1041 } else if (lapic_timer_tsc_deadline) {
1042 lapic_change_mode(et, la, LAT_MODE_DEADLINE);
1043 la->la_timer_period = (et->et_frequency * first) >> 32;
1044 lapic_timer_deadline(la);
1046 lapic_change_mode(et, la, LAT_MODE_ONESHOT);
1047 la->la_timer_period = ((uint32_t)et->et_frequency * first) >>
1049 lapic_timer_oneshot(la);
1055 lapic_et_stop(struct eventtimer *et)
1059 la = &lapics[PCPU_GET(apic_id)];
1060 lapic_timer_stop(la);
1061 la->la_timer_mode = LAT_MODE_UNDEF;
1066 native_lapic_disable(void)
1070 /* Software disable the local APIC. */
1071 value = lapic_read32(LAPIC_SVR);
1072 value &= ~APIC_SVR_SWEN;
1073 lapic_write32(LAPIC_SVR, value);
1081 /* Program the spurious vector to enable the local APIC. */
1082 value = lapic_read32(LAPIC_SVR);
1083 value &= ~(APIC_SVR_VECTOR | APIC_SVR_FOCUS);
1084 value |= APIC_SVR_FEN | APIC_SVR_SWEN | APIC_SPURIOUS_INT;
1085 if (lapic_eoi_suppression)
1086 value |= APIC_SVR_EOI_SUPPRESSION;
1087 lapic_write32(LAPIC_SVR, value);
1090 /* Reset the local APIC on the BSP during resume. */
1092 lapic_resume(struct pic *pic, bool suspend_cancelled)
1099 native_lapic_id(void)
1103 KASSERT(x2apic_mode || lapic_map != NULL, ("local APIC is not mapped"));
1104 v = lapic_read32(LAPIC_ID);
1106 v >>= APIC_ID_SHIFT;
1111 native_lapic_intr_pending(u_int vector)
1116 * The IRR registers are an array of registers each of which
1117 * only describes 32 interrupts in the low 32 bits. Thus, we
1118 * divide the vector by 32 to get the register index.
1119 * Finally, we modulus the vector by 32 to determine the
1120 * individual bit to test.
1122 irr = lapic_read32(LAPIC_IRR0 + vector / 32);
1123 return (irr & 1 << (vector % 32));
1127 native_lapic_set_logical_id(u_int apic_id, u_int cluster, u_int cluster_id)
1131 KASSERT(lapics[apic_id].la_present, ("%s: APIC %u doesn't exist",
1132 __func__, apic_id));
1133 KASSERT(cluster <= APIC_MAX_CLUSTER, ("%s: cluster %u too big",
1134 __func__, cluster));
1135 KASSERT(cluster_id <= APIC_MAX_INTRACLUSTER_ID,
1136 ("%s: intra cluster id %u too big", __func__, cluster_id));
1137 la = &lapics[apic_id];
1138 la->la_cluster = cluster;
1139 la->la_cluster_id = cluster_id;
1143 native_lapic_set_lvt_mask(u_int apic_id, u_int pin, u_char masked)
1146 if (pin > APIC_LVT_MAX)
1148 if (apic_id == APIC_ID_ALL) {
1149 lvts[pin].lvt_masked = masked;
1153 KASSERT(lapics[apic_id].la_present,
1154 ("%s: missing APIC %u", __func__, apic_id));
1155 lapics[apic_id].la_lvts[pin].lvt_masked = masked;
1156 lapics[apic_id].la_lvts[pin].lvt_active = 1;
1158 printf("lapic%u:", apic_id);
1161 printf(" LINT%u %s\n", pin, masked ? "masked" : "unmasked");
1166 native_lapic_set_lvt_mode(u_int apic_id, u_int pin, u_int32_t mode)
1170 if (pin > APIC_LVT_MAX)
1172 if (apic_id == APIC_ID_ALL) {
1177 KASSERT(lapics[apic_id].la_present,
1178 ("%s: missing APIC %u", __func__, apic_id));
1179 lvt = &lapics[apic_id].la_lvts[pin];
1180 lvt->lvt_active = 1;
1182 printf("lapic%u:", apic_id);
1184 lvt->lvt_mode = mode;
1186 case APIC_LVT_DM_NMI:
1187 case APIC_LVT_DM_SMI:
1188 case APIC_LVT_DM_INIT:
1189 case APIC_LVT_DM_EXTINT:
1190 lvt->lvt_edgetrigger = 1;
1191 lvt->lvt_activehi = 1;
1192 if (mode == APIC_LVT_DM_EXTINT)
1193 lvt->lvt_masked = 1;
1195 lvt->lvt_masked = 0;
1198 panic("Unsupported delivery mode: 0x%x\n", mode);
1201 printf(" Routing ");
1203 case APIC_LVT_DM_NMI:
1206 case APIC_LVT_DM_SMI:
1209 case APIC_LVT_DM_INIT:
1212 case APIC_LVT_DM_EXTINT:
1216 printf(" -> LINT%u\n", pin);
1222 native_lapic_set_lvt_polarity(u_int apic_id, u_int pin, enum intr_polarity pol)
1225 if (pin > APIC_LVT_MAX || pol == INTR_POLARITY_CONFORM)
1227 if (apic_id == APIC_ID_ALL) {
1228 lvts[pin].lvt_activehi = (pol == INTR_POLARITY_HIGH);
1232 KASSERT(lapics[apic_id].la_present,
1233 ("%s: missing APIC %u", __func__, apic_id));
1234 lapics[apic_id].la_lvts[pin].lvt_active = 1;
1235 lapics[apic_id].la_lvts[pin].lvt_activehi =
1236 (pol == INTR_POLARITY_HIGH);
1238 printf("lapic%u:", apic_id);
1241 printf(" LINT%u polarity: %s\n", pin,
1242 pol == INTR_POLARITY_HIGH ? "high" : "low");
1247 native_lapic_set_lvt_triggermode(u_int apic_id, u_int pin,
1248 enum intr_trigger trigger)
1251 if (pin > APIC_LVT_MAX || trigger == INTR_TRIGGER_CONFORM)
1253 if (apic_id == APIC_ID_ALL) {
1254 lvts[pin].lvt_edgetrigger = (trigger == INTR_TRIGGER_EDGE);
1258 KASSERT(lapics[apic_id].la_present,
1259 ("%s: missing APIC %u", __func__, apic_id));
1260 lapics[apic_id].la_lvts[pin].lvt_edgetrigger =
1261 (trigger == INTR_TRIGGER_EDGE);
1262 lapics[apic_id].la_lvts[pin].lvt_active = 1;
1264 printf("lapic%u:", apic_id);
1267 printf(" LINT%u trigger: %s\n", pin,
1268 trigger == INTR_TRIGGER_EDGE ? "edge" : "level");
1273 * Adjust the TPR of the current CPU so that it blocks all interrupts below
1274 * the passed in vector.
1277 lapic_set_tpr(u_int vector)
1280 lapic_write32(LAPIC_TPR, vector);
1284 tpr = lapic_read32(LAPIC_TPR) & ~APIC_TPR_PRIO;
1286 lapic_write32(LAPIC_TPR, tpr);
1291 native_lapic_eoi(void)
1294 lapic_write32_nofence(LAPIC_EOI, 0);
1298 lapic_handle_intr(int vector, struct trapframe *frame)
1300 struct intsrc *isrc;
1302 isrc = intr_lookup_source(apic_idt_to_irq(PCPU_GET(apic_id),
1304 intr_execute_handlers(isrc, frame);
1308 lapic_handle_timer(struct trapframe *frame)
1311 struct trapframe *oldframe;
1314 /* Send EOI first thing. */
1317 #if defined(SMP) && !defined(SCHED_ULE)
1319 * Don't do any accounting for the disabled HTT cores, since it
1320 * will provide misleading numbers for the userland.
1322 * No locking is necessary here, since even if we lose the race
1323 * when hlt_cpus_mask changes it is not a big deal, really.
1325 * Don't do that for ULE, since ULE doesn't consider hlt_cpus_mask
1326 * and unlike other schedulers it actually schedules threads to
1329 if (CPU_ISSET(PCPU_GET(cpuid), &hlt_cpus_mask))
1333 /* Look up our local APIC structure for the tick counters. */
1334 la = &lapics[PCPU_GET(apic_id)];
1335 (*la->la_timer_count)++;
1337 if (lapic_et.et_active) {
1339 td->td_intr_nesting_level++;
1340 oldframe = td->td_intr_frame;
1341 td->td_intr_frame = frame;
1342 lapic_et.et_event_cb(&lapic_et, lapic_et.et_arg);
1343 td->td_intr_frame = oldframe;
1344 td->td_intr_nesting_level--;
1350 lapic_timer_set_divisor(u_int divisor)
1353 KASSERT(powerof2(divisor), ("lapic: invalid divisor %u", divisor));
1354 KASSERT(ffs(divisor) <= nitems(lapic_timer_divisors),
1355 ("lapic: invalid divisor %u", divisor));
1356 lapic_write32(LAPIC_DCR_TIMER, lapic_timer_divisors[ffs(divisor) - 1]);
1360 lapic_timer_oneshot(struct lapic *la)
1364 value = la->lvt_timer_base;
1365 value &= ~(APIC_LVTT_TM | APIC_LVT_M);
1366 value |= APIC_LVTT_TM_ONE_SHOT;
1367 la->lvt_timer_last = value;
1368 lapic_write32(LAPIC_LVT_TIMER, value);
1369 lapic_write32(LAPIC_ICR_TIMER, la->la_timer_period);
1373 lapic_timer_oneshot_nointr(struct lapic *la, uint32_t count)
1377 value = la->lvt_timer_base;
1378 value &= ~APIC_LVTT_TM;
1379 value |= APIC_LVTT_TM_ONE_SHOT | APIC_LVT_M;
1380 la->lvt_timer_last = value;
1381 lapic_write32(LAPIC_LVT_TIMER, value);
1382 lapic_write32(LAPIC_ICR_TIMER, count);
1386 lapic_timer_periodic(struct lapic *la)
1390 value = la->lvt_timer_base;
1391 value &= ~(APIC_LVTT_TM | APIC_LVT_M);
1392 value |= APIC_LVTT_TM_PERIODIC;
1393 la->lvt_timer_last = value;
1394 lapic_write32(LAPIC_LVT_TIMER, value);
1395 lapic_write32(LAPIC_ICR_TIMER, la->la_timer_period);
1399 lapic_timer_deadline(struct lapic *la)
1403 value = la->lvt_timer_base;
1404 value &= ~(APIC_LVTT_TM | APIC_LVT_M);
1405 value |= APIC_LVTT_TM_TSCDLT;
1406 if (value != la->lvt_timer_last) {
1407 la->lvt_timer_last = value;
1408 lapic_write32_nofence(LAPIC_LVT_TIMER, value);
1412 wrmsr(MSR_TSC_DEADLINE, la->la_timer_period + rdtsc());
1416 lapic_timer_stop(struct lapic *la)
1420 if (la->la_timer_mode == LAT_MODE_DEADLINE) {
1421 wrmsr(MSR_TSC_DEADLINE, 0);
1424 value = la->lvt_timer_base;
1425 value &= ~APIC_LVTT_TM;
1426 value |= APIC_LVT_M;
1427 la->lvt_timer_last = value;
1428 lapic_write32(LAPIC_LVT_TIMER, value);
1433 lapic_handle_cmc(void)
1441 * Called from the mca_init() to activate the CMC interrupt if this CPU is
1442 * responsible for monitoring any MC banks for CMC events. Since mca_init()
1443 * is called prior to lapic_setup() during boot, this just needs to unmask
1444 * this CPU's LVT_CMCI entry.
1447 native_lapic_enable_cmc(void)
1452 if (!x2apic_mode && lapic_map == NULL)
1455 apic_id = PCPU_GET(apic_id);
1456 KASSERT(lapics[apic_id].la_present,
1457 ("%s: missing APIC %u", __func__, apic_id));
1458 lapics[apic_id].la_lvts[APIC_LVT_CMCI].lvt_masked = 0;
1459 lapics[apic_id].la_lvts[APIC_LVT_CMCI].lvt_active = 1;
1461 printf("lapic%u: CMCI unmasked\n", apic_id);
1465 native_lapic_enable_mca_elvt(void)
1472 if (lapic_map == NULL)
1476 apic_id = PCPU_GET(apic_id);
1477 KASSERT(lapics[apic_id].la_present,
1478 ("%s: missing APIC %u", __func__, apic_id));
1479 elvt_count = amd_read_elvt_count();
1480 if (elvt_count <= APIC_ELVT_MCA)
1483 value = lapic_read32(LAPIC_EXT_LVT0 + APIC_ELVT_MCA);
1484 if ((value & APIC_LVT_M) == 0) {
1486 printf("AMD MCE Thresholding Extended LVT is already active\n");
1487 return (APIC_ELVT_MCA);
1489 lapics[apic_id].la_elvts[APIC_ELVT_MCA].lvt_masked = 0;
1490 lapics[apic_id].la_elvts[APIC_ELVT_MCA].lvt_active = 1;
1492 printf("lapic%u: MCE Thresholding ELVT unmasked\n", apic_id);
1493 return (APIC_ELVT_MCA);
1497 lapic_handle_error(void)
1502 * Read the contents of the error status register. Write to
1503 * the register first before reading from it to force the APIC
1504 * to update its value to indicate any errors that have
1505 * occurred since the previous write to the register.
1507 lapic_write32(LAPIC_ESR, 0);
1508 esr = lapic_read32(LAPIC_ESR);
1510 printf("CPU%d: local APIC error 0x%x\n", PCPU_GET(cpuid), esr);
1515 native_apic_cpuid(u_int apic_id)
1518 return apic_cpuids[apic_id];
1524 /* Request a free IDT vector to be used by the specified IRQ. */
1526 native_apic_alloc_vector(u_int apic_id, u_int irq)
1530 KASSERT(irq < num_io_irqs, ("Invalid IRQ %u", irq));
1533 * Search for a free vector. Currently we just use a very simple
1534 * algorithm to find the first free vector.
1536 mtx_lock_spin(&icu_lock);
1537 for (vector = 0; vector < APIC_NUM_IOINTS; vector++) {
1538 if (lapics[apic_id].la_ioint_irqs[vector] != IRQ_FREE)
1540 lapics[apic_id].la_ioint_irqs[vector] = irq;
1541 mtx_unlock_spin(&icu_lock);
1542 return (vector + APIC_IO_INTS);
1544 mtx_unlock_spin(&icu_lock);
1549 * Request 'count' free contiguous IDT vectors to be used by 'count'
1550 * IRQs. 'count' must be a power of two and the vectors will be
1551 * aligned on a boundary of 'align'. If the request cannot be
1552 * satisfied, 0 is returned.
1555 native_apic_alloc_vectors(u_int apic_id, u_int *irqs, u_int count, u_int align)
1557 u_int first, run, vector;
1559 KASSERT(powerof2(count), ("bad count"));
1560 KASSERT(powerof2(align), ("bad align"));
1561 KASSERT(align >= count, ("align < count"));
1563 for (run = 0; run < count; run++)
1564 KASSERT(irqs[run] < num_io_irqs, ("Invalid IRQ %u at index %u",
1569 * Search for 'count' free vectors. As with apic_alloc_vector(),
1570 * this just uses a simple first fit algorithm.
1574 mtx_lock_spin(&icu_lock);
1575 for (vector = 0; vector < APIC_NUM_IOINTS; vector++) {
1576 /* Vector is in use, end run. */
1577 if (lapics[apic_id].la_ioint_irqs[vector] != IRQ_FREE) {
1583 /* Start a new run if run == 0 and vector is aligned. */
1585 if ((vector & (align - 1)) != 0)
1591 /* Keep looping if the run isn't long enough yet. */
1595 /* Found a run, assign IRQs and return the first vector. */
1596 for (vector = 0; vector < count; vector++)
1597 lapics[apic_id].la_ioint_irqs[first + vector] =
1599 mtx_unlock_spin(&icu_lock);
1600 return (first + APIC_IO_INTS);
1602 mtx_unlock_spin(&icu_lock);
1603 printf("APIC: Couldn't find APIC vectors for %u IRQs\n", count);
1608 * Enable a vector for a particular apic_id. Since all lapics share idt
1609 * entries and ioint_handlers this enables the vector on all lapics. lapics
1610 * which do not have the vector configured would report spurious interrupts
1614 native_apic_enable_vector(u_int apic_id, u_int vector)
1617 KASSERT(vector != IDT_SYSCALL, ("Attempt to overwrite syscall entry"));
1618 KASSERT(ioint_handlers[vector / 32] != NULL,
1619 ("No ISR handler for vector %u", vector));
1620 #ifdef KDTRACE_HOOKS
1621 KASSERT(vector != IDT_DTRACE_RET,
1622 ("Attempt to overwrite DTrace entry"));
1624 setidt(vector, (pti ? ioint_pti_handlers : ioint_handlers)[vector / 32],
1625 SDT_APIC, SEL_KPL, GSEL_APIC);
1629 native_apic_disable_vector(u_int apic_id, u_int vector)
1632 KASSERT(vector != IDT_SYSCALL, ("Attempt to overwrite syscall entry"));
1633 #ifdef KDTRACE_HOOKS
1634 KASSERT(vector != IDT_DTRACE_RET,
1635 ("Attempt to overwrite DTrace entry"));
1637 KASSERT(ioint_handlers[vector / 32] != NULL,
1638 ("No ISR handler for vector %u", vector));
1641 * We can not currently clear the idt entry because other cpus
1642 * may have a valid vector at this offset.
1644 setidt(vector, pti ? &IDTVEC(rsvd_pti) : &IDTVEC(rsvd), SDT_APIC,
1645 SEL_KPL, GSEL_APIC);
1649 /* Release an APIC vector when it's no longer in use. */
1651 native_apic_free_vector(u_int apic_id, u_int vector, u_int irq)
1655 KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
1656 vector <= APIC_IO_INTS + APIC_NUM_IOINTS,
1657 ("Vector %u does not map to an IRQ line", vector));
1658 KASSERT(irq < num_io_irqs, ("Invalid IRQ %u", irq));
1659 KASSERT(lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS] ==
1660 irq, ("IRQ mismatch"));
1661 #ifdef KDTRACE_HOOKS
1662 KASSERT(vector != IDT_DTRACE_RET,
1663 ("Attempt to overwrite DTrace entry"));
1667 * Bind us to the cpu that owned the vector before freeing it so
1668 * we don't lose an interrupt delivery race.
1673 if (sched_is_bound(td))
1674 panic("apic_free_vector: Thread already bound.\n");
1675 sched_bind(td, apic_cpuid(apic_id));
1678 mtx_lock_spin(&icu_lock);
1679 lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS] = IRQ_FREE;
1680 mtx_unlock_spin(&icu_lock);
1688 /* Map an IDT vector (APIC) to an IRQ (interrupt source). */
1690 apic_idt_to_irq(u_int apic_id, u_int vector)
1694 KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
1695 vector <= APIC_IO_INTS + APIC_NUM_IOINTS,
1696 ("Vector %u does not map to an IRQ line", vector));
1697 #ifdef KDTRACE_HOOKS
1698 KASSERT(vector != IDT_DTRACE_RET,
1699 ("Attempt to overwrite DTrace entry"));
1701 irq = lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS];
1709 * Dump data about APIC IDT vector mappings.
1711 DB_SHOW_COMMAND(apic, db_show_apic)
1713 struct intsrc *isrc;
1718 if (strcmp(modif, "vv") == 0)
1720 else if (strcmp(modif, "v") == 0)
1724 for (apic_id = 0; apic_id <= max_apic_id; apic_id++) {
1725 if (lapics[apic_id].la_present == 0)
1727 db_printf("Interrupts bound to lapic %u\n", apic_id);
1728 for (i = 0; i < APIC_NUM_IOINTS + 1 && !db_pager_quit; i++) {
1729 irq = lapics[apic_id].la_ioint_irqs[i];
1730 if (irq == IRQ_FREE || irq == IRQ_SYSCALL)
1732 #ifdef KDTRACE_HOOKS
1733 if (irq == IRQ_DTRACE_RET)
1737 if (irq == IRQ_EVTCHN)
1740 db_printf("vec 0x%2x -> ", i + APIC_IO_INTS);
1741 if (irq == IRQ_TIMER)
1742 db_printf("lapic timer\n");
1743 else if (irq < num_io_irqs) {
1744 isrc = intr_lookup_source(irq);
1745 if (isrc == NULL || verbose == 0)
1746 db_printf("IRQ %u\n", irq);
1748 db_dump_intr_event(isrc->is_event,
1751 db_printf("IRQ %u ???\n", irq);
1757 dump_mask(const char *prefix, uint32_t v, int base)
1762 for (i = 0; i < 32; i++)
1765 db_printf("%s:", prefix);
1768 db_printf(" %02x", base + i);
1774 /* Show info from the lapic regs for this CPU. */
1775 DB_SHOW_COMMAND(lapic, db_show_lapic)
1779 db_printf("lapic ID = %d\n", lapic_id());
1780 v = lapic_read32(LAPIC_VERSION);
1781 db_printf("version = %d.%d\n", (v & APIC_VER_VERSION) >> 4,
1783 db_printf("max LVT = %d\n", (v & APIC_VER_MAXLVT) >> MAXLVTSHIFT);
1784 v = lapic_read32(LAPIC_SVR);
1785 db_printf("SVR = %02x (%s)\n", v & APIC_SVR_VECTOR,
1786 v & APIC_SVR_ENABLE ? "enabled" : "disabled");
1787 db_printf("TPR = %02x\n", lapic_read32(LAPIC_TPR));
1789 #define dump_field(prefix, regn, index) \
1790 dump_mask(__XSTRING(prefix ## index), \
1791 lapic_read32(LAPIC_ ## regn ## index), \
1794 db_printf("In-service Interrupts:\n");
1795 dump_field(isr, ISR, 0);
1796 dump_field(isr, ISR, 1);
1797 dump_field(isr, ISR, 2);
1798 dump_field(isr, ISR, 3);
1799 dump_field(isr, ISR, 4);
1800 dump_field(isr, ISR, 5);
1801 dump_field(isr, ISR, 6);
1802 dump_field(isr, ISR, 7);
1804 db_printf("TMR Interrupts:\n");
1805 dump_field(tmr, TMR, 0);
1806 dump_field(tmr, TMR, 1);
1807 dump_field(tmr, TMR, 2);
1808 dump_field(tmr, TMR, 3);
1809 dump_field(tmr, TMR, 4);
1810 dump_field(tmr, TMR, 5);
1811 dump_field(tmr, TMR, 6);
1812 dump_field(tmr, TMR, 7);
1814 db_printf("IRR Interrupts:\n");
1815 dump_field(irr, IRR, 0);
1816 dump_field(irr, IRR, 1);
1817 dump_field(irr, IRR, 2);
1818 dump_field(irr, IRR, 3);
1819 dump_field(irr, IRR, 4);
1820 dump_field(irr, IRR, 5);
1821 dump_field(irr, IRR, 6);
1822 dump_field(irr, IRR, 7);
1829 * APIC probing support code. This includes code to manage enumerators.
1832 static SLIST_HEAD(, apic_enumerator) enumerators =
1833 SLIST_HEAD_INITIALIZER(enumerators);
1834 static struct apic_enumerator *best_enum;
1837 apic_register_enumerator(struct apic_enumerator *enumerator)
1840 struct apic_enumerator *apic_enum;
1842 SLIST_FOREACH(apic_enum, &enumerators, apic_next) {
1843 if (apic_enum == enumerator)
1844 panic("%s: Duplicate register of %s", __func__,
1845 enumerator->apic_name);
1848 SLIST_INSERT_HEAD(&enumerators, enumerator, apic_next);
1852 * We have to look for CPU's very, very early because certain subsystems
1853 * want to know how many CPU's we have extremely early on in the boot
1857 apic_init(void *dummy __unused)
1859 struct apic_enumerator *enumerator;
1862 /* We only support built in local APICs. */
1863 if (!(cpu_feature & CPUID_APIC))
1866 /* Don't probe if APIC mode is disabled. */
1867 if (resource_disabled("apic", 0))
1870 /* Probe all the enumerators to find the best match. */
1873 SLIST_FOREACH(enumerator, &enumerators, apic_next) {
1874 retval = enumerator->apic_probe();
1877 if (best_enum == NULL || best < retval) {
1878 best_enum = enumerator;
1882 if (best_enum == NULL) {
1884 printf("APIC: Could not find any APICs.\n");
1886 panic("running without device atpic requires a local APIC");
1892 printf("APIC: Using the %s enumerator.\n",
1893 best_enum->apic_name);
1897 * To work around an errata, we disable the local APIC on some
1898 * CPUs during early startup. We need to turn the local APIC back
1899 * on on such CPUs now.
1901 ppro_reenable_apic();
1904 /* Probe the CPU's in the system. */
1905 retval = best_enum->apic_probe_cpus();
1907 printf("%s: Failed to probe CPUs: returned %d\n",
1908 best_enum->apic_name, retval);
1911 SYSINIT(apic_init, SI_SUB_TUNABLES - 1, SI_ORDER_SECOND, apic_init, NULL);
1914 * Setup the local APIC. We have to do this prior to starting up the APs
1918 apic_setup_local(void *dummy __unused)
1922 if (best_enum == NULL)
1925 lapics = malloc(sizeof(*lapics) * (max_apic_id + 1), M_LAPIC,
1928 /* Initialize the local APIC. */
1929 retval = best_enum->apic_setup_local();
1931 printf("%s: Failed to setup the local APIC: returned %d\n",
1932 best_enum->apic_name, retval);
1934 SYSINIT(apic_setup_local, SI_SUB_CPU, SI_ORDER_SECOND, apic_setup_local, NULL);
1937 * Setup the I/O APICs.
1940 apic_setup_io(void *dummy __unused)
1944 if (best_enum == NULL)
1948 * Local APIC must be registered before other PICs and pseudo PICs
1949 * for proper suspend/resume order.
1951 intr_register_pic(&lapic_pic);
1953 retval = best_enum->apic_setup_io();
1955 printf("%s: Failed to setup I/O APICs: returned %d\n",
1956 best_enum->apic_name, retval);
1959 * Finish setting up the local APIC on the BSP once we know
1960 * how to properly program the LINT pins. In particular, this
1961 * enables the EOI suppression mode, if LAPIC supports it and
1962 * user did not disable the mode.
1968 /* Enable the MSI "pic". */
1969 init_ops.msi_init();
1972 xen_intr_alloc_irqs();
1975 SYSINIT(apic_setup_io, SI_SUB_INTR, SI_ORDER_THIRD, apic_setup_io, NULL);
1979 * Inter Processor Interrupt functions. The lapic_ipi_*() functions are
1980 * private to the MD code. The public interface for the rest of the
1981 * kernel is defined in mp_machdep.c.
1985 * Wait delay microseconds for IPI to be sent. If delay is -1, we
1989 native_lapic_ipi_wait(int delay)
1993 /* LAPIC_ICR.APIC_DELSTAT_MASK is undefined in x2APIC mode */
1997 for (rx = 0; delay == -1 || rx < lapic_ipi_wait_mult * delay; rx++) {
1998 if ((lapic_read_icr_lo() & APIC_DELSTAT_MASK) ==
2007 native_lapic_ipi_raw(register_t icrlo, u_int dest)
2011 /* XXX: Need more sanity checking of icrlo? */
2012 KASSERT(x2apic_mode || lapic_map != NULL,
2013 ("%s called too early", __func__));
2014 KASSERT(x2apic_mode ||
2015 (dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
2016 ("%s: invalid dest field", __func__));
2017 KASSERT((icrlo & APIC_ICRLO_RESV_MASK) == 0,
2018 ("%s: reserved bits set in ICR LO register", __func__));
2020 if ((icrlo & APIC_DEST_MASK) == APIC_DEST_DESTFLD) {
2024 icrhi = dest << APIC_ID_SHIFT;
2025 lapic_write_icr(icrhi, icrlo);
2027 lapic_write_icr_lo(icrlo);
2031 #define BEFORE_SPIN 50000
2032 #ifdef DETECT_DEADLOCK
2033 #define AFTER_SPIN 50
2037 native_lapic_ipi_vectored(u_int vector, int dest)
2039 register_t icrlo, destfield;
2041 KASSERT((vector & ~APIC_VECTOR_MASK) == 0,
2042 ("%s: invalid vector %d", __func__, vector));
2046 case APIC_IPI_DEST_SELF:
2047 if (x2apic_mode && vector < IPI_NMI_FIRST) {
2048 lapic_write_self_ipi(vector);
2051 icrlo = APIC_DEST_SELF;
2053 case APIC_IPI_DEST_ALL:
2054 icrlo = APIC_DEST_ALLISELF;
2056 case APIC_IPI_DEST_OTHERS:
2057 icrlo = APIC_DEST_ALLESELF;
2061 KASSERT(x2apic_mode ||
2062 (dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
2063 ("%s: invalid destination 0x%x", __func__, dest));
2068 * NMI IPIs are just fake vectors used to send a NMI. Use special rules
2069 * regarding NMIs if passed, otherwise specify the vector.
2071 if (vector >= IPI_NMI_FIRST)
2072 icrlo |= APIC_DELMODE_NMI;
2074 icrlo |= vector | APIC_DELMODE_FIXED;
2075 icrlo |= APIC_DESTMODE_PHY | APIC_TRIGMOD_EDGE | APIC_LEVEL_ASSERT;
2077 /* Wait for an earlier IPI to finish. */
2078 if (!lapic_ipi_wait(BEFORE_SPIN)) {
2079 if (KERNEL_PANICKED())
2082 panic("APIC: Previous IPI is stuck");
2085 lapic_ipi_raw(icrlo, destfield);
2087 #ifdef DETECT_DEADLOCK
2088 /* Wait for IPI to be delivered. */
2089 if (!lapic_ipi_wait(AFTER_SPIN)) {
2090 #ifdef needsattention
2094 * The above function waits for the message to actually be
2095 * delivered. It breaks out after an arbitrary timeout
2096 * since the message should eventually be delivered (at
2097 * least in theory) and that if it wasn't we would catch
2098 * the failure with the check above when the next IPI is
2101 * We could skip this wait entirely, EXCEPT it probably
2102 * protects us from other routines that assume that the
2103 * message was delivered and acted upon when this function
2106 printf("APIC: IPI might be stuck\n");
2107 #else /* !needsattention */
2108 /* Wait until mesage is sent without a timeout. */
2109 while (lapic_read_icr_lo() & APIC_DELSTAT_PEND)
2111 #endif /* needsattention */
2113 #endif /* DETECT_DEADLOCK */
2119 * Since the IDT is shared by all CPUs the IPI slot update needs to be globally
2122 * Consider the case where an IPI is generated immediately after allocation:
2123 * vector = lapic_ipi_alloc(ipifunc);
2124 * ipi_selected(other_cpus, vector);
2126 * In xAPIC mode a write to ICR_LO has serializing semantics because the
2127 * APIC page is mapped as an uncached region. In x2APIC mode there is an
2128 * explicit 'mfence' before the ICR MSR is written. Therefore in both cases
2129 * the IDT slot update is globally visible before the IPI is delivered.
2132 native_lapic_ipi_alloc(inthand_t *ipifunc)
2134 struct gate_descriptor *ip;
2138 KASSERT(ipifunc != &IDTVEC(rsvd) && ipifunc != &IDTVEC(rsvd_pti),
2139 ("invalid ipifunc %p", ipifunc));
2142 mtx_lock_spin(&icu_lock);
2143 for (idx = IPI_DYN_FIRST; idx <= IPI_DYN_LAST; idx++) {
2145 func = (ip->gd_hioffset << 16) | ip->gd_looffset;
2146 if ((!pti && func == (uintptr_t)&IDTVEC(rsvd)) ||
2147 (pti && func == (uintptr_t)&IDTVEC(rsvd_pti))) {
2149 setidt(vector, ipifunc, SDT_APIC, SEL_KPL, GSEL_APIC);
2153 mtx_unlock_spin(&icu_lock);
2158 native_lapic_ipi_free(int vector)
2160 struct gate_descriptor *ip;
2163 KASSERT(vector >= IPI_DYN_FIRST && vector <= IPI_DYN_LAST,
2164 ("%s: invalid vector %d", __func__, vector));
2166 mtx_lock_spin(&icu_lock);
2168 func = (ip->gd_hioffset << 16) | ip->gd_looffset;
2169 KASSERT(func != (uintptr_t)&IDTVEC(rsvd) &&
2170 func != (uintptr_t)&IDTVEC(rsvd_pti),
2171 ("invalid idtfunc %#lx", func));
2172 setidt(vector, pti ? &IDTVEC(rsvd_pti) : &IDTVEC(rsvd), SDT_APIC,
2173 SEL_KPL, GSEL_APIC);
2174 mtx_unlock_spin(&icu_lock);