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) + 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 int __kvm_apic_update_irr(u32 *pir, void *regs)
367 u32 pir_val, irr_val;
370 for (i = vec = 0; i <= 7; i++, vec += 32) {
371 pir_val = READ_ONCE(pir[i]);
372 irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
374 irr_val |= xchg(&pir[i], 0);
375 *((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
378 max_irr = __fls(irr_val) + vec;
383 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
385 int kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
387 struct kvm_lapic *apic = vcpu->arch.apic;
390 max_irr = __kvm_apic_update_irr(pir, apic->regs);
392 kvm_make_request(KVM_REQ_EVENT, vcpu);
395 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
397 static inline int apic_search_irr(struct kvm_lapic *apic)
399 return find_highest_vector(apic->regs + APIC_IRR);
402 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
407 * Note that irr_pending is just a hint. It will be always
408 * true with virtual interrupt delivery enabled.
410 if (!apic->irr_pending)
413 result = apic_search_irr(apic);
414 ASSERT(result == -1 || result >= 16);
419 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
421 struct kvm_vcpu *vcpu;
425 if (unlikely(vcpu->arch.apicv_active)) {
426 /* try to update RVI */
427 apic_clear_vector(vec, apic->regs + APIC_IRR);
428 kvm_make_request(KVM_REQ_EVENT, vcpu);
430 apic->irr_pending = false;
431 apic_clear_vector(vec, apic->regs + APIC_IRR);
432 if (apic_search_irr(apic) != -1)
433 apic->irr_pending = true;
437 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
439 struct kvm_vcpu *vcpu;
441 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
447 * With APIC virtualization enabled, all caching is disabled
448 * because the processor can modify ISR under the hood. Instead
451 if (unlikely(vcpu->arch.apicv_active))
452 kvm_x86_ops->hwapic_isr_update(vcpu, vec);
455 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
457 * ISR (in service register) bit is set when injecting an interrupt.
458 * The highest vector is injected. Thus the latest bit set matches
459 * the highest bit in ISR.
461 apic->highest_isr_cache = vec;
465 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
470 * Note that isr_count is always 1, and highest_isr_cache
471 * is always -1, with APIC virtualization enabled.
473 if (!apic->isr_count)
475 if (likely(apic->highest_isr_cache != -1))
476 return apic->highest_isr_cache;
478 result = find_highest_vector(apic->regs + APIC_ISR);
479 ASSERT(result == -1 || result >= 16);
484 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
486 struct kvm_vcpu *vcpu;
487 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
493 * We do get here for APIC virtualization enabled if the guest
494 * uses the Hyper-V APIC enlightenment. In this case we may need
495 * to trigger a new interrupt delivery by writing the SVI field;
496 * on the other hand isr_count and highest_isr_cache are unused
497 * and must be left alone.
499 if (unlikely(vcpu->arch.apicv_active))
500 kvm_x86_ops->hwapic_isr_update(vcpu,
501 apic_find_highest_isr(apic));
504 BUG_ON(apic->isr_count < 0);
505 apic->highest_isr_cache = -1;
509 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
511 /* This may race with setting of irr in __apic_accept_irq() and
512 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
513 * will cause vmexit immediately and the value will be recalculated
514 * on the next vmentry.
516 return apic_find_highest_irr(vcpu->arch.apic);
519 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
520 int vector, int level, int trig_mode,
521 struct dest_map *dest_map);
523 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
524 struct dest_map *dest_map)
526 struct kvm_lapic *apic = vcpu->arch.apic;
528 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
529 irq->level, irq->trig_mode, dest_map);
532 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
535 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
539 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
542 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
546 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
548 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
551 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
554 if (pv_eoi_get_user(vcpu, &val) < 0)
555 apic_debug("Can't read EOI MSR value: 0x%llx\n",
556 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
560 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
562 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
563 apic_debug("Can't set EOI MSR value: 0x%llx\n",
564 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
567 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
570 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
572 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
573 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
574 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
577 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
580 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
583 if (apic->vcpu->arch.apicv_active)
584 kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
585 highest_irr = apic_find_highest_irr(apic);
586 if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
591 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
593 u32 tpr, isrv, ppr, old_ppr;
596 old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
597 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
598 isr = apic_find_highest_isr(apic);
599 isrv = (isr != -1) ? isr : 0;
601 if ((tpr & 0xf0) >= (isrv & 0xf0))
606 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
607 apic, ppr, isr, isrv);
611 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
613 return ppr < old_ppr;
616 static void apic_update_ppr(struct kvm_lapic *apic)
620 if (__apic_update_ppr(apic, &ppr) &&
621 apic_has_interrupt_for_ppr(apic, ppr) != -1)
622 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
625 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
627 apic_update_ppr(vcpu->arch.apic);
629 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
631 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
633 kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
634 apic_update_ppr(apic);
637 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
639 return mda == (apic_x2apic_mode(apic) ?
640 X2APIC_BROADCAST : APIC_BROADCAST);
643 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
645 if (kvm_apic_broadcast(apic, mda))
648 if (apic_x2apic_mode(apic))
649 return mda == kvm_x2apic_id(apic);
652 * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
653 * it were in x2APIC mode. Hotplugged VCPUs start in xAPIC mode and
654 * this allows unique addressing of VCPUs with APIC ID over 0xff.
655 * The 0xff condition is needed because writeable xAPIC ID.
657 if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
660 return mda == kvm_xapic_id(apic);
663 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
667 if (kvm_apic_broadcast(apic, mda))
670 logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
672 if (apic_x2apic_mode(apic))
673 return ((logical_id >> 16) == (mda >> 16))
674 && (logical_id & mda & 0xffff) != 0;
676 logical_id = GET_APIC_LOGICAL_ID(logical_id);
678 switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
680 return (logical_id & mda) != 0;
681 case APIC_DFR_CLUSTER:
682 return ((logical_id >> 4) == (mda >> 4))
683 && (logical_id & mda & 0xf) != 0;
685 apic_debug("Bad DFR vcpu %d: %08x\n",
686 apic->vcpu->vcpu_id, kvm_lapic_get_reg(apic, APIC_DFR));
691 /* The KVM local APIC implementation has two quirks:
693 * - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
694 * in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
695 * KVM doesn't do that aliasing.
697 * - in-kernel IOAPIC messages have to be delivered directly to
698 * x2APIC, because the kernel does not support interrupt remapping.
699 * In order to support broadcast without interrupt remapping, x2APIC
700 * rewrites the destination of non-IPI messages from APIC_BROADCAST
701 * to X2APIC_BROADCAST.
703 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API. This is
704 * important when userspace wants to use x2APIC-format MSIs, because
705 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
707 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
708 struct kvm_lapic *source, struct kvm_lapic *target)
710 bool ipi = source != NULL;
712 if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
713 !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
714 return X2APIC_BROADCAST;
719 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
720 int short_hand, unsigned int dest, int dest_mode)
722 struct kvm_lapic *target = vcpu->arch.apic;
723 u32 mda = kvm_apic_mda(vcpu, dest, source, target);
725 apic_debug("target %p, source %p, dest 0x%x, "
726 "dest_mode 0x%x, short_hand 0x%x\n",
727 target, source, dest, dest_mode, short_hand);
730 switch (short_hand) {
731 case APIC_DEST_NOSHORT:
732 if (dest_mode == APIC_DEST_PHYSICAL)
733 return kvm_apic_match_physical_addr(target, mda);
735 return kvm_apic_match_logical_addr(target, mda);
737 return target == source;
738 case APIC_DEST_ALLINC:
740 case APIC_DEST_ALLBUT:
741 return target != source;
743 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
748 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
750 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
751 const unsigned long *bitmap, u32 bitmap_size)
756 mod = vector % dest_vcpus;
758 for (i = 0; i <= mod; i++) {
759 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
760 BUG_ON(idx == bitmap_size);
766 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
768 if (!kvm->arch.disabled_lapic_found) {
769 kvm->arch.disabled_lapic_found = true;
771 "Disabled LAPIC found during irq injection\n");
775 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
776 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
778 if (kvm->arch.x2apic_broadcast_quirk_disabled) {
779 if ((irq->dest_id == APIC_BROADCAST &&
780 map->mode != KVM_APIC_MODE_X2APIC))
782 if (irq->dest_id == X2APIC_BROADCAST)
785 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
786 if (irq->dest_id == (x2apic_ipi ?
787 X2APIC_BROADCAST : APIC_BROADCAST))
794 /* Return true if the interrupt can be handled by using *bitmap as index mask
795 * for valid destinations in *dst array.
796 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
797 * Note: we may have zero kvm_lapic destinations when we return true, which
798 * means that the interrupt should be dropped. In this case, *bitmap would be
799 * zero and *dst undefined.
801 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
802 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
803 struct kvm_apic_map *map, struct kvm_lapic ***dst,
804 unsigned long *bitmap)
808 if (irq->shorthand == APIC_DEST_SELF && src) {
812 } else if (irq->shorthand)
815 if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
818 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
819 if (irq->dest_id > map->max_apic_id) {
822 *dst = &map->phys_map[irq->dest_id];
829 if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
833 if (!kvm_lowest_prio_delivery(irq))
836 if (!kvm_vector_hashing_enabled()) {
838 for_each_set_bit(i, bitmap, 16) {
843 else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
844 (*dst)[lowest]->vcpu) < 0)
851 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
854 if (!(*dst)[lowest]) {
855 kvm_apic_disabled_lapic_found(kvm);
861 *bitmap = (lowest >= 0) ? 1 << lowest : 0;
866 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
867 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
869 struct kvm_apic_map *map;
870 unsigned long bitmap;
871 struct kvm_lapic **dst = NULL;
877 if (irq->shorthand == APIC_DEST_SELF) {
878 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
883 map = rcu_dereference(kvm->arch.apic_map);
885 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
887 for_each_set_bit(i, &bitmap, 16) {
892 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
900 * This routine tries to handler interrupts in posted mode, here is how
901 * it deals with different cases:
902 * - For single-destination interrupts, handle it in posted mode
903 * - Else if vector hashing is enabled and it is a lowest-priority
904 * interrupt, handle it in posted mode and use the following mechanism
905 * to find the destinaiton vCPU.
906 * 1. For lowest-priority interrupts, store all the possible
907 * destination vCPUs in an array.
908 * 2. Use "guest vector % max number of destination vCPUs" to find
909 * the right destination vCPU in the array for the lowest-priority
911 * - Otherwise, use remapped mode to inject the interrupt.
913 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
914 struct kvm_vcpu **dest_vcpu)
916 struct kvm_apic_map *map;
917 unsigned long bitmap;
918 struct kvm_lapic **dst = NULL;
925 map = rcu_dereference(kvm->arch.apic_map);
927 if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
928 hweight16(bitmap) == 1) {
929 unsigned long i = find_first_bit(&bitmap, 16);
932 *dest_vcpu = dst[i]->vcpu;
942 * Add a pending IRQ into lapic.
943 * Return 1 if successfully added and 0 if discarded.
945 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
946 int vector, int level, int trig_mode,
947 struct dest_map *dest_map)
950 struct kvm_vcpu *vcpu = apic->vcpu;
952 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
954 switch (delivery_mode) {
956 vcpu->arch.apic_arb_prio++;
958 if (unlikely(trig_mode && !level))
961 /* FIXME add logic for vcpu on reset */
962 if (unlikely(!apic_enabled(apic)))
968 __set_bit(vcpu->vcpu_id, dest_map->map);
969 dest_map->vectors[vcpu->vcpu_id] = vector;
972 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
974 kvm_lapic_set_vector(vector, apic->regs + APIC_TMR);
976 apic_clear_vector(vector, apic->regs + APIC_TMR);
979 if (vcpu->arch.apicv_active)
980 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
982 kvm_lapic_set_irr(vector, apic);
984 kvm_make_request(KVM_REQ_EVENT, vcpu);
991 vcpu->arch.pv.pv_unhalted = 1;
992 kvm_make_request(KVM_REQ_EVENT, vcpu);
998 kvm_make_request(KVM_REQ_SMI, vcpu);
1004 kvm_inject_nmi(vcpu);
1005 kvm_vcpu_kick(vcpu);
1009 if (!trig_mode || level) {
1011 /* assumes that there are only KVM_APIC_INIT/SIPI */
1012 apic->pending_events = (1UL << KVM_APIC_INIT);
1013 /* make sure pending_events is visible before sending
1016 kvm_make_request(KVM_REQ_EVENT, vcpu);
1017 kvm_vcpu_kick(vcpu);
1019 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
1024 case APIC_DM_STARTUP:
1025 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
1026 vcpu->vcpu_id, vector);
1028 apic->sipi_vector = vector;
1029 /* make sure sipi_vector is visible for the receiver */
1031 set_bit(KVM_APIC_SIPI, &apic->pending_events);
1032 kvm_make_request(KVM_REQ_EVENT, vcpu);
1033 kvm_vcpu_kick(vcpu);
1036 case APIC_DM_EXTINT:
1038 * Should only be called by kvm_apic_local_deliver() with LVT0,
1039 * before NMI watchdog was enabled. Already handled by
1040 * kvm_apic_accept_pic_intr().
1045 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1052 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1054 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1057 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1059 return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1062 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1066 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1067 if (!kvm_ioapic_handles_vector(apic, vector))
1070 /* Request a KVM exit to inform the userspace IOAPIC. */
1071 if (irqchip_split(apic->vcpu->kvm)) {
1072 apic->vcpu->arch.pending_ioapic_eoi = vector;
1073 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1077 if (apic_test_vector(vector, apic->regs + APIC_TMR))
1078 trigger_mode = IOAPIC_LEVEL_TRIG;
1080 trigger_mode = IOAPIC_EDGE_TRIG;
1082 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1085 static int apic_set_eoi(struct kvm_lapic *apic)
1087 int vector = apic_find_highest_isr(apic);
1089 trace_kvm_eoi(apic, vector);
1092 * Not every write EOI will has corresponding ISR,
1093 * one example is when Kernel check timer on setup_IO_APIC
1098 apic_clear_isr(vector, apic);
1099 apic_update_ppr(apic);
1101 if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1102 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1104 kvm_ioapic_send_eoi(apic, vector);
1105 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1110 * this interface assumes a trap-like exit, which has already finished
1111 * desired side effect including vISR and vPPR update.
1113 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1115 struct kvm_lapic *apic = vcpu->arch.apic;
1117 trace_kvm_eoi(apic, vector);
1119 kvm_ioapic_send_eoi(apic, vector);
1120 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1122 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1124 static void apic_send_ipi(struct kvm_lapic *apic)
1126 u32 icr_low = kvm_lapic_get_reg(apic, APIC_ICR);
1127 u32 icr_high = kvm_lapic_get_reg(apic, APIC_ICR2);
1128 struct kvm_lapic_irq irq;
1130 irq.vector = icr_low & APIC_VECTOR_MASK;
1131 irq.delivery_mode = icr_low & APIC_MODE_MASK;
1132 irq.dest_mode = icr_low & APIC_DEST_MASK;
1133 irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1134 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1135 irq.shorthand = icr_low & APIC_SHORT_MASK;
1136 irq.msi_redir_hint = false;
1137 if (apic_x2apic_mode(apic))
1138 irq.dest_id = icr_high;
1140 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1142 trace_kvm_apic_ipi(icr_low, irq.dest_id);
1144 apic_debug("icr_high 0x%x, icr_low 0x%x, "
1145 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1146 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1147 "msi_redir_hint 0x%x\n",
1148 icr_high, icr_low, irq.shorthand, irq.dest_id,
1149 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
1150 irq.vector, irq.msi_redir_hint);
1152 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1155 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1157 ktime_t remaining, now;
1161 ASSERT(apic != NULL);
1163 /* if initial count is 0, current count should also be 0 */
1164 if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1165 apic->lapic_timer.period == 0)
1169 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1170 if (ktime_to_ns(remaining) < 0)
1173 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1174 tmcct = div64_u64(ns,
1175 (APIC_BUS_CYCLE_NS * apic->divide_count));
1180 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1182 struct kvm_vcpu *vcpu = apic->vcpu;
1183 struct kvm_run *run = vcpu->run;
1185 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1186 run->tpr_access.rip = kvm_rip_read(vcpu);
1187 run->tpr_access.is_write = write;
1190 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1192 if (apic->vcpu->arch.tpr_access_reporting)
1193 __report_tpr_access(apic, write);
1196 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1200 if (offset >= LAPIC_MMIO_LENGTH)
1205 apic_debug("Access APIC ARBPRI register which is for P6\n");
1208 case APIC_TMCCT: /* Timer CCR */
1209 if (apic_lvtt_tscdeadline(apic))
1212 val = apic_get_tmcct(apic);
1215 apic_update_ppr(apic);
1216 val = kvm_lapic_get_reg(apic, offset);
1219 report_tpr_access(apic, false);
1222 val = kvm_lapic_get_reg(apic, offset);
1229 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1231 return container_of(dev, struct kvm_lapic, dev);
1234 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1237 unsigned char alignment = offset & 0xf;
1239 /* this bitmask has a bit cleared for each reserved register */
1240 static const u64 rmask = 0x43ff01ffffffe70cULL;
1242 if ((alignment + len) > 4) {
1243 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1248 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1249 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1254 result = __apic_read(apic, offset & ~0xf);
1256 trace_kvm_apic_read(offset, result);
1262 memcpy(data, (char *)&result + alignment, len);
1265 printk(KERN_ERR "Local APIC read with len = %x, "
1266 "should be 1,2, or 4 instead\n", len);
1271 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1273 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1275 return kvm_apic_hw_enabled(apic) &&
1276 addr >= apic->base_address &&
1277 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1280 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1281 gpa_t address, int len, void *data)
1283 struct kvm_lapic *apic = to_lapic(this);
1284 u32 offset = address - apic->base_address;
1286 if (!apic_mmio_in_range(apic, address))
1289 kvm_lapic_reg_read(apic, offset, len, data);
1294 static void update_divide_count(struct kvm_lapic *apic)
1296 u32 tmp1, tmp2, tdcr;
1298 tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1300 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1301 apic->divide_count = 0x1 << (tmp2 & 0x7);
1303 apic_debug("timer divide count is 0x%x\n",
1304 apic->divide_count);
1307 static void apic_update_lvtt(struct kvm_lapic *apic)
1309 u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1310 apic->lapic_timer.timer_mode_mask;
1312 if (apic->lapic_timer.timer_mode != timer_mode) {
1313 apic->lapic_timer.timer_mode = timer_mode;
1314 hrtimer_cancel(&apic->lapic_timer.timer);
1318 static void apic_timer_expired(struct kvm_lapic *apic)
1320 struct kvm_vcpu *vcpu = apic->vcpu;
1321 struct swait_queue_head *q = &vcpu->wq;
1322 struct kvm_timer *ktimer = &apic->lapic_timer;
1324 if (atomic_read(&apic->lapic_timer.pending))
1327 atomic_inc(&apic->lapic_timer.pending);
1328 kvm_set_pending_timer(vcpu);
1330 if (swait_active(q))
1333 if (apic_lvtt_tscdeadline(apic))
1334 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1338 * On APICv, this test will cause a busy wait
1339 * during a higher-priority task.
1342 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1344 struct kvm_lapic *apic = vcpu->arch.apic;
1345 u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1347 if (kvm_apic_hw_enabled(apic)) {
1348 int vec = reg & APIC_VECTOR_MASK;
1349 void *bitmap = apic->regs + APIC_ISR;
1351 if (vcpu->arch.apicv_active)
1352 bitmap = apic->regs + APIC_IRR;
1354 if (apic_test_vector(vec, bitmap))
1360 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1362 struct kvm_lapic *apic = vcpu->arch.apic;
1363 u64 guest_tsc, tsc_deadline;
1365 if (!lapic_in_kernel(vcpu))
1368 if (apic->lapic_timer.expired_tscdeadline == 0)
1371 if (!lapic_timer_int_injected(vcpu))
1374 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1375 apic->lapic_timer.expired_tscdeadline = 0;
1376 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1377 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1379 /* __delay is delay_tsc whenever the hardware has TSC, thus always. */
1380 if (guest_tsc < tsc_deadline)
1381 __delay(min(tsc_deadline - guest_tsc,
1382 nsec_to_cycles(vcpu, lapic_timer_advance_ns)));
1385 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1387 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1390 struct kvm_vcpu *vcpu = apic->vcpu;
1391 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1392 unsigned long flags;
1395 if (unlikely(!tscdeadline || !this_tsc_khz))
1398 local_irq_save(flags);
1401 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1402 if (likely(tscdeadline > guest_tsc)) {
1403 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1404 do_div(ns, this_tsc_khz);
1405 expire = ktime_add_ns(now, ns);
1406 expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1407 hrtimer_start(&apic->lapic_timer.timer,
1408 expire, HRTIMER_MODE_ABS_PINNED);
1410 apic_timer_expired(apic);
1412 local_irq_restore(flags);
1415 static void start_sw_period(struct kvm_lapic *apic)
1417 if (!apic->lapic_timer.period)
1420 if (apic_lvtt_oneshot(apic) &&
1421 ktime_after(ktime_get(),
1422 apic->lapic_timer.target_expiration)) {
1423 apic_timer_expired(apic);
1427 hrtimer_start(&apic->lapic_timer.timer,
1428 apic->lapic_timer.target_expiration,
1429 HRTIMER_MODE_ABS_PINNED);
1432 static bool set_target_expiration(struct kvm_lapic *apic)
1438 apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1439 * APIC_BUS_CYCLE_NS * apic->divide_count;
1441 if (!apic->lapic_timer.period)
1445 * Do not allow the guest to program periodic timers with small
1446 * interval, since the hrtimers are not throttled by the host
1449 if (apic_lvtt_period(apic)) {
1450 s64 min_period = min_timer_period_us * 1000LL;
1452 if (apic->lapic_timer.period < min_period) {
1453 pr_info_ratelimited(
1454 "kvm: vcpu %i: requested %lld ns "
1455 "lapic timer period limited to %lld ns\n",
1456 apic->vcpu->vcpu_id,
1457 apic->lapic_timer.period, min_period);
1458 apic->lapic_timer.period = min_period;
1462 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1464 "timer initial count 0x%x, period %lldns, "
1465 "expire @ 0x%016" PRIx64 ".\n", __func__,
1466 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1467 kvm_lapic_get_reg(apic, APIC_TMICT),
1468 apic->lapic_timer.period,
1469 ktime_to_ns(ktime_add_ns(now,
1470 apic->lapic_timer.period)));
1472 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1473 nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1474 apic->lapic_timer.target_expiration = ktime_add_ns(now, apic->lapic_timer.period);
1479 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1481 apic->lapic_timer.tscdeadline +=
1482 nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1483 apic->lapic_timer.target_expiration =
1484 ktime_add_ns(apic->lapic_timer.target_expiration,
1485 apic->lapic_timer.period);
1488 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1490 if (!lapic_in_kernel(vcpu))
1493 return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1495 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1497 static void cancel_hv_timer(struct kvm_lapic *apic)
1499 kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1500 apic->lapic_timer.hv_timer_in_use = false;
1503 static bool start_hv_timer(struct kvm_lapic *apic)
1505 u64 tscdeadline = apic->lapic_timer.tscdeadline;
1507 if ((atomic_read(&apic->lapic_timer.pending) &&
1508 !apic_lvtt_period(apic)) ||
1509 kvm_x86_ops->set_hv_timer(apic->vcpu, tscdeadline)) {
1510 if (apic->lapic_timer.hv_timer_in_use)
1511 cancel_hv_timer(apic);
1513 apic->lapic_timer.hv_timer_in_use = true;
1514 hrtimer_cancel(&apic->lapic_timer.timer);
1516 /* In case the sw timer triggered in the window */
1517 if (atomic_read(&apic->lapic_timer.pending) &&
1518 !apic_lvtt_period(apic))
1519 cancel_hv_timer(apic);
1521 trace_kvm_hv_timer_state(apic->vcpu->vcpu_id,
1522 apic->lapic_timer.hv_timer_in_use);
1523 return apic->lapic_timer.hv_timer_in_use;
1526 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1528 struct kvm_lapic *apic = vcpu->arch.apic;
1530 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1531 WARN_ON(swait_active(&vcpu->wq));
1532 cancel_hv_timer(apic);
1533 apic_timer_expired(apic);
1535 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1536 advance_periodic_target_expiration(apic);
1537 if (!start_hv_timer(apic))
1538 start_sw_period(apic);
1541 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1543 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1545 struct kvm_lapic *apic = vcpu->arch.apic;
1547 WARN_ON(apic->lapic_timer.hv_timer_in_use);
1549 start_hv_timer(apic);
1551 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1553 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1555 struct kvm_lapic *apic = vcpu->arch.apic;
1557 /* Possibly the TSC deadline timer is not enabled yet */
1558 if (!apic->lapic_timer.hv_timer_in_use)
1561 cancel_hv_timer(apic);
1563 if (atomic_read(&apic->lapic_timer.pending))
1566 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1567 start_sw_period(apic);
1568 else if (apic_lvtt_tscdeadline(apic))
1569 start_sw_tscdeadline(apic);
1571 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1573 static void start_apic_timer(struct kvm_lapic *apic)
1575 atomic_set(&apic->lapic_timer.pending, 0);
1577 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1578 if (set_target_expiration(apic) &&
1579 !(kvm_x86_ops->set_hv_timer && start_hv_timer(apic)))
1580 start_sw_period(apic);
1581 } else if (apic_lvtt_tscdeadline(apic)) {
1582 if (!(kvm_x86_ops->set_hv_timer && start_hv_timer(apic)))
1583 start_sw_tscdeadline(apic);
1587 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1589 bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1591 if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1592 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1593 if (lvt0_in_nmi_mode) {
1594 apic_debug("Receive NMI setting on APIC_LVT0 "
1595 "for cpu %d\n", apic->vcpu->vcpu_id);
1596 atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1598 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1602 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1606 trace_kvm_apic_write(reg, val);
1609 case APIC_ID: /* Local APIC ID */
1610 if (!apic_x2apic_mode(apic))
1611 kvm_apic_set_xapic_id(apic, val >> 24);
1617 report_tpr_access(apic, true);
1618 apic_set_tpr(apic, val & 0xff);
1626 if (!apic_x2apic_mode(apic))
1627 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1633 if (!apic_x2apic_mode(apic)) {
1634 kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1635 recalculate_apic_map(apic->vcpu->kvm);
1642 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1643 mask |= APIC_SPIV_DIRECTED_EOI;
1644 apic_set_spiv(apic, val & mask);
1645 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1649 for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
1650 lvt_val = kvm_lapic_get_reg(apic,
1651 APIC_LVTT + 0x10 * i);
1652 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
1653 lvt_val | APIC_LVT_MASKED);
1655 apic_update_lvtt(apic);
1656 atomic_set(&apic->lapic_timer.pending, 0);
1662 /* No delay here, so we always clear the pending bit */
1663 kvm_lapic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1664 apic_send_ipi(apic);
1668 if (!apic_x2apic_mode(apic))
1670 kvm_lapic_set_reg(apic, APIC_ICR2, val);
1674 apic_manage_nmi_watchdog(apic, val);
1679 /* TODO: Check vector */
1680 if (!kvm_apic_sw_enabled(apic))
1681 val |= APIC_LVT_MASKED;
1683 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1684 kvm_lapic_set_reg(apic, reg, val);
1689 if (!kvm_apic_sw_enabled(apic))
1690 val |= APIC_LVT_MASKED;
1691 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1692 kvm_lapic_set_reg(apic, APIC_LVTT, val);
1693 apic_update_lvtt(apic);
1697 if (apic_lvtt_tscdeadline(apic))
1700 hrtimer_cancel(&apic->lapic_timer.timer);
1701 kvm_lapic_set_reg(apic, APIC_TMICT, val);
1702 start_apic_timer(apic);
1707 apic_debug("KVM_WRITE:TDCR %x\n", val);
1708 kvm_lapic_set_reg(apic, APIC_TDCR, val);
1709 update_divide_count(apic);
1713 if (apic_x2apic_mode(apic) && val != 0) {
1714 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1720 if (apic_x2apic_mode(apic)) {
1721 kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1730 apic_debug("Local APIC Write to read-only register %x\n", reg);
1733 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
1735 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1736 gpa_t address, int len, const void *data)
1738 struct kvm_lapic *apic = to_lapic(this);
1739 unsigned int offset = address - apic->base_address;
1742 if (!apic_mmio_in_range(apic, address))
1746 * APIC register must be aligned on 128-bits boundary.
1747 * 32/64/128 bits registers must be accessed thru 32 bits.
1750 if (len != 4 || (offset & 0xf)) {
1751 /* Don't shout loud, $infamous_os would cause only noise. */
1752 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1758 /* too common printing */
1759 if (offset != APIC_EOI)
1760 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1761 "0x%x\n", __func__, offset, len, val);
1763 kvm_lapic_reg_write(apic, offset & 0xff0, val);
1768 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1770 kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1772 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1774 /* emulate APIC access in a trap manner */
1775 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1779 /* hw has done the conditional check and inst decode */
1782 kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
1784 /* TODO: optimize to just emulate side effect w/o one more write */
1785 kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
1787 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1789 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1791 struct kvm_lapic *apic = vcpu->arch.apic;
1793 if (!vcpu->arch.apic)
1796 hrtimer_cancel(&apic->lapic_timer.timer);
1798 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1799 static_key_slow_dec_deferred(&apic_hw_disabled);
1801 if (!apic->sw_enabled)
1802 static_key_slow_dec_deferred(&apic_sw_disabled);
1805 free_page((unsigned long)apic->regs);
1811 *----------------------------------------------------------------------
1813 *----------------------------------------------------------------------
1815 u64 kvm_get_lapic_target_expiration_tsc(struct kvm_vcpu *vcpu)
1817 struct kvm_lapic *apic = vcpu->arch.apic;
1819 if (!lapic_in_kernel(vcpu))
1822 return apic->lapic_timer.tscdeadline;
1825 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1827 struct kvm_lapic *apic = vcpu->arch.apic;
1829 if (!lapic_in_kernel(vcpu) ||
1830 !apic_lvtt_tscdeadline(apic))
1833 return apic->lapic_timer.tscdeadline;
1836 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1838 struct kvm_lapic *apic = vcpu->arch.apic;
1840 if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
1841 apic_lvtt_period(apic))
1844 hrtimer_cancel(&apic->lapic_timer.timer);
1845 apic->lapic_timer.tscdeadline = data;
1846 start_apic_timer(apic);
1849 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1851 struct kvm_lapic *apic = vcpu->arch.apic;
1853 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1854 | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
1857 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1861 tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1863 return (tpr & 0xf0) >> 4;
1866 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1868 u64 old_value = vcpu->arch.apic_base;
1869 struct kvm_lapic *apic = vcpu->arch.apic;
1872 value |= MSR_IA32_APICBASE_BSP;
1874 vcpu->arch.apic_base = value;
1876 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
1877 kvm_update_cpuid(vcpu);
1882 /* update jump label if enable bit changes */
1883 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1884 if (value & MSR_IA32_APICBASE_ENABLE) {
1885 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1886 static_key_slow_dec_deferred(&apic_hw_disabled);
1888 static_key_slow_inc(&apic_hw_disabled.key);
1889 recalculate_apic_map(vcpu->kvm);
1893 if ((old_value ^ value) & X2APIC_ENABLE) {
1894 if (value & X2APIC_ENABLE) {
1895 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
1896 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1898 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1901 apic->base_address = apic->vcpu->arch.apic_base &
1902 MSR_IA32_APICBASE_BASE;
1904 if ((value & MSR_IA32_APICBASE_ENABLE) &&
1905 apic->base_address != APIC_DEFAULT_PHYS_BASE)
1906 pr_warn_once("APIC base relocation is unsupported by KVM");
1908 /* with FSB delivery interrupt, we can restart APIC functionality */
1909 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1910 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1914 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
1916 struct kvm_lapic *apic;
1919 apic_debug("%s\n", __func__);
1922 apic = vcpu->arch.apic;
1923 ASSERT(apic != NULL);
1925 /* Stop the timer in case it's a reset to an active apic */
1926 hrtimer_cancel(&apic->lapic_timer.timer);
1929 kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
1930 MSR_IA32_APICBASE_ENABLE);
1931 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1933 kvm_apic_set_version(apic->vcpu);
1935 for (i = 0; i < KVM_APIC_LVT_NUM; i++)
1936 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1937 apic_update_lvtt(apic);
1938 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
1939 kvm_lapic_set_reg(apic, APIC_LVT0,
1940 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1941 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
1943 kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
1944 apic_set_spiv(apic, 0xff);
1945 kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
1946 if (!apic_x2apic_mode(apic))
1947 kvm_apic_set_ldr(apic, 0);
1948 kvm_lapic_set_reg(apic, APIC_ESR, 0);
1949 kvm_lapic_set_reg(apic, APIC_ICR, 0);
1950 kvm_lapic_set_reg(apic, APIC_ICR2, 0);
1951 kvm_lapic_set_reg(apic, APIC_TDCR, 0);
1952 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1953 for (i = 0; i < 8; i++) {
1954 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1955 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1956 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1958 apic->irr_pending = vcpu->arch.apicv_active;
1959 apic->isr_count = vcpu->arch.apicv_active ? 1 : 0;
1960 apic->highest_isr_cache = -1;
1961 update_divide_count(apic);
1962 atomic_set(&apic->lapic_timer.pending, 0);
1963 if (kvm_vcpu_is_bsp(vcpu))
1964 kvm_lapic_set_base(vcpu,
1965 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1966 vcpu->arch.pv_eoi.msr_val = 0;
1967 apic_update_ppr(apic);
1969 vcpu->arch.apic_arb_prio = 0;
1970 vcpu->arch.apic_attention = 0;
1972 apic_debug("%s: vcpu=%p, id=0x%x, base_msr="
1973 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1974 vcpu, kvm_lapic_get_reg(apic, APIC_ID),
1975 vcpu->arch.apic_base, apic->base_address);
1979 *----------------------------------------------------------------------
1981 *----------------------------------------------------------------------
1984 static bool lapic_is_periodic(struct kvm_lapic *apic)
1986 return apic_lvtt_period(apic);
1989 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1991 struct kvm_lapic *apic = vcpu->arch.apic;
1993 if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
1994 return atomic_read(&apic->lapic_timer.pending);
1999 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2001 u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2002 int vector, mode, trig_mode;
2004 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2005 vector = reg & APIC_VECTOR_MASK;
2006 mode = reg & APIC_MODE_MASK;
2007 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2008 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2014 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2016 struct kvm_lapic *apic = vcpu->arch.apic;
2019 kvm_apic_local_deliver(apic, APIC_LVT0);
2022 static const struct kvm_io_device_ops apic_mmio_ops = {
2023 .read = apic_mmio_read,
2024 .write = apic_mmio_write,
2027 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2029 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2030 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2032 apic_timer_expired(apic);
2034 if (lapic_is_periodic(apic)) {
2035 advance_periodic_target_expiration(apic);
2036 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2037 return HRTIMER_RESTART;
2039 return HRTIMER_NORESTART;
2042 int kvm_create_lapic(struct kvm_vcpu *vcpu)
2044 struct kvm_lapic *apic;
2046 ASSERT(vcpu != NULL);
2047 apic_debug("apic_init %d\n", vcpu->vcpu_id);
2049 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
2053 vcpu->arch.apic = apic;
2055 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
2057 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2059 goto nomem_free_apic;
2063 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2064 HRTIMER_MODE_ABS_PINNED);
2065 apic->lapic_timer.timer.function = apic_timer_fn;
2068 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2069 * thinking that APIC satet has changed.
2071 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2072 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2073 kvm_lapic_reset(vcpu, false);
2074 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2083 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2085 struct kvm_lapic *apic = vcpu->arch.apic;
2088 if (!apic_enabled(apic))
2091 __apic_update_ppr(apic, &ppr);
2092 return apic_has_interrupt_for_ppr(apic, ppr);
2095 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2097 u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2100 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2102 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2103 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2108 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2110 struct kvm_lapic *apic = vcpu->arch.apic;
2112 if (atomic_read(&apic->lapic_timer.pending) > 0) {
2113 kvm_apic_local_deliver(apic, APIC_LVTT);
2114 if (apic_lvtt_tscdeadline(apic))
2115 apic->lapic_timer.tscdeadline = 0;
2116 if (apic_lvtt_oneshot(apic)) {
2117 apic->lapic_timer.tscdeadline = 0;
2118 apic->lapic_timer.target_expiration = 0;
2120 atomic_set(&apic->lapic_timer.pending, 0);
2124 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2126 int vector = kvm_apic_has_interrupt(vcpu);
2127 struct kvm_lapic *apic = vcpu->arch.apic;
2134 * We get here even with APIC virtualization enabled, if doing
2135 * nested virtualization and L1 runs with the "acknowledge interrupt
2136 * on exit" mode. Then we cannot inject the interrupt via RVI,
2137 * because the process would deliver it through the IDT.
2140 apic_clear_irr(vector, apic);
2141 if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2143 * For auto-EOI interrupts, there might be another pending
2144 * interrupt above PPR, so check whether to raise another
2147 apic_update_ppr(apic);
2150 * For normal interrupts, PPR has been raised and there cannot
2151 * be a higher-priority pending interrupt---except if there was
2152 * a concurrent interrupt injection, but that would have
2153 * triggered KVM_REQ_EVENT already.
2155 apic_set_isr(vector, apic);
2156 __apic_update_ppr(apic, &ppr);
2162 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2163 struct kvm_lapic_state *s, bool set)
2165 if (apic_x2apic_mode(vcpu->arch.apic)) {
2166 u32 *id = (u32 *)(s->regs + APIC_ID);
2168 if (vcpu->kvm->arch.x2apic_format) {
2169 if (*id != vcpu->vcpu_id)
2182 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2184 memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2185 return kvm_apic_state_fixup(vcpu, s, false);
2188 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2190 struct kvm_lapic *apic = vcpu->arch.apic;
2194 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2195 /* set SPIV separately to get count of SW disabled APICs right */
2196 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2198 r = kvm_apic_state_fixup(vcpu, s, true);
2201 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2203 recalculate_apic_map(vcpu->kvm);
2204 kvm_apic_set_version(vcpu);
2206 apic_update_ppr(apic);
2207 hrtimer_cancel(&apic->lapic_timer.timer);
2208 apic_update_lvtt(apic);
2209 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2210 update_divide_count(apic);
2211 start_apic_timer(apic);
2212 apic->irr_pending = true;
2213 apic->isr_count = vcpu->arch.apicv_active ?
2214 1 : count_vectors(apic->regs + APIC_ISR);
2215 apic->highest_isr_cache = -1;
2216 if (vcpu->arch.apicv_active) {
2217 kvm_x86_ops->apicv_post_state_restore(vcpu);
2218 kvm_x86_ops->hwapic_irr_update(vcpu,
2219 apic_find_highest_irr(apic));
2220 kvm_x86_ops->hwapic_isr_update(vcpu,
2221 apic_find_highest_isr(apic));
2223 kvm_make_request(KVM_REQ_EVENT, vcpu);
2224 if (ioapic_in_kernel(vcpu->kvm))
2225 kvm_rtc_eoi_tracking_restore_one(vcpu);
2227 vcpu->arch.apic_arb_prio = 0;
2232 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2234 struct hrtimer *timer;
2236 if (!lapic_in_kernel(vcpu))
2239 timer = &vcpu->arch.apic->lapic_timer.timer;
2240 if (hrtimer_cancel(timer))
2241 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
2245 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2247 * Detect whether guest triggered PV EOI since the
2248 * last entry. If yes, set EOI on guests's behalf.
2249 * Clear PV EOI in guest memory in any case.
2251 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2252 struct kvm_lapic *apic)
2257 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2258 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2260 * KVM_APIC_PV_EOI_PENDING is unset:
2261 * -> host disabled PV EOI.
2262 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2263 * -> host enabled PV EOI, guest did not execute EOI yet.
2264 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2265 * -> host enabled PV EOI, guest executed EOI.
2267 BUG_ON(!pv_eoi_enabled(vcpu));
2268 pending = pv_eoi_get_pending(vcpu);
2270 * Clear pending bit in any case: it will be set again on vmentry.
2271 * While this might not be ideal from performance point of view,
2272 * this makes sure pv eoi is only enabled when we know it's safe.
2274 pv_eoi_clr_pending(vcpu);
2277 vector = apic_set_eoi(apic);
2278 trace_kvm_pv_eoi(apic, vector);
2281 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2285 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2286 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2288 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2291 if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2295 apic_set_tpr(vcpu->arch.apic, data & 0xff);
2299 * apic_sync_pv_eoi_to_guest - called before vmentry
2301 * Detect whether it's safe to enable PV EOI and
2304 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2305 struct kvm_lapic *apic)
2307 if (!pv_eoi_enabled(vcpu) ||
2308 /* IRR set or many bits in ISR: could be nested. */
2309 apic->irr_pending ||
2310 /* Cache not set: could be safe but we don't bother. */
2311 apic->highest_isr_cache == -1 ||
2312 /* Need EOI to update ioapic. */
2313 kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2315 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2316 * so we need not do anything here.
2321 pv_eoi_set_pending(apic->vcpu);
2324 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2327 int max_irr, max_isr;
2328 struct kvm_lapic *apic = vcpu->arch.apic;
2330 apic_sync_pv_eoi_to_guest(vcpu, apic);
2332 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2335 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2336 max_irr = apic_find_highest_irr(apic);
2339 max_isr = apic_find_highest_isr(apic);
2342 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2344 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2348 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2351 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2352 &vcpu->arch.apic->vapic_cache,
2353 vapic_addr, sizeof(u32)))
2355 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2357 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2360 vcpu->arch.apic->vapic_addr = vapic_addr;
2364 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2366 struct kvm_lapic *apic = vcpu->arch.apic;
2367 u32 reg = (msr - APIC_BASE_MSR) << 4;
2369 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2372 if (reg == APIC_ICR2)
2375 /* if this is ICR write vector before command */
2376 if (reg == APIC_ICR)
2377 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2378 return kvm_lapic_reg_write(apic, reg, (u32)data);
2381 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2383 struct kvm_lapic *apic = vcpu->arch.apic;
2384 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2386 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2389 if (reg == APIC_DFR || reg == APIC_ICR2) {
2390 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2395 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2397 if (reg == APIC_ICR)
2398 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2400 *data = (((u64)high) << 32) | low;
2405 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2407 struct kvm_lapic *apic = vcpu->arch.apic;
2409 if (!lapic_in_kernel(vcpu))
2412 /* if this is ICR write vector before command */
2413 if (reg == APIC_ICR)
2414 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2415 return kvm_lapic_reg_write(apic, reg, (u32)data);
2418 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2420 struct kvm_lapic *apic = vcpu->arch.apic;
2423 if (!lapic_in_kernel(vcpu))
2426 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2428 if (reg == APIC_ICR)
2429 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2431 *data = (((u64)high) << 32) | low;
2436 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2438 u64 addr = data & ~KVM_MSR_ENABLED;
2439 if (!IS_ALIGNED(addr, 4))
2442 vcpu->arch.pv_eoi.msr_val = data;
2443 if (!pv_eoi_enabled(vcpu))
2445 return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2449 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2451 struct kvm_lapic *apic = vcpu->arch.apic;
2455 if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2459 * INITs are latched while in SMM. Because an SMM CPU cannot
2460 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2461 * and delay processing of INIT until the next RSM.
2464 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2465 if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2466 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2470 pe = xchg(&apic->pending_events, 0);
2471 if (test_bit(KVM_APIC_INIT, &pe)) {
2472 kvm_lapic_reset(vcpu, true);
2473 kvm_vcpu_reset(vcpu, true);
2474 if (kvm_vcpu_is_bsp(apic->vcpu))
2475 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2477 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2479 if (test_bit(KVM_APIC_SIPI, &pe) &&
2480 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2481 /* evaluate pending_events before reading the vector */
2483 sipi_vector = apic->sipi_vector;
2484 apic_debug("vcpu %d received sipi with vector # %x\n",
2485 vcpu->vcpu_id, sipi_vector);
2486 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2487 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2491 void kvm_lapic_init(void)
2493 /* do not patch jump label more than once per second */
2494 jump_label_rate_limit(&apic_hw_disabled, HZ);
2495 jump_label_rate_limit(&apic_sw_disabled, HZ);
2498 void kvm_lapic_exit(void)
2500 static_key_deferred_flush(&apic_hw_disabled);
2501 static_key_deferred_flush(&apic_sw_disabled);