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"
46 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
48 #define mod_64(x, y) ((x) % (y))
56 /* 14 is the version for Xeon and Pentium 8.4.8*/
57 #define APIC_VERSION (0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
58 #define LAPIC_MMIO_LENGTH (1 << 12)
59 /* followed define is not in apicdef.h */
60 #define MAX_APIC_VECTOR 256
61 #define APIC_VECTORS_PER_REG 32
63 static bool lapic_timer_advance_dynamic __read_mostly;
64 #define LAPIC_TIMER_ADVANCE_ADJUST_MIN 100 /* clock cycles */
65 #define LAPIC_TIMER_ADVANCE_ADJUST_MAX 10000 /* clock cycles */
66 #define LAPIC_TIMER_ADVANCE_NS_INIT 1000
67 #define LAPIC_TIMER_ADVANCE_NS_MAX 5000
68 /* step-by-step approximation to mitigate fluctuation */
69 #define LAPIC_TIMER_ADVANCE_ADJUST_STEP 8
71 static inline int apic_test_vector(int vec, void *bitmap)
73 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
76 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
78 struct kvm_lapic *apic = vcpu->arch.apic;
80 return apic_test_vector(vector, apic->regs + APIC_ISR) ||
81 apic_test_vector(vector, apic->regs + APIC_IRR);
84 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
86 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
89 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
91 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
94 __read_mostly DEFINE_STATIC_KEY_DEFERRED_FALSE(apic_hw_disabled, HZ);
95 __read_mostly DEFINE_STATIC_KEY_DEFERRED_FALSE(apic_sw_disabled, HZ);
97 static inline int apic_enabled(struct kvm_lapic *apic)
99 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
103 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
106 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
107 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
109 static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
111 return apic->vcpu->vcpu_id;
114 static bool kvm_can_post_timer_interrupt(struct kvm_vcpu *vcpu)
116 return pi_inject_timer && kvm_vcpu_apicv_active(vcpu) &&
117 (kvm_mwait_in_guest(vcpu->kvm) || kvm_hlt_in_guest(vcpu->kvm));
120 bool kvm_can_use_hv_timer(struct kvm_vcpu *vcpu)
122 return kvm_x86_ops.set_hv_timer
123 && !(kvm_mwait_in_guest(vcpu->kvm) ||
124 kvm_can_post_timer_interrupt(vcpu));
126 EXPORT_SYMBOL_GPL(kvm_can_use_hv_timer);
128 static bool kvm_use_posted_timer_interrupt(struct kvm_vcpu *vcpu)
130 return kvm_can_post_timer_interrupt(vcpu) && vcpu->mode == IN_GUEST_MODE;
133 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
134 u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
136 case KVM_APIC_MODE_X2APIC: {
137 u32 offset = (dest_id >> 16) * 16;
138 u32 max_apic_id = map->max_apic_id;
140 if (offset <= max_apic_id) {
141 u8 cluster_size = min(max_apic_id - offset + 1, 16U);
143 offset = array_index_nospec(offset, map->max_apic_id + 1);
144 *cluster = &map->phys_map[offset];
145 *mask = dest_id & (0xffff >> (16 - cluster_size));
152 case KVM_APIC_MODE_XAPIC_FLAT:
153 *cluster = map->xapic_flat_map;
154 *mask = dest_id & 0xff;
156 case KVM_APIC_MODE_XAPIC_CLUSTER:
157 *cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
158 *mask = dest_id & 0xf;
166 static void kvm_apic_map_free(struct rcu_head *rcu)
168 struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
174 * CLEAN -> DIRTY and UPDATE_IN_PROGRESS -> DIRTY changes happen without a lock.
176 * DIRTY -> UPDATE_IN_PROGRESS and UPDATE_IN_PROGRESS -> CLEAN happen with
177 * apic_map_lock_held.
185 void kvm_recalculate_apic_map(struct kvm *kvm)
187 struct kvm_apic_map *new, *old = NULL;
188 struct kvm_vcpu *vcpu;
190 u32 max_id = 255; /* enough space for any xAPIC ID */
192 /* Read kvm->arch.apic_map_dirty before kvm->arch.apic_map. */
193 if (atomic_read_acquire(&kvm->arch.apic_map_dirty) == CLEAN)
196 WARN_ONCE(!irqchip_in_kernel(kvm),
197 "Dirty APIC map without an in-kernel local APIC");
199 mutex_lock(&kvm->arch.apic_map_lock);
201 * Read kvm->arch.apic_map_dirty before kvm->arch.apic_map
202 * (if clean) or the APIC registers (if dirty).
204 if (atomic_cmpxchg_acquire(&kvm->arch.apic_map_dirty,
205 DIRTY, UPDATE_IN_PROGRESS) == CLEAN) {
206 /* Someone else has updated the map. */
207 mutex_unlock(&kvm->arch.apic_map_lock);
211 kvm_for_each_vcpu(i, vcpu, kvm)
212 if (kvm_apic_present(vcpu))
213 max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
215 new = kvzalloc(sizeof(struct kvm_apic_map) +
216 sizeof(struct kvm_lapic *) * ((u64)max_id + 1),
222 new->max_apic_id = max_id;
224 kvm_for_each_vcpu(i, vcpu, kvm) {
225 struct kvm_lapic *apic = vcpu->arch.apic;
226 struct kvm_lapic **cluster;
232 if (!kvm_apic_present(vcpu))
235 xapic_id = kvm_xapic_id(apic);
236 x2apic_id = kvm_x2apic_id(apic);
238 /* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
239 if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
240 x2apic_id <= new->max_apic_id)
241 new->phys_map[x2apic_id] = apic;
243 * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
244 * prevent them from masking VCPUs with APIC ID <= 0xff.
246 if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
247 new->phys_map[xapic_id] = apic;
249 if (!kvm_apic_sw_enabled(apic))
252 ldr = kvm_lapic_get_reg(apic, APIC_LDR);
254 if (apic_x2apic_mode(apic)) {
255 new->mode |= KVM_APIC_MODE_X2APIC;
257 ldr = GET_APIC_LOGICAL_ID(ldr);
258 if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
259 new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
261 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
264 if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
268 cluster[ffs(mask) - 1] = apic;
271 old = rcu_dereference_protected(kvm->arch.apic_map,
272 lockdep_is_held(&kvm->arch.apic_map_lock));
273 rcu_assign_pointer(kvm->arch.apic_map, new);
275 * Write kvm->arch.apic_map before clearing apic->apic_map_dirty.
276 * If another update has come in, leave it DIRTY.
278 atomic_cmpxchg_release(&kvm->arch.apic_map_dirty,
279 UPDATE_IN_PROGRESS, CLEAN);
280 mutex_unlock(&kvm->arch.apic_map_lock);
283 call_rcu(&old->rcu, kvm_apic_map_free);
285 kvm_make_scan_ioapic_request(kvm);
288 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
290 bool enabled = val & APIC_SPIV_APIC_ENABLED;
292 kvm_lapic_set_reg(apic, APIC_SPIV, val);
294 if (enabled != apic->sw_enabled) {
295 apic->sw_enabled = enabled;
297 static_branch_slow_dec_deferred(&apic_sw_disabled);
299 static_branch_inc(&apic_sw_disabled.key);
301 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
304 /* Check if there are APF page ready requests pending */
306 kvm_make_request(KVM_REQ_APF_READY, apic->vcpu);
309 static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
311 kvm_lapic_set_reg(apic, APIC_ID, id << 24);
312 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
315 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
317 kvm_lapic_set_reg(apic, APIC_LDR, id);
318 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
321 static inline void kvm_apic_set_dfr(struct kvm_lapic *apic, u32 val)
323 kvm_lapic_set_reg(apic, APIC_DFR, val);
324 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
327 static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
329 return ((id >> 4) << 16) | (1 << (id & 0xf));
332 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
334 u32 ldr = kvm_apic_calc_x2apic_ldr(id);
336 WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
338 kvm_lapic_set_reg(apic, APIC_ID, id);
339 kvm_lapic_set_reg(apic, APIC_LDR, ldr);
340 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
343 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
345 return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
348 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
350 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
353 static inline int apic_lvtt_period(struct kvm_lapic *apic)
355 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
358 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
360 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
363 static inline int apic_lvt_nmi_mode(u32 lvt_val)
365 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
368 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
370 struct kvm_lapic *apic = vcpu->arch.apic;
371 u32 v = APIC_VERSION;
373 if (!lapic_in_kernel(vcpu))
377 * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
378 * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
379 * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
380 * version first and level-triggered interrupts never get EOIed in
383 if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) &&
384 !ioapic_in_kernel(vcpu->kvm))
385 v |= APIC_LVR_DIRECTED_EOI;
386 kvm_lapic_set_reg(apic, APIC_LVR, v);
389 static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
390 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
391 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
392 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
393 LINT_MASK, LINT_MASK, /* LVT0-1 */
394 LVT_MASK /* LVTERR */
397 static int find_highest_vector(void *bitmap)
402 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
403 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
404 reg = bitmap + REG_POS(vec);
406 return __fls(*reg) + vec;
412 static u8 count_vectors(void *bitmap)
418 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
419 reg = bitmap + REG_POS(vec);
420 count += hweight32(*reg);
426 bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
429 u32 pir_val, irr_val, prev_irr_val;
432 max_updated_irr = -1;
435 for (i = vec = 0; i <= 7; i++, vec += 32) {
436 pir_val = READ_ONCE(pir[i]);
437 irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
439 prev_irr_val = irr_val;
440 irr_val |= xchg(&pir[i], 0);
441 *((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
442 if (prev_irr_val != irr_val) {
444 __fls(irr_val ^ prev_irr_val) + vec;
448 *max_irr = __fls(irr_val) + vec;
451 return ((max_updated_irr != -1) &&
452 (max_updated_irr == *max_irr));
454 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
456 bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
458 struct kvm_lapic *apic = vcpu->arch.apic;
460 return __kvm_apic_update_irr(pir, apic->regs, max_irr);
462 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
464 static inline int apic_search_irr(struct kvm_lapic *apic)
466 return find_highest_vector(apic->regs + APIC_IRR);
469 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
474 * Note that irr_pending is just a hint. It will be always
475 * true with virtual interrupt delivery enabled.
477 if (!apic->irr_pending)
480 result = apic_search_irr(apic);
481 ASSERT(result == -1 || result >= 16);
486 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
488 struct kvm_vcpu *vcpu;
492 if (unlikely(vcpu->arch.apicv_active)) {
493 /* need to update RVI */
494 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
495 static_call_cond(kvm_x86_hwapic_irr_update)(vcpu, apic_find_highest_irr(apic));
497 apic->irr_pending = false;
498 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
499 if (apic_search_irr(apic) != -1)
500 apic->irr_pending = true;
504 void kvm_apic_clear_irr(struct kvm_vcpu *vcpu, int vec)
506 apic_clear_irr(vec, vcpu->arch.apic);
508 EXPORT_SYMBOL_GPL(kvm_apic_clear_irr);
510 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
512 struct kvm_vcpu *vcpu;
514 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
520 * With APIC virtualization enabled, all caching is disabled
521 * because the processor can modify ISR under the hood. Instead
524 if (unlikely(vcpu->arch.apicv_active))
525 static_call_cond(kvm_x86_hwapic_isr_update)(vcpu, vec);
528 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
530 * ISR (in service register) bit is set when injecting an interrupt.
531 * The highest vector is injected. Thus the latest bit set matches
532 * the highest bit in ISR.
534 apic->highest_isr_cache = vec;
538 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
543 * Note that isr_count is always 1, and highest_isr_cache
544 * is always -1, with APIC virtualization enabled.
546 if (!apic->isr_count)
548 if (likely(apic->highest_isr_cache != -1))
549 return apic->highest_isr_cache;
551 result = find_highest_vector(apic->regs + APIC_ISR);
552 ASSERT(result == -1 || result >= 16);
557 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
559 struct kvm_vcpu *vcpu;
560 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
566 * We do get here for APIC virtualization enabled if the guest
567 * uses the Hyper-V APIC enlightenment. In this case we may need
568 * to trigger a new interrupt delivery by writing the SVI field;
569 * on the other hand isr_count and highest_isr_cache are unused
570 * and must be left alone.
572 if (unlikely(vcpu->arch.apicv_active))
573 static_call_cond(kvm_x86_hwapic_isr_update)(vcpu, apic_find_highest_isr(apic));
576 BUG_ON(apic->isr_count < 0);
577 apic->highest_isr_cache = -1;
581 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
583 /* This may race with setting of irr in __apic_accept_irq() and
584 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
585 * will cause vmexit immediately and the value will be recalculated
586 * on the next vmentry.
588 return apic_find_highest_irr(vcpu->arch.apic);
590 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
592 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
593 int vector, int level, int trig_mode,
594 struct dest_map *dest_map);
596 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
597 struct dest_map *dest_map)
599 struct kvm_lapic *apic = vcpu->arch.apic;
601 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
602 irq->level, irq->trig_mode, dest_map);
605 static int __pv_send_ipi(unsigned long *ipi_bitmap, struct kvm_apic_map *map,
606 struct kvm_lapic_irq *irq, u32 min)
609 struct kvm_vcpu *vcpu;
611 if (min > map->max_apic_id)
614 for_each_set_bit(i, ipi_bitmap,
615 min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
616 if (map->phys_map[min + i]) {
617 vcpu = map->phys_map[min + i]->vcpu;
618 count += kvm_apic_set_irq(vcpu, irq, NULL);
625 int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
626 unsigned long ipi_bitmap_high, u32 min,
627 unsigned long icr, int op_64_bit)
629 struct kvm_apic_map *map;
630 struct kvm_lapic_irq irq = {0};
631 int cluster_size = op_64_bit ? 64 : 32;
634 if (icr & (APIC_DEST_MASK | APIC_SHORT_MASK))
637 irq.vector = icr & APIC_VECTOR_MASK;
638 irq.delivery_mode = icr & APIC_MODE_MASK;
639 irq.level = (icr & APIC_INT_ASSERT) != 0;
640 irq.trig_mode = icr & APIC_INT_LEVELTRIG;
643 map = rcu_dereference(kvm->arch.apic_map);
647 count = __pv_send_ipi(&ipi_bitmap_low, map, &irq, min);
649 count += __pv_send_ipi(&ipi_bitmap_high, map, &irq, min);
656 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
659 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
663 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
666 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
670 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
672 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
675 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
677 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0)
680 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
683 static bool pv_eoi_test_and_clr_pending(struct kvm_vcpu *vcpu)
687 if (pv_eoi_get_user(vcpu, &val) < 0)
690 val &= KVM_PV_EOI_ENABLED;
692 if (val && pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0)
696 * Clear pending bit in any case: it will be set again on vmentry.
697 * While this might not be ideal from performance point of view,
698 * this makes sure pv eoi is only enabled when we know it's safe.
700 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
705 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
708 if (kvm_x86_ops.sync_pir_to_irr)
709 highest_irr = static_call(kvm_x86_sync_pir_to_irr)(apic->vcpu);
711 highest_irr = apic_find_highest_irr(apic);
712 if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
717 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
719 u32 tpr, isrv, ppr, old_ppr;
722 old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
723 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
724 isr = apic_find_highest_isr(apic);
725 isrv = (isr != -1) ? isr : 0;
727 if ((tpr & 0xf0) >= (isrv & 0xf0))
734 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
736 return ppr < old_ppr;
739 static void apic_update_ppr(struct kvm_lapic *apic)
743 if (__apic_update_ppr(apic, &ppr) &&
744 apic_has_interrupt_for_ppr(apic, ppr) != -1)
745 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
748 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
750 apic_update_ppr(vcpu->arch.apic);
752 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
754 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
756 kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
757 apic_update_ppr(apic);
760 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
762 return mda == (apic_x2apic_mode(apic) ?
763 X2APIC_BROADCAST : APIC_BROADCAST);
766 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
768 if (kvm_apic_broadcast(apic, mda))
771 if (apic_x2apic_mode(apic))
772 return mda == kvm_x2apic_id(apic);
775 * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
776 * it were in x2APIC mode. Hotplugged VCPUs start in xAPIC mode and
777 * this allows unique addressing of VCPUs with APIC ID over 0xff.
778 * The 0xff condition is needed because writeable xAPIC ID.
780 if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
783 return mda == kvm_xapic_id(apic);
786 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
790 if (kvm_apic_broadcast(apic, mda))
793 logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
795 if (apic_x2apic_mode(apic))
796 return ((logical_id >> 16) == (mda >> 16))
797 && (logical_id & mda & 0xffff) != 0;
799 logical_id = GET_APIC_LOGICAL_ID(logical_id);
801 switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
803 return (logical_id & mda) != 0;
804 case APIC_DFR_CLUSTER:
805 return ((logical_id >> 4) == (mda >> 4))
806 && (logical_id & mda & 0xf) != 0;
812 /* The KVM local APIC implementation has two quirks:
814 * - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
815 * in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
816 * KVM doesn't do that aliasing.
818 * - in-kernel IOAPIC messages have to be delivered directly to
819 * x2APIC, because the kernel does not support interrupt remapping.
820 * In order to support broadcast without interrupt remapping, x2APIC
821 * rewrites the destination of non-IPI messages from APIC_BROADCAST
822 * to X2APIC_BROADCAST.
824 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API. This is
825 * important when userspace wants to use x2APIC-format MSIs, because
826 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
828 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
829 struct kvm_lapic *source, struct kvm_lapic *target)
831 bool ipi = source != NULL;
833 if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
834 !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
835 return X2APIC_BROADCAST;
840 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
841 int shorthand, unsigned int dest, int dest_mode)
843 struct kvm_lapic *target = vcpu->arch.apic;
844 u32 mda = kvm_apic_mda(vcpu, dest, source, target);
848 case APIC_DEST_NOSHORT:
849 if (dest_mode == APIC_DEST_PHYSICAL)
850 return kvm_apic_match_physical_addr(target, mda);
852 return kvm_apic_match_logical_addr(target, mda);
854 return target == source;
855 case APIC_DEST_ALLINC:
857 case APIC_DEST_ALLBUT:
858 return target != source;
863 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
865 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
866 const unsigned long *bitmap, u32 bitmap_size)
871 mod = vector % dest_vcpus;
873 for (i = 0; i <= mod; i++) {
874 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
875 BUG_ON(idx == bitmap_size);
881 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
883 if (!kvm->arch.disabled_lapic_found) {
884 kvm->arch.disabled_lapic_found = true;
886 "Disabled LAPIC found during irq injection\n");
890 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
891 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
893 if (kvm->arch.x2apic_broadcast_quirk_disabled) {
894 if ((irq->dest_id == APIC_BROADCAST &&
895 map->mode != KVM_APIC_MODE_X2APIC))
897 if (irq->dest_id == X2APIC_BROADCAST)
900 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
901 if (irq->dest_id == (x2apic_ipi ?
902 X2APIC_BROADCAST : APIC_BROADCAST))
909 /* Return true if the interrupt can be handled by using *bitmap as index mask
910 * for valid destinations in *dst array.
911 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
912 * Note: we may have zero kvm_lapic destinations when we return true, which
913 * means that the interrupt should be dropped. In this case, *bitmap would be
914 * zero and *dst undefined.
916 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
917 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
918 struct kvm_apic_map *map, struct kvm_lapic ***dst,
919 unsigned long *bitmap)
923 if (irq->shorthand == APIC_DEST_SELF && src) {
927 } else if (irq->shorthand)
930 if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
933 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
934 if (irq->dest_id > map->max_apic_id) {
937 u32 dest_id = array_index_nospec(irq->dest_id, map->max_apic_id + 1);
938 *dst = &map->phys_map[dest_id];
945 if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
949 if (!kvm_lowest_prio_delivery(irq))
952 if (!kvm_vector_hashing_enabled()) {
954 for_each_set_bit(i, bitmap, 16) {
959 else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
960 (*dst)[lowest]->vcpu) < 0)
967 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
970 if (!(*dst)[lowest]) {
971 kvm_apic_disabled_lapic_found(kvm);
977 *bitmap = (lowest >= 0) ? 1 << lowest : 0;
982 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
983 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
985 struct kvm_apic_map *map;
986 unsigned long bitmap;
987 struct kvm_lapic **dst = NULL;
993 if (irq->shorthand == APIC_DEST_SELF) {
994 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
999 map = rcu_dereference(kvm->arch.apic_map);
1001 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
1004 for_each_set_bit(i, &bitmap, 16) {
1007 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
1016 * This routine tries to handle interrupts in posted mode, here is how
1017 * it deals with different cases:
1018 * - For single-destination interrupts, handle it in posted mode
1019 * - Else if vector hashing is enabled and it is a lowest-priority
1020 * interrupt, handle it in posted mode and use the following mechanism
1021 * to find the destination vCPU.
1022 * 1. For lowest-priority interrupts, store all the possible
1023 * destination vCPUs in an array.
1024 * 2. Use "guest vector % max number of destination vCPUs" to find
1025 * the right destination vCPU in the array for the lowest-priority
1027 * - Otherwise, use remapped mode to inject the interrupt.
1029 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
1030 struct kvm_vcpu **dest_vcpu)
1032 struct kvm_apic_map *map;
1033 unsigned long bitmap;
1034 struct kvm_lapic **dst = NULL;
1041 map = rcu_dereference(kvm->arch.apic_map);
1043 if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
1044 hweight16(bitmap) == 1) {
1045 unsigned long i = find_first_bit(&bitmap, 16);
1048 *dest_vcpu = dst[i]->vcpu;
1058 * Add a pending IRQ into lapic.
1059 * Return 1 if successfully added and 0 if discarded.
1061 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
1062 int vector, int level, int trig_mode,
1063 struct dest_map *dest_map)
1066 struct kvm_vcpu *vcpu = apic->vcpu;
1068 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
1070 switch (delivery_mode) {
1071 case APIC_DM_LOWEST:
1072 vcpu->arch.apic_arb_prio++;
1075 if (unlikely(trig_mode && !level))
1078 /* FIXME add logic for vcpu on reset */
1079 if (unlikely(!apic_enabled(apic)))
1085 __set_bit(vcpu->vcpu_id, dest_map->map);
1086 dest_map->vectors[vcpu->vcpu_id] = vector;
1089 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
1091 kvm_lapic_set_vector(vector,
1092 apic->regs + APIC_TMR);
1094 kvm_lapic_clear_vector(vector,
1095 apic->regs + APIC_TMR);
1098 static_call(kvm_x86_deliver_interrupt)(apic, delivery_mode,
1104 vcpu->arch.pv.pv_unhalted = 1;
1105 kvm_make_request(KVM_REQ_EVENT, vcpu);
1106 kvm_vcpu_kick(vcpu);
1111 kvm_make_request(KVM_REQ_SMI, vcpu);
1112 kvm_vcpu_kick(vcpu);
1117 kvm_inject_nmi(vcpu);
1118 kvm_vcpu_kick(vcpu);
1122 if (!trig_mode || level) {
1124 /* assumes that there are only KVM_APIC_INIT/SIPI */
1125 apic->pending_events = (1UL << KVM_APIC_INIT);
1126 kvm_make_request(KVM_REQ_EVENT, vcpu);
1127 kvm_vcpu_kick(vcpu);
1131 case APIC_DM_STARTUP:
1133 apic->sipi_vector = vector;
1134 /* make sure sipi_vector is visible for the receiver */
1136 set_bit(KVM_APIC_SIPI, &apic->pending_events);
1137 kvm_make_request(KVM_REQ_EVENT, vcpu);
1138 kvm_vcpu_kick(vcpu);
1141 case APIC_DM_EXTINT:
1143 * Should only be called by kvm_apic_local_deliver() with LVT0,
1144 * before NMI watchdog was enabled. Already handled by
1145 * kvm_apic_accept_pic_intr().
1150 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1158 * This routine identifies the destination vcpus mask meant to receive the
1159 * IOAPIC interrupts. It either uses kvm_apic_map_get_dest_lapic() to find
1160 * out the destination vcpus array and set the bitmap or it traverses to
1161 * each available vcpu to identify the same.
1163 void kvm_bitmap_or_dest_vcpus(struct kvm *kvm, struct kvm_lapic_irq *irq,
1164 unsigned long *vcpu_bitmap)
1166 struct kvm_lapic **dest_vcpu = NULL;
1167 struct kvm_lapic *src = NULL;
1168 struct kvm_apic_map *map;
1169 struct kvm_vcpu *vcpu;
1170 unsigned long bitmap, i;
1175 map = rcu_dereference(kvm->arch.apic_map);
1177 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dest_vcpu,
1180 for_each_set_bit(i, &bitmap, 16) {
1183 vcpu_idx = dest_vcpu[i]->vcpu->vcpu_idx;
1184 __set_bit(vcpu_idx, vcpu_bitmap);
1187 kvm_for_each_vcpu(i, vcpu, kvm) {
1188 if (!kvm_apic_present(vcpu))
1190 if (!kvm_apic_match_dest(vcpu, NULL,
1195 __set_bit(i, vcpu_bitmap);
1201 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1203 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1206 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1208 return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1211 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1215 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1216 if (!kvm_ioapic_handles_vector(apic, vector))
1219 /* Request a KVM exit to inform the userspace IOAPIC. */
1220 if (irqchip_split(apic->vcpu->kvm)) {
1221 apic->vcpu->arch.pending_ioapic_eoi = vector;
1222 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1226 if (apic_test_vector(vector, apic->regs + APIC_TMR))
1227 trigger_mode = IOAPIC_LEVEL_TRIG;
1229 trigger_mode = IOAPIC_EDGE_TRIG;
1231 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1234 static int apic_set_eoi(struct kvm_lapic *apic)
1236 int vector = apic_find_highest_isr(apic);
1238 trace_kvm_eoi(apic, vector);
1241 * Not every write EOI will has corresponding ISR,
1242 * one example is when Kernel check timer on setup_IO_APIC
1247 apic_clear_isr(vector, apic);
1248 apic_update_ppr(apic);
1250 if (to_hv_vcpu(apic->vcpu) &&
1251 test_bit(vector, to_hv_synic(apic->vcpu)->vec_bitmap))
1252 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1254 kvm_ioapic_send_eoi(apic, vector);
1255 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1260 * this interface assumes a trap-like exit, which has already finished
1261 * desired side effect including vISR and vPPR update.
1263 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1265 struct kvm_lapic *apic = vcpu->arch.apic;
1267 trace_kvm_eoi(apic, vector);
1269 kvm_ioapic_send_eoi(apic, vector);
1270 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1272 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1274 void kvm_apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
1276 struct kvm_lapic_irq irq;
1278 irq.vector = icr_low & APIC_VECTOR_MASK;
1279 irq.delivery_mode = icr_low & APIC_MODE_MASK;
1280 irq.dest_mode = icr_low & APIC_DEST_MASK;
1281 irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1282 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1283 irq.shorthand = icr_low & APIC_SHORT_MASK;
1284 irq.msi_redir_hint = false;
1285 if (apic_x2apic_mode(apic))
1286 irq.dest_id = icr_high;
1288 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1290 trace_kvm_apic_ipi(icr_low, irq.dest_id);
1292 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1295 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1297 ktime_t remaining, now;
1301 ASSERT(apic != NULL);
1303 /* if initial count is 0, current count should also be 0 */
1304 if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1305 apic->lapic_timer.period == 0)
1309 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1310 if (ktime_to_ns(remaining) < 0)
1313 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1314 tmcct = div64_u64(ns,
1315 (APIC_BUS_CYCLE_NS * apic->divide_count));
1320 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1322 struct kvm_vcpu *vcpu = apic->vcpu;
1323 struct kvm_run *run = vcpu->run;
1325 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1326 run->tpr_access.rip = kvm_rip_read(vcpu);
1327 run->tpr_access.is_write = write;
1330 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1332 if (apic->vcpu->arch.tpr_access_reporting)
1333 __report_tpr_access(apic, write);
1336 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1340 if (offset >= LAPIC_MMIO_LENGTH)
1347 case APIC_TMCCT: /* Timer CCR */
1348 if (apic_lvtt_tscdeadline(apic))
1351 val = apic_get_tmcct(apic);
1354 apic_update_ppr(apic);
1355 val = kvm_lapic_get_reg(apic, offset);
1358 report_tpr_access(apic, false);
1361 val = kvm_lapic_get_reg(apic, offset);
1368 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1370 return container_of(dev, struct kvm_lapic, dev);
1373 #define APIC_REG_MASK(reg) (1ull << ((reg) >> 4))
1374 #define APIC_REGS_MASK(first, count) \
1375 (APIC_REG_MASK(first) * ((1ull << (count)) - 1))
1377 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1380 unsigned char alignment = offset & 0xf;
1382 /* this bitmask has a bit cleared for each reserved register */
1383 u64 valid_reg_mask =
1384 APIC_REG_MASK(APIC_ID) |
1385 APIC_REG_MASK(APIC_LVR) |
1386 APIC_REG_MASK(APIC_TASKPRI) |
1387 APIC_REG_MASK(APIC_PROCPRI) |
1388 APIC_REG_MASK(APIC_LDR) |
1389 APIC_REG_MASK(APIC_DFR) |
1390 APIC_REG_MASK(APIC_SPIV) |
1391 APIC_REGS_MASK(APIC_ISR, APIC_ISR_NR) |
1392 APIC_REGS_MASK(APIC_TMR, APIC_ISR_NR) |
1393 APIC_REGS_MASK(APIC_IRR, APIC_ISR_NR) |
1394 APIC_REG_MASK(APIC_ESR) |
1395 APIC_REG_MASK(APIC_ICR) |
1396 APIC_REG_MASK(APIC_ICR2) |
1397 APIC_REG_MASK(APIC_LVTT) |
1398 APIC_REG_MASK(APIC_LVTTHMR) |
1399 APIC_REG_MASK(APIC_LVTPC) |
1400 APIC_REG_MASK(APIC_LVT0) |
1401 APIC_REG_MASK(APIC_LVT1) |
1402 APIC_REG_MASK(APIC_LVTERR) |
1403 APIC_REG_MASK(APIC_TMICT) |
1404 APIC_REG_MASK(APIC_TMCCT) |
1405 APIC_REG_MASK(APIC_TDCR);
1407 /* ARBPRI is not valid on x2APIC */
1408 if (!apic_x2apic_mode(apic))
1409 valid_reg_mask |= APIC_REG_MASK(APIC_ARBPRI);
1411 if (alignment + len > 4)
1414 if (offset > 0x3f0 || !(valid_reg_mask & APIC_REG_MASK(offset)))
1417 result = __apic_read(apic, offset & ~0xf);
1419 trace_kvm_apic_read(offset, result);
1425 memcpy(data, (char *)&result + alignment, len);
1428 printk(KERN_ERR "Local APIC read with len = %x, "
1429 "should be 1,2, or 4 instead\n", len);
1434 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1436 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1438 return addr >= apic->base_address &&
1439 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1442 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1443 gpa_t address, int len, void *data)
1445 struct kvm_lapic *apic = to_lapic(this);
1446 u32 offset = address - apic->base_address;
1448 if (!apic_mmio_in_range(apic, address))
1451 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1452 if (!kvm_check_has_quirk(vcpu->kvm,
1453 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1456 memset(data, 0xff, len);
1460 kvm_lapic_reg_read(apic, offset, len, data);
1465 static void update_divide_count(struct kvm_lapic *apic)
1467 u32 tmp1, tmp2, tdcr;
1469 tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1471 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1472 apic->divide_count = 0x1 << (tmp2 & 0x7);
1475 static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1478 * Do not allow the guest to program periodic timers with small
1479 * interval, since the hrtimers are not throttled by the host
1482 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1483 s64 min_period = min_timer_period_us * 1000LL;
1485 if (apic->lapic_timer.period < min_period) {
1486 pr_info_ratelimited(
1487 "kvm: vcpu %i: requested %lld ns "
1488 "lapic timer period limited to %lld ns\n",
1489 apic->vcpu->vcpu_id,
1490 apic->lapic_timer.period, min_period);
1491 apic->lapic_timer.period = min_period;
1496 static void cancel_hv_timer(struct kvm_lapic *apic);
1498 static void cancel_apic_timer(struct kvm_lapic *apic)
1500 hrtimer_cancel(&apic->lapic_timer.timer);
1502 if (apic->lapic_timer.hv_timer_in_use)
1503 cancel_hv_timer(apic);
1507 static void apic_update_lvtt(struct kvm_lapic *apic)
1509 u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1510 apic->lapic_timer.timer_mode_mask;
1512 if (apic->lapic_timer.timer_mode != timer_mode) {
1513 if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1514 APIC_LVT_TIMER_TSCDEADLINE)) {
1515 cancel_apic_timer(apic);
1516 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1517 apic->lapic_timer.period = 0;
1518 apic->lapic_timer.tscdeadline = 0;
1520 apic->lapic_timer.timer_mode = timer_mode;
1521 limit_periodic_timer_frequency(apic);
1526 * On APICv, this test will cause a busy wait
1527 * during a higher-priority task.
1530 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1532 struct kvm_lapic *apic = vcpu->arch.apic;
1533 u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1535 if (kvm_apic_hw_enabled(apic)) {
1536 int vec = reg & APIC_VECTOR_MASK;
1537 void *bitmap = apic->regs + APIC_ISR;
1539 if (vcpu->arch.apicv_active)
1540 bitmap = apic->regs + APIC_IRR;
1542 if (apic_test_vector(vec, bitmap))
1548 static inline void __wait_lapic_expire(struct kvm_vcpu *vcpu, u64 guest_cycles)
1550 u64 timer_advance_ns = vcpu->arch.apic->lapic_timer.timer_advance_ns;
1553 * If the guest TSC is running at a different ratio than the host, then
1554 * convert the delay to nanoseconds to achieve an accurate delay. Note
1555 * that __delay() uses delay_tsc whenever the hardware has TSC, thus
1556 * always for VMX enabled hardware.
1558 if (vcpu->arch.tsc_scaling_ratio == kvm_default_tsc_scaling_ratio) {
1559 __delay(min(guest_cycles,
1560 nsec_to_cycles(vcpu, timer_advance_ns)));
1562 u64 delay_ns = guest_cycles * 1000000ULL;
1563 do_div(delay_ns, vcpu->arch.virtual_tsc_khz);
1564 ndelay(min_t(u32, delay_ns, timer_advance_ns));
1568 static inline void adjust_lapic_timer_advance(struct kvm_vcpu *vcpu,
1569 s64 advance_expire_delta)
1571 struct kvm_lapic *apic = vcpu->arch.apic;
1572 u32 timer_advance_ns = apic->lapic_timer.timer_advance_ns;
1575 /* Do not adjust for tiny fluctuations or large random spikes. */
1576 if (abs(advance_expire_delta) > LAPIC_TIMER_ADVANCE_ADJUST_MAX ||
1577 abs(advance_expire_delta) < LAPIC_TIMER_ADVANCE_ADJUST_MIN)
1581 if (advance_expire_delta < 0) {
1582 ns = -advance_expire_delta * 1000000ULL;
1583 do_div(ns, vcpu->arch.virtual_tsc_khz);
1584 timer_advance_ns -= ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1587 ns = advance_expire_delta * 1000000ULL;
1588 do_div(ns, vcpu->arch.virtual_tsc_khz);
1589 timer_advance_ns += ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1592 if (unlikely(timer_advance_ns > LAPIC_TIMER_ADVANCE_NS_MAX))
1593 timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
1594 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
1597 static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1599 struct kvm_lapic *apic = vcpu->arch.apic;
1600 u64 guest_tsc, tsc_deadline;
1602 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1603 apic->lapic_timer.expired_tscdeadline = 0;
1604 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1605 apic->lapic_timer.advance_expire_delta = guest_tsc - tsc_deadline;
1607 if (lapic_timer_advance_dynamic) {
1608 adjust_lapic_timer_advance(vcpu, apic->lapic_timer.advance_expire_delta);
1610 * If the timer fired early, reread the TSC to account for the
1611 * overhead of the above adjustment to avoid waiting longer
1612 * than is necessary.
1614 if (guest_tsc < tsc_deadline)
1615 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1618 if (guest_tsc < tsc_deadline)
1619 __wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
1622 void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1624 if (lapic_in_kernel(vcpu) &&
1625 vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1626 vcpu->arch.apic->lapic_timer.timer_advance_ns &&
1627 lapic_timer_int_injected(vcpu))
1628 __kvm_wait_lapic_expire(vcpu);
1630 EXPORT_SYMBOL_GPL(kvm_wait_lapic_expire);
1632 static void kvm_apic_inject_pending_timer_irqs(struct kvm_lapic *apic)
1634 struct kvm_timer *ktimer = &apic->lapic_timer;
1636 kvm_apic_local_deliver(apic, APIC_LVTT);
1637 if (apic_lvtt_tscdeadline(apic)) {
1638 ktimer->tscdeadline = 0;
1639 } else if (apic_lvtt_oneshot(apic)) {
1640 ktimer->tscdeadline = 0;
1641 ktimer->target_expiration = 0;
1645 static void apic_timer_expired(struct kvm_lapic *apic, bool from_timer_fn)
1647 struct kvm_vcpu *vcpu = apic->vcpu;
1648 struct kvm_timer *ktimer = &apic->lapic_timer;
1650 if (atomic_read(&apic->lapic_timer.pending))
1653 if (apic_lvtt_tscdeadline(apic) || ktimer->hv_timer_in_use)
1654 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1656 if (!from_timer_fn && vcpu->arch.apicv_active) {
1657 WARN_ON(kvm_get_running_vcpu() != vcpu);
1658 kvm_apic_inject_pending_timer_irqs(apic);
1662 if (kvm_use_posted_timer_interrupt(apic->vcpu)) {
1664 * Ensure the guest's timer has truly expired before posting an
1665 * interrupt. Open code the relevant checks to avoid querying
1666 * lapic_timer_int_injected(), which will be false since the
1667 * interrupt isn't yet injected. Waiting until after injecting
1668 * is not an option since that won't help a posted interrupt.
1670 if (vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1671 vcpu->arch.apic->lapic_timer.timer_advance_ns)
1672 __kvm_wait_lapic_expire(vcpu);
1673 kvm_apic_inject_pending_timer_irqs(apic);
1677 atomic_inc(&apic->lapic_timer.pending);
1678 kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
1680 kvm_vcpu_kick(vcpu);
1683 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1685 struct kvm_timer *ktimer = &apic->lapic_timer;
1686 u64 guest_tsc, tscdeadline = ktimer->tscdeadline;
1689 struct kvm_vcpu *vcpu = apic->vcpu;
1690 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1691 unsigned long flags;
1694 if (unlikely(!tscdeadline || !this_tsc_khz))
1697 local_irq_save(flags);
1700 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1702 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1703 do_div(ns, this_tsc_khz);
1705 if (likely(tscdeadline > guest_tsc) &&
1706 likely(ns > apic->lapic_timer.timer_advance_ns)) {
1707 expire = ktime_add_ns(now, ns);
1708 expire = ktime_sub_ns(expire, ktimer->timer_advance_ns);
1709 hrtimer_start(&ktimer->timer, expire, HRTIMER_MODE_ABS_HARD);
1711 apic_timer_expired(apic, false);
1713 local_irq_restore(flags);
1716 static inline u64 tmict_to_ns(struct kvm_lapic *apic, u32 tmict)
1718 return (u64)tmict * APIC_BUS_CYCLE_NS * (u64)apic->divide_count;
1721 static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1723 ktime_t now, remaining;
1724 u64 ns_remaining_old, ns_remaining_new;
1726 apic->lapic_timer.period =
1727 tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1728 limit_periodic_timer_frequency(apic);
1731 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1732 if (ktime_to_ns(remaining) < 0)
1735 ns_remaining_old = ktime_to_ns(remaining);
1736 ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1737 apic->divide_count, old_divisor);
1739 apic->lapic_timer.tscdeadline +=
1740 nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1741 nsec_to_cycles(apic->vcpu, ns_remaining_old);
1742 apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1745 static bool set_target_expiration(struct kvm_lapic *apic, u32 count_reg)
1752 apic->lapic_timer.period =
1753 tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1755 if (!apic->lapic_timer.period) {
1756 apic->lapic_timer.tscdeadline = 0;
1760 limit_periodic_timer_frequency(apic);
1761 deadline = apic->lapic_timer.period;
1763 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1764 if (unlikely(count_reg != APIC_TMICT)) {
1765 deadline = tmict_to_ns(apic,
1766 kvm_lapic_get_reg(apic, count_reg));
1767 if (unlikely(deadline <= 0))
1768 deadline = apic->lapic_timer.period;
1769 else if (unlikely(deadline > apic->lapic_timer.period)) {
1770 pr_info_ratelimited(
1771 "kvm: vcpu %i: requested lapic timer restore with "
1772 "starting count register %#x=%u (%lld ns) > initial count (%lld ns). "
1773 "Using initial count to start timer.\n",
1774 apic->vcpu->vcpu_id,
1776 kvm_lapic_get_reg(apic, count_reg),
1777 deadline, apic->lapic_timer.period);
1778 kvm_lapic_set_reg(apic, count_reg, 0);
1779 deadline = apic->lapic_timer.period;
1784 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1785 nsec_to_cycles(apic->vcpu, deadline);
1786 apic->lapic_timer.target_expiration = ktime_add_ns(now, deadline);
1791 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1793 ktime_t now = ktime_get();
1798 * Synchronize both deadlines to the same time source or
1799 * differences in the periods (caused by differences in the
1800 * underlying clocks or numerical approximation errors) will
1801 * cause the two to drift apart over time as the errors
1804 apic->lapic_timer.target_expiration =
1805 ktime_add_ns(apic->lapic_timer.target_expiration,
1806 apic->lapic_timer.period);
1807 delta = ktime_sub(apic->lapic_timer.target_expiration, now);
1808 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1809 nsec_to_cycles(apic->vcpu, delta);
1812 static void start_sw_period(struct kvm_lapic *apic)
1814 if (!apic->lapic_timer.period)
1817 if (ktime_after(ktime_get(),
1818 apic->lapic_timer.target_expiration)) {
1819 apic_timer_expired(apic, false);
1821 if (apic_lvtt_oneshot(apic))
1824 advance_periodic_target_expiration(apic);
1827 hrtimer_start(&apic->lapic_timer.timer,
1828 apic->lapic_timer.target_expiration,
1829 HRTIMER_MODE_ABS_HARD);
1832 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1834 if (!lapic_in_kernel(vcpu))
1837 return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1839 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1841 static void cancel_hv_timer(struct kvm_lapic *apic)
1843 WARN_ON(preemptible());
1844 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1845 static_call(kvm_x86_cancel_hv_timer)(apic->vcpu);
1846 apic->lapic_timer.hv_timer_in_use = false;
1849 static bool start_hv_timer(struct kvm_lapic *apic)
1851 struct kvm_timer *ktimer = &apic->lapic_timer;
1852 struct kvm_vcpu *vcpu = apic->vcpu;
1855 WARN_ON(preemptible());
1856 if (!kvm_can_use_hv_timer(vcpu))
1859 if (!ktimer->tscdeadline)
1862 if (static_call(kvm_x86_set_hv_timer)(vcpu, ktimer->tscdeadline, &expired))
1865 ktimer->hv_timer_in_use = true;
1866 hrtimer_cancel(&ktimer->timer);
1869 * To simplify handling the periodic timer, leave the hv timer running
1870 * even if the deadline timer has expired, i.e. rely on the resulting
1871 * VM-Exit to recompute the periodic timer's target expiration.
1873 if (!apic_lvtt_period(apic)) {
1875 * Cancel the hv timer if the sw timer fired while the hv timer
1876 * was being programmed, or if the hv timer itself expired.
1878 if (atomic_read(&ktimer->pending)) {
1879 cancel_hv_timer(apic);
1880 } else if (expired) {
1881 apic_timer_expired(apic, false);
1882 cancel_hv_timer(apic);
1886 trace_kvm_hv_timer_state(vcpu->vcpu_id, ktimer->hv_timer_in_use);
1891 static void start_sw_timer(struct kvm_lapic *apic)
1893 struct kvm_timer *ktimer = &apic->lapic_timer;
1895 WARN_ON(preemptible());
1896 if (apic->lapic_timer.hv_timer_in_use)
1897 cancel_hv_timer(apic);
1898 if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1901 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1902 start_sw_period(apic);
1903 else if (apic_lvtt_tscdeadline(apic))
1904 start_sw_tscdeadline(apic);
1905 trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1908 static void restart_apic_timer(struct kvm_lapic *apic)
1912 if (!apic_lvtt_period(apic) && atomic_read(&apic->lapic_timer.pending))
1915 if (!start_hv_timer(apic))
1916 start_sw_timer(apic);
1921 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1923 struct kvm_lapic *apic = vcpu->arch.apic;
1926 /* If the preempt notifier has already run, it also called apic_timer_expired */
1927 if (!apic->lapic_timer.hv_timer_in_use)
1929 WARN_ON(kvm_vcpu_is_blocking(vcpu));
1930 apic_timer_expired(apic, false);
1931 cancel_hv_timer(apic);
1933 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1934 advance_periodic_target_expiration(apic);
1935 restart_apic_timer(apic);
1940 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1942 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1944 restart_apic_timer(vcpu->arch.apic);
1947 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1949 struct kvm_lapic *apic = vcpu->arch.apic;
1952 /* Possibly the TSC deadline timer is not enabled yet */
1953 if (apic->lapic_timer.hv_timer_in_use)
1954 start_sw_timer(apic);
1958 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
1960 struct kvm_lapic *apic = vcpu->arch.apic;
1962 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1963 restart_apic_timer(apic);
1966 static void __start_apic_timer(struct kvm_lapic *apic, u32 count_reg)
1968 atomic_set(&apic->lapic_timer.pending, 0);
1970 if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1971 && !set_target_expiration(apic, count_reg))
1974 restart_apic_timer(apic);
1977 static void start_apic_timer(struct kvm_lapic *apic)
1979 __start_apic_timer(apic, APIC_TMICT);
1982 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1984 bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1986 if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1987 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1988 if (lvt0_in_nmi_mode) {
1989 atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1991 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1995 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1999 trace_kvm_apic_write(reg, val);
2002 case APIC_ID: /* Local APIC ID */
2003 if (!apic_x2apic_mode(apic))
2004 kvm_apic_set_xapic_id(apic, val >> 24);
2010 report_tpr_access(apic, true);
2011 apic_set_tpr(apic, val & 0xff);
2019 if (!apic_x2apic_mode(apic))
2020 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
2026 if (!apic_x2apic_mode(apic))
2027 kvm_apic_set_dfr(apic, val | 0x0FFFFFFF);
2034 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
2035 mask |= APIC_SPIV_DIRECTED_EOI;
2036 apic_set_spiv(apic, val & mask);
2037 if (!(val & APIC_SPIV_APIC_ENABLED)) {
2041 for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
2042 lvt_val = kvm_lapic_get_reg(apic,
2043 APIC_LVTT + 0x10 * i);
2044 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
2045 lvt_val | APIC_LVT_MASKED);
2047 apic_update_lvtt(apic);
2048 atomic_set(&apic->lapic_timer.pending, 0);
2054 /* No delay here, so we always clear the pending bit */
2056 kvm_apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
2057 kvm_lapic_set_reg(apic, APIC_ICR, val);
2061 if (!apic_x2apic_mode(apic))
2063 kvm_lapic_set_reg(apic, APIC_ICR2, val);
2067 apic_manage_nmi_watchdog(apic, val);
2073 /* TODO: Check vector */
2077 if (!kvm_apic_sw_enabled(apic))
2078 val |= APIC_LVT_MASKED;
2079 size = ARRAY_SIZE(apic_lvt_mask);
2080 index = array_index_nospec(
2081 (reg - APIC_LVTT) >> 4, size);
2082 val &= apic_lvt_mask[index];
2083 kvm_lapic_set_reg(apic, reg, val);
2088 if (!kvm_apic_sw_enabled(apic))
2089 val |= APIC_LVT_MASKED;
2090 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
2091 kvm_lapic_set_reg(apic, APIC_LVTT, val);
2092 apic_update_lvtt(apic);
2096 if (apic_lvtt_tscdeadline(apic))
2099 cancel_apic_timer(apic);
2100 kvm_lapic_set_reg(apic, APIC_TMICT, val);
2101 start_apic_timer(apic);
2105 uint32_t old_divisor = apic->divide_count;
2107 kvm_lapic_set_reg(apic, APIC_TDCR, val & 0xb);
2108 update_divide_count(apic);
2109 if (apic->divide_count != old_divisor &&
2110 apic->lapic_timer.period) {
2111 hrtimer_cancel(&apic->lapic_timer.timer);
2112 update_target_expiration(apic, old_divisor);
2113 restart_apic_timer(apic);
2118 if (apic_x2apic_mode(apic) && val != 0)
2123 if (apic_x2apic_mode(apic)) {
2124 kvm_lapic_reg_write(apic, APIC_ICR,
2125 APIC_DEST_SELF | (val & APIC_VECTOR_MASK));
2134 kvm_recalculate_apic_map(apic->vcpu->kvm);
2138 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
2140 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
2141 gpa_t address, int len, const void *data)
2143 struct kvm_lapic *apic = to_lapic(this);
2144 unsigned int offset = address - apic->base_address;
2147 if (!apic_mmio_in_range(apic, address))
2150 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
2151 if (!kvm_check_has_quirk(vcpu->kvm,
2152 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
2159 * APIC register must be aligned on 128-bits boundary.
2160 * 32/64/128 bits registers must be accessed thru 32 bits.
2163 if (len != 4 || (offset & 0xf))
2168 kvm_lapic_reg_write(apic, offset & 0xff0, val);
2173 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
2175 kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
2177 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
2179 /* emulate APIC access in a trap manner */
2180 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
2184 /* hw has done the conditional check and inst decode */
2187 kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
2189 /* TODO: optimize to just emulate side effect w/o one more write */
2190 kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
2192 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
2194 void kvm_free_lapic(struct kvm_vcpu *vcpu)
2196 struct kvm_lapic *apic = vcpu->arch.apic;
2198 if (!vcpu->arch.apic)
2201 hrtimer_cancel(&apic->lapic_timer.timer);
2203 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
2204 static_branch_slow_dec_deferred(&apic_hw_disabled);
2206 if (!apic->sw_enabled)
2207 static_branch_slow_dec_deferred(&apic_sw_disabled);
2210 free_page((unsigned long)apic->regs);
2216 *----------------------------------------------------------------------
2218 *----------------------------------------------------------------------
2220 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
2222 struct kvm_lapic *apic = vcpu->arch.apic;
2224 if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2227 return apic->lapic_timer.tscdeadline;
2230 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
2232 struct kvm_lapic *apic = vcpu->arch.apic;
2234 if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2237 hrtimer_cancel(&apic->lapic_timer.timer);
2238 apic->lapic_timer.tscdeadline = data;
2239 start_apic_timer(apic);
2242 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
2244 apic_set_tpr(vcpu->arch.apic, (cr8 & 0x0f) << 4);
2247 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
2251 tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
2253 return (tpr & 0xf0) >> 4;
2256 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
2258 u64 old_value = vcpu->arch.apic_base;
2259 struct kvm_lapic *apic = vcpu->arch.apic;
2261 vcpu->arch.apic_base = value;
2263 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
2264 kvm_update_cpuid_runtime(vcpu);
2269 /* update jump label if enable bit changes */
2270 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
2271 if (value & MSR_IA32_APICBASE_ENABLE) {
2272 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2273 static_branch_slow_dec_deferred(&apic_hw_disabled);
2274 /* Check if there are APF page ready requests pending */
2275 kvm_make_request(KVM_REQ_APF_READY, vcpu);
2277 static_branch_inc(&apic_hw_disabled.key);
2278 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2282 if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
2283 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
2285 if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE))
2286 static_call_cond(kvm_x86_set_virtual_apic_mode)(vcpu);
2288 apic->base_address = apic->vcpu->arch.apic_base &
2289 MSR_IA32_APICBASE_BASE;
2291 if ((value & MSR_IA32_APICBASE_ENABLE) &&
2292 apic->base_address != APIC_DEFAULT_PHYS_BASE)
2293 pr_warn_once("APIC base relocation is unsupported by KVM");
2296 void kvm_apic_update_apicv(struct kvm_vcpu *vcpu)
2298 struct kvm_lapic *apic = vcpu->arch.apic;
2300 if (vcpu->arch.apicv_active) {
2301 /* irr_pending is always true when apicv is activated. */
2302 apic->irr_pending = true;
2303 apic->isr_count = 1;
2306 * Don't clear irr_pending, searching the IRR can race with
2307 * updates from the CPU as APICv is still active from hardware's
2308 * perspective. The flag will be cleared as appropriate when
2309 * KVM injects the interrupt.
2311 apic->isr_count = count_vectors(apic->regs + APIC_ISR);
2314 EXPORT_SYMBOL_GPL(kvm_apic_update_apicv);
2316 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2318 struct kvm_lapic *apic = vcpu->arch.apic;
2323 msr_val = APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE;
2324 if (kvm_vcpu_is_reset_bsp(vcpu))
2325 msr_val |= MSR_IA32_APICBASE_BSP;
2326 kvm_lapic_set_base(vcpu, msr_val);
2332 /* Stop the timer in case it's a reset to an active apic */
2333 hrtimer_cancel(&apic->lapic_timer.timer);
2335 /* The xAPIC ID is set at RESET even if the APIC was already enabled. */
2337 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2338 kvm_apic_set_version(apic->vcpu);
2340 for (i = 0; i < KVM_APIC_LVT_NUM; i++)
2341 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
2342 apic_update_lvtt(apic);
2343 if (kvm_vcpu_is_reset_bsp(vcpu) &&
2344 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2345 kvm_lapic_set_reg(apic, APIC_LVT0,
2346 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2347 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2349 kvm_apic_set_dfr(apic, 0xffffffffU);
2350 apic_set_spiv(apic, 0xff);
2351 kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2352 if (!apic_x2apic_mode(apic))
2353 kvm_apic_set_ldr(apic, 0);
2354 kvm_lapic_set_reg(apic, APIC_ESR, 0);
2355 kvm_lapic_set_reg(apic, APIC_ICR, 0);
2356 kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2357 kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2358 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2359 for (i = 0; i < 8; i++) {
2360 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2361 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2362 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2364 kvm_apic_update_apicv(vcpu);
2365 apic->highest_isr_cache = -1;
2366 update_divide_count(apic);
2367 atomic_set(&apic->lapic_timer.pending, 0);
2369 vcpu->arch.pv_eoi.msr_val = 0;
2370 apic_update_ppr(apic);
2371 if (vcpu->arch.apicv_active) {
2372 static_call_cond(kvm_x86_apicv_post_state_restore)(vcpu);
2373 static_call_cond(kvm_x86_hwapic_irr_update)(vcpu, -1);
2374 static_call_cond(kvm_x86_hwapic_isr_update)(vcpu, -1);
2377 vcpu->arch.apic_arb_prio = 0;
2378 vcpu->arch.apic_attention = 0;
2380 kvm_recalculate_apic_map(vcpu->kvm);
2384 *----------------------------------------------------------------------
2386 *----------------------------------------------------------------------
2389 static bool lapic_is_periodic(struct kvm_lapic *apic)
2391 return apic_lvtt_period(apic);
2394 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2396 struct kvm_lapic *apic = vcpu->arch.apic;
2398 if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2399 return atomic_read(&apic->lapic_timer.pending);
2404 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2406 u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2407 int vector, mode, trig_mode;
2409 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2410 vector = reg & APIC_VECTOR_MASK;
2411 mode = reg & APIC_MODE_MASK;
2412 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2413 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2419 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2421 struct kvm_lapic *apic = vcpu->arch.apic;
2424 kvm_apic_local_deliver(apic, APIC_LVT0);
2427 static const struct kvm_io_device_ops apic_mmio_ops = {
2428 .read = apic_mmio_read,
2429 .write = apic_mmio_write,
2432 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2434 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2435 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2437 apic_timer_expired(apic, true);
2439 if (lapic_is_periodic(apic)) {
2440 advance_periodic_target_expiration(apic);
2441 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2442 return HRTIMER_RESTART;
2444 return HRTIMER_NORESTART;
2447 int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
2449 struct kvm_lapic *apic;
2451 ASSERT(vcpu != NULL);
2453 apic = kzalloc(sizeof(*apic), GFP_KERNEL_ACCOUNT);
2457 vcpu->arch.apic = apic;
2459 apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
2461 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2463 goto nomem_free_apic;
2467 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2468 HRTIMER_MODE_ABS_HARD);
2469 apic->lapic_timer.timer.function = apic_timer_fn;
2470 if (timer_advance_ns == -1) {
2471 apic->lapic_timer.timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
2472 lapic_timer_advance_dynamic = true;
2474 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
2475 lapic_timer_advance_dynamic = false;
2479 * Stuff the APIC ENABLE bit in lieu of temporarily incrementing
2480 * apic_hw_disabled; the full RESET value is set by kvm_lapic_reset().
2482 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2483 static_branch_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2484 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2489 vcpu->arch.apic = NULL;
2494 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2496 struct kvm_lapic *apic = vcpu->arch.apic;
2499 if (!kvm_apic_present(vcpu))
2502 __apic_update_ppr(apic, &ppr);
2503 return apic_has_interrupt_for_ppr(apic, ppr);
2505 EXPORT_SYMBOL_GPL(kvm_apic_has_interrupt);
2507 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2509 u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2511 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2513 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2514 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2519 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2521 struct kvm_lapic *apic = vcpu->arch.apic;
2523 if (atomic_read(&apic->lapic_timer.pending) > 0) {
2524 kvm_apic_inject_pending_timer_irqs(apic);
2525 atomic_set(&apic->lapic_timer.pending, 0);
2529 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2531 int vector = kvm_apic_has_interrupt(vcpu);
2532 struct kvm_lapic *apic = vcpu->arch.apic;
2539 * We get here even with APIC virtualization enabled, if doing
2540 * nested virtualization and L1 runs with the "acknowledge interrupt
2541 * on exit" mode. Then we cannot inject the interrupt via RVI,
2542 * because the process would deliver it through the IDT.
2545 apic_clear_irr(vector, apic);
2546 if (to_hv_vcpu(vcpu) && test_bit(vector, to_hv_synic(vcpu)->auto_eoi_bitmap)) {
2548 * For auto-EOI interrupts, there might be another pending
2549 * interrupt above PPR, so check whether to raise another
2552 apic_update_ppr(apic);
2555 * For normal interrupts, PPR has been raised and there cannot
2556 * be a higher-priority pending interrupt---except if there was
2557 * a concurrent interrupt injection, but that would have
2558 * triggered KVM_REQ_EVENT already.
2560 apic_set_isr(vector, apic);
2561 __apic_update_ppr(apic, &ppr);
2567 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2568 struct kvm_lapic_state *s, bool set)
2570 if (apic_x2apic_mode(vcpu->arch.apic)) {
2571 u32 *id = (u32 *)(s->regs + APIC_ID);
2572 u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2574 if (vcpu->kvm->arch.x2apic_format) {
2575 if (*id != vcpu->vcpu_id)
2584 /* In x2APIC mode, the LDR is fixed and based on the id */
2586 *ldr = kvm_apic_calc_x2apic_ldr(*id);
2592 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2594 memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2597 * Get calculated timer current count for remaining timer period (if
2598 * any) and store it in the returned register set.
2600 __kvm_lapic_set_reg(s->regs, APIC_TMCCT,
2601 __apic_read(vcpu->arch.apic, APIC_TMCCT));
2603 return kvm_apic_state_fixup(vcpu, s, false);
2606 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2608 struct kvm_lapic *apic = vcpu->arch.apic;
2611 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2612 /* set SPIV separately to get count of SW disabled APICs right */
2613 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2615 r = kvm_apic_state_fixup(vcpu, s, true);
2617 kvm_recalculate_apic_map(vcpu->kvm);
2620 memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
2622 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2623 kvm_recalculate_apic_map(vcpu->kvm);
2624 kvm_apic_set_version(vcpu);
2626 apic_update_ppr(apic);
2627 cancel_apic_timer(apic);
2628 apic->lapic_timer.expired_tscdeadline = 0;
2629 apic_update_lvtt(apic);
2630 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2631 update_divide_count(apic);
2632 __start_apic_timer(apic, APIC_TMCCT);
2633 kvm_lapic_set_reg(apic, APIC_TMCCT, 0);
2634 kvm_apic_update_apicv(vcpu);
2635 apic->highest_isr_cache = -1;
2636 if (vcpu->arch.apicv_active) {
2637 static_call_cond(kvm_x86_apicv_post_state_restore)(vcpu);
2638 static_call_cond(kvm_x86_hwapic_irr_update)(vcpu, apic_find_highest_irr(apic));
2639 static_call_cond(kvm_x86_hwapic_isr_update)(vcpu, apic_find_highest_isr(apic));
2641 kvm_make_request(KVM_REQ_EVENT, vcpu);
2642 if (ioapic_in_kernel(vcpu->kvm))
2643 kvm_rtc_eoi_tracking_restore_one(vcpu);
2645 vcpu->arch.apic_arb_prio = 0;
2650 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2652 struct hrtimer *timer;
2654 if (!lapic_in_kernel(vcpu) ||
2655 kvm_can_post_timer_interrupt(vcpu))
2658 timer = &vcpu->arch.apic->lapic_timer.timer;
2659 if (hrtimer_cancel(timer))
2660 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_HARD);
2664 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2666 * Detect whether guest triggered PV EOI since the
2667 * last entry. If yes, set EOI on guests's behalf.
2668 * Clear PV EOI in guest memory in any case.
2670 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2671 struct kvm_lapic *apic)
2675 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2676 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2678 * KVM_APIC_PV_EOI_PENDING is unset:
2679 * -> host disabled PV EOI.
2680 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2681 * -> host enabled PV EOI, guest did not execute EOI yet.
2682 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2683 * -> host enabled PV EOI, guest executed EOI.
2685 BUG_ON(!pv_eoi_enabled(vcpu));
2687 if (pv_eoi_test_and_clr_pending(vcpu))
2689 vector = apic_set_eoi(apic);
2690 trace_kvm_pv_eoi(apic, vector);
2693 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2697 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2698 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2700 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2703 if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2707 apic_set_tpr(vcpu->arch.apic, data & 0xff);
2711 * apic_sync_pv_eoi_to_guest - called before vmentry
2713 * Detect whether it's safe to enable PV EOI and
2716 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2717 struct kvm_lapic *apic)
2719 if (!pv_eoi_enabled(vcpu) ||
2720 /* IRR set or many bits in ISR: could be nested. */
2721 apic->irr_pending ||
2722 /* Cache not set: could be safe but we don't bother. */
2723 apic->highest_isr_cache == -1 ||
2724 /* Need EOI to update ioapic. */
2725 kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2727 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2728 * so we need not do anything here.
2733 pv_eoi_set_pending(apic->vcpu);
2736 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2739 int max_irr, max_isr;
2740 struct kvm_lapic *apic = vcpu->arch.apic;
2742 apic_sync_pv_eoi_to_guest(vcpu, apic);
2744 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2747 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2748 max_irr = apic_find_highest_irr(apic);
2751 max_isr = apic_find_highest_isr(apic);
2754 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2756 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2760 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2763 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2764 &vcpu->arch.apic->vapic_cache,
2765 vapic_addr, sizeof(u32)))
2767 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2769 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2772 vcpu->arch.apic->vapic_addr = vapic_addr;
2776 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2778 struct kvm_lapic *apic = vcpu->arch.apic;
2779 u32 reg = (msr - APIC_BASE_MSR) << 4;
2781 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2784 if (reg == APIC_ICR2)
2787 /* if this is ICR write vector before command */
2788 if (reg == APIC_ICR)
2789 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2790 return kvm_lapic_reg_write(apic, reg, (u32)data);
2793 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2795 struct kvm_lapic *apic = vcpu->arch.apic;
2796 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2798 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2801 if (reg == APIC_DFR || reg == APIC_ICR2)
2804 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2806 if (reg == APIC_ICR)
2807 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2809 *data = (((u64)high) << 32) | low;
2814 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2816 struct kvm_lapic *apic = vcpu->arch.apic;
2818 if (!lapic_in_kernel(vcpu))
2821 /* if this is ICR write vector before command */
2822 if (reg == APIC_ICR)
2823 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2824 return kvm_lapic_reg_write(apic, reg, (u32)data);
2827 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2829 struct kvm_lapic *apic = vcpu->arch.apic;
2832 if (!lapic_in_kernel(vcpu))
2835 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2837 if (reg == APIC_ICR)
2838 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2840 *data = (((u64)high) << 32) | low;
2845 int kvm_lapic_set_pv_eoi(struct kvm_vcpu *vcpu, u64 data, unsigned long len)
2847 u64 addr = data & ~KVM_MSR_ENABLED;
2848 struct gfn_to_hva_cache *ghc = &vcpu->arch.pv_eoi.data;
2849 unsigned long new_len;
2852 if (!IS_ALIGNED(addr, 4))
2855 if (data & KVM_MSR_ENABLED) {
2856 if (addr == ghc->gpa && len <= ghc->len)
2861 ret = kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, addr, new_len);
2866 vcpu->arch.pv_eoi.msr_val = data;
2871 int kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2873 struct kvm_lapic *apic = vcpu->arch.apic;
2878 if (!lapic_in_kernel(vcpu))
2882 * Read pending events before calling the check_events
2885 pe = smp_load_acquire(&apic->pending_events);
2889 if (is_guest_mode(vcpu)) {
2890 r = kvm_check_nested_events(vcpu);
2892 return r == -EBUSY ? 0 : r;
2894 * If an event has happened and caused a vmexit,
2895 * we know INITs are latched and therefore
2896 * we will not incorrectly deliver an APIC
2897 * event instead of a vmexit.
2902 * INITs are latched while CPU is in specific states
2903 * (SMM, VMX root mode, SVM with GIF=0).
2904 * Because a CPU cannot be in these states immediately
2905 * after it has processed an INIT signal (and thus in
2906 * KVM_MP_STATE_INIT_RECEIVED state), just eat SIPIs
2907 * and leave the INIT pending.
2909 if (kvm_vcpu_latch_init(vcpu)) {
2910 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2911 if (test_bit(KVM_APIC_SIPI, &pe))
2912 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2916 if (test_bit(KVM_APIC_INIT, &pe)) {
2917 clear_bit(KVM_APIC_INIT, &apic->pending_events);
2918 kvm_vcpu_reset(vcpu, true);
2919 if (kvm_vcpu_is_bsp(apic->vcpu))
2920 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2922 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2924 if (test_bit(KVM_APIC_SIPI, &pe)) {
2925 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2926 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2927 /* evaluate pending_events before reading the vector */
2929 sipi_vector = apic->sipi_vector;
2930 static_call(kvm_x86_vcpu_deliver_sipi_vector)(vcpu, sipi_vector);
2931 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2937 void kvm_lapic_exit(void)
2939 static_key_deferred_flush(&apic_hw_disabled);
2940 WARN_ON(static_branch_unlikely(&apic_hw_disabled.key));
2941 static_key_deferred_flush(&apic_sw_disabled);
2942 WARN_ON(static_branch_unlikely(&apic_sw_disabled.key));