1 // SPDX-License-Identifier: GPL-2.0-only
4 * Local APIC virtualization
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2007 Novell
8 * Copyright (C) 2007 Intel
9 * Copyright 2009 Red Hat, Inc. and/or its affiliates.
12 * Dor Laor <dor.laor@qumranet.com>
13 * Gregory Haskins <ghaskins@novell.com>
14 * Yaozu (Eddie) Dong <eddie.dong@intel.com>
16 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
19 #include <linux/kvm_host.h>
20 #include <linux/kvm.h>
22 #include <linux/highmem.h>
23 #include <linux/smp.h>
24 #include <linux/hrtimer.h>
26 #include <linux/export.h>
27 #include <linux/math64.h>
28 #include <linux/slab.h>
29 #include <asm/processor.h>
32 #include <asm/current.h>
33 #include <asm/apicdef.h>
34 #include <asm/delay.h>
35 #include <linux/atomic.h>
36 #include <linux/jump_label.h>
37 #include "kvm_cache_regs.h"
45 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
47 #define mod_64(x, y) ((x) % (y))
55 /* 14 is the version for Xeon and Pentium 8.4.8*/
56 #define APIC_VERSION (0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
57 #define LAPIC_MMIO_LENGTH (1 << 12)
58 /* followed define is not in apicdef.h */
59 #define APIC_SHORT_MASK 0xc0000
60 #define APIC_DEST_NOSHORT 0x0
61 #define APIC_DEST_MASK 0x800
62 #define MAX_APIC_VECTOR 256
63 #define APIC_VECTORS_PER_REG 32
65 #define APIC_BROADCAST 0xFF
66 #define X2APIC_BROADCAST 0xFFFFFFFFul
68 static bool lapic_timer_advance_dynamic __read_mostly;
69 #define LAPIC_TIMER_ADVANCE_ADJUST_MIN 100 /* clock cycles */
70 #define LAPIC_TIMER_ADVANCE_ADJUST_MAX 10000 /* clock cycles */
71 #define LAPIC_TIMER_ADVANCE_NS_INIT 1000
72 #define LAPIC_TIMER_ADVANCE_NS_MAX 5000
73 /* step-by-step approximation to mitigate fluctuation */
74 #define LAPIC_TIMER_ADVANCE_ADJUST_STEP 8
76 static inline int apic_test_vector(int vec, void *bitmap)
78 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
81 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
83 struct kvm_lapic *apic = vcpu->arch.apic;
85 return apic_test_vector(vector, apic->regs + APIC_ISR) ||
86 apic_test_vector(vector, apic->regs + APIC_IRR);
89 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
91 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
94 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
96 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
99 struct static_key_deferred apic_hw_disabled __read_mostly;
100 struct static_key_deferred apic_sw_disabled __read_mostly;
102 static inline int apic_enabled(struct kvm_lapic *apic)
104 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
108 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
111 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
112 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
114 static inline u8 kvm_xapic_id(struct kvm_lapic *apic)
116 return kvm_lapic_get_reg(apic, APIC_ID) >> 24;
119 static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
121 return apic->vcpu->vcpu_id;
124 bool kvm_can_post_timer_interrupt(struct kvm_vcpu *vcpu)
126 return pi_inject_timer && kvm_vcpu_apicv_active(vcpu);
128 EXPORT_SYMBOL_GPL(kvm_can_post_timer_interrupt);
130 static bool kvm_use_posted_timer_interrupt(struct kvm_vcpu *vcpu)
132 return kvm_can_post_timer_interrupt(vcpu) && vcpu->mode == IN_GUEST_MODE;
135 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
136 u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
138 case KVM_APIC_MODE_X2APIC: {
139 u32 offset = (dest_id >> 16) * 16;
140 u32 max_apic_id = map->max_apic_id;
142 if (offset <= max_apic_id) {
143 u8 cluster_size = min(max_apic_id - offset + 1, 16U);
145 offset = array_index_nospec(offset, map->max_apic_id + 1);
146 *cluster = &map->phys_map[offset];
147 *mask = dest_id & (0xffff >> (16 - cluster_size));
154 case KVM_APIC_MODE_XAPIC_FLAT:
155 *cluster = map->xapic_flat_map;
156 *mask = dest_id & 0xff;
158 case KVM_APIC_MODE_XAPIC_CLUSTER:
159 *cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
160 *mask = dest_id & 0xf;
168 static void kvm_apic_map_free(struct rcu_head *rcu)
170 struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
175 static void recalculate_apic_map(struct kvm *kvm)
177 struct kvm_apic_map *new, *old = NULL;
178 struct kvm_vcpu *vcpu;
180 u32 max_id = 255; /* enough space for any xAPIC ID */
182 mutex_lock(&kvm->arch.apic_map_lock);
184 kvm_for_each_vcpu(i, vcpu, kvm)
185 if (kvm_apic_present(vcpu))
186 max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
188 new = kvzalloc(sizeof(struct kvm_apic_map) +
189 sizeof(struct kvm_lapic *) * ((u64)max_id + 1),
195 new->max_apic_id = max_id;
197 kvm_for_each_vcpu(i, vcpu, kvm) {
198 struct kvm_lapic *apic = vcpu->arch.apic;
199 struct kvm_lapic **cluster;
205 if (!kvm_apic_present(vcpu))
208 xapic_id = kvm_xapic_id(apic);
209 x2apic_id = kvm_x2apic_id(apic);
211 /* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
212 if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
213 x2apic_id <= new->max_apic_id)
214 new->phys_map[x2apic_id] = apic;
216 * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
217 * prevent them from masking VCPUs with APIC ID <= 0xff.
219 if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
220 new->phys_map[xapic_id] = apic;
222 if (!kvm_apic_sw_enabled(apic))
225 ldr = kvm_lapic_get_reg(apic, APIC_LDR);
227 if (apic_x2apic_mode(apic)) {
228 new->mode |= KVM_APIC_MODE_X2APIC;
230 ldr = GET_APIC_LOGICAL_ID(ldr);
231 if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
232 new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
234 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
237 if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
241 cluster[ffs(mask) - 1] = apic;
244 old = rcu_dereference_protected(kvm->arch.apic_map,
245 lockdep_is_held(&kvm->arch.apic_map_lock));
246 rcu_assign_pointer(kvm->arch.apic_map, new);
247 mutex_unlock(&kvm->arch.apic_map_lock);
250 call_rcu(&old->rcu, kvm_apic_map_free);
252 kvm_make_scan_ioapic_request(kvm);
255 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
257 bool enabled = val & APIC_SPIV_APIC_ENABLED;
259 kvm_lapic_set_reg(apic, APIC_SPIV, val);
261 if (enabled != apic->sw_enabled) {
262 apic->sw_enabled = enabled;
264 static_key_slow_dec_deferred(&apic_sw_disabled);
266 static_key_slow_inc(&apic_sw_disabled.key);
268 recalculate_apic_map(apic->vcpu->kvm);
272 static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
274 kvm_lapic_set_reg(apic, APIC_ID, id << 24);
275 recalculate_apic_map(apic->vcpu->kvm);
278 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
280 kvm_lapic_set_reg(apic, APIC_LDR, id);
281 recalculate_apic_map(apic->vcpu->kvm);
284 static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
286 return ((id >> 4) << 16) | (1 << (id & 0xf));
289 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
291 u32 ldr = kvm_apic_calc_x2apic_ldr(id);
293 WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
295 kvm_lapic_set_reg(apic, APIC_ID, id);
296 kvm_lapic_set_reg(apic, APIC_LDR, ldr);
297 recalculate_apic_map(apic->vcpu->kvm);
300 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
302 return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
305 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
307 return kvm_lapic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
310 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
312 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
315 static inline int apic_lvtt_period(struct kvm_lapic *apic)
317 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
320 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
322 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
325 static inline int apic_lvt_nmi_mode(u32 lvt_val)
327 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
330 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
332 struct kvm_lapic *apic = vcpu->arch.apic;
333 struct kvm_cpuid_entry2 *feat;
334 u32 v = APIC_VERSION;
336 if (!lapic_in_kernel(vcpu))
340 * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
341 * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
342 * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
343 * version first and level-triggered interrupts never get EOIed in
346 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
347 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))) &&
348 !ioapic_in_kernel(vcpu->kvm))
349 v |= APIC_LVR_DIRECTED_EOI;
350 kvm_lapic_set_reg(apic, APIC_LVR, v);
353 static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
354 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
355 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
356 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
357 LINT_MASK, LINT_MASK, /* LVT0-1 */
358 LVT_MASK /* LVTERR */
361 static int find_highest_vector(void *bitmap)
366 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
367 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
368 reg = bitmap + REG_POS(vec);
370 return __fls(*reg) + vec;
376 static u8 count_vectors(void *bitmap)
382 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
383 reg = bitmap + REG_POS(vec);
384 count += hweight32(*reg);
390 bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
393 u32 pir_val, irr_val, prev_irr_val;
396 max_updated_irr = -1;
399 for (i = vec = 0; i <= 7; i++, vec += 32) {
400 pir_val = READ_ONCE(pir[i]);
401 irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
403 prev_irr_val = irr_val;
404 irr_val |= xchg(&pir[i], 0);
405 *((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
406 if (prev_irr_val != irr_val) {
408 __fls(irr_val ^ prev_irr_val) + vec;
412 *max_irr = __fls(irr_val) + vec;
415 return ((max_updated_irr != -1) &&
416 (max_updated_irr == *max_irr));
418 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
420 bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
422 struct kvm_lapic *apic = vcpu->arch.apic;
424 return __kvm_apic_update_irr(pir, apic->regs, max_irr);
426 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
428 static inline int apic_search_irr(struct kvm_lapic *apic)
430 return find_highest_vector(apic->regs + APIC_IRR);
433 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
438 * Note that irr_pending is just a hint. It will be always
439 * true with virtual interrupt delivery enabled.
441 if (!apic->irr_pending)
444 result = apic_search_irr(apic);
445 ASSERT(result == -1 || result >= 16);
450 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
452 struct kvm_vcpu *vcpu;
456 if (unlikely(vcpu->arch.apicv_active)) {
457 /* need to update RVI */
458 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
459 kvm_x86_ops->hwapic_irr_update(vcpu,
460 apic_find_highest_irr(apic));
462 apic->irr_pending = false;
463 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
464 if (apic_search_irr(apic) != -1)
465 apic->irr_pending = true;
469 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
471 struct kvm_vcpu *vcpu;
473 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
479 * With APIC virtualization enabled, all caching is disabled
480 * because the processor can modify ISR under the hood. Instead
483 if (unlikely(vcpu->arch.apicv_active))
484 kvm_x86_ops->hwapic_isr_update(vcpu, vec);
487 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
489 * ISR (in service register) bit is set when injecting an interrupt.
490 * The highest vector is injected. Thus the latest bit set matches
491 * the highest bit in ISR.
493 apic->highest_isr_cache = vec;
497 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
502 * Note that isr_count is always 1, and highest_isr_cache
503 * is always -1, with APIC virtualization enabled.
505 if (!apic->isr_count)
507 if (likely(apic->highest_isr_cache != -1))
508 return apic->highest_isr_cache;
510 result = find_highest_vector(apic->regs + APIC_ISR);
511 ASSERT(result == -1 || result >= 16);
516 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
518 struct kvm_vcpu *vcpu;
519 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
525 * We do get here for APIC virtualization enabled if the guest
526 * uses the Hyper-V APIC enlightenment. In this case we may need
527 * to trigger a new interrupt delivery by writing the SVI field;
528 * on the other hand isr_count and highest_isr_cache are unused
529 * and must be left alone.
531 if (unlikely(vcpu->arch.apicv_active))
532 kvm_x86_ops->hwapic_isr_update(vcpu,
533 apic_find_highest_isr(apic));
536 BUG_ON(apic->isr_count < 0);
537 apic->highest_isr_cache = -1;
541 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
543 /* This may race with setting of irr in __apic_accept_irq() and
544 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
545 * will cause vmexit immediately and the value will be recalculated
546 * on the next vmentry.
548 return apic_find_highest_irr(vcpu->arch.apic);
550 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
552 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
553 int vector, int level, int trig_mode,
554 struct dest_map *dest_map);
556 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
557 struct dest_map *dest_map)
559 struct kvm_lapic *apic = vcpu->arch.apic;
561 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
562 irq->level, irq->trig_mode, dest_map);
565 static int __pv_send_ipi(unsigned long *ipi_bitmap, struct kvm_apic_map *map,
566 struct kvm_lapic_irq *irq, u32 min)
569 struct kvm_vcpu *vcpu;
571 if (min > map->max_apic_id)
574 for_each_set_bit(i, ipi_bitmap,
575 min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
576 if (map->phys_map[min + i]) {
577 vcpu = map->phys_map[min + i]->vcpu;
578 count += kvm_apic_set_irq(vcpu, irq, NULL);
585 int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
586 unsigned long ipi_bitmap_high, u32 min,
587 unsigned long icr, int op_64_bit)
589 struct kvm_apic_map *map;
590 struct kvm_lapic_irq irq = {0};
591 int cluster_size = op_64_bit ? 64 : 32;
594 if (icr & (APIC_DEST_MASK | APIC_SHORT_MASK))
597 irq.vector = icr & APIC_VECTOR_MASK;
598 irq.delivery_mode = icr & APIC_MODE_MASK;
599 irq.level = (icr & APIC_INT_ASSERT) != 0;
600 irq.trig_mode = icr & APIC_INT_LEVELTRIG;
603 map = rcu_dereference(kvm->arch.apic_map);
607 count = __pv_send_ipi(&ipi_bitmap_low, map, &irq, min);
609 count += __pv_send_ipi(&ipi_bitmap_high, map, &irq, min);
616 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
619 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
623 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
626 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
630 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
632 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
635 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
638 if (pv_eoi_get_user(vcpu, &val) < 0)
639 printk(KERN_WARNING "Can't read EOI MSR value: 0x%llx\n",
640 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
644 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
646 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
647 printk(KERN_WARNING "Can't set EOI MSR value: 0x%llx\n",
648 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
651 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
654 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
656 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
657 printk(KERN_WARNING "Can't clear EOI MSR value: 0x%llx\n",
658 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
661 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
664 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
667 if (apic->vcpu->arch.apicv_active)
668 highest_irr = kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
670 highest_irr = apic_find_highest_irr(apic);
671 if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
676 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
678 u32 tpr, isrv, ppr, old_ppr;
681 old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
682 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
683 isr = apic_find_highest_isr(apic);
684 isrv = (isr != -1) ? isr : 0;
686 if ((tpr & 0xf0) >= (isrv & 0xf0))
693 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
695 return ppr < old_ppr;
698 static void apic_update_ppr(struct kvm_lapic *apic)
702 if (__apic_update_ppr(apic, &ppr) &&
703 apic_has_interrupt_for_ppr(apic, ppr) != -1)
704 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
707 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
709 apic_update_ppr(vcpu->arch.apic);
711 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
713 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
715 kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
716 apic_update_ppr(apic);
719 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
721 return mda == (apic_x2apic_mode(apic) ?
722 X2APIC_BROADCAST : APIC_BROADCAST);
725 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
727 if (kvm_apic_broadcast(apic, mda))
730 if (apic_x2apic_mode(apic))
731 return mda == kvm_x2apic_id(apic);
734 * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
735 * it were in x2APIC mode. Hotplugged VCPUs start in xAPIC mode and
736 * this allows unique addressing of VCPUs with APIC ID over 0xff.
737 * The 0xff condition is needed because writeable xAPIC ID.
739 if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
742 return mda == kvm_xapic_id(apic);
745 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
749 if (kvm_apic_broadcast(apic, mda))
752 logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
754 if (apic_x2apic_mode(apic))
755 return ((logical_id >> 16) == (mda >> 16))
756 && (logical_id & mda & 0xffff) != 0;
758 logical_id = GET_APIC_LOGICAL_ID(logical_id);
760 switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
762 return (logical_id & mda) != 0;
763 case APIC_DFR_CLUSTER:
764 return ((logical_id >> 4) == (mda >> 4))
765 && (logical_id & mda & 0xf) != 0;
771 /* The KVM local APIC implementation has two quirks:
773 * - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
774 * in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
775 * KVM doesn't do that aliasing.
777 * - in-kernel IOAPIC messages have to be delivered directly to
778 * x2APIC, because the kernel does not support interrupt remapping.
779 * In order to support broadcast without interrupt remapping, x2APIC
780 * rewrites the destination of non-IPI messages from APIC_BROADCAST
781 * to X2APIC_BROADCAST.
783 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API. This is
784 * important when userspace wants to use x2APIC-format MSIs, because
785 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
787 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
788 struct kvm_lapic *source, struct kvm_lapic *target)
790 bool ipi = source != NULL;
792 if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
793 !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
794 return X2APIC_BROADCAST;
799 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
800 int short_hand, unsigned int dest, int dest_mode)
802 struct kvm_lapic *target = vcpu->arch.apic;
803 u32 mda = kvm_apic_mda(vcpu, dest, source, target);
806 switch (short_hand) {
807 case APIC_DEST_NOSHORT:
808 if (dest_mode == APIC_DEST_PHYSICAL)
809 return kvm_apic_match_physical_addr(target, mda);
811 return kvm_apic_match_logical_addr(target, mda);
813 return target == source;
814 case APIC_DEST_ALLINC:
816 case APIC_DEST_ALLBUT:
817 return target != source;
822 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
824 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
825 const unsigned long *bitmap, u32 bitmap_size)
830 mod = vector % dest_vcpus;
832 for (i = 0; i <= mod; i++) {
833 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
834 BUG_ON(idx == bitmap_size);
840 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
842 if (!kvm->arch.disabled_lapic_found) {
843 kvm->arch.disabled_lapic_found = true;
845 "Disabled LAPIC found during irq injection\n");
849 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
850 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
852 if (kvm->arch.x2apic_broadcast_quirk_disabled) {
853 if ((irq->dest_id == APIC_BROADCAST &&
854 map->mode != KVM_APIC_MODE_X2APIC))
856 if (irq->dest_id == X2APIC_BROADCAST)
859 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
860 if (irq->dest_id == (x2apic_ipi ?
861 X2APIC_BROADCAST : APIC_BROADCAST))
868 /* Return true if the interrupt can be handled by using *bitmap as index mask
869 * for valid destinations in *dst array.
870 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
871 * Note: we may have zero kvm_lapic destinations when we return true, which
872 * means that the interrupt should be dropped. In this case, *bitmap would be
873 * zero and *dst undefined.
875 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
876 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
877 struct kvm_apic_map *map, struct kvm_lapic ***dst,
878 unsigned long *bitmap)
882 if (irq->shorthand == APIC_DEST_SELF && src) {
886 } else if (irq->shorthand)
889 if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
892 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
893 if (irq->dest_id > map->max_apic_id) {
896 u32 dest_id = array_index_nospec(irq->dest_id, map->max_apic_id + 1);
897 *dst = &map->phys_map[dest_id];
904 if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
908 if (!kvm_lowest_prio_delivery(irq))
911 if (!kvm_vector_hashing_enabled()) {
913 for_each_set_bit(i, bitmap, 16) {
918 else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
919 (*dst)[lowest]->vcpu) < 0)
926 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
929 if (!(*dst)[lowest]) {
930 kvm_apic_disabled_lapic_found(kvm);
936 *bitmap = (lowest >= 0) ? 1 << lowest : 0;
941 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
942 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
944 struct kvm_apic_map *map;
945 unsigned long bitmap;
946 struct kvm_lapic **dst = NULL;
952 if (irq->shorthand == APIC_DEST_SELF) {
953 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
958 map = rcu_dereference(kvm->arch.apic_map);
960 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
963 for_each_set_bit(i, &bitmap, 16) {
966 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
975 * This routine tries to handler interrupts in posted mode, here is how
976 * it deals with different cases:
977 * - For single-destination interrupts, handle it in posted mode
978 * - Else if vector hashing is enabled and it is a lowest-priority
979 * interrupt, handle it in posted mode and use the following mechanism
980 * to find the destinaiton vCPU.
981 * 1. For lowest-priority interrupts, store all the possible
982 * destination vCPUs in an array.
983 * 2. Use "guest vector % max number of destination vCPUs" to find
984 * the right destination vCPU in the array for the lowest-priority
986 * - Otherwise, use remapped mode to inject the interrupt.
988 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
989 struct kvm_vcpu **dest_vcpu)
991 struct kvm_apic_map *map;
992 unsigned long bitmap;
993 struct kvm_lapic **dst = NULL;
1000 map = rcu_dereference(kvm->arch.apic_map);
1002 if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
1003 hweight16(bitmap) == 1) {
1004 unsigned long i = find_first_bit(&bitmap, 16);
1007 *dest_vcpu = dst[i]->vcpu;
1017 * Add a pending IRQ into lapic.
1018 * Return 1 if successfully added and 0 if discarded.
1020 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
1021 int vector, int level, int trig_mode,
1022 struct dest_map *dest_map)
1025 struct kvm_vcpu *vcpu = apic->vcpu;
1027 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
1029 switch (delivery_mode) {
1030 case APIC_DM_LOWEST:
1031 vcpu->arch.apic_arb_prio++;
1034 if (unlikely(trig_mode && !level))
1037 /* FIXME add logic for vcpu on reset */
1038 if (unlikely(!apic_enabled(apic)))
1044 __set_bit(vcpu->vcpu_id, dest_map->map);
1045 dest_map->vectors[vcpu->vcpu_id] = vector;
1048 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
1050 kvm_lapic_set_vector(vector,
1051 apic->regs + APIC_TMR);
1053 kvm_lapic_clear_vector(vector,
1054 apic->regs + APIC_TMR);
1057 if (vcpu->arch.apicv_active)
1058 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
1060 kvm_lapic_set_irr(vector, apic);
1062 kvm_make_request(KVM_REQ_EVENT, vcpu);
1063 kvm_vcpu_kick(vcpu);
1069 vcpu->arch.pv.pv_unhalted = 1;
1070 kvm_make_request(KVM_REQ_EVENT, vcpu);
1071 kvm_vcpu_kick(vcpu);
1076 kvm_make_request(KVM_REQ_SMI, vcpu);
1077 kvm_vcpu_kick(vcpu);
1082 kvm_inject_nmi(vcpu);
1083 kvm_vcpu_kick(vcpu);
1087 if (!trig_mode || level) {
1089 /* assumes that there are only KVM_APIC_INIT/SIPI */
1090 apic->pending_events = (1UL << KVM_APIC_INIT);
1091 /* make sure pending_events is visible before sending
1094 kvm_make_request(KVM_REQ_EVENT, vcpu);
1095 kvm_vcpu_kick(vcpu);
1099 case APIC_DM_STARTUP:
1101 apic->sipi_vector = vector;
1102 /* make sure sipi_vector is visible for the receiver */
1104 set_bit(KVM_APIC_SIPI, &apic->pending_events);
1105 kvm_make_request(KVM_REQ_EVENT, vcpu);
1106 kvm_vcpu_kick(vcpu);
1109 case APIC_DM_EXTINT:
1111 * Should only be called by kvm_apic_local_deliver() with LVT0,
1112 * before NMI watchdog was enabled. Already handled by
1113 * kvm_apic_accept_pic_intr().
1118 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1126 * This routine identifies the destination vcpus mask meant to receive the
1127 * IOAPIC interrupts. It either uses kvm_apic_map_get_dest_lapic() to find
1128 * out the destination vcpus array and set the bitmap or it traverses to
1129 * each available vcpu to identify the same.
1131 void kvm_bitmap_or_dest_vcpus(struct kvm *kvm, struct kvm_lapic_irq *irq,
1132 unsigned long *vcpu_bitmap)
1134 struct kvm_lapic **dest_vcpu = NULL;
1135 struct kvm_lapic *src = NULL;
1136 struct kvm_apic_map *map;
1137 struct kvm_vcpu *vcpu;
1138 unsigned long bitmap;
1143 map = rcu_dereference(kvm->arch.apic_map);
1145 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dest_vcpu,
1148 for_each_set_bit(i, &bitmap, 16) {
1151 vcpu_idx = dest_vcpu[i]->vcpu->vcpu_idx;
1152 __set_bit(vcpu_idx, vcpu_bitmap);
1155 kvm_for_each_vcpu(i, vcpu, kvm) {
1156 if (!kvm_apic_present(vcpu))
1158 if (!kvm_apic_match_dest(vcpu, NULL,
1163 __set_bit(i, vcpu_bitmap);
1169 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1171 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1174 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1176 return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1179 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1183 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1184 if (!kvm_ioapic_handles_vector(apic, vector))
1187 /* Request a KVM exit to inform the userspace IOAPIC. */
1188 if (irqchip_split(apic->vcpu->kvm)) {
1189 apic->vcpu->arch.pending_ioapic_eoi = vector;
1190 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1194 if (apic_test_vector(vector, apic->regs + APIC_TMR))
1195 trigger_mode = IOAPIC_LEVEL_TRIG;
1197 trigger_mode = IOAPIC_EDGE_TRIG;
1199 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1202 static int apic_set_eoi(struct kvm_lapic *apic)
1204 int vector = apic_find_highest_isr(apic);
1206 trace_kvm_eoi(apic, vector);
1209 * Not every write EOI will has corresponding ISR,
1210 * one example is when Kernel check timer on setup_IO_APIC
1215 apic_clear_isr(vector, apic);
1216 apic_update_ppr(apic);
1218 if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1219 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1221 kvm_ioapic_send_eoi(apic, vector);
1222 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1227 * this interface assumes a trap-like exit, which has already finished
1228 * desired side effect including vISR and vPPR update.
1230 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1232 struct kvm_lapic *apic = vcpu->arch.apic;
1234 trace_kvm_eoi(apic, vector);
1236 kvm_ioapic_send_eoi(apic, vector);
1237 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1239 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1241 static void apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
1243 struct kvm_lapic_irq irq;
1245 irq.vector = icr_low & APIC_VECTOR_MASK;
1246 irq.delivery_mode = icr_low & APIC_MODE_MASK;
1247 irq.dest_mode = icr_low & APIC_DEST_MASK;
1248 irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1249 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1250 irq.shorthand = icr_low & APIC_SHORT_MASK;
1251 irq.msi_redir_hint = false;
1252 if (apic_x2apic_mode(apic))
1253 irq.dest_id = icr_high;
1255 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1257 trace_kvm_apic_ipi(icr_low, irq.dest_id);
1259 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1262 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1264 ktime_t remaining, now;
1268 ASSERT(apic != NULL);
1270 /* if initial count is 0, current count should also be 0 */
1271 if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1272 apic->lapic_timer.period == 0)
1276 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1277 if (ktime_to_ns(remaining) < 0)
1280 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1281 tmcct = div64_u64(ns,
1282 (APIC_BUS_CYCLE_NS * apic->divide_count));
1287 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1289 struct kvm_vcpu *vcpu = apic->vcpu;
1290 struct kvm_run *run = vcpu->run;
1292 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1293 run->tpr_access.rip = kvm_rip_read(vcpu);
1294 run->tpr_access.is_write = write;
1297 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1299 if (apic->vcpu->arch.tpr_access_reporting)
1300 __report_tpr_access(apic, write);
1303 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1307 if (offset >= LAPIC_MMIO_LENGTH)
1314 case APIC_TMCCT: /* Timer CCR */
1315 if (apic_lvtt_tscdeadline(apic))
1318 val = apic_get_tmcct(apic);
1321 apic_update_ppr(apic);
1322 val = kvm_lapic_get_reg(apic, offset);
1325 report_tpr_access(apic, false);
1328 val = kvm_lapic_get_reg(apic, offset);
1335 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1337 return container_of(dev, struct kvm_lapic, dev);
1340 #define APIC_REG_MASK(reg) (1ull << ((reg) >> 4))
1341 #define APIC_REGS_MASK(first, count) \
1342 (APIC_REG_MASK(first) * ((1ull << (count)) - 1))
1344 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1347 unsigned char alignment = offset & 0xf;
1349 /* this bitmask has a bit cleared for each reserved register */
1350 u64 valid_reg_mask =
1351 APIC_REG_MASK(APIC_ID) |
1352 APIC_REG_MASK(APIC_LVR) |
1353 APIC_REG_MASK(APIC_TASKPRI) |
1354 APIC_REG_MASK(APIC_PROCPRI) |
1355 APIC_REG_MASK(APIC_LDR) |
1356 APIC_REG_MASK(APIC_DFR) |
1357 APIC_REG_MASK(APIC_SPIV) |
1358 APIC_REGS_MASK(APIC_ISR, APIC_ISR_NR) |
1359 APIC_REGS_MASK(APIC_TMR, APIC_ISR_NR) |
1360 APIC_REGS_MASK(APIC_IRR, APIC_ISR_NR) |
1361 APIC_REG_MASK(APIC_ESR) |
1362 APIC_REG_MASK(APIC_ICR) |
1363 APIC_REG_MASK(APIC_ICR2) |
1364 APIC_REG_MASK(APIC_LVTT) |
1365 APIC_REG_MASK(APIC_LVTTHMR) |
1366 APIC_REG_MASK(APIC_LVTPC) |
1367 APIC_REG_MASK(APIC_LVT0) |
1368 APIC_REG_MASK(APIC_LVT1) |
1369 APIC_REG_MASK(APIC_LVTERR) |
1370 APIC_REG_MASK(APIC_TMICT) |
1371 APIC_REG_MASK(APIC_TMCCT) |
1372 APIC_REG_MASK(APIC_TDCR);
1374 /* ARBPRI is not valid on x2APIC */
1375 if (!apic_x2apic_mode(apic))
1376 valid_reg_mask |= APIC_REG_MASK(APIC_ARBPRI);
1378 if (offset > 0x3f0 || !(valid_reg_mask & APIC_REG_MASK(offset)))
1381 result = __apic_read(apic, offset & ~0xf);
1383 trace_kvm_apic_read(offset, result);
1389 memcpy(data, (char *)&result + alignment, len);
1392 printk(KERN_ERR "Local APIC read with len = %x, "
1393 "should be 1,2, or 4 instead\n", len);
1398 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1400 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1402 return addr >= apic->base_address &&
1403 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1406 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1407 gpa_t address, int len, void *data)
1409 struct kvm_lapic *apic = to_lapic(this);
1410 u32 offset = address - apic->base_address;
1412 if (!apic_mmio_in_range(apic, address))
1415 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1416 if (!kvm_check_has_quirk(vcpu->kvm,
1417 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1420 memset(data, 0xff, len);
1424 kvm_lapic_reg_read(apic, offset, len, data);
1429 static void update_divide_count(struct kvm_lapic *apic)
1431 u32 tmp1, tmp2, tdcr;
1433 tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1435 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1436 apic->divide_count = 0x1 << (tmp2 & 0x7);
1439 static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1442 * Do not allow the guest to program periodic timers with small
1443 * interval, since the hrtimers are not throttled by the host
1446 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1447 s64 min_period = min_timer_period_us * 1000LL;
1449 if (apic->lapic_timer.period < min_period) {
1450 pr_info_ratelimited(
1451 "kvm: vcpu %i: requested %lld ns "
1452 "lapic timer period limited to %lld ns\n",
1453 apic->vcpu->vcpu_id,
1454 apic->lapic_timer.period, min_period);
1455 apic->lapic_timer.period = min_period;
1460 static void apic_update_lvtt(struct kvm_lapic *apic)
1462 u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1463 apic->lapic_timer.timer_mode_mask;
1465 if (apic->lapic_timer.timer_mode != timer_mode) {
1466 if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1467 APIC_LVT_TIMER_TSCDEADLINE)) {
1468 hrtimer_cancel(&apic->lapic_timer.timer);
1469 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1470 apic->lapic_timer.period = 0;
1471 apic->lapic_timer.tscdeadline = 0;
1473 apic->lapic_timer.timer_mode = timer_mode;
1474 limit_periodic_timer_frequency(apic);
1479 * On APICv, this test will cause a busy wait
1480 * during a higher-priority task.
1483 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1485 struct kvm_lapic *apic = vcpu->arch.apic;
1486 u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1488 if (kvm_apic_hw_enabled(apic)) {
1489 int vec = reg & APIC_VECTOR_MASK;
1490 void *bitmap = apic->regs + APIC_ISR;
1492 if (vcpu->arch.apicv_active)
1493 bitmap = apic->regs + APIC_IRR;
1495 if (apic_test_vector(vec, bitmap))
1501 static inline void __wait_lapic_expire(struct kvm_vcpu *vcpu, u64 guest_cycles)
1503 u64 timer_advance_ns = vcpu->arch.apic->lapic_timer.timer_advance_ns;
1506 * If the guest TSC is running at a different ratio than the host, then
1507 * convert the delay to nanoseconds to achieve an accurate delay. Note
1508 * that __delay() uses delay_tsc whenever the hardware has TSC, thus
1509 * always for VMX enabled hardware.
1511 if (vcpu->arch.tsc_scaling_ratio == kvm_default_tsc_scaling_ratio) {
1512 __delay(min(guest_cycles,
1513 nsec_to_cycles(vcpu, timer_advance_ns)));
1515 u64 delay_ns = guest_cycles * 1000000ULL;
1516 do_div(delay_ns, vcpu->arch.virtual_tsc_khz);
1517 ndelay(min_t(u32, delay_ns, timer_advance_ns));
1521 static inline void adjust_lapic_timer_advance(struct kvm_vcpu *vcpu,
1522 s64 advance_expire_delta)
1524 struct kvm_lapic *apic = vcpu->arch.apic;
1525 u32 timer_advance_ns = apic->lapic_timer.timer_advance_ns;
1528 /* Do not adjust for tiny fluctuations or large random spikes. */
1529 if (abs(advance_expire_delta) > LAPIC_TIMER_ADVANCE_ADJUST_MAX ||
1530 abs(advance_expire_delta) < LAPIC_TIMER_ADVANCE_ADJUST_MIN)
1534 if (advance_expire_delta < 0) {
1535 ns = -advance_expire_delta * 1000000ULL;
1536 do_div(ns, vcpu->arch.virtual_tsc_khz);
1537 timer_advance_ns -= ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1540 ns = advance_expire_delta * 1000000ULL;
1541 do_div(ns, vcpu->arch.virtual_tsc_khz);
1542 timer_advance_ns += ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1545 if (unlikely(timer_advance_ns > LAPIC_TIMER_ADVANCE_NS_MAX))
1546 timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
1547 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
1550 static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1552 struct kvm_lapic *apic = vcpu->arch.apic;
1553 u64 guest_tsc, tsc_deadline;
1555 if (apic->lapic_timer.expired_tscdeadline == 0)
1558 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1559 apic->lapic_timer.expired_tscdeadline = 0;
1560 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1561 apic->lapic_timer.advance_expire_delta = guest_tsc - tsc_deadline;
1563 if (guest_tsc < tsc_deadline)
1564 __wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
1566 if (lapic_timer_advance_dynamic)
1567 adjust_lapic_timer_advance(vcpu, apic->lapic_timer.advance_expire_delta);
1570 void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1572 if (lapic_timer_int_injected(vcpu))
1573 __kvm_wait_lapic_expire(vcpu);
1575 EXPORT_SYMBOL_GPL(kvm_wait_lapic_expire);
1577 static void kvm_apic_inject_pending_timer_irqs(struct kvm_lapic *apic)
1579 struct kvm_timer *ktimer = &apic->lapic_timer;
1581 kvm_apic_local_deliver(apic, APIC_LVTT);
1582 if (apic_lvtt_tscdeadline(apic))
1583 ktimer->tscdeadline = 0;
1584 if (apic_lvtt_oneshot(apic)) {
1585 ktimer->tscdeadline = 0;
1586 ktimer->target_expiration = 0;
1590 static void apic_timer_expired(struct kvm_lapic *apic)
1592 struct kvm_vcpu *vcpu = apic->vcpu;
1593 struct kvm_timer *ktimer = &apic->lapic_timer;
1595 if (atomic_read(&apic->lapic_timer.pending))
1598 if (apic_lvtt_tscdeadline(apic) || ktimer->hv_timer_in_use)
1599 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1601 if (kvm_use_posted_timer_interrupt(apic->vcpu)) {
1602 if (apic->lapic_timer.timer_advance_ns)
1603 __kvm_wait_lapic_expire(vcpu);
1604 kvm_apic_inject_pending_timer_irqs(apic);
1608 atomic_inc(&apic->lapic_timer.pending);
1609 kvm_set_pending_timer(vcpu);
1612 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1614 struct kvm_timer *ktimer = &apic->lapic_timer;
1615 u64 guest_tsc, tscdeadline = ktimer->tscdeadline;
1618 struct kvm_vcpu *vcpu = apic->vcpu;
1619 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1620 unsigned long flags;
1623 if (unlikely(!tscdeadline || !this_tsc_khz))
1626 local_irq_save(flags);
1629 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1631 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1632 do_div(ns, this_tsc_khz);
1634 if (likely(tscdeadline > guest_tsc) &&
1635 likely(ns > apic->lapic_timer.timer_advance_ns)) {
1636 expire = ktime_add_ns(now, ns);
1637 expire = ktime_sub_ns(expire, ktimer->timer_advance_ns);
1638 hrtimer_start(&ktimer->timer, expire, HRTIMER_MODE_ABS_HARD);
1640 apic_timer_expired(apic);
1642 local_irq_restore(flags);
1645 static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1647 ktime_t now, remaining;
1648 u64 ns_remaining_old, ns_remaining_new;
1650 apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1651 * APIC_BUS_CYCLE_NS * apic->divide_count;
1652 limit_periodic_timer_frequency(apic);
1655 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1656 if (ktime_to_ns(remaining) < 0)
1659 ns_remaining_old = ktime_to_ns(remaining);
1660 ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1661 apic->divide_count, old_divisor);
1663 apic->lapic_timer.tscdeadline +=
1664 nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1665 nsec_to_cycles(apic->vcpu, ns_remaining_old);
1666 apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1669 static bool set_target_expiration(struct kvm_lapic *apic)
1675 apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1676 * APIC_BUS_CYCLE_NS * apic->divide_count;
1678 if (!apic->lapic_timer.period) {
1679 apic->lapic_timer.tscdeadline = 0;
1683 limit_periodic_timer_frequency(apic);
1685 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1686 nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1687 apic->lapic_timer.target_expiration = ktime_add_ns(now, apic->lapic_timer.period);
1692 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1694 ktime_t now = ktime_get();
1699 * Synchronize both deadlines to the same time source or
1700 * differences in the periods (caused by differences in the
1701 * underlying clocks or numerical approximation errors) will
1702 * cause the two to drift apart over time as the errors
1705 apic->lapic_timer.target_expiration =
1706 ktime_add_ns(apic->lapic_timer.target_expiration,
1707 apic->lapic_timer.period);
1708 delta = ktime_sub(apic->lapic_timer.target_expiration, now);
1709 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1710 nsec_to_cycles(apic->vcpu, delta);
1713 static void start_sw_period(struct kvm_lapic *apic)
1715 if (!apic->lapic_timer.period)
1718 if (ktime_after(ktime_get(),
1719 apic->lapic_timer.target_expiration)) {
1720 apic_timer_expired(apic);
1722 if (apic_lvtt_oneshot(apic))
1725 advance_periodic_target_expiration(apic);
1728 hrtimer_start(&apic->lapic_timer.timer,
1729 apic->lapic_timer.target_expiration,
1733 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1735 if (!lapic_in_kernel(vcpu))
1738 return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1740 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1742 static void cancel_hv_timer(struct kvm_lapic *apic)
1744 WARN_ON(preemptible());
1745 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1746 kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1747 apic->lapic_timer.hv_timer_in_use = false;
1750 static bool start_hv_timer(struct kvm_lapic *apic)
1752 struct kvm_timer *ktimer = &apic->lapic_timer;
1753 struct kvm_vcpu *vcpu = apic->vcpu;
1756 WARN_ON(preemptible());
1757 if (!kvm_x86_ops->set_hv_timer)
1760 if (!ktimer->tscdeadline)
1763 if (kvm_x86_ops->set_hv_timer(vcpu, ktimer->tscdeadline, &expired))
1766 ktimer->hv_timer_in_use = true;
1767 hrtimer_cancel(&ktimer->timer);
1770 * To simplify handling the periodic timer, leave the hv timer running
1771 * even if the deadline timer has expired, i.e. rely on the resulting
1772 * VM-Exit to recompute the periodic timer's target expiration.
1774 if (!apic_lvtt_period(apic)) {
1776 * Cancel the hv timer if the sw timer fired while the hv timer
1777 * was being programmed, or if the hv timer itself expired.
1779 if (atomic_read(&ktimer->pending)) {
1780 cancel_hv_timer(apic);
1781 } else if (expired) {
1782 apic_timer_expired(apic);
1783 cancel_hv_timer(apic);
1787 trace_kvm_hv_timer_state(vcpu->vcpu_id, ktimer->hv_timer_in_use);
1792 static void start_sw_timer(struct kvm_lapic *apic)
1794 struct kvm_timer *ktimer = &apic->lapic_timer;
1796 WARN_ON(preemptible());
1797 if (apic->lapic_timer.hv_timer_in_use)
1798 cancel_hv_timer(apic);
1799 if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1802 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1803 start_sw_period(apic);
1804 else if (apic_lvtt_tscdeadline(apic))
1805 start_sw_tscdeadline(apic);
1806 trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1809 static void restart_apic_timer(struct kvm_lapic *apic)
1813 if (!apic_lvtt_period(apic) && atomic_read(&apic->lapic_timer.pending))
1816 if (!start_hv_timer(apic))
1817 start_sw_timer(apic);
1822 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1824 struct kvm_lapic *apic = vcpu->arch.apic;
1827 /* If the preempt notifier has already run, it also called apic_timer_expired */
1828 if (!apic->lapic_timer.hv_timer_in_use)
1830 WARN_ON(swait_active(&vcpu->wq));
1831 cancel_hv_timer(apic);
1832 apic_timer_expired(apic);
1834 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1835 advance_periodic_target_expiration(apic);
1836 restart_apic_timer(apic);
1841 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1843 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1845 restart_apic_timer(vcpu->arch.apic);
1847 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1849 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1851 struct kvm_lapic *apic = vcpu->arch.apic;
1854 /* Possibly the TSC deadline timer is not enabled yet */
1855 if (apic->lapic_timer.hv_timer_in_use)
1856 start_sw_timer(apic);
1859 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1861 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
1863 struct kvm_lapic *apic = vcpu->arch.apic;
1865 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1866 restart_apic_timer(apic);
1869 static void start_apic_timer(struct kvm_lapic *apic)
1871 atomic_set(&apic->lapic_timer.pending, 0);
1873 if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1874 && !set_target_expiration(apic))
1877 restart_apic_timer(apic);
1880 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1882 bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1884 if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1885 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1886 if (lvt0_in_nmi_mode) {
1887 atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1889 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1893 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1897 trace_kvm_apic_write(reg, val);
1900 case APIC_ID: /* Local APIC ID */
1901 if (!apic_x2apic_mode(apic))
1902 kvm_apic_set_xapic_id(apic, val >> 24);
1908 report_tpr_access(apic, true);
1909 apic_set_tpr(apic, val & 0xff);
1917 if (!apic_x2apic_mode(apic))
1918 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1924 if (!apic_x2apic_mode(apic)) {
1925 kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1926 recalculate_apic_map(apic->vcpu->kvm);
1933 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1934 mask |= APIC_SPIV_DIRECTED_EOI;
1935 apic_set_spiv(apic, val & mask);
1936 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1940 for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
1941 lvt_val = kvm_lapic_get_reg(apic,
1942 APIC_LVTT + 0x10 * i);
1943 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
1944 lvt_val | APIC_LVT_MASKED);
1946 apic_update_lvtt(apic);
1947 atomic_set(&apic->lapic_timer.pending, 0);
1953 /* No delay here, so we always clear the pending bit */
1955 apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
1956 kvm_lapic_set_reg(apic, APIC_ICR, val);
1960 if (!apic_x2apic_mode(apic))
1962 kvm_lapic_set_reg(apic, APIC_ICR2, val);
1966 apic_manage_nmi_watchdog(apic, val);
1972 /* TODO: Check vector */
1973 if (!kvm_apic_sw_enabled(apic))
1974 val |= APIC_LVT_MASKED;
1976 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1977 kvm_lapic_set_reg(apic, reg, val);
1982 if (!kvm_apic_sw_enabled(apic))
1983 val |= APIC_LVT_MASKED;
1984 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1985 kvm_lapic_set_reg(apic, APIC_LVTT, val);
1986 apic_update_lvtt(apic);
1990 if (apic_lvtt_tscdeadline(apic))
1993 hrtimer_cancel(&apic->lapic_timer.timer);
1994 kvm_lapic_set_reg(apic, APIC_TMICT, val);
1995 start_apic_timer(apic);
1999 uint32_t old_divisor = apic->divide_count;
2001 kvm_lapic_set_reg(apic, APIC_TDCR, val);
2002 update_divide_count(apic);
2003 if (apic->divide_count != old_divisor &&
2004 apic->lapic_timer.period) {
2005 hrtimer_cancel(&apic->lapic_timer.timer);
2006 update_target_expiration(apic, old_divisor);
2007 restart_apic_timer(apic);
2012 if (apic_x2apic_mode(apic) && val != 0)
2017 if (apic_x2apic_mode(apic)) {
2018 kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
2029 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
2031 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
2032 gpa_t address, int len, const void *data)
2034 struct kvm_lapic *apic = to_lapic(this);
2035 unsigned int offset = address - apic->base_address;
2038 if (!apic_mmio_in_range(apic, address))
2041 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
2042 if (!kvm_check_has_quirk(vcpu->kvm,
2043 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
2050 * APIC register must be aligned on 128-bits boundary.
2051 * 32/64/128 bits registers must be accessed thru 32 bits.
2054 if (len != 4 || (offset & 0xf))
2059 kvm_lapic_reg_write(apic, offset & 0xff0, val);
2064 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
2066 kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
2068 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
2070 /* emulate APIC access in a trap manner */
2071 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
2075 /* hw has done the conditional check and inst decode */
2078 kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
2080 /* TODO: optimize to just emulate side effect w/o one more write */
2081 kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
2083 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
2085 void kvm_free_lapic(struct kvm_vcpu *vcpu)
2087 struct kvm_lapic *apic = vcpu->arch.apic;
2089 if (!vcpu->arch.apic)
2092 hrtimer_cancel(&apic->lapic_timer.timer);
2094 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
2095 static_key_slow_dec_deferred(&apic_hw_disabled);
2097 if (!apic->sw_enabled)
2098 static_key_slow_dec_deferred(&apic_sw_disabled);
2101 free_page((unsigned long)apic->regs);
2107 *----------------------------------------------------------------------
2109 *----------------------------------------------------------------------
2111 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
2113 struct kvm_lapic *apic = vcpu->arch.apic;
2115 if (!lapic_in_kernel(vcpu) ||
2116 !apic_lvtt_tscdeadline(apic))
2119 return apic->lapic_timer.tscdeadline;
2122 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
2124 struct kvm_lapic *apic = vcpu->arch.apic;
2126 if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
2127 apic_lvtt_period(apic))
2130 hrtimer_cancel(&apic->lapic_timer.timer);
2131 apic->lapic_timer.tscdeadline = data;
2132 start_apic_timer(apic);
2135 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
2137 struct kvm_lapic *apic = vcpu->arch.apic;
2139 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
2140 | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
2143 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
2147 tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
2149 return (tpr & 0xf0) >> 4;
2152 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
2154 u64 old_value = vcpu->arch.apic_base;
2155 struct kvm_lapic *apic = vcpu->arch.apic;
2158 value |= MSR_IA32_APICBASE_BSP;
2160 vcpu->arch.apic_base = value;
2162 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
2163 kvm_update_cpuid(vcpu);
2168 /* update jump label if enable bit changes */
2169 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
2170 if (value & MSR_IA32_APICBASE_ENABLE) {
2171 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2172 static_key_slow_dec_deferred(&apic_hw_disabled);
2174 static_key_slow_inc(&apic_hw_disabled.key);
2175 recalculate_apic_map(vcpu->kvm);
2179 if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
2180 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
2182 if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE))
2183 kvm_x86_ops->set_virtual_apic_mode(vcpu);
2185 apic->base_address = apic->vcpu->arch.apic_base &
2186 MSR_IA32_APICBASE_BASE;
2188 if ((value & MSR_IA32_APICBASE_ENABLE) &&
2189 apic->base_address != APIC_DEFAULT_PHYS_BASE)
2190 pr_warn_once("APIC base relocation is unsupported by KVM");
2193 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2195 struct kvm_lapic *apic = vcpu->arch.apic;
2201 /* Stop the timer in case it's a reset to an active apic */
2202 hrtimer_cancel(&apic->lapic_timer.timer);
2205 kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
2206 MSR_IA32_APICBASE_ENABLE);
2207 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2209 kvm_apic_set_version(apic->vcpu);
2211 for (i = 0; i < KVM_APIC_LVT_NUM; i++)
2212 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
2213 apic_update_lvtt(apic);
2214 if (kvm_vcpu_is_reset_bsp(vcpu) &&
2215 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2216 kvm_lapic_set_reg(apic, APIC_LVT0,
2217 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2218 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2220 kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
2221 apic_set_spiv(apic, 0xff);
2222 kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2223 if (!apic_x2apic_mode(apic))
2224 kvm_apic_set_ldr(apic, 0);
2225 kvm_lapic_set_reg(apic, APIC_ESR, 0);
2226 kvm_lapic_set_reg(apic, APIC_ICR, 0);
2227 kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2228 kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2229 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2230 for (i = 0; i < 8; i++) {
2231 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2232 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2233 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2235 apic->irr_pending = vcpu->arch.apicv_active;
2236 apic->isr_count = vcpu->arch.apicv_active ? 1 : 0;
2237 apic->highest_isr_cache = -1;
2238 update_divide_count(apic);
2239 atomic_set(&apic->lapic_timer.pending, 0);
2240 if (kvm_vcpu_is_bsp(vcpu))
2241 kvm_lapic_set_base(vcpu,
2242 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
2243 vcpu->arch.pv_eoi.msr_val = 0;
2244 apic_update_ppr(apic);
2245 if (vcpu->arch.apicv_active) {
2246 kvm_x86_ops->apicv_post_state_restore(vcpu);
2247 kvm_x86_ops->hwapic_irr_update(vcpu, -1);
2248 kvm_x86_ops->hwapic_isr_update(vcpu, -1);
2251 vcpu->arch.apic_arb_prio = 0;
2252 vcpu->arch.apic_attention = 0;
2256 *----------------------------------------------------------------------
2258 *----------------------------------------------------------------------
2261 static bool lapic_is_periodic(struct kvm_lapic *apic)
2263 return apic_lvtt_period(apic);
2266 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2268 struct kvm_lapic *apic = vcpu->arch.apic;
2270 if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2271 return atomic_read(&apic->lapic_timer.pending);
2276 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2278 u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2279 int vector, mode, trig_mode;
2281 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2282 vector = reg & APIC_VECTOR_MASK;
2283 mode = reg & APIC_MODE_MASK;
2284 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2285 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2291 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2293 struct kvm_lapic *apic = vcpu->arch.apic;
2296 kvm_apic_local_deliver(apic, APIC_LVT0);
2299 static const struct kvm_io_device_ops apic_mmio_ops = {
2300 .read = apic_mmio_read,
2301 .write = apic_mmio_write,
2304 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2306 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2307 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2309 apic_timer_expired(apic);
2311 if (lapic_is_periodic(apic)) {
2312 advance_periodic_target_expiration(apic);
2313 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2314 return HRTIMER_RESTART;
2316 return HRTIMER_NORESTART;
2319 int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
2321 struct kvm_lapic *apic;
2323 ASSERT(vcpu != NULL);
2325 apic = kzalloc(sizeof(*apic), GFP_KERNEL_ACCOUNT);
2329 vcpu->arch.apic = apic;
2331 apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
2333 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2335 goto nomem_free_apic;
2339 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2340 HRTIMER_MODE_ABS_HARD);
2341 apic->lapic_timer.timer.function = apic_timer_fn;
2342 if (timer_advance_ns == -1) {
2343 apic->lapic_timer.timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
2344 lapic_timer_advance_dynamic = true;
2346 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
2347 lapic_timer_advance_dynamic = false;
2351 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2352 * thinking that APIC state has changed.
2354 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2355 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2356 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2361 vcpu->arch.apic = NULL;
2366 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2368 struct kvm_lapic *apic = vcpu->arch.apic;
2371 if (!kvm_apic_hw_enabled(apic))
2374 __apic_update_ppr(apic, &ppr);
2375 return apic_has_interrupt_for_ppr(apic, ppr);
2378 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2380 u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2383 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2385 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2386 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2391 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2393 struct kvm_lapic *apic = vcpu->arch.apic;
2395 if (atomic_read(&apic->lapic_timer.pending) > 0) {
2396 kvm_apic_inject_pending_timer_irqs(apic);
2397 atomic_set(&apic->lapic_timer.pending, 0);
2401 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2403 int vector = kvm_apic_has_interrupt(vcpu);
2404 struct kvm_lapic *apic = vcpu->arch.apic;
2411 * We get here even with APIC virtualization enabled, if doing
2412 * nested virtualization and L1 runs with the "acknowledge interrupt
2413 * on exit" mode. Then we cannot inject the interrupt via RVI,
2414 * because the process would deliver it through the IDT.
2417 apic_clear_irr(vector, apic);
2418 if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2420 * For auto-EOI interrupts, there might be another pending
2421 * interrupt above PPR, so check whether to raise another
2424 apic_update_ppr(apic);
2427 * For normal interrupts, PPR has been raised and there cannot
2428 * be a higher-priority pending interrupt---except if there was
2429 * a concurrent interrupt injection, but that would have
2430 * triggered KVM_REQ_EVENT already.
2432 apic_set_isr(vector, apic);
2433 __apic_update_ppr(apic, &ppr);
2439 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2440 struct kvm_lapic_state *s, bool set)
2442 if (apic_x2apic_mode(vcpu->arch.apic)) {
2443 u32 *id = (u32 *)(s->regs + APIC_ID);
2444 u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2446 if (vcpu->kvm->arch.x2apic_format) {
2447 if (*id != vcpu->vcpu_id)
2456 /* In x2APIC mode, the LDR is fixed and based on the id */
2458 *ldr = kvm_apic_calc_x2apic_ldr(*id);
2464 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2466 memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2467 return kvm_apic_state_fixup(vcpu, s, false);
2470 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2472 struct kvm_lapic *apic = vcpu->arch.apic;
2476 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2477 /* set SPIV separately to get count of SW disabled APICs right */
2478 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2480 r = kvm_apic_state_fixup(vcpu, s, true);
2483 memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
2485 recalculate_apic_map(vcpu->kvm);
2486 kvm_apic_set_version(vcpu);
2488 apic_update_ppr(apic);
2489 hrtimer_cancel(&apic->lapic_timer.timer);
2490 apic_update_lvtt(apic);
2491 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2492 update_divide_count(apic);
2493 start_apic_timer(apic);
2494 apic->irr_pending = true;
2495 apic->isr_count = vcpu->arch.apicv_active ?
2496 1 : count_vectors(apic->regs + APIC_ISR);
2497 apic->highest_isr_cache = -1;
2498 if (vcpu->arch.apicv_active) {
2499 kvm_x86_ops->apicv_post_state_restore(vcpu);
2500 kvm_x86_ops->hwapic_irr_update(vcpu,
2501 apic_find_highest_irr(apic));
2502 kvm_x86_ops->hwapic_isr_update(vcpu,
2503 apic_find_highest_isr(apic));
2505 kvm_make_request(KVM_REQ_EVENT, vcpu);
2506 if (ioapic_in_kernel(vcpu->kvm))
2507 kvm_rtc_eoi_tracking_restore_one(vcpu);
2509 vcpu->arch.apic_arb_prio = 0;
2514 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2516 struct hrtimer *timer;
2518 if (!lapic_in_kernel(vcpu) ||
2519 kvm_can_post_timer_interrupt(vcpu))
2522 timer = &vcpu->arch.apic->lapic_timer.timer;
2523 if (hrtimer_cancel(timer))
2524 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_HARD);
2528 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2530 * Detect whether guest triggered PV EOI since the
2531 * last entry. If yes, set EOI on guests's behalf.
2532 * Clear PV EOI in guest memory in any case.
2534 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2535 struct kvm_lapic *apic)
2540 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2541 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2543 * KVM_APIC_PV_EOI_PENDING is unset:
2544 * -> host disabled PV EOI.
2545 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2546 * -> host enabled PV EOI, guest did not execute EOI yet.
2547 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2548 * -> host enabled PV EOI, guest executed EOI.
2550 BUG_ON(!pv_eoi_enabled(vcpu));
2551 pending = pv_eoi_get_pending(vcpu);
2553 * Clear pending bit in any case: it will be set again on vmentry.
2554 * While this might not be ideal from performance point of view,
2555 * this makes sure pv eoi is only enabled when we know it's safe.
2557 pv_eoi_clr_pending(vcpu);
2560 vector = apic_set_eoi(apic);
2561 trace_kvm_pv_eoi(apic, vector);
2564 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2568 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2569 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2571 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2574 if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2578 apic_set_tpr(vcpu->arch.apic, data & 0xff);
2582 * apic_sync_pv_eoi_to_guest - called before vmentry
2584 * Detect whether it's safe to enable PV EOI and
2587 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2588 struct kvm_lapic *apic)
2590 if (!pv_eoi_enabled(vcpu) ||
2591 /* IRR set or many bits in ISR: could be nested. */
2592 apic->irr_pending ||
2593 /* Cache not set: could be safe but we don't bother. */
2594 apic->highest_isr_cache == -1 ||
2595 /* Need EOI to update ioapic. */
2596 kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2598 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2599 * so we need not do anything here.
2604 pv_eoi_set_pending(apic->vcpu);
2607 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2610 int max_irr, max_isr;
2611 struct kvm_lapic *apic = vcpu->arch.apic;
2613 apic_sync_pv_eoi_to_guest(vcpu, apic);
2615 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2618 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2619 max_irr = apic_find_highest_irr(apic);
2622 max_isr = apic_find_highest_isr(apic);
2625 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2627 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2631 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2634 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2635 &vcpu->arch.apic->vapic_cache,
2636 vapic_addr, sizeof(u32)))
2638 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2640 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2643 vcpu->arch.apic->vapic_addr = vapic_addr;
2647 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2649 struct kvm_lapic *apic = vcpu->arch.apic;
2650 u32 reg = (msr - APIC_BASE_MSR) << 4;
2652 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2655 if (reg == APIC_ICR2)
2658 /* if this is ICR write vector before command */
2659 if (reg == APIC_ICR)
2660 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2661 return kvm_lapic_reg_write(apic, reg, (u32)data);
2664 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2666 struct kvm_lapic *apic = vcpu->arch.apic;
2667 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2669 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2672 if (reg == APIC_DFR || reg == APIC_ICR2)
2675 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2677 if (reg == APIC_ICR)
2678 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2680 *data = (((u64)high) << 32) | low;
2685 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2687 struct kvm_lapic *apic = vcpu->arch.apic;
2689 if (!lapic_in_kernel(vcpu))
2692 /* if this is ICR write vector before command */
2693 if (reg == APIC_ICR)
2694 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2695 return kvm_lapic_reg_write(apic, reg, (u32)data);
2698 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2700 struct kvm_lapic *apic = vcpu->arch.apic;
2703 if (!lapic_in_kernel(vcpu))
2706 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2708 if (reg == APIC_ICR)
2709 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2711 *data = (((u64)high) << 32) | low;
2716 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data, unsigned long len)
2718 u64 addr = data & ~KVM_MSR_ENABLED;
2719 struct gfn_to_hva_cache *ghc = &vcpu->arch.pv_eoi.data;
2720 unsigned long new_len;
2722 if (!IS_ALIGNED(addr, 4))
2725 vcpu->arch.pv_eoi.msr_val = data;
2726 if (!pv_eoi_enabled(vcpu))
2729 if (addr == ghc->gpa && len <= ghc->len)
2734 return kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, addr, new_len);
2737 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2739 struct kvm_lapic *apic = vcpu->arch.apic;
2743 if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2747 * INITs are latched while CPU is in specific states
2748 * (SMM, VMX non-root mode, SVM with GIF=0).
2749 * Because a CPU cannot be in these states immediately
2750 * after it has processed an INIT signal (and thus in
2751 * KVM_MP_STATE_INIT_RECEIVED state), just eat SIPIs
2752 * and leave the INIT pending.
2754 if (kvm_vcpu_latch_init(vcpu)) {
2755 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2756 if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2757 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2761 pe = xchg(&apic->pending_events, 0);
2762 if (test_bit(KVM_APIC_INIT, &pe)) {
2763 kvm_vcpu_reset(vcpu, true);
2764 if (kvm_vcpu_is_bsp(apic->vcpu))
2765 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2767 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2769 if (test_bit(KVM_APIC_SIPI, &pe) &&
2770 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2771 /* evaluate pending_events before reading the vector */
2773 sipi_vector = apic->sipi_vector;
2774 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2775 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2779 void kvm_lapic_init(void)
2781 /* do not patch jump label more than once per second */
2782 jump_label_rate_limit(&apic_hw_disabled, HZ);
2783 jump_label_rate_limit(&apic_sw_disabled, HZ);
2786 void kvm_lapic_exit(void)
2788 static_key_deferred_flush(&apic_hw_disabled);
2789 static_key_deferred_flush(&apic_sw_disabled);