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/export.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"
47 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
49 #define mod_64(x, y) ((x) % (y))
57 #define APIC_BUS_CYCLE_NS 1
59 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
60 #define apic_debug(fmt, arg...)
62 /* 14 is the version for Xeon and Pentium 8.4.8*/
63 #define APIC_VERSION (0x14UL | ((KVM_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 static inline int apic_test_vector(int vec, void *bitmap)
77 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
80 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
82 struct kvm_lapic *apic = vcpu->arch.apic;
84 return apic_test_vector(vector, apic->regs + APIC_ISR) ||
85 apic_test_vector(vector, apic->regs + APIC_IRR);
88 static inline void apic_clear_vector(int vec, void *bitmap)
90 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
93 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
95 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
98 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
100 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
103 struct static_key_deferred apic_hw_disabled __read_mostly;
104 struct static_key_deferred apic_sw_disabled __read_mostly;
106 static inline int apic_enabled(struct kvm_lapic *apic)
108 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
112 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
115 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
116 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
118 static inline u8 kvm_xapic_id(struct kvm_lapic *apic)
120 return kvm_lapic_get_reg(apic, APIC_ID) >> 24;
123 static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
125 return apic->vcpu->vcpu_id;
128 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
129 u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
131 case KVM_APIC_MODE_X2APIC: {
132 u32 offset = (dest_id >> 16) * 16;
133 u32 max_apic_id = map->max_apic_id;
135 if (offset <= max_apic_id) {
136 u8 cluster_size = min(max_apic_id - offset + 1, 16U);
138 *cluster = &map->phys_map[offset];
139 *mask = dest_id & (0xffff >> (16 - cluster_size));
146 case KVM_APIC_MODE_XAPIC_FLAT:
147 *cluster = map->xapic_flat_map;
148 *mask = dest_id & 0xff;
150 case KVM_APIC_MODE_XAPIC_CLUSTER:
151 *cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
152 *mask = dest_id & 0xf;
160 static void kvm_apic_map_free(struct rcu_head *rcu)
162 struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
167 static void recalculate_apic_map(struct kvm *kvm)
169 struct kvm_apic_map *new, *old = NULL;
170 struct kvm_vcpu *vcpu;
172 u32 max_id = 255; /* enough space for any xAPIC ID */
174 mutex_lock(&kvm->arch.apic_map_lock);
176 kvm_for_each_vcpu(i, vcpu, kvm)
177 if (kvm_apic_present(vcpu))
178 max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
180 new = kvm_kvzalloc(sizeof(struct kvm_apic_map) +
181 sizeof(struct kvm_lapic *) * ((u64)max_id + 1));
186 new->max_apic_id = max_id;
188 kvm_for_each_vcpu(i, vcpu, kvm) {
189 struct kvm_lapic *apic = vcpu->arch.apic;
190 struct kvm_lapic **cluster;
196 if (!kvm_apic_present(vcpu))
199 xapic_id = kvm_xapic_id(apic);
200 x2apic_id = kvm_x2apic_id(apic);
202 /* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
203 if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
204 x2apic_id <= new->max_apic_id)
205 new->phys_map[x2apic_id] = apic;
207 * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
208 * prevent them from masking VCPUs with APIC ID <= 0xff.
210 if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
211 new->phys_map[xapic_id] = apic;
213 ldr = kvm_lapic_get_reg(apic, APIC_LDR);
215 if (apic_x2apic_mode(apic)) {
216 new->mode |= KVM_APIC_MODE_X2APIC;
218 ldr = GET_APIC_LOGICAL_ID(ldr);
219 if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
220 new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
222 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
225 if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
229 cluster[ffs(mask) - 1] = apic;
232 old = rcu_dereference_protected(kvm->arch.apic_map,
233 lockdep_is_held(&kvm->arch.apic_map_lock));
234 rcu_assign_pointer(kvm->arch.apic_map, new);
235 mutex_unlock(&kvm->arch.apic_map_lock);
238 call_rcu(&old->rcu, kvm_apic_map_free);
240 kvm_make_scan_ioapic_request(kvm);
243 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
245 bool enabled = val & APIC_SPIV_APIC_ENABLED;
247 kvm_lapic_set_reg(apic, APIC_SPIV, val);
249 if (enabled != apic->sw_enabled) {
250 apic->sw_enabled = enabled;
252 static_key_slow_dec_deferred(&apic_sw_disabled);
253 recalculate_apic_map(apic->vcpu->kvm);
255 static_key_slow_inc(&apic_sw_disabled.key);
259 static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
261 kvm_lapic_set_reg(apic, APIC_ID, id << 24);
262 recalculate_apic_map(apic->vcpu->kvm);
265 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
267 kvm_lapic_set_reg(apic, APIC_LDR, id);
268 recalculate_apic_map(apic->vcpu->kvm);
271 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
273 u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
275 WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
277 kvm_lapic_set_reg(apic, APIC_ID, id);
278 kvm_lapic_set_reg(apic, APIC_LDR, ldr);
279 recalculate_apic_map(apic->vcpu->kvm);
282 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
284 return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
287 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
289 return kvm_lapic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
292 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
294 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
297 static inline int apic_lvtt_period(struct kvm_lapic *apic)
299 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
302 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
304 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
307 static inline int apic_lvt_nmi_mode(u32 lvt_val)
309 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
312 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
314 struct kvm_lapic *apic = vcpu->arch.apic;
315 struct kvm_cpuid_entry2 *feat;
316 u32 v = APIC_VERSION;
318 if (!lapic_in_kernel(vcpu))
321 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
322 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
323 v |= APIC_LVR_DIRECTED_EOI;
324 kvm_lapic_set_reg(apic, APIC_LVR, v);
327 static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
328 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
329 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
330 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
331 LINT_MASK, LINT_MASK, /* LVT0-1 */
332 LVT_MASK /* LVTERR */
335 static int find_highest_vector(void *bitmap)
340 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
341 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
342 reg = bitmap + REG_POS(vec);
344 return fls(*reg) - 1 + vec;
350 static u8 count_vectors(void *bitmap)
356 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
357 reg = bitmap + REG_POS(vec);
358 count += hweight32(*reg);
364 void __kvm_apic_update_irr(u32 *pir, void *regs)
368 for (i = 0; i <= 7; i++) {
369 pir_val = READ_ONCE(pir[i]);
371 pir_val = xchg(&pir[i], 0);
372 *((u32 *)(regs + APIC_IRR + i * 0x10)) |= pir_val;
376 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
378 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
380 struct kvm_lapic *apic = vcpu->arch.apic;
382 __kvm_apic_update_irr(pir, apic->regs);
384 kvm_make_request(KVM_REQ_EVENT, vcpu);
386 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
388 static inline int apic_search_irr(struct kvm_lapic *apic)
390 return find_highest_vector(apic->regs + APIC_IRR);
393 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
398 * Note that irr_pending is just a hint. It will be always
399 * true with virtual interrupt delivery enabled.
401 if (!apic->irr_pending)
404 if (apic->vcpu->arch.apicv_active)
405 kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
406 result = apic_search_irr(apic);
407 ASSERT(result == -1 || result >= 16);
412 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
414 struct kvm_vcpu *vcpu;
418 if (unlikely(vcpu->arch.apicv_active)) {
419 /* try to update RVI */
420 apic_clear_vector(vec, apic->regs + APIC_IRR);
421 kvm_make_request(KVM_REQ_EVENT, vcpu);
423 apic->irr_pending = false;
424 apic_clear_vector(vec, apic->regs + APIC_IRR);
425 if (apic_search_irr(apic) != -1)
426 apic->irr_pending = true;
430 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
432 struct kvm_vcpu *vcpu;
434 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
440 * With APIC virtualization enabled, all caching is disabled
441 * because the processor can modify ISR under the hood. Instead
444 if (unlikely(vcpu->arch.apicv_active))
445 kvm_x86_ops->hwapic_isr_update(vcpu, vec);
448 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
450 * ISR (in service register) bit is set when injecting an interrupt.
451 * The highest vector is injected. Thus the latest bit set matches
452 * the highest bit in ISR.
454 apic->highest_isr_cache = vec;
458 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
463 * Note that isr_count is always 1, and highest_isr_cache
464 * is always -1, with APIC virtualization enabled.
466 if (!apic->isr_count)
468 if (likely(apic->highest_isr_cache != -1))
469 return apic->highest_isr_cache;
471 result = find_highest_vector(apic->regs + APIC_ISR);
472 ASSERT(result == -1 || result >= 16);
477 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
479 struct kvm_vcpu *vcpu;
480 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
486 * We do get here for APIC virtualization enabled if the guest
487 * uses the Hyper-V APIC enlightenment. In this case we may need
488 * to trigger a new interrupt delivery by writing the SVI field;
489 * on the other hand isr_count and highest_isr_cache are unused
490 * and must be left alone.
492 if (unlikely(vcpu->arch.apicv_active))
493 kvm_x86_ops->hwapic_isr_update(vcpu,
494 apic_find_highest_isr(apic));
497 BUG_ON(apic->isr_count < 0);
498 apic->highest_isr_cache = -1;
502 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
504 /* This may race with setting of irr in __apic_accept_irq() and
505 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
506 * will cause vmexit immediately and the value will be recalculated
507 * on the next vmentry.
509 return apic_find_highest_irr(vcpu->arch.apic);
512 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
513 int vector, int level, int trig_mode,
514 struct dest_map *dest_map);
516 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
517 struct dest_map *dest_map)
519 struct kvm_lapic *apic = vcpu->arch.apic;
521 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
522 irq->level, irq->trig_mode, dest_map);
525 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
528 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
532 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
535 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
539 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
541 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
544 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
547 if (pv_eoi_get_user(vcpu, &val) < 0)
548 apic_debug("Can't read EOI MSR value: 0x%llx\n",
549 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
553 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
555 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
556 apic_debug("Can't set EOI MSR value: 0x%llx\n",
557 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
560 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
563 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
565 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
566 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
567 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
570 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
573 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
575 int highest_irr = apic_find_highest_irr(apic);
576 if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
581 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
583 u32 tpr, isrv, ppr, old_ppr;
586 old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
587 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
588 isr = apic_find_highest_isr(apic);
589 isrv = (isr != -1) ? isr : 0;
591 if ((tpr & 0xf0) >= (isrv & 0xf0))
596 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
597 apic, ppr, isr, isrv);
601 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
603 return ppr < old_ppr;
606 static void apic_update_ppr(struct kvm_lapic *apic)
610 if (__apic_update_ppr(apic, &ppr))
611 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
614 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
616 apic_update_ppr(vcpu->arch.apic);
618 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
620 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
622 kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
623 apic_update_ppr(apic);
626 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
628 return mda == (apic_x2apic_mode(apic) ?
629 X2APIC_BROADCAST : APIC_BROADCAST);
632 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
634 if (kvm_apic_broadcast(apic, mda))
637 if (apic_x2apic_mode(apic))
638 return mda == kvm_x2apic_id(apic);
641 * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
642 * it were in x2APIC mode. Hotplugged VCPUs start in xAPIC mode and
643 * this allows unique addressing of VCPUs with APIC ID over 0xff.
644 * The 0xff condition is needed because writeable xAPIC ID.
646 if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
649 return mda == kvm_xapic_id(apic);
652 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
656 if (kvm_apic_broadcast(apic, mda))
659 logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
661 if (apic_x2apic_mode(apic))
662 return ((logical_id >> 16) == (mda >> 16))
663 && (logical_id & mda & 0xffff) != 0;
665 logical_id = GET_APIC_LOGICAL_ID(logical_id);
667 switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
669 return (logical_id & mda) != 0;
670 case APIC_DFR_CLUSTER:
671 return ((logical_id >> 4) == (mda >> 4))
672 && (logical_id & mda & 0xf) != 0;
674 apic_debug("Bad DFR vcpu %d: %08x\n",
675 apic->vcpu->vcpu_id, kvm_lapic_get_reg(apic, APIC_DFR));
680 /* The KVM local APIC implementation has two quirks:
682 * - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
683 * in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
684 * KVM doesn't do that aliasing.
686 * - in-kernel IOAPIC messages have to be delivered directly to
687 * x2APIC, because the kernel does not support interrupt remapping.
688 * In order to support broadcast without interrupt remapping, x2APIC
689 * rewrites the destination of non-IPI messages from APIC_BROADCAST
690 * to X2APIC_BROADCAST.
692 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API. This is
693 * important when userspace wants to use x2APIC-format MSIs, because
694 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
696 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
697 struct kvm_lapic *source, struct kvm_lapic *target)
699 bool ipi = source != NULL;
701 if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
702 !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
703 return X2APIC_BROADCAST;
708 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
709 int short_hand, unsigned int dest, int dest_mode)
711 struct kvm_lapic *target = vcpu->arch.apic;
712 u32 mda = kvm_apic_mda(vcpu, dest, source, target);
714 apic_debug("target %p, source %p, dest 0x%x, "
715 "dest_mode 0x%x, short_hand 0x%x\n",
716 target, source, dest, dest_mode, short_hand);
719 switch (short_hand) {
720 case APIC_DEST_NOSHORT:
721 if (dest_mode == APIC_DEST_PHYSICAL)
722 return kvm_apic_match_physical_addr(target, mda);
724 return kvm_apic_match_logical_addr(target, mda);
726 return target == source;
727 case APIC_DEST_ALLINC:
729 case APIC_DEST_ALLBUT:
730 return target != source;
732 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
737 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
739 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
740 const unsigned long *bitmap, u32 bitmap_size)
745 mod = vector % dest_vcpus;
747 for (i = 0; i <= mod; i++) {
748 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
749 BUG_ON(idx == bitmap_size);
755 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
757 if (!kvm->arch.disabled_lapic_found) {
758 kvm->arch.disabled_lapic_found = true;
760 "Disabled LAPIC found during irq injection\n");
764 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
765 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
767 if (kvm->arch.x2apic_broadcast_quirk_disabled) {
768 if ((irq->dest_id == APIC_BROADCAST &&
769 map->mode != KVM_APIC_MODE_X2APIC))
771 if (irq->dest_id == X2APIC_BROADCAST)
774 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
775 if (irq->dest_id == (x2apic_ipi ?
776 X2APIC_BROADCAST : APIC_BROADCAST))
783 /* Return true if the interrupt can be handled by using *bitmap as index mask
784 * for valid destinations in *dst array.
785 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
786 * Note: we may have zero kvm_lapic destinations when we return true, which
787 * means that the interrupt should be dropped. In this case, *bitmap would be
788 * zero and *dst undefined.
790 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
791 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
792 struct kvm_apic_map *map, struct kvm_lapic ***dst,
793 unsigned long *bitmap)
797 if (irq->shorthand == APIC_DEST_SELF && src) {
801 } else if (irq->shorthand)
804 if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
807 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
808 if (irq->dest_id > map->max_apic_id) {
811 *dst = &map->phys_map[irq->dest_id];
818 if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
822 if (!kvm_lowest_prio_delivery(irq))
825 if (!kvm_vector_hashing_enabled()) {
827 for_each_set_bit(i, bitmap, 16) {
832 else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
833 (*dst)[lowest]->vcpu) < 0)
840 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
843 if (!(*dst)[lowest]) {
844 kvm_apic_disabled_lapic_found(kvm);
850 *bitmap = (lowest >= 0) ? 1 << lowest : 0;
855 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
856 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
858 struct kvm_apic_map *map;
859 unsigned long bitmap;
860 struct kvm_lapic **dst = NULL;
866 if (irq->shorthand == APIC_DEST_SELF) {
867 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
872 map = rcu_dereference(kvm->arch.apic_map);
874 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
876 for_each_set_bit(i, &bitmap, 16) {
881 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
889 * This routine tries to handler interrupts in posted mode, here is how
890 * it deals with different cases:
891 * - For single-destination interrupts, handle it in posted mode
892 * - Else if vector hashing is enabled and it is a lowest-priority
893 * interrupt, handle it in posted mode and use the following mechanism
894 * to find the destinaiton vCPU.
895 * 1. For lowest-priority interrupts, store all the possible
896 * destination vCPUs in an array.
897 * 2. Use "guest vector % max number of destination vCPUs" to find
898 * the right destination vCPU in the array for the lowest-priority
900 * - Otherwise, use remapped mode to inject the interrupt.
902 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
903 struct kvm_vcpu **dest_vcpu)
905 struct kvm_apic_map *map;
906 unsigned long bitmap;
907 struct kvm_lapic **dst = NULL;
914 map = rcu_dereference(kvm->arch.apic_map);
916 if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
917 hweight16(bitmap) == 1) {
918 unsigned long i = find_first_bit(&bitmap, 16);
921 *dest_vcpu = dst[i]->vcpu;
931 * Add a pending IRQ into lapic.
932 * Return 1 if successfully added and 0 if discarded.
934 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
935 int vector, int level, int trig_mode,
936 struct dest_map *dest_map)
939 struct kvm_vcpu *vcpu = apic->vcpu;
941 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
943 switch (delivery_mode) {
945 vcpu->arch.apic_arb_prio++;
947 if (unlikely(trig_mode && !level))
950 /* FIXME add logic for vcpu on reset */
951 if (unlikely(!apic_enabled(apic)))
957 __set_bit(vcpu->vcpu_id, dest_map->map);
958 dest_map->vectors[vcpu->vcpu_id] = vector;
961 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
963 kvm_lapic_set_vector(vector, apic->regs + APIC_TMR);
965 apic_clear_vector(vector, apic->regs + APIC_TMR);
968 if (vcpu->arch.apicv_active)
969 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
971 kvm_lapic_set_irr(vector, apic);
973 kvm_make_request(KVM_REQ_EVENT, vcpu);
980 vcpu->arch.pv.pv_unhalted = 1;
981 kvm_make_request(KVM_REQ_EVENT, vcpu);
987 kvm_make_request(KVM_REQ_SMI, vcpu);
993 kvm_inject_nmi(vcpu);
998 if (!trig_mode || level) {
1000 /* assumes that there are only KVM_APIC_INIT/SIPI */
1001 apic->pending_events = (1UL << KVM_APIC_INIT);
1002 /* make sure pending_events is visible before sending
1005 kvm_make_request(KVM_REQ_EVENT, vcpu);
1006 kvm_vcpu_kick(vcpu);
1008 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
1013 case APIC_DM_STARTUP:
1014 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
1015 vcpu->vcpu_id, vector);
1017 apic->sipi_vector = vector;
1018 /* make sure sipi_vector is visible for the receiver */
1020 set_bit(KVM_APIC_SIPI, &apic->pending_events);
1021 kvm_make_request(KVM_REQ_EVENT, vcpu);
1022 kvm_vcpu_kick(vcpu);
1025 case APIC_DM_EXTINT:
1027 * Should only be called by kvm_apic_local_deliver() with LVT0,
1028 * before NMI watchdog was enabled. Already handled by
1029 * kvm_apic_accept_pic_intr().
1034 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1041 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1043 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1046 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1048 return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1051 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1055 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1056 if (!kvm_ioapic_handles_vector(apic, vector))
1059 /* Request a KVM exit to inform the userspace IOAPIC. */
1060 if (irqchip_split(apic->vcpu->kvm)) {
1061 apic->vcpu->arch.pending_ioapic_eoi = vector;
1062 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1066 if (apic_test_vector(vector, apic->regs + APIC_TMR))
1067 trigger_mode = IOAPIC_LEVEL_TRIG;
1069 trigger_mode = IOAPIC_EDGE_TRIG;
1071 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1074 static int apic_set_eoi(struct kvm_lapic *apic)
1076 int vector = apic_find_highest_isr(apic);
1078 trace_kvm_eoi(apic, vector);
1081 * Not every write EOI will has corresponding ISR,
1082 * one example is when Kernel check timer on setup_IO_APIC
1087 apic_clear_isr(vector, apic);
1088 apic_update_ppr(apic);
1090 if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1091 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1093 kvm_ioapic_send_eoi(apic, vector);
1094 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1099 * this interface assumes a trap-like exit, which has already finished
1100 * desired side effect including vISR and vPPR update.
1102 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1104 struct kvm_lapic *apic = vcpu->arch.apic;
1106 trace_kvm_eoi(apic, vector);
1108 kvm_ioapic_send_eoi(apic, vector);
1109 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1111 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1113 static void apic_send_ipi(struct kvm_lapic *apic)
1115 u32 icr_low = kvm_lapic_get_reg(apic, APIC_ICR);
1116 u32 icr_high = kvm_lapic_get_reg(apic, APIC_ICR2);
1117 struct kvm_lapic_irq irq;
1119 irq.vector = icr_low & APIC_VECTOR_MASK;
1120 irq.delivery_mode = icr_low & APIC_MODE_MASK;
1121 irq.dest_mode = icr_low & APIC_DEST_MASK;
1122 irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1123 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1124 irq.shorthand = icr_low & APIC_SHORT_MASK;
1125 irq.msi_redir_hint = false;
1126 if (apic_x2apic_mode(apic))
1127 irq.dest_id = icr_high;
1129 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1131 trace_kvm_apic_ipi(icr_low, irq.dest_id);
1133 apic_debug("icr_high 0x%x, icr_low 0x%x, "
1134 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1135 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1136 "msi_redir_hint 0x%x\n",
1137 icr_high, icr_low, irq.shorthand, irq.dest_id,
1138 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
1139 irq.vector, irq.msi_redir_hint);
1141 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1144 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1146 ktime_t remaining, now;
1150 ASSERT(apic != NULL);
1152 /* if initial count is 0, current count should also be 0 */
1153 if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1154 apic->lapic_timer.period == 0)
1158 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1159 if (ktime_to_ns(remaining) < 0)
1162 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1163 tmcct = div64_u64(ns,
1164 (APIC_BUS_CYCLE_NS * apic->divide_count));
1169 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1171 struct kvm_vcpu *vcpu = apic->vcpu;
1172 struct kvm_run *run = vcpu->run;
1174 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1175 run->tpr_access.rip = kvm_rip_read(vcpu);
1176 run->tpr_access.is_write = write;
1179 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1181 if (apic->vcpu->arch.tpr_access_reporting)
1182 __report_tpr_access(apic, write);
1185 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1189 if (offset >= LAPIC_MMIO_LENGTH)
1194 apic_debug("Access APIC ARBPRI register which is for P6\n");
1197 case APIC_TMCCT: /* Timer CCR */
1198 if (apic_lvtt_tscdeadline(apic))
1201 val = apic_get_tmcct(apic);
1204 apic_update_ppr(apic);
1205 val = kvm_lapic_get_reg(apic, offset);
1208 report_tpr_access(apic, false);
1211 val = kvm_lapic_get_reg(apic, offset);
1218 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1220 return container_of(dev, struct kvm_lapic, dev);
1223 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1226 unsigned char alignment = offset & 0xf;
1228 /* this bitmask has a bit cleared for each reserved register */
1229 static const u64 rmask = 0x43ff01ffffffe70cULL;
1231 if ((alignment + len) > 4) {
1232 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1237 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1238 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1243 result = __apic_read(apic, offset & ~0xf);
1245 trace_kvm_apic_read(offset, result);
1251 memcpy(data, (char *)&result + alignment, len);
1254 printk(KERN_ERR "Local APIC read with len = %x, "
1255 "should be 1,2, or 4 instead\n", len);
1260 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1262 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1264 return kvm_apic_hw_enabled(apic) &&
1265 addr >= apic->base_address &&
1266 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1269 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1270 gpa_t address, int len, void *data)
1272 struct kvm_lapic *apic = to_lapic(this);
1273 u32 offset = address - apic->base_address;
1275 if (!apic_mmio_in_range(apic, address))
1278 kvm_lapic_reg_read(apic, offset, len, data);
1283 static void update_divide_count(struct kvm_lapic *apic)
1285 u32 tmp1, tmp2, tdcr;
1287 tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1289 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1290 apic->divide_count = 0x1 << (tmp2 & 0x7);
1292 apic_debug("timer divide count is 0x%x\n",
1293 apic->divide_count);
1296 static void apic_update_lvtt(struct kvm_lapic *apic)
1298 u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1299 apic->lapic_timer.timer_mode_mask;
1301 if (apic->lapic_timer.timer_mode != timer_mode) {
1302 apic->lapic_timer.timer_mode = timer_mode;
1303 hrtimer_cancel(&apic->lapic_timer.timer);
1307 static void apic_timer_expired(struct kvm_lapic *apic)
1309 struct kvm_vcpu *vcpu = apic->vcpu;
1310 struct swait_queue_head *q = &vcpu->wq;
1311 struct kvm_timer *ktimer = &apic->lapic_timer;
1313 if (atomic_read(&apic->lapic_timer.pending))
1316 atomic_inc(&apic->lapic_timer.pending);
1317 kvm_set_pending_timer(vcpu);
1319 if (swait_active(q))
1322 if (apic_lvtt_tscdeadline(apic))
1323 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1327 * On APICv, this test will cause a busy wait
1328 * during a higher-priority task.
1331 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1333 struct kvm_lapic *apic = vcpu->arch.apic;
1334 u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1336 if (kvm_apic_hw_enabled(apic)) {
1337 int vec = reg & APIC_VECTOR_MASK;
1338 void *bitmap = apic->regs + APIC_ISR;
1340 if (vcpu->arch.apicv_active)
1341 bitmap = apic->regs + APIC_IRR;
1343 if (apic_test_vector(vec, bitmap))
1349 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1351 struct kvm_lapic *apic = vcpu->arch.apic;
1352 u64 guest_tsc, tsc_deadline;
1354 if (!lapic_in_kernel(vcpu))
1357 if (apic->lapic_timer.expired_tscdeadline == 0)
1360 if (!lapic_timer_int_injected(vcpu))
1363 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1364 apic->lapic_timer.expired_tscdeadline = 0;
1365 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1366 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1368 /* __delay is delay_tsc whenever the hardware has TSC, thus always. */
1369 if (guest_tsc < tsc_deadline)
1370 __delay(min(tsc_deadline - guest_tsc,
1371 nsec_to_cycles(vcpu, lapic_timer_advance_ns)));
1374 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1376 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1379 struct kvm_vcpu *vcpu = apic->vcpu;
1380 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1381 unsigned long flags;
1384 if (unlikely(!tscdeadline || !this_tsc_khz))
1387 local_irq_save(flags);
1390 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1391 if (likely(tscdeadline > guest_tsc)) {
1392 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1393 do_div(ns, this_tsc_khz);
1394 expire = ktime_add_ns(now, ns);
1395 expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1396 hrtimer_start(&apic->lapic_timer.timer,
1397 expire, HRTIMER_MODE_ABS_PINNED);
1399 apic_timer_expired(apic);
1401 local_irq_restore(flags);
1404 static void start_sw_period(struct kvm_lapic *apic)
1406 if (!apic->lapic_timer.period)
1409 if (apic_lvtt_oneshot(apic) &&
1410 ktime_after(ktime_get(),
1411 apic->lapic_timer.target_expiration)) {
1412 apic_timer_expired(apic);
1416 hrtimer_start(&apic->lapic_timer.timer,
1417 apic->lapic_timer.target_expiration,
1418 HRTIMER_MODE_ABS_PINNED);
1421 static bool set_target_expiration(struct kvm_lapic *apic)
1427 apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1428 * APIC_BUS_CYCLE_NS * apic->divide_count;
1430 if (!apic->lapic_timer.period)
1434 * Do not allow the guest to program periodic timers with small
1435 * interval, since the hrtimers are not throttled by the host
1438 if (apic_lvtt_period(apic)) {
1439 s64 min_period = min_timer_period_us * 1000LL;
1441 if (apic->lapic_timer.period < min_period) {
1442 pr_info_ratelimited(
1443 "kvm: vcpu %i: requested %lld ns "
1444 "lapic timer period limited to %lld ns\n",
1445 apic->vcpu->vcpu_id,
1446 apic->lapic_timer.period, min_period);
1447 apic->lapic_timer.period = min_period;
1451 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1453 "timer initial count 0x%x, period %lldns, "
1454 "expire @ 0x%016" PRIx64 ".\n", __func__,
1455 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1456 kvm_lapic_get_reg(apic, APIC_TMICT),
1457 apic->lapic_timer.period,
1458 ktime_to_ns(ktime_add_ns(now,
1459 apic->lapic_timer.period)));
1461 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1462 nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1463 apic->lapic_timer.target_expiration = ktime_add_ns(now, apic->lapic_timer.period);
1468 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1470 apic->lapic_timer.tscdeadline +=
1471 nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1472 apic->lapic_timer.target_expiration =
1473 ktime_add_ns(apic->lapic_timer.target_expiration,
1474 apic->lapic_timer.period);
1477 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1479 if (!lapic_in_kernel(vcpu))
1482 return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1484 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1486 static void cancel_hv_timer(struct kvm_lapic *apic)
1488 kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1489 apic->lapic_timer.hv_timer_in_use = false;
1492 static bool start_hv_timer(struct kvm_lapic *apic)
1494 u64 tscdeadline = apic->lapic_timer.tscdeadline;
1496 if ((atomic_read(&apic->lapic_timer.pending) &&
1497 !apic_lvtt_period(apic)) ||
1498 kvm_x86_ops->set_hv_timer(apic->vcpu, tscdeadline)) {
1499 if (apic->lapic_timer.hv_timer_in_use)
1500 cancel_hv_timer(apic);
1502 apic->lapic_timer.hv_timer_in_use = true;
1503 hrtimer_cancel(&apic->lapic_timer.timer);
1505 /* In case the sw timer triggered in the window */
1506 if (atomic_read(&apic->lapic_timer.pending) &&
1507 !apic_lvtt_period(apic))
1508 cancel_hv_timer(apic);
1510 trace_kvm_hv_timer_state(apic->vcpu->vcpu_id,
1511 apic->lapic_timer.hv_timer_in_use);
1512 return apic->lapic_timer.hv_timer_in_use;
1515 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1517 struct kvm_lapic *apic = vcpu->arch.apic;
1519 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1520 WARN_ON(swait_active(&vcpu->wq));
1521 cancel_hv_timer(apic);
1522 apic_timer_expired(apic);
1524 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1525 advance_periodic_target_expiration(apic);
1526 if (!start_hv_timer(apic))
1527 start_sw_period(apic);
1530 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1532 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1534 struct kvm_lapic *apic = vcpu->arch.apic;
1536 WARN_ON(apic->lapic_timer.hv_timer_in_use);
1538 start_hv_timer(apic);
1540 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1542 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1544 struct kvm_lapic *apic = vcpu->arch.apic;
1546 /* Possibly the TSC deadline timer is not enabled yet */
1547 if (!apic->lapic_timer.hv_timer_in_use)
1550 cancel_hv_timer(apic);
1552 if (atomic_read(&apic->lapic_timer.pending))
1555 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1556 start_sw_period(apic);
1557 else if (apic_lvtt_tscdeadline(apic))
1558 start_sw_tscdeadline(apic);
1560 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1562 static void start_apic_timer(struct kvm_lapic *apic)
1564 atomic_set(&apic->lapic_timer.pending, 0);
1566 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1567 if (set_target_expiration(apic) &&
1568 !(kvm_x86_ops->set_hv_timer && start_hv_timer(apic)))
1569 start_sw_period(apic);
1570 } else if (apic_lvtt_tscdeadline(apic)) {
1571 if (!(kvm_x86_ops->set_hv_timer && start_hv_timer(apic)))
1572 start_sw_tscdeadline(apic);
1576 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1578 bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1580 if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1581 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1582 if (lvt0_in_nmi_mode) {
1583 apic_debug("Receive NMI setting on APIC_LVT0 "
1584 "for cpu %d\n", apic->vcpu->vcpu_id);
1585 atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1587 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1591 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1595 trace_kvm_apic_write(reg, val);
1598 case APIC_ID: /* Local APIC ID */
1599 if (!apic_x2apic_mode(apic))
1600 kvm_apic_set_xapic_id(apic, val >> 24);
1606 report_tpr_access(apic, true);
1607 apic_set_tpr(apic, val & 0xff);
1615 if (!apic_x2apic_mode(apic))
1616 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1622 if (!apic_x2apic_mode(apic)) {
1623 kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1624 recalculate_apic_map(apic->vcpu->kvm);
1631 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1632 mask |= APIC_SPIV_DIRECTED_EOI;
1633 apic_set_spiv(apic, val & mask);
1634 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1638 for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
1639 lvt_val = kvm_lapic_get_reg(apic,
1640 APIC_LVTT + 0x10 * i);
1641 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
1642 lvt_val | APIC_LVT_MASKED);
1644 apic_update_lvtt(apic);
1645 atomic_set(&apic->lapic_timer.pending, 0);
1651 /* No delay here, so we always clear the pending bit */
1652 kvm_lapic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1653 apic_send_ipi(apic);
1657 if (!apic_x2apic_mode(apic))
1659 kvm_lapic_set_reg(apic, APIC_ICR2, val);
1663 apic_manage_nmi_watchdog(apic, val);
1668 /* TODO: Check vector */
1669 if (!kvm_apic_sw_enabled(apic))
1670 val |= APIC_LVT_MASKED;
1672 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1673 kvm_lapic_set_reg(apic, reg, val);
1678 if (!kvm_apic_sw_enabled(apic))
1679 val |= APIC_LVT_MASKED;
1680 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1681 kvm_lapic_set_reg(apic, APIC_LVTT, val);
1682 apic_update_lvtt(apic);
1686 if (apic_lvtt_tscdeadline(apic))
1689 hrtimer_cancel(&apic->lapic_timer.timer);
1690 kvm_lapic_set_reg(apic, APIC_TMICT, val);
1691 start_apic_timer(apic);
1696 apic_debug("KVM_WRITE:TDCR %x\n", val);
1697 kvm_lapic_set_reg(apic, APIC_TDCR, val);
1698 update_divide_count(apic);
1702 if (apic_x2apic_mode(apic) && val != 0) {
1703 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1709 if (apic_x2apic_mode(apic)) {
1710 kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1719 apic_debug("Local APIC Write to read-only register %x\n", reg);
1722 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
1724 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1725 gpa_t address, int len, const void *data)
1727 struct kvm_lapic *apic = to_lapic(this);
1728 unsigned int offset = address - apic->base_address;
1731 if (!apic_mmio_in_range(apic, address))
1735 * APIC register must be aligned on 128-bits boundary.
1736 * 32/64/128 bits registers must be accessed thru 32 bits.
1739 if (len != 4 || (offset & 0xf)) {
1740 /* Don't shout loud, $infamous_os would cause only noise. */
1741 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1747 /* too common printing */
1748 if (offset != APIC_EOI)
1749 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1750 "0x%x\n", __func__, offset, len, val);
1752 kvm_lapic_reg_write(apic, offset & 0xff0, val);
1757 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1759 kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1761 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1763 /* emulate APIC access in a trap manner */
1764 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1768 /* hw has done the conditional check and inst decode */
1771 kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
1773 /* TODO: optimize to just emulate side effect w/o one more write */
1774 kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
1776 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1778 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1780 struct kvm_lapic *apic = vcpu->arch.apic;
1782 if (!vcpu->arch.apic)
1785 hrtimer_cancel(&apic->lapic_timer.timer);
1787 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1788 static_key_slow_dec_deferred(&apic_hw_disabled);
1790 if (!apic->sw_enabled)
1791 static_key_slow_dec_deferred(&apic_sw_disabled);
1794 free_page((unsigned long)apic->regs);
1800 *----------------------------------------------------------------------
1802 *----------------------------------------------------------------------
1804 u64 kvm_get_lapic_target_expiration_tsc(struct kvm_vcpu *vcpu)
1806 struct kvm_lapic *apic = vcpu->arch.apic;
1808 if (!lapic_in_kernel(vcpu))
1811 return apic->lapic_timer.tscdeadline;
1814 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1816 struct kvm_lapic *apic = vcpu->arch.apic;
1818 if (!lapic_in_kernel(vcpu) ||
1819 !apic_lvtt_tscdeadline(apic))
1822 return apic->lapic_timer.tscdeadline;
1825 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1827 struct kvm_lapic *apic = vcpu->arch.apic;
1829 if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
1830 apic_lvtt_period(apic))
1833 hrtimer_cancel(&apic->lapic_timer.timer);
1834 apic->lapic_timer.tscdeadline = data;
1835 start_apic_timer(apic);
1838 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1840 struct kvm_lapic *apic = vcpu->arch.apic;
1842 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1843 | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
1846 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1850 tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1852 return (tpr & 0xf0) >> 4;
1855 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1857 u64 old_value = vcpu->arch.apic_base;
1858 struct kvm_lapic *apic = vcpu->arch.apic;
1861 value |= MSR_IA32_APICBASE_BSP;
1863 vcpu->arch.apic_base = value;
1865 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
1866 kvm_update_cpuid(vcpu);
1871 /* update jump label if enable bit changes */
1872 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1873 if (value & MSR_IA32_APICBASE_ENABLE) {
1874 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1875 static_key_slow_dec_deferred(&apic_hw_disabled);
1877 static_key_slow_inc(&apic_hw_disabled.key);
1878 recalculate_apic_map(vcpu->kvm);
1882 if ((old_value ^ value) & X2APIC_ENABLE) {
1883 if (value & X2APIC_ENABLE) {
1884 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
1885 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1887 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1890 apic->base_address = apic->vcpu->arch.apic_base &
1891 MSR_IA32_APICBASE_BASE;
1893 if ((value & MSR_IA32_APICBASE_ENABLE) &&
1894 apic->base_address != APIC_DEFAULT_PHYS_BASE)
1895 pr_warn_once("APIC base relocation is unsupported by KVM");
1897 /* with FSB delivery interrupt, we can restart APIC functionality */
1898 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1899 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1903 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
1905 struct kvm_lapic *apic;
1908 apic_debug("%s\n", __func__);
1911 apic = vcpu->arch.apic;
1912 ASSERT(apic != NULL);
1914 /* Stop the timer in case it's a reset to an active apic */
1915 hrtimer_cancel(&apic->lapic_timer.timer);
1918 kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
1919 MSR_IA32_APICBASE_ENABLE);
1920 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1922 kvm_apic_set_version(apic->vcpu);
1924 for (i = 0; i < KVM_APIC_LVT_NUM; i++)
1925 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1926 apic_update_lvtt(apic);
1927 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
1928 kvm_lapic_set_reg(apic, APIC_LVT0,
1929 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1930 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
1932 kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
1933 apic_set_spiv(apic, 0xff);
1934 kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
1935 if (!apic_x2apic_mode(apic))
1936 kvm_apic_set_ldr(apic, 0);
1937 kvm_lapic_set_reg(apic, APIC_ESR, 0);
1938 kvm_lapic_set_reg(apic, APIC_ICR, 0);
1939 kvm_lapic_set_reg(apic, APIC_ICR2, 0);
1940 kvm_lapic_set_reg(apic, APIC_TDCR, 0);
1941 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1942 for (i = 0; i < 8; i++) {
1943 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1944 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1945 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1947 apic->irr_pending = vcpu->arch.apicv_active;
1948 apic->isr_count = vcpu->arch.apicv_active ? 1 : 0;
1949 apic->highest_isr_cache = -1;
1950 update_divide_count(apic);
1951 atomic_set(&apic->lapic_timer.pending, 0);
1952 if (kvm_vcpu_is_bsp(vcpu))
1953 kvm_lapic_set_base(vcpu,
1954 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1955 vcpu->arch.pv_eoi.msr_val = 0;
1956 apic_update_ppr(apic);
1958 vcpu->arch.apic_arb_prio = 0;
1959 vcpu->arch.apic_attention = 0;
1961 apic_debug("%s: vcpu=%p, id=0x%x, base_msr="
1962 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1963 vcpu, kvm_lapic_get_reg(apic, APIC_ID),
1964 vcpu->arch.apic_base, apic->base_address);
1968 *----------------------------------------------------------------------
1970 *----------------------------------------------------------------------
1973 static bool lapic_is_periodic(struct kvm_lapic *apic)
1975 return apic_lvtt_period(apic);
1978 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1980 struct kvm_lapic *apic = vcpu->arch.apic;
1982 if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
1983 return atomic_read(&apic->lapic_timer.pending);
1988 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1990 u32 reg = kvm_lapic_get_reg(apic, lvt_type);
1991 int vector, mode, trig_mode;
1993 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1994 vector = reg & APIC_VECTOR_MASK;
1995 mode = reg & APIC_MODE_MASK;
1996 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1997 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2003 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2005 struct kvm_lapic *apic = vcpu->arch.apic;
2008 kvm_apic_local_deliver(apic, APIC_LVT0);
2011 static const struct kvm_io_device_ops apic_mmio_ops = {
2012 .read = apic_mmio_read,
2013 .write = apic_mmio_write,
2016 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2018 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2019 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2021 apic_timer_expired(apic);
2023 if (lapic_is_periodic(apic)) {
2024 advance_periodic_target_expiration(apic);
2025 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2026 return HRTIMER_RESTART;
2028 return HRTIMER_NORESTART;
2031 int kvm_create_lapic(struct kvm_vcpu *vcpu)
2033 struct kvm_lapic *apic;
2035 ASSERT(vcpu != NULL);
2036 apic_debug("apic_init %d\n", vcpu->vcpu_id);
2038 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
2042 vcpu->arch.apic = apic;
2044 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
2046 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2048 goto nomem_free_apic;
2052 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2053 HRTIMER_MODE_ABS_PINNED);
2054 apic->lapic_timer.timer.function = apic_timer_fn;
2057 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2058 * thinking that APIC satet has changed.
2060 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2061 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2062 kvm_lapic_reset(vcpu, false);
2063 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2072 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2074 struct kvm_lapic *apic = vcpu->arch.apic;
2077 if (!apic_enabled(apic))
2080 __apic_update_ppr(apic, &ppr);
2081 return apic_has_interrupt_for_ppr(apic, ppr);
2084 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2086 u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2089 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2091 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2092 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2097 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2099 struct kvm_lapic *apic = vcpu->arch.apic;
2101 if (atomic_read(&apic->lapic_timer.pending) > 0) {
2102 kvm_apic_local_deliver(apic, APIC_LVTT);
2103 if (apic_lvtt_tscdeadline(apic))
2104 apic->lapic_timer.tscdeadline = 0;
2105 if (apic_lvtt_oneshot(apic)) {
2106 apic->lapic_timer.tscdeadline = 0;
2107 apic->lapic_timer.target_expiration = 0;
2109 atomic_set(&apic->lapic_timer.pending, 0);
2113 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2115 int vector = kvm_apic_has_interrupt(vcpu);
2116 struct kvm_lapic *apic = vcpu->arch.apic;
2122 * We get here even with APIC virtualization enabled, if doing
2123 * nested virtualization and L1 runs with the "acknowledge interrupt
2124 * on exit" mode. Then we cannot inject the interrupt via RVI,
2125 * because the process would deliver it through the IDT.
2128 apic_set_isr(vector, apic);
2129 apic_update_ppr(apic);
2130 apic_clear_irr(vector, apic);
2132 if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2133 apic_clear_isr(vector, apic);
2134 apic_update_ppr(apic);
2140 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2141 struct kvm_lapic_state *s, bool set)
2143 if (apic_x2apic_mode(vcpu->arch.apic)) {
2144 u32 *id = (u32 *)(s->regs + APIC_ID);
2146 if (vcpu->kvm->arch.x2apic_format) {
2147 if (*id != vcpu->vcpu_id)
2160 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2162 memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2163 return kvm_apic_state_fixup(vcpu, s, false);
2166 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2168 struct kvm_lapic *apic = vcpu->arch.apic;
2172 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2173 /* set SPIV separately to get count of SW disabled APICs right */
2174 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2176 r = kvm_apic_state_fixup(vcpu, s, true);
2179 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2181 recalculate_apic_map(vcpu->kvm);
2182 kvm_apic_set_version(vcpu);
2184 apic_update_ppr(apic);
2185 hrtimer_cancel(&apic->lapic_timer.timer);
2186 apic_update_lvtt(apic);
2187 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2188 update_divide_count(apic);
2189 start_apic_timer(apic);
2190 apic->irr_pending = true;
2191 apic->isr_count = vcpu->arch.apicv_active ?
2192 1 : count_vectors(apic->regs + APIC_ISR);
2193 apic->highest_isr_cache = -1;
2194 if (vcpu->arch.apicv_active) {
2195 if (kvm_x86_ops->apicv_post_state_restore)
2196 kvm_x86_ops->apicv_post_state_restore(vcpu);
2197 kvm_x86_ops->hwapic_irr_update(vcpu,
2198 apic_find_highest_irr(apic));
2199 kvm_x86_ops->hwapic_isr_update(vcpu,
2200 apic_find_highest_isr(apic));
2202 kvm_make_request(KVM_REQ_EVENT, vcpu);
2203 if (ioapic_in_kernel(vcpu->kvm))
2204 kvm_rtc_eoi_tracking_restore_one(vcpu);
2206 vcpu->arch.apic_arb_prio = 0;
2211 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2213 struct hrtimer *timer;
2215 if (!lapic_in_kernel(vcpu))
2218 timer = &vcpu->arch.apic->lapic_timer.timer;
2219 if (hrtimer_cancel(timer))
2220 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
2224 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2226 * Detect whether guest triggered PV EOI since the
2227 * last entry. If yes, set EOI on guests's behalf.
2228 * Clear PV EOI in guest memory in any case.
2230 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2231 struct kvm_lapic *apic)
2236 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2237 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2239 * KVM_APIC_PV_EOI_PENDING is unset:
2240 * -> host disabled PV EOI.
2241 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2242 * -> host enabled PV EOI, guest did not execute EOI yet.
2243 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2244 * -> host enabled PV EOI, guest executed EOI.
2246 BUG_ON(!pv_eoi_enabled(vcpu));
2247 pending = pv_eoi_get_pending(vcpu);
2249 * Clear pending bit in any case: it will be set again on vmentry.
2250 * While this might not be ideal from performance point of view,
2251 * this makes sure pv eoi is only enabled when we know it's safe.
2253 pv_eoi_clr_pending(vcpu);
2256 vector = apic_set_eoi(apic);
2257 trace_kvm_pv_eoi(apic, vector);
2260 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2264 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2265 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2267 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2270 if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2274 apic_set_tpr(vcpu->arch.apic, data & 0xff);
2278 * apic_sync_pv_eoi_to_guest - called before vmentry
2280 * Detect whether it's safe to enable PV EOI and
2283 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2284 struct kvm_lapic *apic)
2286 if (!pv_eoi_enabled(vcpu) ||
2287 /* IRR set or many bits in ISR: could be nested. */
2288 apic->irr_pending ||
2289 /* Cache not set: could be safe but we don't bother. */
2290 apic->highest_isr_cache == -1 ||
2291 /* Need EOI to update ioapic. */
2292 kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2294 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2295 * so we need not do anything here.
2300 pv_eoi_set_pending(apic->vcpu);
2303 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2306 int max_irr, max_isr;
2307 struct kvm_lapic *apic = vcpu->arch.apic;
2309 apic_sync_pv_eoi_to_guest(vcpu, apic);
2311 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2314 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2315 max_irr = apic_find_highest_irr(apic);
2318 max_isr = apic_find_highest_isr(apic);
2321 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2323 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2327 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2330 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2331 &vcpu->arch.apic->vapic_cache,
2332 vapic_addr, sizeof(u32)))
2334 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2336 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2339 vcpu->arch.apic->vapic_addr = vapic_addr;
2343 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2345 struct kvm_lapic *apic = vcpu->arch.apic;
2346 u32 reg = (msr - APIC_BASE_MSR) << 4;
2348 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2351 if (reg == APIC_ICR2)
2354 /* if this is ICR write vector before command */
2355 if (reg == APIC_ICR)
2356 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2357 return kvm_lapic_reg_write(apic, reg, (u32)data);
2360 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2362 struct kvm_lapic *apic = vcpu->arch.apic;
2363 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2365 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2368 if (reg == APIC_DFR || reg == APIC_ICR2) {
2369 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2374 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2376 if (reg == APIC_ICR)
2377 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2379 *data = (((u64)high) << 32) | low;
2384 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2386 struct kvm_lapic *apic = vcpu->arch.apic;
2388 if (!lapic_in_kernel(vcpu))
2391 /* if this is ICR write vector before command */
2392 if (reg == APIC_ICR)
2393 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2394 return kvm_lapic_reg_write(apic, reg, (u32)data);
2397 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2399 struct kvm_lapic *apic = vcpu->arch.apic;
2402 if (!lapic_in_kernel(vcpu))
2405 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2407 if (reg == APIC_ICR)
2408 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2410 *data = (((u64)high) << 32) | low;
2415 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2417 u64 addr = data & ~KVM_MSR_ENABLED;
2418 if (!IS_ALIGNED(addr, 4))
2421 vcpu->arch.pv_eoi.msr_val = data;
2422 if (!pv_eoi_enabled(vcpu))
2424 return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2428 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2430 struct kvm_lapic *apic = vcpu->arch.apic;
2434 if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2438 * INITs are latched while in SMM. Because an SMM CPU cannot
2439 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2440 * and delay processing of INIT until the next RSM.
2443 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2444 if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2445 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2449 pe = xchg(&apic->pending_events, 0);
2450 if (test_bit(KVM_APIC_INIT, &pe)) {
2451 kvm_lapic_reset(vcpu, true);
2452 kvm_vcpu_reset(vcpu, true);
2453 if (kvm_vcpu_is_bsp(apic->vcpu))
2454 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2456 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2458 if (test_bit(KVM_APIC_SIPI, &pe) &&
2459 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2460 /* evaluate pending_events before reading the vector */
2462 sipi_vector = apic->sipi_vector;
2463 apic_debug("vcpu %d received sipi with vector # %x\n",
2464 vcpu->vcpu_id, sipi_vector);
2465 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2466 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2470 void kvm_lapic_init(void)
2472 /* do not patch jump label more than once per second */
2473 jump_label_rate_limit(&apic_hw_disabled, HZ);
2474 jump_label_rate_limit(&apic_sw_disabled, HZ);