3 * Local APIC virtualization
5 * Copyright (C) 2006 Qumranet, Inc.
6 * Copyright (C) 2007 Novell
7 * Copyright (C) 2007 Intel
8 * Copyright 2009 Red Hat, Inc. and/or its affiliates.
11 * Dor Laor <dor.laor@qumranet.com>
12 * Gregory Haskins <ghaskins@novell.com>
13 * Yaozu (Eddie) Dong <eddie.dong@intel.com>
15 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
24 #include <linux/highmem.h>
25 #include <linux/smp.h>
26 #include <linux/hrtimer.h>
28 #include <linux/module.h>
29 #include <linux/math64.h>
30 #include <linux/slab.h>
31 #include <asm/processor.h>
34 #include <asm/current.h>
35 #include <asm/apicdef.h>
36 #include <asm/delay.h>
37 #include <linux/atomic.h>
38 #include <linux/jump_label.h>
39 #include "kvm_cache_regs.h"
46 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
48 #define mod_64(x, y) ((x) % (y))
56 #define APIC_BUS_CYCLE_NS 1
58 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
59 #define apic_debug(fmt, arg...) do {} while (0)
61 #define APIC_LVT_NUM 6
62 /* 14 is the version for Xeon and Pentium 8.4.8*/
63 #define APIC_VERSION (0x14UL | ((APIC_LVT_NUM - 1) << 16))
64 #define LAPIC_MMIO_LENGTH (1 << 12)
65 /* followed define is not in apicdef.h */
66 #define APIC_SHORT_MASK 0xc0000
67 #define APIC_DEST_NOSHORT 0x0
68 #define APIC_DEST_MASK 0x800
69 #define MAX_APIC_VECTOR 256
70 #define APIC_VECTORS_PER_REG 32
72 #define APIC_BROADCAST 0xFF
73 #define X2APIC_BROADCAST 0xFFFFFFFFul
75 #define VEC_POS(v) ((v) & (32 - 1))
76 #define REG_POS(v) (((v) >> 5) << 4)
78 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
80 *((u32 *) (apic->regs + reg_off)) = val;
83 static inline int apic_test_vector(int vec, void *bitmap)
85 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
88 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
90 struct kvm_lapic *apic = vcpu->arch.apic;
92 return apic_test_vector(vector, apic->regs + APIC_ISR) ||
93 apic_test_vector(vector, apic->regs + APIC_IRR);
96 static inline void apic_set_vector(int vec, void *bitmap)
98 set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
101 static inline void apic_clear_vector(int vec, void *bitmap)
103 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
106 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
108 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
111 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
113 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
116 struct static_key_deferred apic_hw_disabled __read_mostly;
117 struct static_key_deferred apic_sw_disabled __read_mostly;
119 static inline int apic_enabled(struct kvm_lapic *apic)
121 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
125 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
128 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
129 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
131 static inline int kvm_apic_id(struct kvm_lapic *apic)
133 return (kvm_apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
136 /* The logical map is definitely wrong if we have multiple
137 * modes at the same time. (Physical map is always right.)
139 static inline bool kvm_apic_logical_map_valid(struct kvm_apic_map *map)
141 return !(map->mode & (map->mode - 1));
145 apic_logical_id(struct kvm_apic_map *map, u32 dest_id, u16 *cid, u16 *lid)
149 BUILD_BUG_ON(KVM_APIC_MODE_XAPIC_CLUSTER != 4);
150 BUILD_BUG_ON(KVM_APIC_MODE_XAPIC_FLAT != 8);
151 BUILD_BUG_ON(KVM_APIC_MODE_X2APIC != 16);
152 lid_bits = map->mode;
154 *cid = dest_id >> lid_bits;
155 *lid = dest_id & ((1 << lid_bits) - 1);
158 static void recalculate_apic_map(struct kvm *kvm)
160 struct kvm_apic_map *new, *old = NULL;
161 struct kvm_vcpu *vcpu;
164 new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
166 mutex_lock(&kvm->arch.apic_map_lock);
171 kvm_for_each_vcpu(i, vcpu, kvm) {
172 struct kvm_lapic *apic = vcpu->arch.apic;
176 if (!kvm_apic_present(vcpu))
179 aid = kvm_apic_id(apic);
180 ldr = kvm_apic_get_reg(apic, APIC_LDR);
182 if (aid < ARRAY_SIZE(new->phys_map))
183 new->phys_map[aid] = apic;
185 if (apic_x2apic_mode(apic)) {
186 new->mode |= KVM_APIC_MODE_X2APIC;
188 ldr = GET_APIC_LOGICAL_ID(ldr);
189 if (kvm_apic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
190 new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
192 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
195 if (!kvm_apic_logical_map_valid(new))
198 apic_logical_id(new, ldr, &cid, &lid);
200 if (lid && cid < ARRAY_SIZE(new->logical_map))
201 new->logical_map[cid][ffs(lid) - 1] = apic;
204 old = rcu_dereference_protected(kvm->arch.apic_map,
205 lockdep_is_held(&kvm->arch.apic_map_lock));
206 rcu_assign_pointer(kvm->arch.apic_map, new);
207 mutex_unlock(&kvm->arch.apic_map_lock);
212 kvm_make_scan_ioapic_request(kvm);
215 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
217 bool enabled = val & APIC_SPIV_APIC_ENABLED;
219 apic_set_reg(apic, APIC_SPIV, val);
221 if (enabled != apic->sw_enabled) {
222 apic->sw_enabled = enabled;
224 static_key_slow_dec_deferred(&apic_sw_disabled);
225 recalculate_apic_map(apic->vcpu->kvm);
227 static_key_slow_inc(&apic_sw_disabled.key);
231 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
233 apic_set_reg(apic, APIC_ID, id << 24);
234 recalculate_apic_map(apic->vcpu->kvm);
237 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
239 apic_set_reg(apic, APIC_LDR, id);
240 recalculate_apic_map(apic->vcpu->kvm);
243 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u8 id)
245 u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
247 apic_set_reg(apic, APIC_ID, id << 24);
248 apic_set_reg(apic, APIC_LDR, ldr);
249 recalculate_apic_map(apic->vcpu->kvm);
252 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
254 return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
257 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
259 return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
262 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
264 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
267 static inline int apic_lvtt_period(struct kvm_lapic *apic)
269 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
272 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
274 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
277 static inline int apic_lvt_nmi_mode(u32 lvt_val)
279 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
282 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
284 struct kvm_lapic *apic = vcpu->arch.apic;
285 struct kvm_cpuid_entry2 *feat;
286 u32 v = APIC_VERSION;
288 if (!kvm_vcpu_has_lapic(vcpu))
292 * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
293 * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
294 * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
295 * version first and level-triggered interrupts never get EOIed in
298 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
299 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))) &&
300 !ioapic_in_kernel(vcpu->kvm))
301 v |= APIC_LVR_DIRECTED_EOI;
302 apic_set_reg(apic, APIC_LVR, v);
305 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
306 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
307 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
308 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
309 LINT_MASK, LINT_MASK, /* LVT0-1 */
310 LVT_MASK /* LVTERR */
313 static int find_highest_vector(void *bitmap)
318 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
319 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
320 reg = bitmap + REG_POS(vec);
322 return fls(*reg) - 1 + vec;
328 static u8 count_vectors(void *bitmap)
334 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
335 reg = bitmap + REG_POS(vec);
336 count += hweight32(*reg);
342 void __kvm_apic_update_irr(u32 *pir, void *regs)
346 for (i = 0; i <= 7; i++) {
347 pir_val = xchg(&pir[i], 0);
349 *((u32 *)(regs + APIC_IRR + i * 0x10)) |= pir_val;
352 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
354 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
356 struct kvm_lapic *apic = vcpu->arch.apic;
358 __kvm_apic_update_irr(pir, apic->regs);
360 kvm_make_request(KVM_REQ_EVENT, vcpu);
362 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
364 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
366 apic_set_vector(vec, apic->regs + APIC_IRR);
368 * irr_pending must be true if any interrupt is pending; set it after
369 * APIC_IRR to avoid race with apic_clear_irr
371 apic->irr_pending = true;
374 static inline int apic_search_irr(struct kvm_lapic *apic)
376 return find_highest_vector(apic->regs + APIC_IRR);
379 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
384 * Note that irr_pending is just a hint. It will be always
385 * true with virtual interrupt delivery enabled.
387 if (!apic->irr_pending)
390 kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
391 result = apic_search_irr(apic);
392 ASSERT(result == -1 || result >= 16);
397 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
399 struct kvm_vcpu *vcpu;
403 if (unlikely(kvm_vcpu_apic_vid_enabled(vcpu))) {
404 /* try to update RVI */
405 apic_clear_vector(vec, apic->regs + APIC_IRR);
406 kvm_make_request(KVM_REQ_EVENT, vcpu);
408 apic->irr_pending = false;
409 apic_clear_vector(vec, apic->regs + APIC_IRR);
410 if (apic_search_irr(apic) != -1)
411 apic->irr_pending = true;
415 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
417 struct kvm_vcpu *vcpu;
419 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
425 * With APIC virtualization enabled, all caching is disabled
426 * because the processor can modify ISR under the hood. Instead
429 if (unlikely(kvm_x86_ops->hwapic_isr_update))
430 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
433 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
435 * ISR (in service register) bit is set when injecting an interrupt.
436 * The highest vector is injected. Thus the latest bit set matches
437 * the highest bit in ISR.
439 apic->highest_isr_cache = vec;
443 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
448 * Note that isr_count is always 1, and highest_isr_cache
449 * is always -1, with APIC virtualization enabled.
451 if (!apic->isr_count)
453 if (likely(apic->highest_isr_cache != -1))
454 return apic->highest_isr_cache;
456 result = find_highest_vector(apic->regs + APIC_ISR);
457 ASSERT(result == -1 || result >= 16);
462 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
464 struct kvm_vcpu *vcpu;
465 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
471 * We do get here for APIC virtualization enabled if the guest
472 * uses the Hyper-V APIC enlightenment. In this case we may need
473 * to trigger a new interrupt delivery by writing the SVI field;
474 * on the other hand isr_count and highest_isr_cache are unused
475 * and must be left alone.
477 if (unlikely(kvm_x86_ops->hwapic_isr_update))
478 kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
479 apic_find_highest_isr(apic));
482 BUG_ON(apic->isr_count < 0);
483 apic->highest_isr_cache = -1;
487 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
491 /* This may race with setting of irr in __apic_accept_irq() and
492 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
493 * will cause vmexit immediately and the value will be recalculated
494 * on the next vmentry.
496 if (!kvm_vcpu_has_lapic(vcpu))
498 highest_irr = apic_find_highest_irr(vcpu->arch.apic);
503 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
504 int vector, int level, int trig_mode,
505 unsigned long *dest_map);
507 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
508 unsigned long *dest_map)
510 struct kvm_lapic *apic = vcpu->arch.apic;
512 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
513 irq->level, irq->trig_mode, dest_map);
516 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
519 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
523 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
526 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
530 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
532 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
535 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
538 if (pv_eoi_get_user(vcpu, &val) < 0)
539 apic_debug("Can't read EOI MSR value: 0x%llx\n",
540 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
544 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
546 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
547 apic_debug("Can't set EOI MSR value: 0x%llx\n",
548 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
551 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
554 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
556 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
557 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
558 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
561 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
564 static void apic_update_ppr(struct kvm_lapic *apic)
566 u32 tpr, isrv, ppr, old_ppr;
569 old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
570 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
571 isr = apic_find_highest_isr(apic);
572 isrv = (isr != -1) ? isr : 0;
574 if ((tpr & 0xf0) >= (isrv & 0xf0))
579 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
580 apic, ppr, isr, isrv);
582 if (old_ppr != ppr) {
583 apic_set_reg(apic, APIC_PROCPRI, ppr);
585 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
589 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
591 apic_set_reg(apic, APIC_TASKPRI, tpr);
592 apic_update_ppr(apic);
595 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
597 if (apic_x2apic_mode(apic))
598 return mda == X2APIC_BROADCAST;
600 return GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST;
603 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
605 if (kvm_apic_broadcast(apic, mda))
608 if (apic_x2apic_mode(apic))
609 return mda == kvm_apic_id(apic);
611 return mda == SET_APIC_DEST_FIELD(kvm_apic_id(apic));
614 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
618 if (kvm_apic_broadcast(apic, mda))
621 logical_id = kvm_apic_get_reg(apic, APIC_LDR);
623 if (apic_x2apic_mode(apic))
624 return ((logical_id >> 16) == (mda >> 16))
625 && (logical_id & mda & 0xffff) != 0;
627 logical_id = GET_APIC_LOGICAL_ID(logical_id);
628 mda = GET_APIC_DEST_FIELD(mda);
630 switch (kvm_apic_get_reg(apic, APIC_DFR)) {
632 return (logical_id & mda) != 0;
633 case APIC_DFR_CLUSTER:
634 return ((logical_id >> 4) == (mda >> 4))
635 && (logical_id & mda & 0xf) != 0;
637 apic_debug("Bad DFR vcpu %d: %08x\n",
638 apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
643 /* KVM APIC implementation has two quirks
644 * - dest always begins at 0 while xAPIC MDA has offset 24,
645 * - IOxAPIC messages have to be delivered (directly) to x2APIC.
647 static u32 kvm_apic_mda(unsigned int dest_id, struct kvm_lapic *source,
648 struct kvm_lapic *target)
650 bool ipi = source != NULL;
651 bool x2apic_mda = apic_x2apic_mode(ipi ? source : target);
653 if (!ipi && dest_id == APIC_BROADCAST && x2apic_mda)
654 return X2APIC_BROADCAST;
656 return x2apic_mda ? dest_id : SET_APIC_DEST_FIELD(dest_id);
659 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
660 int short_hand, unsigned int dest, int dest_mode)
662 struct kvm_lapic *target = vcpu->arch.apic;
663 u32 mda = kvm_apic_mda(dest, source, target);
665 apic_debug("target %p, source %p, dest 0x%x, "
666 "dest_mode 0x%x, short_hand 0x%x\n",
667 target, source, dest, dest_mode, short_hand);
670 switch (short_hand) {
671 case APIC_DEST_NOSHORT:
672 if (dest_mode == APIC_DEST_PHYSICAL)
673 return kvm_apic_match_physical_addr(target, mda);
675 return kvm_apic_match_logical_addr(target, mda);
677 return target == source;
678 case APIC_DEST_ALLINC:
680 case APIC_DEST_ALLBUT:
681 return target != source;
683 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
689 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
690 struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
692 struct kvm_apic_map *map;
693 unsigned long bitmap = 1;
694 struct kvm_lapic **dst;
696 bool ret, x2apic_ipi;
700 if (irq->shorthand == APIC_DEST_SELF) {
701 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
708 x2apic_ipi = src && apic_x2apic_mode(src);
709 if (irq->dest_id == (x2apic_ipi ? X2APIC_BROADCAST : APIC_BROADCAST))
714 map = rcu_dereference(kvm->arch.apic_map);
721 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
722 if (irq->dest_id >= ARRAY_SIZE(map->phys_map))
725 dst = &map->phys_map[irq->dest_id];
729 if (!kvm_apic_logical_map_valid(map)) {
734 apic_logical_id(map, irq->dest_id, &cid, (u16 *)&bitmap);
736 if (cid >= ARRAY_SIZE(map->logical_map))
739 dst = map->logical_map[cid];
741 if (kvm_lowest_prio_delivery(irq)) {
743 for_each_set_bit(i, &bitmap, 16) {
748 else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
752 bitmap = (l >= 0) ? 1 << l : 0;
756 for_each_set_bit(i, &bitmap, 16) {
761 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
768 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
769 struct kvm_vcpu **dest_vcpu)
771 struct kvm_apic_map *map;
773 struct kvm_lapic *dst = NULL;
779 map = rcu_dereference(kvm->arch.apic_map);
784 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
785 if (irq->dest_id == 0xFF)
788 if (irq->dest_id >= ARRAY_SIZE(map->phys_map))
791 dst = map->phys_map[irq->dest_id];
792 if (dst && kvm_apic_present(dst->vcpu))
793 *dest_vcpu = dst->vcpu;
798 unsigned long bitmap = 1;
801 if (!kvm_apic_logical_map_valid(map))
804 apic_logical_id(map, irq->dest_id, &cid, (u16 *)&bitmap);
806 if (cid >= ARRAY_SIZE(map->logical_map))
809 for_each_set_bit(i, &bitmap, 16) {
810 dst = map->logical_map[cid][i];
815 if (dst && kvm_apic_present(dst->vcpu))
816 *dest_vcpu = dst->vcpu;
828 * Add a pending IRQ into lapic.
829 * Return 1 if successfully added and 0 if discarded.
831 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
832 int vector, int level, int trig_mode,
833 unsigned long *dest_map)
836 struct kvm_vcpu *vcpu = apic->vcpu;
838 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
840 switch (delivery_mode) {
842 vcpu->arch.apic_arb_prio++;
844 if (unlikely(trig_mode && !level))
847 /* FIXME add logic for vcpu on reset */
848 if (unlikely(!apic_enabled(apic)))
854 __set_bit(vcpu->vcpu_id, dest_map);
856 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
858 apic_set_vector(vector, apic->regs + APIC_TMR);
860 apic_clear_vector(vector, apic->regs + APIC_TMR);
863 if (kvm_x86_ops->deliver_posted_interrupt)
864 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
866 apic_set_irr(vector, apic);
868 kvm_make_request(KVM_REQ_EVENT, vcpu);
875 vcpu->arch.pv.pv_unhalted = 1;
876 kvm_make_request(KVM_REQ_EVENT, vcpu);
882 kvm_make_request(KVM_REQ_SMI, vcpu);
888 kvm_inject_nmi(vcpu);
893 if (!trig_mode || level) {
895 /* assumes that there are only KVM_APIC_INIT/SIPI */
896 apic->pending_events = (1UL << KVM_APIC_INIT);
897 /* make sure pending_events is visible before sending
900 kvm_make_request(KVM_REQ_EVENT, vcpu);
903 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
908 case APIC_DM_STARTUP:
909 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
910 vcpu->vcpu_id, vector);
912 apic->sipi_vector = vector;
913 /* make sure sipi_vector is visible for the receiver */
915 set_bit(KVM_APIC_SIPI, &apic->pending_events);
916 kvm_make_request(KVM_REQ_EVENT, vcpu);
922 * Should only be called by kvm_apic_local_deliver() with LVT0,
923 * before NMI watchdog was enabled. Already handled by
924 * kvm_apic_accept_pic_intr().
929 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
936 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
938 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
941 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
943 return test_bit(vector, (ulong *)apic->vcpu->arch.eoi_exit_bitmap);
946 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
950 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
951 if (!kvm_ioapic_handles_vector(apic, vector))
954 /* Request a KVM exit to inform the userspace IOAPIC. */
955 if (irqchip_split(apic->vcpu->kvm)) {
956 apic->vcpu->arch.pending_ioapic_eoi = vector;
957 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
961 if (apic_test_vector(vector, apic->regs + APIC_TMR))
962 trigger_mode = IOAPIC_LEVEL_TRIG;
964 trigger_mode = IOAPIC_EDGE_TRIG;
966 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
969 static int apic_set_eoi(struct kvm_lapic *apic)
971 int vector = apic_find_highest_isr(apic);
973 trace_kvm_eoi(apic, vector);
976 * Not every write EOI will has corresponding ISR,
977 * one example is when Kernel check timer on setup_IO_APIC
982 apic_clear_isr(vector, apic);
983 apic_update_ppr(apic);
985 kvm_ioapic_send_eoi(apic, vector);
986 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
991 * this interface assumes a trap-like exit, which has already finished
992 * desired side effect including vISR and vPPR update.
994 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
996 struct kvm_lapic *apic = vcpu->arch.apic;
998 trace_kvm_eoi(apic, vector);
1000 kvm_ioapic_send_eoi(apic, vector);
1001 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1003 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1005 static void apic_send_ipi(struct kvm_lapic *apic)
1007 u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
1008 u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
1009 struct kvm_lapic_irq irq;
1011 irq.vector = icr_low & APIC_VECTOR_MASK;
1012 irq.delivery_mode = icr_low & APIC_MODE_MASK;
1013 irq.dest_mode = icr_low & APIC_DEST_MASK;
1014 irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1015 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1016 irq.shorthand = icr_low & APIC_SHORT_MASK;
1017 irq.msi_redir_hint = false;
1018 if (apic_x2apic_mode(apic))
1019 irq.dest_id = icr_high;
1021 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1023 trace_kvm_apic_ipi(icr_low, irq.dest_id);
1025 apic_debug("icr_high 0x%x, icr_low 0x%x, "
1026 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1027 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1028 "msi_redir_hint 0x%x\n",
1029 icr_high, icr_low, irq.shorthand, irq.dest_id,
1030 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
1031 irq.vector, irq.msi_redir_hint);
1033 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1036 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1042 ASSERT(apic != NULL);
1044 /* if initial count is 0, current count should also be 0 */
1045 if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
1046 apic->lapic_timer.period == 0)
1049 remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
1050 if (ktime_to_ns(remaining) < 0)
1051 remaining = ktime_set(0, 0);
1053 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1054 tmcct = div64_u64(ns,
1055 (APIC_BUS_CYCLE_NS * apic->divide_count));
1060 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1062 struct kvm_vcpu *vcpu = apic->vcpu;
1063 struct kvm_run *run = vcpu->run;
1065 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1066 run->tpr_access.rip = kvm_rip_read(vcpu);
1067 run->tpr_access.is_write = write;
1070 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1072 if (apic->vcpu->arch.tpr_access_reporting)
1073 __report_tpr_access(apic, write);
1076 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1080 if (offset >= LAPIC_MMIO_LENGTH)
1085 if (apic_x2apic_mode(apic))
1086 val = kvm_apic_id(apic);
1088 val = kvm_apic_id(apic) << 24;
1091 apic_debug("Access APIC ARBPRI register which is for P6\n");
1094 case APIC_TMCCT: /* Timer CCR */
1095 if (apic_lvtt_tscdeadline(apic))
1098 val = apic_get_tmcct(apic);
1101 apic_update_ppr(apic);
1102 val = kvm_apic_get_reg(apic, offset);
1105 report_tpr_access(apic, false);
1108 val = kvm_apic_get_reg(apic, offset);
1115 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1117 return container_of(dev, struct kvm_lapic, dev);
1120 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1123 unsigned char alignment = offset & 0xf;
1125 /* this bitmask has a bit cleared for each reserved register */
1126 static const u64 rmask = 0x43ff01ffffffe70cULL;
1128 if ((alignment + len) > 4) {
1129 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1134 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1135 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1140 result = __apic_read(apic, offset & ~0xf);
1142 trace_kvm_apic_read(offset, result);
1148 memcpy(data, (char *)&result + alignment, len);
1151 printk(KERN_ERR "Local APIC read with len = %x, "
1152 "should be 1,2, or 4 instead\n", len);
1158 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1160 return kvm_apic_hw_enabled(apic) &&
1161 addr >= apic->base_address &&
1162 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1165 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1166 gpa_t address, int len, void *data)
1168 struct kvm_lapic *apic = to_lapic(this);
1169 u32 offset = address - apic->base_address;
1171 if (!apic_mmio_in_range(apic, address))
1174 apic_reg_read(apic, offset, len, data);
1179 static void update_divide_count(struct kvm_lapic *apic)
1181 u32 tmp1, tmp2, tdcr;
1183 tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1185 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1186 apic->divide_count = 0x1 << (tmp2 & 0x7);
1188 apic_debug("timer divide count is 0x%x\n",
1189 apic->divide_count);
1192 static void apic_update_lvtt(struct kvm_lapic *apic)
1194 u32 timer_mode = kvm_apic_get_reg(apic, APIC_LVTT) &
1195 apic->lapic_timer.timer_mode_mask;
1197 if (apic->lapic_timer.timer_mode != timer_mode) {
1198 apic->lapic_timer.timer_mode = timer_mode;
1199 hrtimer_cancel(&apic->lapic_timer.timer);
1203 static void apic_timer_expired(struct kvm_lapic *apic)
1205 struct kvm_vcpu *vcpu = apic->vcpu;
1206 wait_queue_head_t *q = &vcpu->wq;
1207 struct kvm_timer *ktimer = &apic->lapic_timer;
1209 if (atomic_read(&apic->lapic_timer.pending))
1212 atomic_inc(&apic->lapic_timer.pending);
1213 kvm_set_pending_timer(vcpu);
1215 if (waitqueue_active(q))
1216 wake_up_interruptible(q);
1218 if (apic_lvtt_tscdeadline(apic))
1219 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1223 * On APICv, this test will cause a busy wait
1224 * during a higher-priority task.
1227 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1229 struct kvm_lapic *apic = vcpu->arch.apic;
1230 u32 reg = kvm_apic_get_reg(apic, APIC_LVTT);
1232 if (kvm_apic_hw_enabled(apic)) {
1233 int vec = reg & APIC_VECTOR_MASK;
1234 void *bitmap = apic->regs + APIC_ISR;
1236 if (kvm_x86_ops->deliver_posted_interrupt)
1237 bitmap = apic->regs + APIC_IRR;
1239 if (apic_test_vector(vec, bitmap))
1245 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1247 struct kvm_lapic *apic = vcpu->arch.apic;
1248 u64 guest_tsc, tsc_deadline;
1250 if (!kvm_vcpu_has_lapic(vcpu))
1253 if (apic->lapic_timer.expired_tscdeadline == 0)
1256 if (!lapic_timer_int_injected(vcpu))
1259 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1260 apic->lapic_timer.expired_tscdeadline = 0;
1261 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1262 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1264 /* __delay is delay_tsc whenever the hardware has TSC, thus always. */
1265 if (guest_tsc < tsc_deadline)
1266 __delay(tsc_deadline - guest_tsc);
1269 static void start_apic_timer(struct kvm_lapic *apic)
1273 atomic_set(&apic->lapic_timer.pending, 0);
1275 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1276 /* lapic timer in oneshot or periodic mode */
1277 now = apic->lapic_timer.timer.base->get_time();
1278 apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1279 * APIC_BUS_CYCLE_NS * apic->divide_count;
1281 if (!apic->lapic_timer.period)
1284 * Do not allow the guest to program periodic timers with small
1285 * interval, since the hrtimers are not throttled by the host
1288 if (apic_lvtt_period(apic)) {
1289 s64 min_period = min_timer_period_us * 1000LL;
1291 if (apic->lapic_timer.period < min_period) {
1292 pr_info_ratelimited(
1293 "kvm: vcpu %i: requested %lld ns "
1294 "lapic timer period limited to %lld ns\n",
1295 apic->vcpu->vcpu_id,
1296 apic->lapic_timer.period, min_period);
1297 apic->lapic_timer.period = min_period;
1301 hrtimer_start(&apic->lapic_timer.timer,
1302 ktime_add_ns(now, apic->lapic_timer.period),
1305 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1307 "timer initial count 0x%x, period %lldns, "
1308 "expire @ 0x%016" PRIx64 ".\n", __func__,
1309 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1310 kvm_apic_get_reg(apic, APIC_TMICT),
1311 apic->lapic_timer.period,
1312 ktime_to_ns(ktime_add_ns(now,
1313 apic->lapic_timer.period)));
1314 } else if (apic_lvtt_tscdeadline(apic)) {
1315 /* lapic timer in tsc deadline mode */
1316 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1319 struct kvm_vcpu *vcpu = apic->vcpu;
1320 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1321 unsigned long flags;
1323 if (unlikely(!tscdeadline || !this_tsc_khz))
1326 local_irq_save(flags);
1328 now = apic->lapic_timer.timer.base->get_time();
1329 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1330 if (likely(tscdeadline > guest_tsc)) {
1331 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1332 do_div(ns, this_tsc_khz);
1333 expire = ktime_add_ns(now, ns);
1334 expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1335 hrtimer_start(&apic->lapic_timer.timer,
1336 expire, HRTIMER_MODE_ABS);
1338 apic_timer_expired(apic);
1340 local_irq_restore(flags);
1344 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1346 bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1348 if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1349 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1350 if (lvt0_in_nmi_mode) {
1351 apic_debug("Receive NMI setting on APIC_LVT0 "
1352 "for cpu %d\n", apic->vcpu->vcpu_id);
1353 atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1355 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1359 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1363 trace_kvm_apic_write(reg, val);
1366 case APIC_ID: /* Local APIC ID */
1367 if (!apic_x2apic_mode(apic))
1368 kvm_apic_set_id(apic, val >> 24);
1374 report_tpr_access(apic, true);
1375 apic_set_tpr(apic, val & 0xff);
1383 if (!apic_x2apic_mode(apic))
1384 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1390 if (!apic_x2apic_mode(apic)) {
1391 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1392 recalculate_apic_map(apic->vcpu->kvm);
1399 if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1400 mask |= APIC_SPIV_DIRECTED_EOI;
1401 apic_set_spiv(apic, val & mask);
1402 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1406 for (i = 0; i < APIC_LVT_NUM; i++) {
1407 lvt_val = kvm_apic_get_reg(apic,
1408 APIC_LVTT + 0x10 * i);
1409 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1410 lvt_val | APIC_LVT_MASKED);
1412 apic_update_lvtt(apic);
1413 atomic_set(&apic->lapic_timer.pending, 0);
1419 /* No delay here, so we always clear the pending bit */
1420 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1421 apic_send_ipi(apic);
1425 if (!apic_x2apic_mode(apic))
1427 apic_set_reg(apic, APIC_ICR2, val);
1431 apic_manage_nmi_watchdog(apic, val);
1436 /* TODO: Check vector */
1437 if (!kvm_apic_sw_enabled(apic))
1438 val |= APIC_LVT_MASKED;
1440 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1441 apic_set_reg(apic, reg, val);
1446 if (!kvm_apic_sw_enabled(apic))
1447 val |= APIC_LVT_MASKED;
1448 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1449 apic_set_reg(apic, APIC_LVTT, val);
1450 apic_update_lvtt(apic);
1454 if (apic_lvtt_tscdeadline(apic))
1457 hrtimer_cancel(&apic->lapic_timer.timer);
1458 apic_set_reg(apic, APIC_TMICT, val);
1459 start_apic_timer(apic);
1464 apic_debug("KVM_WRITE:TDCR %x\n", val);
1465 apic_set_reg(apic, APIC_TDCR, val);
1466 update_divide_count(apic);
1470 if (apic_x2apic_mode(apic) && val != 0) {
1471 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1477 if (apic_x2apic_mode(apic)) {
1478 apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1487 apic_debug("Local APIC Write to read-only register %x\n", reg);
1491 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1492 gpa_t address, int len, const void *data)
1494 struct kvm_lapic *apic = to_lapic(this);
1495 unsigned int offset = address - apic->base_address;
1498 if (!apic_mmio_in_range(apic, address))
1502 * APIC register must be aligned on 128-bits boundary.
1503 * 32/64/128 bits registers must be accessed thru 32 bits.
1506 if (len != 4 || (offset & 0xf)) {
1507 /* Don't shout loud, $infamous_os would cause only noise. */
1508 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1514 /* too common printing */
1515 if (offset != APIC_EOI)
1516 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1517 "0x%x\n", __func__, offset, len, val);
1519 apic_reg_write(apic, offset & 0xff0, val);
1524 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1526 if (kvm_vcpu_has_lapic(vcpu))
1527 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1529 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1531 /* emulate APIC access in a trap manner */
1532 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1536 /* hw has done the conditional check and inst decode */
1539 apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1541 /* TODO: optimize to just emulate side effect w/o one more write */
1542 apic_reg_write(vcpu->arch.apic, offset, val);
1544 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1546 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1548 struct kvm_lapic *apic = vcpu->arch.apic;
1550 if (!vcpu->arch.apic)
1553 hrtimer_cancel(&apic->lapic_timer.timer);
1555 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1556 static_key_slow_dec_deferred(&apic_hw_disabled);
1558 if (!apic->sw_enabled)
1559 static_key_slow_dec_deferred(&apic_sw_disabled);
1562 free_page((unsigned long)apic->regs);
1568 *----------------------------------------------------------------------
1570 *----------------------------------------------------------------------
1573 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1575 struct kvm_lapic *apic = vcpu->arch.apic;
1577 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1578 apic_lvtt_period(apic))
1581 return apic->lapic_timer.tscdeadline;
1584 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1586 struct kvm_lapic *apic = vcpu->arch.apic;
1588 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1589 apic_lvtt_period(apic))
1592 hrtimer_cancel(&apic->lapic_timer.timer);
1593 apic->lapic_timer.tscdeadline = data;
1594 start_apic_timer(apic);
1597 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1599 struct kvm_lapic *apic = vcpu->arch.apic;
1601 if (!kvm_vcpu_has_lapic(vcpu))
1604 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1605 | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1608 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1612 if (!kvm_vcpu_has_lapic(vcpu))
1615 tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1617 return (tpr & 0xf0) >> 4;
1620 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1622 u64 old_value = vcpu->arch.apic_base;
1623 struct kvm_lapic *apic = vcpu->arch.apic;
1626 value |= MSR_IA32_APICBASE_BSP;
1627 vcpu->arch.apic_base = value;
1631 vcpu->arch.apic_base = value;
1633 /* update jump label if enable bit changes */
1634 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1635 if (value & MSR_IA32_APICBASE_ENABLE)
1636 static_key_slow_dec_deferred(&apic_hw_disabled);
1638 static_key_slow_inc(&apic_hw_disabled.key);
1639 recalculate_apic_map(vcpu->kvm);
1642 if ((old_value ^ value) & X2APIC_ENABLE) {
1643 if (value & X2APIC_ENABLE) {
1644 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
1645 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1647 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1650 apic->base_address = apic->vcpu->arch.apic_base &
1651 MSR_IA32_APICBASE_BASE;
1653 if ((value & MSR_IA32_APICBASE_ENABLE) &&
1654 apic->base_address != APIC_DEFAULT_PHYS_BASE)
1655 pr_warn_once("APIC base relocation is unsupported by KVM");
1657 /* with FSB delivery interrupt, we can restart APIC functionality */
1658 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1659 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1663 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
1665 struct kvm_lapic *apic;
1668 apic_debug("%s\n", __func__);
1671 apic = vcpu->arch.apic;
1672 ASSERT(apic != NULL);
1674 /* Stop the timer in case it's a reset to an active apic */
1675 hrtimer_cancel(&apic->lapic_timer.timer);
1678 kvm_apic_set_id(apic, vcpu->vcpu_id);
1679 kvm_apic_set_version(apic->vcpu);
1681 for (i = 0; i < APIC_LVT_NUM; i++)
1682 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1683 apic_update_lvtt(apic);
1684 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
1685 apic_set_reg(apic, APIC_LVT0,
1686 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1687 apic_manage_nmi_watchdog(apic, kvm_apic_get_reg(apic, APIC_LVT0));
1689 apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1690 apic_set_spiv(apic, 0xff);
1691 apic_set_reg(apic, APIC_TASKPRI, 0);
1692 if (!apic_x2apic_mode(apic))
1693 kvm_apic_set_ldr(apic, 0);
1694 apic_set_reg(apic, APIC_ESR, 0);
1695 apic_set_reg(apic, APIC_ICR, 0);
1696 apic_set_reg(apic, APIC_ICR2, 0);
1697 apic_set_reg(apic, APIC_TDCR, 0);
1698 apic_set_reg(apic, APIC_TMICT, 0);
1699 for (i = 0; i < 8; i++) {
1700 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1701 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1702 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1704 apic->irr_pending = kvm_vcpu_apic_vid_enabled(vcpu);
1705 apic->isr_count = kvm_x86_ops->hwapic_isr_update ? 1 : 0;
1706 apic->highest_isr_cache = -1;
1707 update_divide_count(apic);
1708 atomic_set(&apic->lapic_timer.pending, 0);
1709 if (kvm_vcpu_is_bsp(vcpu))
1710 kvm_lapic_set_base(vcpu,
1711 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1712 vcpu->arch.pv_eoi.msr_val = 0;
1713 apic_update_ppr(apic);
1715 vcpu->arch.apic_arb_prio = 0;
1716 vcpu->arch.apic_attention = 0;
1718 apic_debug("%s: vcpu=%p, id=%d, base_msr="
1719 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1720 vcpu, kvm_apic_id(apic),
1721 vcpu->arch.apic_base, apic->base_address);
1725 *----------------------------------------------------------------------
1727 *----------------------------------------------------------------------
1730 static bool lapic_is_periodic(struct kvm_lapic *apic)
1732 return apic_lvtt_period(apic);
1735 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1737 struct kvm_lapic *apic = vcpu->arch.apic;
1739 if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1740 apic_lvt_enabled(apic, APIC_LVTT))
1741 return atomic_read(&apic->lapic_timer.pending);
1746 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1748 u32 reg = kvm_apic_get_reg(apic, lvt_type);
1749 int vector, mode, trig_mode;
1751 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1752 vector = reg & APIC_VECTOR_MASK;
1753 mode = reg & APIC_MODE_MASK;
1754 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1755 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1761 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1763 struct kvm_lapic *apic = vcpu->arch.apic;
1766 kvm_apic_local_deliver(apic, APIC_LVT0);
1769 static const struct kvm_io_device_ops apic_mmio_ops = {
1770 .read = apic_mmio_read,
1771 .write = apic_mmio_write,
1774 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1776 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1777 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1779 apic_timer_expired(apic);
1781 if (lapic_is_periodic(apic)) {
1782 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1783 return HRTIMER_RESTART;
1785 return HRTIMER_NORESTART;
1788 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1790 struct kvm_lapic *apic;
1792 ASSERT(vcpu != NULL);
1793 apic_debug("apic_init %d\n", vcpu->vcpu_id);
1795 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1799 vcpu->arch.apic = apic;
1801 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1803 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1805 goto nomem_free_apic;
1809 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1811 apic->lapic_timer.timer.function = apic_timer_fn;
1814 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1815 * thinking that APIC satet has changed.
1817 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1818 kvm_lapic_set_base(vcpu,
1819 APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1821 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1822 kvm_lapic_reset(vcpu, false);
1823 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1832 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1834 struct kvm_lapic *apic = vcpu->arch.apic;
1837 if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1840 apic_update_ppr(apic);
1841 highest_irr = apic_find_highest_irr(apic);
1842 if ((highest_irr == -1) ||
1843 ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1848 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1850 u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1853 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1855 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1856 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1861 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1863 struct kvm_lapic *apic = vcpu->arch.apic;
1865 if (!kvm_vcpu_has_lapic(vcpu))
1868 if (atomic_read(&apic->lapic_timer.pending) > 0) {
1869 kvm_apic_local_deliver(apic, APIC_LVTT);
1870 if (apic_lvtt_tscdeadline(apic))
1871 apic->lapic_timer.tscdeadline = 0;
1872 atomic_set(&apic->lapic_timer.pending, 0);
1876 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1878 int vector = kvm_apic_has_interrupt(vcpu);
1879 struct kvm_lapic *apic = vcpu->arch.apic;
1885 * We get here even with APIC virtualization enabled, if doing
1886 * nested virtualization and L1 runs with the "acknowledge interrupt
1887 * on exit" mode. Then we cannot inject the interrupt via RVI,
1888 * because the process would deliver it through the IDT.
1891 apic_set_isr(vector, apic);
1892 apic_update_ppr(apic);
1893 apic_clear_irr(vector, apic);
1897 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1898 struct kvm_lapic_state *s)
1900 struct kvm_lapic *apic = vcpu->arch.apic;
1902 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1903 /* set SPIV separately to get count of SW disabled APICs right */
1904 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1905 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1906 /* call kvm_apic_set_id() to put apic into apic_map */
1907 kvm_apic_set_id(apic, kvm_apic_id(apic));
1908 kvm_apic_set_version(vcpu);
1910 apic_update_ppr(apic);
1911 hrtimer_cancel(&apic->lapic_timer.timer);
1912 apic_update_lvtt(apic);
1913 apic_manage_nmi_watchdog(apic, kvm_apic_get_reg(apic, APIC_LVT0));
1914 update_divide_count(apic);
1915 start_apic_timer(apic);
1916 apic->irr_pending = true;
1917 apic->isr_count = kvm_x86_ops->hwapic_isr_update ?
1918 1 : count_vectors(apic->regs + APIC_ISR);
1919 apic->highest_isr_cache = -1;
1920 if (kvm_x86_ops->hwapic_irr_update)
1921 kvm_x86_ops->hwapic_irr_update(vcpu,
1922 apic_find_highest_irr(apic));
1923 if (unlikely(kvm_x86_ops->hwapic_isr_update))
1924 kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
1925 apic_find_highest_isr(apic));
1926 kvm_make_request(KVM_REQ_EVENT, vcpu);
1927 if (ioapic_in_kernel(vcpu->kvm))
1928 kvm_rtc_eoi_tracking_restore_one(vcpu);
1930 vcpu->arch.apic_arb_prio = 0;
1933 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1935 struct hrtimer *timer;
1937 if (!kvm_vcpu_has_lapic(vcpu))
1940 timer = &vcpu->arch.apic->lapic_timer.timer;
1941 if (hrtimer_cancel(timer))
1942 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1946 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1948 * Detect whether guest triggered PV EOI since the
1949 * last entry. If yes, set EOI on guests's behalf.
1950 * Clear PV EOI in guest memory in any case.
1952 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1953 struct kvm_lapic *apic)
1958 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1959 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1961 * KVM_APIC_PV_EOI_PENDING is unset:
1962 * -> host disabled PV EOI.
1963 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1964 * -> host enabled PV EOI, guest did not execute EOI yet.
1965 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1966 * -> host enabled PV EOI, guest executed EOI.
1968 BUG_ON(!pv_eoi_enabled(vcpu));
1969 pending = pv_eoi_get_pending(vcpu);
1971 * Clear pending bit in any case: it will be set again on vmentry.
1972 * While this might not be ideal from performance point of view,
1973 * this makes sure pv eoi is only enabled when we know it's safe.
1975 pv_eoi_clr_pending(vcpu);
1978 vector = apic_set_eoi(apic);
1979 trace_kvm_pv_eoi(apic, vector);
1982 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1986 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1987 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1989 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1992 if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1996 apic_set_tpr(vcpu->arch.apic, data & 0xff);
2000 * apic_sync_pv_eoi_to_guest - called before vmentry
2002 * Detect whether it's safe to enable PV EOI and
2005 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2006 struct kvm_lapic *apic)
2008 if (!pv_eoi_enabled(vcpu) ||
2009 /* IRR set or many bits in ISR: could be nested. */
2010 apic->irr_pending ||
2011 /* Cache not set: could be safe but we don't bother. */
2012 apic->highest_isr_cache == -1 ||
2013 /* Need EOI to update ioapic. */
2014 kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2016 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2017 * so we need not do anything here.
2022 pv_eoi_set_pending(apic->vcpu);
2025 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2028 int max_irr, max_isr;
2029 struct kvm_lapic *apic = vcpu->arch.apic;
2031 apic_sync_pv_eoi_to_guest(vcpu, apic);
2033 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2036 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
2037 max_irr = apic_find_highest_irr(apic);
2040 max_isr = apic_find_highest_isr(apic);
2043 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2045 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2049 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2052 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2053 &vcpu->arch.apic->vapic_cache,
2054 vapic_addr, sizeof(u32)))
2056 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2058 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2061 vcpu->arch.apic->vapic_addr = vapic_addr;
2065 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2067 struct kvm_lapic *apic = vcpu->arch.apic;
2068 u32 reg = (msr - APIC_BASE_MSR) << 4;
2070 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2073 if (reg == APIC_ICR2)
2076 /* if this is ICR write vector before command */
2077 if (reg == APIC_ICR)
2078 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2079 return apic_reg_write(apic, reg, (u32)data);
2082 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2084 struct kvm_lapic *apic = vcpu->arch.apic;
2085 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2087 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2090 if (reg == APIC_DFR || reg == APIC_ICR2) {
2091 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2096 if (apic_reg_read(apic, reg, 4, &low))
2098 if (reg == APIC_ICR)
2099 apic_reg_read(apic, APIC_ICR2, 4, &high);
2101 *data = (((u64)high) << 32) | low;
2106 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2108 struct kvm_lapic *apic = vcpu->arch.apic;
2110 if (!kvm_vcpu_has_lapic(vcpu))
2113 /* if this is ICR write vector before command */
2114 if (reg == APIC_ICR)
2115 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2116 return apic_reg_write(apic, reg, (u32)data);
2119 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2121 struct kvm_lapic *apic = vcpu->arch.apic;
2124 if (!kvm_vcpu_has_lapic(vcpu))
2127 if (apic_reg_read(apic, reg, 4, &low))
2129 if (reg == APIC_ICR)
2130 apic_reg_read(apic, APIC_ICR2, 4, &high);
2132 *data = (((u64)high) << 32) | low;
2137 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2139 u64 addr = data & ~KVM_MSR_ENABLED;
2140 if (!IS_ALIGNED(addr, 4))
2143 vcpu->arch.pv_eoi.msr_val = data;
2144 if (!pv_eoi_enabled(vcpu))
2146 return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2150 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2152 struct kvm_lapic *apic = vcpu->arch.apic;
2156 if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
2160 * INITs are latched while in SMM. Because an SMM CPU cannot
2161 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2162 * and delay processing of INIT until the next RSM.
2165 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2166 if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2167 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2171 pe = xchg(&apic->pending_events, 0);
2172 if (test_bit(KVM_APIC_INIT, &pe)) {
2173 kvm_lapic_reset(vcpu, true);
2174 kvm_vcpu_reset(vcpu, true);
2175 if (kvm_vcpu_is_bsp(apic->vcpu))
2176 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2178 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2180 if (test_bit(KVM_APIC_SIPI, &pe) &&
2181 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2182 /* evaluate pending_events before reading the vector */
2184 sipi_vector = apic->sipi_vector;
2185 apic_debug("vcpu %d received sipi with vector # %x\n",
2186 vcpu->vcpu_id, sipi_vector);
2187 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2188 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2192 void kvm_lapic_init(void)
2194 /* do not patch jump label more than once per second */
2195 jump_label_rate_limit(&apic_hw_disabled, HZ);
2196 jump_label_rate_limit(&apic_sw_disabled, HZ);
2199 void kvm_lapic_exit(void)
2201 static_key_deferred_flush(&apic_hw_disabled);
2202 static_key_deferred_flush(&apic_sw_disabled);