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);
119 bool kvm_can_use_hv_timer(struct kvm_vcpu *vcpu)
121 return kvm_x86_ops.set_hv_timer
122 && !(kvm_mwait_in_guest(vcpu->kvm) ||
123 kvm_can_post_timer_interrupt(vcpu));
125 EXPORT_SYMBOL_GPL(kvm_can_use_hv_timer);
127 static bool kvm_use_posted_timer_interrupt(struct kvm_vcpu *vcpu)
129 return kvm_can_post_timer_interrupt(vcpu) && vcpu->mode == IN_GUEST_MODE;
132 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
133 u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
135 case KVM_APIC_MODE_X2APIC: {
136 u32 offset = (dest_id >> 16) * 16;
137 u32 max_apic_id = map->max_apic_id;
139 if (offset <= max_apic_id) {
140 u8 cluster_size = min(max_apic_id - offset + 1, 16U);
142 offset = array_index_nospec(offset, map->max_apic_id + 1);
143 *cluster = &map->phys_map[offset];
144 *mask = dest_id & (0xffff >> (16 - cluster_size));
151 case KVM_APIC_MODE_XAPIC_FLAT:
152 *cluster = map->xapic_flat_map;
153 *mask = dest_id & 0xff;
155 case KVM_APIC_MODE_XAPIC_CLUSTER:
156 *cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
157 *mask = dest_id & 0xf;
165 static void kvm_apic_map_free(struct rcu_head *rcu)
167 struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
173 * CLEAN -> DIRTY and UPDATE_IN_PROGRESS -> DIRTY changes happen without a lock.
175 * DIRTY -> UPDATE_IN_PROGRESS and UPDATE_IN_PROGRESS -> CLEAN happen with
176 * apic_map_lock_held.
184 void kvm_recalculate_apic_map(struct kvm *kvm)
186 struct kvm_apic_map *new, *old = NULL;
187 struct kvm_vcpu *vcpu;
189 u32 max_id = 255; /* enough space for any xAPIC ID */
191 /* Read kvm->arch.apic_map_dirty before kvm->arch.apic_map. */
192 if (atomic_read_acquire(&kvm->arch.apic_map_dirty) == CLEAN)
195 mutex_lock(&kvm->arch.apic_map_lock);
197 * Read kvm->arch.apic_map_dirty before kvm->arch.apic_map
198 * (if clean) or the APIC registers (if dirty).
200 if (atomic_cmpxchg_acquire(&kvm->arch.apic_map_dirty,
201 DIRTY, UPDATE_IN_PROGRESS) == CLEAN) {
202 /* Someone else has updated the map. */
203 mutex_unlock(&kvm->arch.apic_map_lock);
207 kvm_for_each_vcpu(i, vcpu, kvm)
208 if (kvm_apic_present(vcpu))
209 max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
211 new = kvzalloc(sizeof(struct kvm_apic_map) +
212 sizeof(struct kvm_lapic *) * ((u64)max_id + 1),
218 new->max_apic_id = max_id;
220 kvm_for_each_vcpu(i, vcpu, kvm) {
221 struct kvm_lapic *apic = vcpu->arch.apic;
222 struct kvm_lapic **cluster;
228 if (!kvm_apic_present(vcpu))
231 xapic_id = kvm_xapic_id(apic);
232 x2apic_id = kvm_x2apic_id(apic);
234 /* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
235 if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
236 x2apic_id <= new->max_apic_id)
237 new->phys_map[x2apic_id] = apic;
239 * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
240 * prevent them from masking VCPUs with APIC ID <= 0xff.
242 if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
243 new->phys_map[xapic_id] = apic;
245 if (!kvm_apic_sw_enabled(apic))
248 ldr = kvm_lapic_get_reg(apic, APIC_LDR);
250 if (apic_x2apic_mode(apic)) {
251 new->mode |= KVM_APIC_MODE_X2APIC;
253 ldr = GET_APIC_LOGICAL_ID(ldr);
254 if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
255 new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
257 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
260 if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
264 cluster[ffs(mask) - 1] = apic;
267 old = rcu_dereference_protected(kvm->arch.apic_map,
268 lockdep_is_held(&kvm->arch.apic_map_lock));
269 rcu_assign_pointer(kvm->arch.apic_map, new);
271 * Write kvm->arch.apic_map before clearing apic->apic_map_dirty.
272 * If another update has come in, leave it DIRTY.
274 atomic_cmpxchg_release(&kvm->arch.apic_map_dirty,
275 UPDATE_IN_PROGRESS, CLEAN);
276 mutex_unlock(&kvm->arch.apic_map_lock);
279 call_rcu(&old->rcu, kvm_apic_map_free);
281 kvm_make_scan_ioapic_request(kvm);
284 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
286 bool enabled = val & APIC_SPIV_APIC_ENABLED;
288 kvm_lapic_set_reg(apic, APIC_SPIV, val);
290 if (enabled != apic->sw_enabled) {
291 apic->sw_enabled = enabled;
293 static_branch_slow_dec_deferred(&apic_sw_disabled);
295 static_branch_inc(&apic_sw_disabled.key);
297 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
301 static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
303 kvm_lapic_set_reg(apic, APIC_ID, id << 24);
304 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
307 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
309 kvm_lapic_set_reg(apic, APIC_LDR, id);
310 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
313 static inline void kvm_apic_set_dfr(struct kvm_lapic *apic, u32 val)
315 kvm_lapic_set_reg(apic, APIC_DFR, val);
316 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
319 static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
321 return ((id >> 4) << 16) | (1 << (id & 0xf));
324 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
326 u32 ldr = kvm_apic_calc_x2apic_ldr(id);
328 WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
330 kvm_lapic_set_reg(apic, APIC_ID, id);
331 kvm_lapic_set_reg(apic, APIC_LDR, ldr);
332 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
335 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
337 return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
340 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
342 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
345 static inline int apic_lvtt_period(struct kvm_lapic *apic)
347 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
350 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
352 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
355 static inline int apic_lvt_nmi_mode(u32 lvt_val)
357 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
360 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
362 struct kvm_lapic *apic = vcpu->arch.apic;
363 u32 v = APIC_VERSION;
365 if (!lapic_in_kernel(vcpu))
369 * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
370 * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
371 * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
372 * version first and level-triggered interrupts never get EOIed in
375 if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) &&
376 !ioapic_in_kernel(vcpu->kvm))
377 v |= APIC_LVR_DIRECTED_EOI;
378 kvm_lapic_set_reg(apic, APIC_LVR, v);
381 static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
382 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
383 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
384 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
385 LINT_MASK, LINT_MASK, /* LVT0-1 */
386 LVT_MASK /* LVTERR */
389 static int find_highest_vector(void *bitmap)
394 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
395 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
396 reg = bitmap + REG_POS(vec);
398 return __fls(*reg) + vec;
404 static u8 count_vectors(void *bitmap)
410 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
411 reg = bitmap + REG_POS(vec);
412 count += hweight32(*reg);
418 bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
421 u32 pir_val, irr_val, prev_irr_val;
424 max_updated_irr = -1;
427 for (i = vec = 0; i <= 7; i++, vec += 32) {
428 pir_val = READ_ONCE(pir[i]);
429 irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
431 prev_irr_val = irr_val;
432 irr_val |= xchg(&pir[i], 0);
433 *((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
434 if (prev_irr_val != irr_val) {
436 __fls(irr_val ^ prev_irr_val) + vec;
440 *max_irr = __fls(irr_val) + vec;
443 return ((max_updated_irr != -1) &&
444 (max_updated_irr == *max_irr));
446 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
448 bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
450 struct kvm_lapic *apic = vcpu->arch.apic;
452 return __kvm_apic_update_irr(pir, apic->regs, max_irr);
454 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
456 static inline int apic_search_irr(struct kvm_lapic *apic)
458 return find_highest_vector(apic->regs + APIC_IRR);
461 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
466 * Note that irr_pending is just a hint. It will be always
467 * true with virtual interrupt delivery enabled.
469 if (!apic->irr_pending)
472 result = apic_search_irr(apic);
473 ASSERT(result == -1 || result >= 16);
478 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
480 struct kvm_vcpu *vcpu;
484 if (unlikely(vcpu->arch.apicv_active)) {
485 /* need to update RVI */
486 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
487 static_call(kvm_x86_hwapic_irr_update)(vcpu,
488 apic_find_highest_irr(apic));
490 apic->irr_pending = false;
491 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
492 if (apic_search_irr(apic) != -1)
493 apic->irr_pending = true;
497 void kvm_apic_clear_irr(struct kvm_vcpu *vcpu, int vec)
499 apic_clear_irr(vec, vcpu->arch.apic);
501 EXPORT_SYMBOL_GPL(kvm_apic_clear_irr);
503 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
505 struct kvm_vcpu *vcpu;
507 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
513 * With APIC virtualization enabled, all caching is disabled
514 * because the processor can modify ISR under the hood. Instead
517 if (unlikely(vcpu->arch.apicv_active))
518 static_call(kvm_x86_hwapic_isr_update)(vcpu, vec);
521 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
523 * ISR (in service register) bit is set when injecting an interrupt.
524 * The highest vector is injected. Thus the latest bit set matches
525 * the highest bit in ISR.
527 apic->highest_isr_cache = vec;
531 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
536 * Note that isr_count is always 1, and highest_isr_cache
537 * is always -1, with APIC virtualization enabled.
539 if (!apic->isr_count)
541 if (likely(apic->highest_isr_cache != -1))
542 return apic->highest_isr_cache;
544 result = find_highest_vector(apic->regs + APIC_ISR);
545 ASSERT(result == -1 || result >= 16);
550 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
552 struct kvm_vcpu *vcpu;
553 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
559 * We do get here for APIC virtualization enabled if the guest
560 * uses the Hyper-V APIC enlightenment. In this case we may need
561 * to trigger a new interrupt delivery by writing the SVI field;
562 * on the other hand isr_count and highest_isr_cache are unused
563 * and must be left alone.
565 if (unlikely(vcpu->arch.apicv_active))
566 static_call(kvm_x86_hwapic_isr_update)(vcpu,
567 apic_find_highest_isr(apic));
570 BUG_ON(apic->isr_count < 0);
571 apic->highest_isr_cache = -1;
575 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
577 /* This may race with setting of irr in __apic_accept_irq() and
578 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
579 * will cause vmexit immediately and the value will be recalculated
580 * on the next vmentry.
582 return apic_find_highest_irr(vcpu->arch.apic);
584 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
586 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
587 int vector, int level, int trig_mode,
588 struct dest_map *dest_map);
590 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
591 struct dest_map *dest_map)
593 struct kvm_lapic *apic = vcpu->arch.apic;
595 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
596 irq->level, irq->trig_mode, dest_map);
599 static int __pv_send_ipi(unsigned long *ipi_bitmap, struct kvm_apic_map *map,
600 struct kvm_lapic_irq *irq, u32 min)
603 struct kvm_vcpu *vcpu;
605 if (min > map->max_apic_id)
608 for_each_set_bit(i, ipi_bitmap,
609 min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
610 if (map->phys_map[min + i]) {
611 vcpu = map->phys_map[min + i]->vcpu;
612 count += kvm_apic_set_irq(vcpu, irq, NULL);
619 int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
620 unsigned long ipi_bitmap_high, u32 min,
621 unsigned long icr, int op_64_bit)
623 struct kvm_apic_map *map;
624 struct kvm_lapic_irq irq = {0};
625 int cluster_size = op_64_bit ? 64 : 32;
628 if (icr & (APIC_DEST_MASK | APIC_SHORT_MASK))
631 irq.vector = icr & APIC_VECTOR_MASK;
632 irq.delivery_mode = icr & APIC_MODE_MASK;
633 irq.level = (icr & APIC_INT_ASSERT) != 0;
634 irq.trig_mode = icr & APIC_INT_LEVELTRIG;
637 map = rcu_dereference(kvm->arch.apic_map);
641 count = __pv_send_ipi(&ipi_bitmap_low, map, &irq, min);
643 count += __pv_send_ipi(&ipi_bitmap_high, map, &irq, min);
650 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
653 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
657 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
660 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
664 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
666 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
669 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
672 if (pv_eoi_get_user(vcpu, &val) < 0) {
673 printk(KERN_WARNING "Can't read EOI MSR value: 0x%llx\n",
674 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
677 return val & KVM_PV_EOI_ENABLED;
680 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
682 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
683 printk(KERN_WARNING "Can't set EOI MSR value: 0x%llx\n",
684 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
687 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
690 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
692 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
693 printk(KERN_WARNING "Can't clear EOI MSR value: 0x%llx\n",
694 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
697 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
700 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
703 if (apic->vcpu->arch.apicv_active)
704 highest_irr = static_call(kvm_x86_sync_pir_to_irr)(apic->vcpu);
706 highest_irr = apic_find_highest_irr(apic);
707 if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
712 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
714 u32 tpr, isrv, ppr, old_ppr;
717 old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
718 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
719 isr = apic_find_highest_isr(apic);
720 isrv = (isr != -1) ? isr : 0;
722 if ((tpr & 0xf0) >= (isrv & 0xf0))
729 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
731 return ppr < old_ppr;
734 static void apic_update_ppr(struct kvm_lapic *apic)
738 if (__apic_update_ppr(apic, &ppr) &&
739 apic_has_interrupt_for_ppr(apic, ppr) != -1)
740 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
743 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
745 apic_update_ppr(vcpu->arch.apic);
747 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
749 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
751 kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
752 apic_update_ppr(apic);
755 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
757 return mda == (apic_x2apic_mode(apic) ?
758 X2APIC_BROADCAST : APIC_BROADCAST);
761 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
763 if (kvm_apic_broadcast(apic, mda))
766 if (apic_x2apic_mode(apic))
767 return mda == kvm_x2apic_id(apic);
770 * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
771 * it were in x2APIC mode. Hotplugged VCPUs start in xAPIC mode and
772 * this allows unique addressing of VCPUs with APIC ID over 0xff.
773 * The 0xff condition is needed because writeable xAPIC ID.
775 if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
778 return mda == kvm_xapic_id(apic);
781 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
785 if (kvm_apic_broadcast(apic, mda))
788 logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
790 if (apic_x2apic_mode(apic))
791 return ((logical_id >> 16) == (mda >> 16))
792 && (logical_id & mda & 0xffff) != 0;
794 logical_id = GET_APIC_LOGICAL_ID(logical_id);
796 switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
798 return (logical_id & mda) != 0;
799 case APIC_DFR_CLUSTER:
800 return ((logical_id >> 4) == (mda >> 4))
801 && (logical_id & mda & 0xf) != 0;
807 /* The KVM local APIC implementation has two quirks:
809 * - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
810 * in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
811 * KVM doesn't do that aliasing.
813 * - in-kernel IOAPIC messages have to be delivered directly to
814 * x2APIC, because the kernel does not support interrupt remapping.
815 * In order to support broadcast without interrupt remapping, x2APIC
816 * rewrites the destination of non-IPI messages from APIC_BROADCAST
817 * to X2APIC_BROADCAST.
819 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API. This is
820 * important when userspace wants to use x2APIC-format MSIs, because
821 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
823 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
824 struct kvm_lapic *source, struct kvm_lapic *target)
826 bool ipi = source != NULL;
828 if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
829 !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
830 return X2APIC_BROADCAST;
835 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
836 int shorthand, unsigned int dest, int dest_mode)
838 struct kvm_lapic *target = vcpu->arch.apic;
839 u32 mda = kvm_apic_mda(vcpu, dest, source, target);
843 case APIC_DEST_NOSHORT:
844 if (dest_mode == APIC_DEST_PHYSICAL)
845 return kvm_apic_match_physical_addr(target, mda);
847 return kvm_apic_match_logical_addr(target, mda);
849 return target == source;
850 case APIC_DEST_ALLINC:
852 case APIC_DEST_ALLBUT:
853 return target != source;
858 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
860 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
861 const unsigned long *bitmap, u32 bitmap_size)
866 mod = vector % dest_vcpus;
868 for (i = 0; i <= mod; i++) {
869 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
870 BUG_ON(idx == bitmap_size);
876 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
878 if (!kvm->arch.disabled_lapic_found) {
879 kvm->arch.disabled_lapic_found = true;
881 "Disabled LAPIC found during irq injection\n");
885 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
886 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
888 if (kvm->arch.x2apic_broadcast_quirk_disabled) {
889 if ((irq->dest_id == APIC_BROADCAST &&
890 map->mode != KVM_APIC_MODE_X2APIC))
892 if (irq->dest_id == X2APIC_BROADCAST)
895 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
896 if (irq->dest_id == (x2apic_ipi ?
897 X2APIC_BROADCAST : APIC_BROADCAST))
904 /* Return true if the interrupt can be handled by using *bitmap as index mask
905 * for valid destinations in *dst array.
906 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
907 * Note: we may have zero kvm_lapic destinations when we return true, which
908 * means that the interrupt should be dropped. In this case, *bitmap would be
909 * zero and *dst undefined.
911 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
912 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
913 struct kvm_apic_map *map, struct kvm_lapic ***dst,
914 unsigned long *bitmap)
918 if (irq->shorthand == APIC_DEST_SELF && src) {
922 } else if (irq->shorthand)
925 if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
928 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
929 if (irq->dest_id > map->max_apic_id) {
932 u32 dest_id = array_index_nospec(irq->dest_id, map->max_apic_id + 1);
933 *dst = &map->phys_map[dest_id];
940 if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
944 if (!kvm_lowest_prio_delivery(irq))
947 if (!kvm_vector_hashing_enabled()) {
949 for_each_set_bit(i, bitmap, 16) {
954 else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
955 (*dst)[lowest]->vcpu) < 0)
962 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
965 if (!(*dst)[lowest]) {
966 kvm_apic_disabled_lapic_found(kvm);
972 *bitmap = (lowest >= 0) ? 1 << lowest : 0;
977 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
978 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
980 struct kvm_apic_map *map;
981 unsigned long bitmap;
982 struct kvm_lapic **dst = NULL;
988 if (irq->shorthand == APIC_DEST_SELF) {
989 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
994 map = rcu_dereference(kvm->arch.apic_map);
996 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
999 for_each_set_bit(i, &bitmap, 16) {
1002 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
1011 * This routine tries to handle interrupts in posted mode, here is how
1012 * it deals with different cases:
1013 * - For single-destination interrupts, handle it in posted mode
1014 * - Else if vector hashing is enabled and it is a lowest-priority
1015 * interrupt, handle it in posted mode and use the following mechanism
1016 * to find the destination vCPU.
1017 * 1. For lowest-priority interrupts, store all the possible
1018 * destination vCPUs in an array.
1019 * 2. Use "guest vector % max number of destination vCPUs" to find
1020 * the right destination vCPU in the array for the lowest-priority
1022 * - Otherwise, use remapped mode to inject the interrupt.
1024 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
1025 struct kvm_vcpu **dest_vcpu)
1027 struct kvm_apic_map *map;
1028 unsigned long bitmap;
1029 struct kvm_lapic **dst = NULL;
1036 map = rcu_dereference(kvm->arch.apic_map);
1038 if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
1039 hweight16(bitmap) == 1) {
1040 unsigned long i = find_first_bit(&bitmap, 16);
1043 *dest_vcpu = dst[i]->vcpu;
1053 * Add a pending IRQ into lapic.
1054 * Return 1 if successfully added and 0 if discarded.
1056 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
1057 int vector, int level, int trig_mode,
1058 struct dest_map *dest_map)
1061 struct kvm_vcpu *vcpu = apic->vcpu;
1063 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
1065 switch (delivery_mode) {
1066 case APIC_DM_LOWEST:
1067 vcpu->arch.apic_arb_prio++;
1070 if (unlikely(trig_mode && !level))
1073 /* FIXME add logic for vcpu on reset */
1074 if (unlikely(!apic_enabled(apic)))
1080 __set_bit(vcpu->vcpu_id, dest_map->map);
1081 dest_map->vectors[vcpu->vcpu_id] = vector;
1084 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
1086 kvm_lapic_set_vector(vector,
1087 apic->regs + APIC_TMR);
1089 kvm_lapic_clear_vector(vector,
1090 apic->regs + APIC_TMR);
1093 if (static_call(kvm_x86_deliver_posted_interrupt)(vcpu, vector)) {
1094 kvm_lapic_set_irr(vector, apic);
1095 kvm_make_request(KVM_REQ_EVENT, vcpu);
1096 kvm_vcpu_kick(vcpu);
1102 vcpu->arch.pv.pv_unhalted = 1;
1103 kvm_make_request(KVM_REQ_EVENT, vcpu);
1104 kvm_vcpu_kick(vcpu);
1109 kvm_make_request(KVM_REQ_SMI, vcpu);
1110 kvm_vcpu_kick(vcpu);
1115 kvm_inject_nmi(vcpu);
1116 kvm_vcpu_kick(vcpu);
1120 if (!trig_mode || level) {
1122 /* assumes that there are only KVM_APIC_INIT/SIPI */
1123 apic->pending_events = (1UL << KVM_APIC_INIT);
1124 kvm_make_request(KVM_REQ_EVENT, vcpu);
1125 kvm_vcpu_kick(vcpu);
1129 case APIC_DM_STARTUP:
1131 apic->sipi_vector = vector;
1132 /* make sure sipi_vector is visible for the receiver */
1134 set_bit(KVM_APIC_SIPI, &apic->pending_events);
1135 kvm_make_request(KVM_REQ_EVENT, vcpu);
1136 kvm_vcpu_kick(vcpu);
1139 case APIC_DM_EXTINT:
1141 * Should only be called by kvm_apic_local_deliver() with LVT0,
1142 * before NMI watchdog was enabled. Already handled by
1143 * kvm_apic_accept_pic_intr().
1148 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1156 * This routine identifies the destination vcpus mask meant to receive the
1157 * IOAPIC interrupts. It either uses kvm_apic_map_get_dest_lapic() to find
1158 * out the destination vcpus array and set the bitmap or it traverses to
1159 * each available vcpu to identify the same.
1161 void kvm_bitmap_or_dest_vcpus(struct kvm *kvm, struct kvm_lapic_irq *irq,
1162 unsigned long *vcpu_bitmap)
1164 struct kvm_lapic **dest_vcpu = NULL;
1165 struct kvm_lapic *src = NULL;
1166 struct kvm_apic_map *map;
1167 struct kvm_vcpu *vcpu;
1168 unsigned long bitmap;
1173 map = rcu_dereference(kvm->arch.apic_map);
1175 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dest_vcpu,
1178 for_each_set_bit(i, &bitmap, 16) {
1181 vcpu_idx = dest_vcpu[i]->vcpu->vcpu_idx;
1182 __set_bit(vcpu_idx, vcpu_bitmap);
1185 kvm_for_each_vcpu(i, vcpu, kvm) {
1186 if (!kvm_apic_present(vcpu))
1188 if (!kvm_apic_match_dest(vcpu, NULL,
1193 __set_bit(i, vcpu_bitmap);
1199 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1201 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1204 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1206 return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1209 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1213 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1214 if (!kvm_ioapic_handles_vector(apic, vector))
1217 /* Request a KVM exit to inform the userspace IOAPIC. */
1218 if (irqchip_split(apic->vcpu->kvm)) {
1219 apic->vcpu->arch.pending_ioapic_eoi = vector;
1220 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1224 if (apic_test_vector(vector, apic->regs + APIC_TMR))
1225 trigger_mode = IOAPIC_LEVEL_TRIG;
1227 trigger_mode = IOAPIC_EDGE_TRIG;
1229 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1232 static int apic_set_eoi(struct kvm_lapic *apic)
1234 int vector = apic_find_highest_isr(apic);
1236 trace_kvm_eoi(apic, vector);
1239 * Not every write EOI will has corresponding ISR,
1240 * one example is when Kernel check timer on setup_IO_APIC
1245 apic_clear_isr(vector, apic);
1246 apic_update_ppr(apic);
1248 if (to_hv_vcpu(apic->vcpu) &&
1249 test_bit(vector, to_hv_synic(apic->vcpu)->vec_bitmap))
1250 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1252 kvm_ioapic_send_eoi(apic, vector);
1253 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1258 * this interface assumes a trap-like exit, which has already finished
1259 * desired side effect including vISR and vPPR update.
1261 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1263 struct kvm_lapic *apic = vcpu->arch.apic;
1265 trace_kvm_eoi(apic, vector);
1267 kvm_ioapic_send_eoi(apic, vector);
1268 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1270 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1272 void kvm_apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
1274 struct kvm_lapic_irq irq;
1276 irq.vector = icr_low & APIC_VECTOR_MASK;
1277 irq.delivery_mode = icr_low & APIC_MODE_MASK;
1278 irq.dest_mode = icr_low & APIC_DEST_MASK;
1279 irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1280 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1281 irq.shorthand = icr_low & APIC_SHORT_MASK;
1282 irq.msi_redir_hint = false;
1283 if (apic_x2apic_mode(apic))
1284 irq.dest_id = icr_high;
1286 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1288 trace_kvm_apic_ipi(icr_low, irq.dest_id);
1290 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1293 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1295 ktime_t remaining, now;
1299 ASSERT(apic != NULL);
1301 /* if initial count is 0, current count should also be 0 */
1302 if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1303 apic->lapic_timer.period == 0)
1307 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1308 if (ktime_to_ns(remaining) < 0)
1311 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1312 tmcct = div64_u64(ns,
1313 (APIC_BUS_CYCLE_NS * apic->divide_count));
1318 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1320 struct kvm_vcpu *vcpu = apic->vcpu;
1321 struct kvm_run *run = vcpu->run;
1323 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1324 run->tpr_access.rip = kvm_rip_read(vcpu);
1325 run->tpr_access.is_write = write;
1328 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1330 if (apic->vcpu->arch.tpr_access_reporting)
1331 __report_tpr_access(apic, write);
1334 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1338 if (offset >= LAPIC_MMIO_LENGTH)
1345 case APIC_TMCCT: /* Timer CCR */
1346 if (apic_lvtt_tscdeadline(apic))
1349 val = apic_get_tmcct(apic);
1352 apic_update_ppr(apic);
1353 val = kvm_lapic_get_reg(apic, offset);
1356 report_tpr_access(apic, false);
1359 val = kvm_lapic_get_reg(apic, offset);
1366 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1368 return container_of(dev, struct kvm_lapic, dev);
1371 #define APIC_REG_MASK(reg) (1ull << ((reg) >> 4))
1372 #define APIC_REGS_MASK(first, count) \
1373 (APIC_REG_MASK(first) * ((1ull << (count)) - 1))
1375 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1378 unsigned char alignment = offset & 0xf;
1380 /* this bitmask has a bit cleared for each reserved register */
1381 u64 valid_reg_mask =
1382 APIC_REG_MASK(APIC_ID) |
1383 APIC_REG_MASK(APIC_LVR) |
1384 APIC_REG_MASK(APIC_TASKPRI) |
1385 APIC_REG_MASK(APIC_PROCPRI) |
1386 APIC_REG_MASK(APIC_LDR) |
1387 APIC_REG_MASK(APIC_DFR) |
1388 APIC_REG_MASK(APIC_SPIV) |
1389 APIC_REGS_MASK(APIC_ISR, APIC_ISR_NR) |
1390 APIC_REGS_MASK(APIC_TMR, APIC_ISR_NR) |
1391 APIC_REGS_MASK(APIC_IRR, APIC_ISR_NR) |
1392 APIC_REG_MASK(APIC_ESR) |
1393 APIC_REG_MASK(APIC_ICR) |
1394 APIC_REG_MASK(APIC_ICR2) |
1395 APIC_REG_MASK(APIC_LVTT) |
1396 APIC_REG_MASK(APIC_LVTTHMR) |
1397 APIC_REG_MASK(APIC_LVTPC) |
1398 APIC_REG_MASK(APIC_LVT0) |
1399 APIC_REG_MASK(APIC_LVT1) |
1400 APIC_REG_MASK(APIC_LVTERR) |
1401 APIC_REG_MASK(APIC_TMICT) |
1402 APIC_REG_MASK(APIC_TMCCT) |
1403 APIC_REG_MASK(APIC_TDCR);
1405 /* ARBPRI is not valid on x2APIC */
1406 if (!apic_x2apic_mode(apic))
1407 valid_reg_mask |= APIC_REG_MASK(APIC_ARBPRI);
1409 if (offset > 0x3f0 || !(valid_reg_mask & APIC_REG_MASK(offset)))
1412 result = __apic_read(apic, offset & ~0xf);
1414 trace_kvm_apic_read(offset, result);
1420 memcpy(data, (char *)&result + alignment, len);
1423 printk(KERN_ERR "Local APIC read with len = %x, "
1424 "should be 1,2, or 4 instead\n", len);
1429 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1431 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1433 return addr >= apic->base_address &&
1434 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1437 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1438 gpa_t address, int len, void *data)
1440 struct kvm_lapic *apic = to_lapic(this);
1441 u32 offset = address - apic->base_address;
1443 if (!apic_mmio_in_range(apic, address))
1446 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1447 if (!kvm_check_has_quirk(vcpu->kvm,
1448 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1451 memset(data, 0xff, len);
1455 kvm_lapic_reg_read(apic, offset, len, data);
1460 static void update_divide_count(struct kvm_lapic *apic)
1462 u32 tmp1, tmp2, tdcr;
1464 tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1466 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1467 apic->divide_count = 0x1 << (tmp2 & 0x7);
1470 static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1473 * Do not allow the guest to program periodic timers with small
1474 * interval, since the hrtimers are not throttled by the host
1477 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1478 s64 min_period = min_timer_period_us * 1000LL;
1480 if (apic->lapic_timer.period < min_period) {
1481 pr_info_ratelimited(
1482 "kvm: vcpu %i: requested %lld ns "
1483 "lapic timer period limited to %lld ns\n",
1484 apic->vcpu->vcpu_id,
1485 apic->lapic_timer.period, min_period);
1486 apic->lapic_timer.period = min_period;
1491 static void cancel_hv_timer(struct kvm_lapic *apic);
1493 static void apic_update_lvtt(struct kvm_lapic *apic)
1495 u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1496 apic->lapic_timer.timer_mode_mask;
1498 if (apic->lapic_timer.timer_mode != timer_mode) {
1499 if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1500 APIC_LVT_TIMER_TSCDEADLINE)) {
1501 hrtimer_cancel(&apic->lapic_timer.timer);
1503 if (apic->lapic_timer.hv_timer_in_use)
1504 cancel_hv_timer(apic);
1506 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1507 apic->lapic_timer.period = 0;
1508 apic->lapic_timer.tscdeadline = 0;
1510 apic->lapic_timer.timer_mode = timer_mode;
1511 limit_periodic_timer_frequency(apic);
1516 * On APICv, this test will cause a busy wait
1517 * during a higher-priority task.
1520 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1522 struct kvm_lapic *apic = vcpu->arch.apic;
1523 u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1525 if (kvm_apic_hw_enabled(apic)) {
1526 int vec = reg & APIC_VECTOR_MASK;
1527 void *bitmap = apic->regs + APIC_ISR;
1529 if (vcpu->arch.apicv_active)
1530 bitmap = apic->regs + APIC_IRR;
1532 if (apic_test_vector(vec, bitmap))
1538 static inline void __wait_lapic_expire(struct kvm_vcpu *vcpu, u64 guest_cycles)
1540 u64 timer_advance_ns = vcpu->arch.apic->lapic_timer.timer_advance_ns;
1543 * If the guest TSC is running at a different ratio than the host, then
1544 * convert the delay to nanoseconds to achieve an accurate delay. Note
1545 * that __delay() uses delay_tsc whenever the hardware has TSC, thus
1546 * always for VMX enabled hardware.
1548 if (vcpu->arch.tsc_scaling_ratio == kvm_default_tsc_scaling_ratio) {
1549 __delay(min(guest_cycles,
1550 nsec_to_cycles(vcpu, timer_advance_ns)));
1552 u64 delay_ns = guest_cycles * 1000000ULL;
1553 do_div(delay_ns, vcpu->arch.virtual_tsc_khz);
1554 ndelay(min_t(u32, delay_ns, timer_advance_ns));
1558 static inline void adjust_lapic_timer_advance(struct kvm_vcpu *vcpu,
1559 s64 advance_expire_delta)
1561 struct kvm_lapic *apic = vcpu->arch.apic;
1562 u32 timer_advance_ns = apic->lapic_timer.timer_advance_ns;
1565 /* Do not adjust for tiny fluctuations or large random spikes. */
1566 if (abs(advance_expire_delta) > LAPIC_TIMER_ADVANCE_ADJUST_MAX ||
1567 abs(advance_expire_delta) < LAPIC_TIMER_ADVANCE_ADJUST_MIN)
1571 if (advance_expire_delta < 0) {
1572 ns = -advance_expire_delta * 1000000ULL;
1573 do_div(ns, vcpu->arch.virtual_tsc_khz);
1574 timer_advance_ns -= ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1577 ns = advance_expire_delta * 1000000ULL;
1578 do_div(ns, vcpu->arch.virtual_tsc_khz);
1579 timer_advance_ns += ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1582 if (unlikely(timer_advance_ns > LAPIC_TIMER_ADVANCE_NS_MAX))
1583 timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
1584 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
1587 static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1589 struct kvm_lapic *apic = vcpu->arch.apic;
1590 u64 guest_tsc, tsc_deadline;
1592 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1593 apic->lapic_timer.expired_tscdeadline = 0;
1594 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1595 apic->lapic_timer.advance_expire_delta = guest_tsc - tsc_deadline;
1597 if (guest_tsc < tsc_deadline)
1598 __wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
1600 if (lapic_timer_advance_dynamic)
1601 adjust_lapic_timer_advance(vcpu, apic->lapic_timer.advance_expire_delta);
1604 void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1606 if (lapic_in_kernel(vcpu) &&
1607 vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1608 vcpu->arch.apic->lapic_timer.timer_advance_ns &&
1609 lapic_timer_int_injected(vcpu))
1610 __kvm_wait_lapic_expire(vcpu);
1612 EXPORT_SYMBOL_GPL(kvm_wait_lapic_expire);
1614 static void kvm_apic_inject_pending_timer_irqs(struct kvm_lapic *apic)
1616 struct kvm_timer *ktimer = &apic->lapic_timer;
1618 kvm_apic_local_deliver(apic, APIC_LVTT);
1619 if (apic_lvtt_tscdeadline(apic)) {
1620 ktimer->tscdeadline = 0;
1621 } else if (apic_lvtt_oneshot(apic)) {
1622 ktimer->tscdeadline = 0;
1623 ktimer->target_expiration = 0;
1627 static void apic_timer_expired(struct kvm_lapic *apic, bool from_timer_fn)
1629 struct kvm_vcpu *vcpu = apic->vcpu;
1630 struct kvm_timer *ktimer = &apic->lapic_timer;
1632 if (atomic_read(&apic->lapic_timer.pending))
1635 if (apic_lvtt_tscdeadline(apic) || ktimer->hv_timer_in_use)
1636 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1638 if (!from_timer_fn && vcpu->arch.apicv_active) {
1639 WARN_ON(kvm_get_running_vcpu() != vcpu);
1640 kvm_apic_inject_pending_timer_irqs(apic);
1644 if (kvm_use_posted_timer_interrupt(apic->vcpu)) {
1646 * Ensure the guest's timer has truly expired before posting an
1647 * interrupt. Open code the relevant checks to avoid querying
1648 * lapic_timer_int_injected(), which will be false since the
1649 * interrupt isn't yet injected. Waiting until after injecting
1650 * is not an option since that won't help a posted interrupt.
1652 if (vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1653 vcpu->arch.apic->lapic_timer.timer_advance_ns)
1654 __kvm_wait_lapic_expire(vcpu);
1655 kvm_apic_inject_pending_timer_irqs(apic);
1659 atomic_inc(&apic->lapic_timer.pending);
1660 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1662 kvm_vcpu_kick(vcpu);
1665 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1667 struct kvm_timer *ktimer = &apic->lapic_timer;
1668 u64 guest_tsc, tscdeadline = ktimer->tscdeadline;
1671 struct kvm_vcpu *vcpu = apic->vcpu;
1672 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1673 unsigned long flags;
1676 if (unlikely(!tscdeadline || !this_tsc_khz))
1679 local_irq_save(flags);
1682 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1684 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1685 do_div(ns, this_tsc_khz);
1687 if (likely(tscdeadline > guest_tsc) &&
1688 likely(ns > apic->lapic_timer.timer_advance_ns)) {
1689 expire = ktime_add_ns(now, ns);
1690 expire = ktime_sub_ns(expire, ktimer->timer_advance_ns);
1691 hrtimer_start(&ktimer->timer, expire, HRTIMER_MODE_ABS_HARD);
1693 apic_timer_expired(apic, false);
1695 local_irq_restore(flags);
1698 static inline u64 tmict_to_ns(struct kvm_lapic *apic, u32 tmict)
1700 return (u64)tmict * APIC_BUS_CYCLE_NS * (u64)apic->divide_count;
1703 static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1705 ktime_t now, remaining;
1706 u64 ns_remaining_old, ns_remaining_new;
1708 apic->lapic_timer.period =
1709 tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1710 limit_periodic_timer_frequency(apic);
1713 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1714 if (ktime_to_ns(remaining) < 0)
1717 ns_remaining_old = ktime_to_ns(remaining);
1718 ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1719 apic->divide_count, old_divisor);
1721 apic->lapic_timer.tscdeadline +=
1722 nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1723 nsec_to_cycles(apic->vcpu, ns_remaining_old);
1724 apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1727 static bool set_target_expiration(struct kvm_lapic *apic, u32 count_reg)
1734 apic->lapic_timer.period =
1735 tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1737 if (!apic->lapic_timer.period) {
1738 apic->lapic_timer.tscdeadline = 0;
1742 limit_periodic_timer_frequency(apic);
1743 deadline = apic->lapic_timer.period;
1745 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1746 if (unlikely(count_reg != APIC_TMICT)) {
1747 deadline = tmict_to_ns(apic,
1748 kvm_lapic_get_reg(apic, count_reg));
1749 if (unlikely(deadline <= 0))
1750 deadline = apic->lapic_timer.period;
1751 else if (unlikely(deadline > apic->lapic_timer.period)) {
1752 pr_info_ratelimited(
1753 "kvm: vcpu %i: requested lapic timer restore with "
1754 "starting count register %#x=%u (%lld ns) > initial count (%lld ns). "
1755 "Using initial count to start timer.\n",
1756 apic->vcpu->vcpu_id,
1758 kvm_lapic_get_reg(apic, count_reg),
1759 deadline, apic->lapic_timer.period);
1760 kvm_lapic_set_reg(apic, count_reg, 0);
1761 deadline = apic->lapic_timer.period;
1766 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1767 nsec_to_cycles(apic->vcpu, deadline);
1768 apic->lapic_timer.target_expiration = ktime_add_ns(now, deadline);
1773 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1775 ktime_t now = ktime_get();
1780 * Synchronize both deadlines to the same time source or
1781 * differences in the periods (caused by differences in the
1782 * underlying clocks or numerical approximation errors) will
1783 * cause the two to drift apart over time as the errors
1786 apic->lapic_timer.target_expiration =
1787 ktime_add_ns(apic->lapic_timer.target_expiration,
1788 apic->lapic_timer.period);
1789 delta = ktime_sub(apic->lapic_timer.target_expiration, now);
1790 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1791 nsec_to_cycles(apic->vcpu, delta);
1794 static void start_sw_period(struct kvm_lapic *apic)
1796 if (!apic->lapic_timer.period)
1799 if (ktime_after(ktime_get(),
1800 apic->lapic_timer.target_expiration)) {
1801 apic_timer_expired(apic, false);
1803 if (apic_lvtt_oneshot(apic))
1806 advance_periodic_target_expiration(apic);
1809 hrtimer_start(&apic->lapic_timer.timer,
1810 apic->lapic_timer.target_expiration,
1811 HRTIMER_MODE_ABS_HARD);
1814 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1816 if (!lapic_in_kernel(vcpu))
1819 return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1821 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1823 static void cancel_hv_timer(struct kvm_lapic *apic)
1825 WARN_ON(preemptible());
1826 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1827 static_call(kvm_x86_cancel_hv_timer)(apic->vcpu);
1828 apic->lapic_timer.hv_timer_in_use = false;
1831 static bool start_hv_timer(struct kvm_lapic *apic)
1833 struct kvm_timer *ktimer = &apic->lapic_timer;
1834 struct kvm_vcpu *vcpu = apic->vcpu;
1837 WARN_ON(preemptible());
1838 if (!kvm_can_use_hv_timer(vcpu))
1841 if (!ktimer->tscdeadline)
1844 if (static_call(kvm_x86_set_hv_timer)(vcpu, ktimer->tscdeadline, &expired))
1847 ktimer->hv_timer_in_use = true;
1848 hrtimer_cancel(&ktimer->timer);
1851 * To simplify handling the periodic timer, leave the hv timer running
1852 * even if the deadline timer has expired, i.e. rely on the resulting
1853 * VM-Exit to recompute the periodic timer's target expiration.
1855 if (!apic_lvtt_period(apic)) {
1857 * Cancel the hv timer if the sw timer fired while the hv timer
1858 * was being programmed, or if the hv timer itself expired.
1860 if (atomic_read(&ktimer->pending)) {
1861 cancel_hv_timer(apic);
1862 } else if (expired) {
1863 apic_timer_expired(apic, false);
1864 cancel_hv_timer(apic);
1868 trace_kvm_hv_timer_state(vcpu->vcpu_id, ktimer->hv_timer_in_use);
1873 static void start_sw_timer(struct kvm_lapic *apic)
1875 struct kvm_timer *ktimer = &apic->lapic_timer;
1877 WARN_ON(preemptible());
1878 if (apic->lapic_timer.hv_timer_in_use)
1879 cancel_hv_timer(apic);
1880 if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1883 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1884 start_sw_period(apic);
1885 else if (apic_lvtt_tscdeadline(apic))
1886 start_sw_tscdeadline(apic);
1887 trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1890 static void restart_apic_timer(struct kvm_lapic *apic)
1894 if (!apic_lvtt_period(apic) && atomic_read(&apic->lapic_timer.pending))
1897 if (!start_hv_timer(apic))
1898 start_sw_timer(apic);
1903 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1905 struct kvm_lapic *apic = vcpu->arch.apic;
1908 /* If the preempt notifier has already run, it also called apic_timer_expired */
1909 if (!apic->lapic_timer.hv_timer_in_use)
1911 WARN_ON(rcuwait_active(&vcpu->wait));
1912 cancel_hv_timer(apic);
1913 apic_timer_expired(apic, false);
1915 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1916 advance_periodic_target_expiration(apic);
1917 restart_apic_timer(apic);
1922 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1924 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1926 restart_apic_timer(vcpu->arch.apic);
1928 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1930 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1932 struct kvm_lapic *apic = vcpu->arch.apic;
1935 /* Possibly the TSC deadline timer is not enabled yet */
1936 if (apic->lapic_timer.hv_timer_in_use)
1937 start_sw_timer(apic);
1940 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1942 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
1944 struct kvm_lapic *apic = vcpu->arch.apic;
1946 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1947 restart_apic_timer(apic);
1950 static void __start_apic_timer(struct kvm_lapic *apic, u32 count_reg)
1952 atomic_set(&apic->lapic_timer.pending, 0);
1954 if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1955 && !set_target_expiration(apic, count_reg))
1958 restart_apic_timer(apic);
1961 static void start_apic_timer(struct kvm_lapic *apic)
1963 __start_apic_timer(apic, APIC_TMICT);
1966 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1968 bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1970 if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1971 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1972 if (lvt0_in_nmi_mode) {
1973 atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1975 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1979 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1983 trace_kvm_apic_write(reg, val);
1986 case APIC_ID: /* Local APIC ID */
1987 if (!apic_x2apic_mode(apic))
1988 kvm_apic_set_xapic_id(apic, val >> 24);
1994 report_tpr_access(apic, true);
1995 apic_set_tpr(apic, val & 0xff);
2003 if (!apic_x2apic_mode(apic))
2004 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
2010 if (!apic_x2apic_mode(apic))
2011 kvm_apic_set_dfr(apic, val | 0x0FFFFFFF);
2018 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
2019 mask |= APIC_SPIV_DIRECTED_EOI;
2020 apic_set_spiv(apic, val & mask);
2021 if (!(val & APIC_SPIV_APIC_ENABLED)) {
2025 for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
2026 lvt_val = kvm_lapic_get_reg(apic,
2027 APIC_LVTT + 0x10 * i);
2028 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
2029 lvt_val | APIC_LVT_MASKED);
2031 apic_update_lvtt(apic);
2032 atomic_set(&apic->lapic_timer.pending, 0);
2038 /* No delay here, so we always clear the pending bit */
2040 kvm_apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
2041 kvm_lapic_set_reg(apic, APIC_ICR, val);
2045 if (!apic_x2apic_mode(apic))
2047 kvm_lapic_set_reg(apic, APIC_ICR2, val);
2051 apic_manage_nmi_watchdog(apic, val);
2057 /* TODO: Check vector */
2061 if (!kvm_apic_sw_enabled(apic))
2062 val |= APIC_LVT_MASKED;
2063 size = ARRAY_SIZE(apic_lvt_mask);
2064 index = array_index_nospec(
2065 (reg - APIC_LVTT) >> 4, size);
2066 val &= apic_lvt_mask[index];
2067 kvm_lapic_set_reg(apic, reg, val);
2072 if (!kvm_apic_sw_enabled(apic))
2073 val |= APIC_LVT_MASKED;
2074 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
2075 kvm_lapic_set_reg(apic, APIC_LVTT, val);
2076 apic_update_lvtt(apic);
2080 if (apic_lvtt_tscdeadline(apic))
2083 hrtimer_cancel(&apic->lapic_timer.timer);
2084 kvm_lapic_set_reg(apic, APIC_TMICT, val);
2085 start_apic_timer(apic);
2089 uint32_t old_divisor = apic->divide_count;
2091 kvm_lapic_set_reg(apic, APIC_TDCR, val & 0xb);
2092 update_divide_count(apic);
2093 if (apic->divide_count != old_divisor &&
2094 apic->lapic_timer.period) {
2095 hrtimer_cancel(&apic->lapic_timer.timer);
2096 update_target_expiration(apic, old_divisor);
2097 restart_apic_timer(apic);
2102 if (apic_x2apic_mode(apic) && val != 0)
2107 if (apic_x2apic_mode(apic)) {
2108 kvm_lapic_reg_write(apic, APIC_ICR,
2109 APIC_DEST_SELF | (val & APIC_VECTOR_MASK));
2118 kvm_recalculate_apic_map(apic->vcpu->kvm);
2122 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
2124 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
2125 gpa_t address, int len, const void *data)
2127 struct kvm_lapic *apic = to_lapic(this);
2128 unsigned int offset = address - apic->base_address;
2131 if (!apic_mmio_in_range(apic, address))
2134 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
2135 if (!kvm_check_has_quirk(vcpu->kvm,
2136 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
2143 * APIC register must be aligned on 128-bits boundary.
2144 * 32/64/128 bits registers must be accessed thru 32 bits.
2147 if (len != 4 || (offset & 0xf))
2152 kvm_lapic_reg_write(apic, offset & 0xff0, val);
2157 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
2159 kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
2161 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
2163 /* emulate APIC access in a trap manner */
2164 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
2168 /* hw has done the conditional check and inst decode */
2171 kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
2173 /* TODO: optimize to just emulate side effect w/o one more write */
2174 kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
2176 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
2178 void kvm_free_lapic(struct kvm_vcpu *vcpu)
2180 struct kvm_lapic *apic = vcpu->arch.apic;
2182 if (!vcpu->arch.apic)
2185 hrtimer_cancel(&apic->lapic_timer.timer);
2187 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
2188 static_branch_slow_dec_deferred(&apic_hw_disabled);
2190 if (!apic->sw_enabled)
2191 static_branch_slow_dec_deferred(&apic_sw_disabled);
2194 free_page((unsigned long)apic->regs);
2200 *----------------------------------------------------------------------
2202 *----------------------------------------------------------------------
2204 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
2206 struct kvm_lapic *apic = vcpu->arch.apic;
2208 if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2211 return apic->lapic_timer.tscdeadline;
2214 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
2216 struct kvm_lapic *apic = vcpu->arch.apic;
2218 if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2221 hrtimer_cancel(&apic->lapic_timer.timer);
2222 apic->lapic_timer.tscdeadline = data;
2223 start_apic_timer(apic);
2226 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
2228 struct kvm_lapic *apic = vcpu->arch.apic;
2230 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
2231 | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
2234 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
2238 tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
2240 return (tpr & 0xf0) >> 4;
2243 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
2245 u64 old_value = vcpu->arch.apic_base;
2246 struct kvm_lapic *apic = vcpu->arch.apic;
2249 value |= MSR_IA32_APICBASE_BSP;
2251 vcpu->arch.apic_base = value;
2253 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
2254 kvm_update_cpuid_runtime(vcpu);
2259 /* update jump label if enable bit changes */
2260 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
2261 if (value & MSR_IA32_APICBASE_ENABLE) {
2262 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2263 static_branch_slow_dec_deferred(&apic_hw_disabled);
2265 static_branch_inc(&apic_hw_disabled.key);
2266 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2270 if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
2271 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
2273 if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE))
2274 static_call(kvm_x86_set_virtual_apic_mode)(vcpu);
2276 apic->base_address = apic->vcpu->arch.apic_base &
2277 MSR_IA32_APICBASE_BASE;
2279 if ((value & MSR_IA32_APICBASE_ENABLE) &&
2280 apic->base_address != APIC_DEFAULT_PHYS_BASE)
2281 pr_warn_once("APIC base relocation is unsupported by KVM");
2284 void kvm_apic_update_apicv(struct kvm_vcpu *vcpu)
2286 struct kvm_lapic *apic = vcpu->arch.apic;
2288 if (vcpu->arch.apicv_active) {
2289 /* irr_pending is always true when apicv is activated. */
2290 apic->irr_pending = true;
2291 apic->isr_count = 1;
2293 apic->irr_pending = (apic_search_irr(apic) != -1);
2294 apic->isr_count = count_vectors(apic->regs + APIC_ISR);
2297 EXPORT_SYMBOL_GPL(kvm_apic_update_apicv);
2299 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2301 struct kvm_lapic *apic = vcpu->arch.apic;
2307 /* Stop the timer in case it's a reset to an active apic */
2308 hrtimer_cancel(&apic->lapic_timer.timer);
2311 kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
2312 MSR_IA32_APICBASE_ENABLE);
2313 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2315 kvm_apic_set_version(apic->vcpu);
2317 for (i = 0; i < KVM_APIC_LVT_NUM; i++)
2318 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
2319 apic_update_lvtt(apic);
2320 if (kvm_vcpu_is_reset_bsp(vcpu) &&
2321 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2322 kvm_lapic_set_reg(apic, APIC_LVT0,
2323 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2324 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2326 kvm_apic_set_dfr(apic, 0xffffffffU);
2327 apic_set_spiv(apic, 0xff);
2328 kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2329 if (!apic_x2apic_mode(apic))
2330 kvm_apic_set_ldr(apic, 0);
2331 kvm_lapic_set_reg(apic, APIC_ESR, 0);
2332 kvm_lapic_set_reg(apic, APIC_ICR, 0);
2333 kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2334 kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2335 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2336 for (i = 0; i < 8; i++) {
2337 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2338 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2339 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2341 kvm_apic_update_apicv(vcpu);
2342 apic->highest_isr_cache = -1;
2343 update_divide_count(apic);
2344 atomic_set(&apic->lapic_timer.pending, 0);
2345 if (kvm_vcpu_is_bsp(vcpu))
2346 kvm_lapic_set_base(vcpu,
2347 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
2348 vcpu->arch.pv_eoi.msr_val = 0;
2349 apic_update_ppr(apic);
2350 if (vcpu->arch.apicv_active) {
2351 static_call(kvm_x86_apicv_post_state_restore)(vcpu);
2352 static_call(kvm_x86_hwapic_irr_update)(vcpu, -1);
2353 static_call(kvm_x86_hwapic_isr_update)(vcpu, -1);
2356 vcpu->arch.apic_arb_prio = 0;
2357 vcpu->arch.apic_attention = 0;
2359 kvm_recalculate_apic_map(vcpu->kvm);
2363 *----------------------------------------------------------------------
2365 *----------------------------------------------------------------------
2368 static bool lapic_is_periodic(struct kvm_lapic *apic)
2370 return apic_lvtt_period(apic);
2373 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2375 struct kvm_lapic *apic = vcpu->arch.apic;
2377 if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2378 return atomic_read(&apic->lapic_timer.pending);
2383 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2385 u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2386 int vector, mode, trig_mode;
2388 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2389 vector = reg & APIC_VECTOR_MASK;
2390 mode = reg & APIC_MODE_MASK;
2391 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2392 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2398 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2400 struct kvm_lapic *apic = vcpu->arch.apic;
2403 kvm_apic_local_deliver(apic, APIC_LVT0);
2406 static const struct kvm_io_device_ops apic_mmio_ops = {
2407 .read = apic_mmio_read,
2408 .write = apic_mmio_write,
2411 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2413 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2414 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2416 apic_timer_expired(apic, true);
2418 if (lapic_is_periodic(apic)) {
2419 advance_periodic_target_expiration(apic);
2420 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2421 return HRTIMER_RESTART;
2423 return HRTIMER_NORESTART;
2426 int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
2428 struct kvm_lapic *apic;
2430 ASSERT(vcpu != NULL);
2432 apic = kzalloc(sizeof(*apic), GFP_KERNEL_ACCOUNT);
2436 vcpu->arch.apic = apic;
2438 apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
2440 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2442 goto nomem_free_apic;
2446 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2447 HRTIMER_MODE_ABS_HARD);
2448 apic->lapic_timer.timer.function = apic_timer_fn;
2449 if (timer_advance_ns == -1) {
2450 apic->lapic_timer.timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
2451 lapic_timer_advance_dynamic = true;
2453 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
2454 lapic_timer_advance_dynamic = false;
2458 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2459 * thinking that APIC state has changed.
2461 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2462 static_branch_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2463 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2468 vcpu->arch.apic = NULL;
2473 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2475 struct kvm_lapic *apic = vcpu->arch.apic;
2478 if (!kvm_apic_present(vcpu))
2481 __apic_update_ppr(apic, &ppr);
2482 return apic_has_interrupt_for_ppr(apic, ppr);
2484 EXPORT_SYMBOL_GPL(kvm_apic_has_interrupt);
2486 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2488 u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2490 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2492 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2493 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2498 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2500 struct kvm_lapic *apic = vcpu->arch.apic;
2502 if (atomic_read(&apic->lapic_timer.pending) > 0) {
2503 kvm_apic_inject_pending_timer_irqs(apic);
2504 atomic_set(&apic->lapic_timer.pending, 0);
2508 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2510 int vector = kvm_apic_has_interrupt(vcpu);
2511 struct kvm_lapic *apic = vcpu->arch.apic;
2518 * We get here even with APIC virtualization enabled, if doing
2519 * nested virtualization and L1 runs with the "acknowledge interrupt
2520 * on exit" mode. Then we cannot inject the interrupt via RVI,
2521 * because the process would deliver it through the IDT.
2524 apic_clear_irr(vector, apic);
2525 if (to_hv_vcpu(vcpu) && test_bit(vector, to_hv_synic(vcpu)->auto_eoi_bitmap)) {
2527 * For auto-EOI interrupts, there might be another pending
2528 * interrupt above PPR, so check whether to raise another
2531 apic_update_ppr(apic);
2534 * For normal interrupts, PPR has been raised and there cannot
2535 * be a higher-priority pending interrupt---except if there was
2536 * a concurrent interrupt injection, but that would have
2537 * triggered KVM_REQ_EVENT already.
2539 apic_set_isr(vector, apic);
2540 __apic_update_ppr(apic, &ppr);
2546 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2547 struct kvm_lapic_state *s, bool set)
2549 if (apic_x2apic_mode(vcpu->arch.apic)) {
2550 u32 *id = (u32 *)(s->regs + APIC_ID);
2551 u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2553 if (vcpu->kvm->arch.x2apic_format) {
2554 if (*id != vcpu->vcpu_id)
2563 /* In x2APIC mode, the LDR is fixed and based on the id */
2565 *ldr = kvm_apic_calc_x2apic_ldr(*id);
2571 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2573 memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2576 * Get calculated timer current count for remaining timer period (if
2577 * any) and store it in the returned register set.
2579 __kvm_lapic_set_reg(s->regs, APIC_TMCCT,
2580 __apic_read(vcpu->arch.apic, APIC_TMCCT));
2582 return kvm_apic_state_fixup(vcpu, s, false);
2585 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2587 struct kvm_lapic *apic = vcpu->arch.apic;
2590 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2591 /* set SPIV separately to get count of SW disabled APICs right */
2592 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2594 r = kvm_apic_state_fixup(vcpu, s, true);
2596 kvm_recalculate_apic_map(vcpu->kvm);
2599 memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
2601 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2602 kvm_recalculate_apic_map(vcpu->kvm);
2603 kvm_apic_set_version(vcpu);
2605 apic_update_ppr(apic);
2606 hrtimer_cancel(&apic->lapic_timer.timer);
2607 apic->lapic_timer.expired_tscdeadline = 0;
2608 apic_update_lvtt(apic);
2609 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2610 update_divide_count(apic);
2611 __start_apic_timer(apic, APIC_TMCCT);
2612 kvm_apic_update_apicv(vcpu);
2613 apic->highest_isr_cache = -1;
2614 if (vcpu->arch.apicv_active) {
2615 static_call(kvm_x86_apicv_post_state_restore)(vcpu);
2616 static_call(kvm_x86_hwapic_irr_update)(vcpu,
2617 apic_find_highest_irr(apic));
2618 static_call(kvm_x86_hwapic_isr_update)(vcpu,
2619 apic_find_highest_isr(apic));
2621 kvm_make_request(KVM_REQ_EVENT, vcpu);
2622 if (ioapic_in_kernel(vcpu->kvm))
2623 kvm_rtc_eoi_tracking_restore_one(vcpu);
2625 vcpu->arch.apic_arb_prio = 0;
2630 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2632 struct hrtimer *timer;
2634 if (!lapic_in_kernel(vcpu) ||
2635 kvm_can_post_timer_interrupt(vcpu))
2638 timer = &vcpu->arch.apic->lapic_timer.timer;
2639 if (hrtimer_cancel(timer))
2640 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_HARD);
2644 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2646 * Detect whether guest triggered PV EOI since the
2647 * last entry. If yes, set EOI on guests's behalf.
2648 * Clear PV EOI in guest memory in any case.
2650 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2651 struct kvm_lapic *apic)
2656 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2657 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2659 * KVM_APIC_PV_EOI_PENDING is unset:
2660 * -> host disabled PV EOI.
2661 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2662 * -> host enabled PV EOI, guest did not execute EOI yet.
2663 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2664 * -> host enabled PV EOI, guest executed EOI.
2666 BUG_ON(!pv_eoi_enabled(vcpu));
2667 pending = pv_eoi_get_pending(vcpu);
2669 * Clear pending bit in any case: it will be set again on vmentry.
2670 * While this might not be ideal from performance point of view,
2671 * this makes sure pv eoi is only enabled when we know it's safe.
2673 pv_eoi_clr_pending(vcpu);
2676 vector = apic_set_eoi(apic);
2677 trace_kvm_pv_eoi(apic, vector);
2680 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2684 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2685 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2687 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2690 if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2694 apic_set_tpr(vcpu->arch.apic, data & 0xff);
2698 * apic_sync_pv_eoi_to_guest - called before vmentry
2700 * Detect whether it's safe to enable PV EOI and
2703 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2704 struct kvm_lapic *apic)
2706 if (!pv_eoi_enabled(vcpu) ||
2707 /* IRR set or many bits in ISR: could be nested. */
2708 apic->irr_pending ||
2709 /* Cache not set: could be safe but we don't bother. */
2710 apic->highest_isr_cache == -1 ||
2711 /* Need EOI to update ioapic. */
2712 kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2714 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2715 * so we need not do anything here.
2720 pv_eoi_set_pending(apic->vcpu);
2723 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2726 int max_irr, max_isr;
2727 struct kvm_lapic *apic = vcpu->arch.apic;
2729 apic_sync_pv_eoi_to_guest(vcpu, apic);
2731 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2734 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2735 max_irr = apic_find_highest_irr(apic);
2738 max_isr = apic_find_highest_isr(apic);
2741 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2743 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2747 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2750 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2751 &vcpu->arch.apic->vapic_cache,
2752 vapic_addr, sizeof(u32)))
2754 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2756 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2759 vcpu->arch.apic->vapic_addr = vapic_addr;
2763 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2765 struct kvm_lapic *apic = vcpu->arch.apic;
2766 u32 reg = (msr - APIC_BASE_MSR) << 4;
2768 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2771 if (reg == APIC_ICR2)
2774 /* if this is ICR write vector before command */
2775 if (reg == APIC_ICR)
2776 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2777 return kvm_lapic_reg_write(apic, reg, (u32)data);
2780 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2782 struct kvm_lapic *apic = vcpu->arch.apic;
2783 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2785 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2788 if (reg == APIC_DFR || reg == APIC_ICR2)
2791 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2793 if (reg == APIC_ICR)
2794 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2796 *data = (((u64)high) << 32) | low;
2801 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2803 struct kvm_lapic *apic = vcpu->arch.apic;
2805 if (!lapic_in_kernel(vcpu))
2808 /* if this is ICR write vector before command */
2809 if (reg == APIC_ICR)
2810 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2811 return kvm_lapic_reg_write(apic, reg, (u32)data);
2814 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2816 struct kvm_lapic *apic = vcpu->arch.apic;
2819 if (!lapic_in_kernel(vcpu))
2822 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2824 if (reg == APIC_ICR)
2825 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2827 *data = (((u64)high) << 32) | low;
2832 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data, unsigned long len)
2834 u64 addr = data & ~KVM_MSR_ENABLED;
2835 struct gfn_to_hva_cache *ghc = &vcpu->arch.pv_eoi.data;
2836 unsigned long new_len;
2838 if (!IS_ALIGNED(addr, 4))
2841 vcpu->arch.pv_eoi.msr_val = data;
2842 if (!pv_eoi_enabled(vcpu))
2845 if (addr == ghc->gpa && len <= ghc->len)
2850 return kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, addr, new_len);
2853 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2855 struct kvm_lapic *apic = vcpu->arch.apic;
2860 if (!lapic_in_kernel(vcpu))
2864 * Read pending events before calling the check_events
2867 pe = smp_load_acquire(&apic->pending_events);
2871 if (is_guest_mode(vcpu)) {
2872 r = kvm_x86_ops.nested_ops->check_events(vcpu);
2876 * If an event has happened and caused a vmexit,
2877 * we know INITs are latched and therefore
2878 * we will not incorrectly deliver an APIC
2879 * event instead of a vmexit.
2884 * INITs are latched while CPU is in specific states
2885 * (SMM, VMX root mode, SVM with GIF=0).
2886 * Because a CPU cannot be in these states immediately
2887 * after it has processed an INIT signal (and thus in
2888 * KVM_MP_STATE_INIT_RECEIVED state), just eat SIPIs
2889 * and leave the INIT pending.
2891 if (kvm_vcpu_latch_init(vcpu)) {
2892 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2893 if (test_bit(KVM_APIC_SIPI, &pe))
2894 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2898 if (test_bit(KVM_APIC_INIT, &pe)) {
2899 clear_bit(KVM_APIC_INIT, &apic->pending_events);
2900 kvm_vcpu_reset(vcpu, true);
2901 if (kvm_vcpu_is_bsp(apic->vcpu))
2902 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2904 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2906 if (test_bit(KVM_APIC_SIPI, &pe)) {
2907 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2908 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2909 /* evaluate pending_events before reading the vector */
2911 sipi_vector = apic->sipi_vector;
2912 kvm_x86_ops.vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2913 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2918 void kvm_lapic_exit(void)
2920 static_key_deferred_flush(&apic_hw_disabled);
2921 static_key_deferred_flush(&apic_sw_disabled);