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 struct static_key_deferred apic_hw_disabled __read_mostly;
95 struct static_key_deferred apic_sw_disabled __read_mostly;
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_key_slow_dec_deferred(&apic_sw_disabled);
295 static_key_slow_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 kvm_x86_ops.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 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
499 struct kvm_vcpu *vcpu;
501 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
507 * With APIC virtualization enabled, all caching is disabled
508 * because the processor can modify ISR under the hood. Instead
511 if (unlikely(vcpu->arch.apicv_active))
512 kvm_x86_ops.hwapic_isr_update(vcpu, vec);
515 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
517 * ISR (in service register) bit is set when injecting an interrupt.
518 * The highest vector is injected. Thus the latest bit set matches
519 * the highest bit in ISR.
521 apic->highest_isr_cache = vec;
525 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
530 * Note that isr_count is always 1, and highest_isr_cache
531 * is always -1, with APIC virtualization enabled.
533 if (!apic->isr_count)
535 if (likely(apic->highest_isr_cache != -1))
536 return apic->highest_isr_cache;
538 result = find_highest_vector(apic->regs + APIC_ISR);
539 ASSERT(result == -1 || result >= 16);
544 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
546 struct kvm_vcpu *vcpu;
547 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
553 * We do get here for APIC virtualization enabled if the guest
554 * uses the Hyper-V APIC enlightenment. In this case we may need
555 * to trigger a new interrupt delivery by writing the SVI field;
556 * on the other hand isr_count and highest_isr_cache are unused
557 * and must be left alone.
559 if (unlikely(vcpu->arch.apicv_active))
560 kvm_x86_ops.hwapic_isr_update(vcpu,
561 apic_find_highest_isr(apic));
564 BUG_ON(apic->isr_count < 0);
565 apic->highest_isr_cache = -1;
569 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
571 /* This may race with setting of irr in __apic_accept_irq() and
572 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
573 * will cause vmexit immediately and the value will be recalculated
574 * on the next vmentry.
576 return apic_find_highest_irr(vcpu->arch.apic);
578 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
580 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
581 int vector, int level, int trig_mode,
582 struct dest_map *dest_map);
584 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
585 struct dest_map *dest_map)
587 struct kvm_lapic *apic = vcpu->arch.apic;
589 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
590 irq->level, irq->trig_mode, dest_map);
593 static int __pv_send_ipi(unsigned long *ipi_bitmap, struct kvm_apic_map *map,
594 struct kvm_lapic_irq *irq, u32 min)
597 struct kvm_vcpu *vcpu;
599 if (min > map->max_apic_id)
602 for_each_set_bit(i, ipi_bitmap,
603 min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
604 if (map->phys_map[min + i]) {
605 vcpu = map->phys_map[min + i]->vcpu;
606 count += kvm_apic_set_irq(vcpu, irq, NULL);
613 int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
614 unsigned long ipi_bitmap_high, u32 min,
615 unsigned long icr, int op_64_bit)
617 struct kvm_apic_map *map;
618 struct kvm_lapic_irq irq = {0};
619 int cluster_size = op_64_bit ? 64 : 32;
622 if (icr & (APIC_DEST_MASK | APIC_SHORT_MASK))
625 irq.vector = icr & APIC_VECTOR_MASK;
626 irq.delivery_mode = icr & APIC_MODE_MASK;
627 irq.level = (icr & APIC_INT_ASSERT) != 0;
628 irq.trig_mode = icr & APIC_INT_LEVELTRIG;
631 map = rcu_dereference(kvm->arch.apic_map);
635 count = __pv_send_ipi(&ipi_bitmap_low, map, &irq, min);
637 count += __pv_send_ipi(&ipi_bitmap_high, map, &irq, min);
644 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
647 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
651 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
654 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
658 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
660 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
663 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
666 if (pv_eoi_get_user(vcpu, &val) < 0) {
667 printk(KERN_WARNING "Can't read EOI MSR value: 0x%llx\n",
668 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
674 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
676 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
677 printk(KERN_WARNING "Can't set EOI MSR value: 0x%llx\n",
678 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
681 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
684 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
686 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
687 printk(KERN_WARNING "Can't clear EOI MSR value: 0x%llx\n",
688 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
691 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
694 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
697 if (apic->vcpu->arch.apicv_active)
698 highest_irr = kvm_x86_ops.sync_pir_to_irr(apic->vcpu);
700 highest_irr = apic_find_highest_irr(apic);
701 if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
706 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
708 u32 tpr, isrv, ppr, old_ppr;
711 old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
712 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
713 isr = apic_find_highest_isr(apic);
714 isrv = (isr != -1) ? isr : 0;
716 if ((tpr & 0xf0) >= (isrv & 0xf0))
723 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
725 return ppr < old_ppr;
728 static void apic_update_ppr(struct kvm_lapic *apic)
732 if (__apic_update_ppr(apic, &ppr) &&
733 apic_has_interrupt_for_ppr(apic, ppr) != -1)
734 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
737 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
739 apic_update_ppr(vcpu->arch.apic);
741 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
743 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
745 kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
746 apic_update_ppr(apic);
749 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
751 return mda == (apic_x2apic_mode(apic) ?
752 X2APIC_BROADCAST : APIC_BROADCAST);
755 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
757 if (kvm_apic_broadcast(apic, mda))
760 if (apic_x2apic_mode(apic))
761 return mda == kvm_x2apic_id(apic);
764 * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
765 * it were in x2APIC mode. Hotplugged VCPUs start in xAPIC mode and
766 * this allows unique addressing of VCPUs with APIC ID over 0xff.
767 * The 0xff condition is needed because writeable xAPIC ID.
769 if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
772 return mda == kvm_xapic_id(apic);
775 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
779 if (kvm_apic_broadcast(apic, mda))
782 logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
784 if (apic_x2apic_mode(apic))
785 return ((logical_id >> 16) == (mda >> 16))
786 && (logical_id & mda & 0xffff) != 0;
788 logical_id = GET_APIC_LOGICAL_ID(logical_id);
790 switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
792 return (logical_id & mda) != 0;
793 case APIC_DFR_CLUSTER:
794 return ((logical_id >> 4) == (mda >> 4))
795 && (logical_id & mda & 0xf) != 0;
801 /* The KVM local APIC implementation has two quirks:
803 * - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
804 * in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
805 * KVM doesn't do that aliasing.
807 * - in-kernel IOAPIC messages have to be delivered directly to
808 * x2APIC, because the kernel does not support interrupt remapping.
809 * In order to support broadcast without interrupt remapping, x2APIC
810 * rewrites the destination of non-IPI messages from APIC_BROADCAST
811 * to X2APIC_BROADCAST.
813 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API. This is
814 * important when userspace wants to use x2APIC-format MSIs, because
815 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
817 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
818 struct kvm_lapic *source, struct kvm_lapic *target)
820 bool ipi = source != NULL;
822 if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
823 !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
824 return X2APIC_BROADCAST;
829 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
830 int shorthand, unsigned int dest, int dest_mode)
832 struct kvm_lapic *target = vcpu->arch.apic;
833 u32 mda = kvm_apic_mda(vcpu, dest, source, target);
837 case APIC_DEST_NOSHORT:
838 if (dest_mode == APIC_DEST_PHYSICAL)
839 return kvm_apic_match_physical_addr(target, mda);
841 return kvm_apic_match_logical_addr(target, mda);
843 return target == source;
844 case APIC_DEST_ALLINC:
846 case APIC_DEST_ALLBUT:
847 return target != source;
852 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
854 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
855 const unsigned long *bitmap, u32 bitmap_size)
860 mod = vector % dest_vcpus;
862 for (i = 0; i <= mod; i++) {
863 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
864 BUG_ON(idx == bitmap_size);
870 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
872 if (!kvm->arch.disabled_lapic_found) {
873 kvm->arch.disabled_lapic_found = true;
875 "Disabled LAPIC found during irq injection\n");
879 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
880 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
882 if (kvm->arch.x2apic_broadcast_quirk_disabled) {
883 if ((irq->dest_id == APIC_BROADCAST &&
884 map->mode != KVM_APIC_MODE_X2APIC))
886 if (irq->dest_id == X2APIC_BROADCAST)
889 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
890 if (irq->dest_id == (x2apic_ipi ?
891 X2APIC_BROADCAST : APIC_BROADCAST))
898 /* Return true if the interrupt can be handled by using *bitmap as index mask
899 * for valid destinations in *dst array.
900 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
901 * Note: we may have zero kvm_lapic destinations when we return true, which
902 * means that the interrupt should be dropped. In this case, *bitmap would be
903 * zero and *dst undefined.
905 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
906 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
907 struct kvm_apic_map *map, struct kvm_lapic ***dst,
908 unsigned long *bitmap)
912 if (irq->shorthand == APIC_DEST_SELF && src) {
916 } else if (irq->shorthand)
919 if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
922 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
923 if (irq->dest_id > map->max_apic_id) {
926 u32 dest_id = array_index_nospec(irq->dest_id, map->max_apic_id + 1);
927 *dst = &map->phys_map[dest_id];
934 if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
938 if (!kvm_lowest_prio_delivery(irq))
941 if (!kvm_vector_hashing_enabled()) {
943 for_each_set_bit(i, bitmap, 16) {
948 else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
949 (*dst)[lowest]->vcpu) < 0)
956 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
959 if (!(*dst)[lowest]) {
960 kvm_apic_disabled_lapic_found(kvm);
966 *bitmap = (lowest >= 0) ? 1 << lowest : 0;
971 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
972 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
974 struct kvm_apic_map *map;
975 unsigned long bitmap;
976 struct kvm_lapic **dst = NULL;
982 if (irq->shorthand == APIC_DEST_SELF) {
983 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
988 map = rcu_dereference(kvm->arch.apic_map);
990 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
993 for_each_set_bit(i, &bitmap, 16) {
996 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
1005 * This routine tries to handle interrupts in posted mode, here is how
1006 * it deals with different cases:
1007 * - For single-destination interrupts, handle it in posted mode
1008 * - Else if vector hashing is enabled and it is a lowest-priority
1009 * interrupt, handle it in posted mode and use the following mechanism
1010 * to find the destination vCPU.
1011 * 1. For lowest-priority interrupts, store all the possible
1012 * destination vCPUs in an array.
1013 * 2. Use "guest vector % max number of destination vCPUs" to find
1014 * the right destination vCPU in the array for the lowest-priority
1016 * - Otherwise, use remapped mode to inject the interrupt.
1018 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
1019 struct kvm_vcpu **dest_vcpu)
1021 struct kvm_apic_map *map;
1022 unsigned long bitmap;
1023 struct kvm_lapic **dst = NULL;
1030 map = rcu_dereference(kvm->arch.apic_map);
1032 if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
1033 hweight16(bitmap) == 1) {
1034 unsigned long i = find_first_bit(&bitmap, 16);
1037 *dest_vcpu = dst[i]->vcpu;
1047 * Add a pending IRQ into lapic.
1048 * Return 1 if successfully added and 0 if discarded.
1050 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
1051 int vector, int level, int trig_mode,
1052 struct dest_map *dest_map)
1055 struct kvm_vcpu *vcpu = apic->vcpu;
1057 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
1059 switch (delivery_mode) {
1060 case APIC_DM_LOWEST:
1061 vcpu->arch.apic_arb_prio++;
1064 if (unlikely(trig_mode && !level))
1067 /* FIXME add logic for vcpu on reset */
1068 if (unlikely(!apic_enabled(apic)))
1074 __set_bit(vcpu->vcpu_id, dest_map->map);
1075 dest_map->vectors[vcpu->vcpu_id] = vector;
1078 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
1080 kvm_lapic_set_vector(vector,
1081 apic->regs + APIC_TMR);
1083 kvm_lapic_clear_vector(vector,
1084 apic->regs + APIC_TMR);
1087 if (kvm_x86_ops.deliver_posted_interrupt(vcpu, vector)) {
1088 kvm_lapic_set_irr(vector, apic);
1089 kvm_make_request(KVM_REQ_EVENT, vcpu);
1090 kvm_vcpu_kick(vcpu);
1096 vcpu->arch.pv.pv_unhalted = 1;
1097 kvm_make_request(KVM_REQ_EVENT, vcpu);
1098 kvm_vcpu_kick(vcpu);
1103 kvm_make_request(KVM_REQ_SMI, vcpu);
1104 kvm_vcpu_kick(vcpu);
1109 kvm_inject_nmi(vcpu);
1110 kvm_vcpu_kick(vcpu);
1114 if (!trig_mode || level) {
1116 /* assumes that there are only KVM_APIC_INIT/SIPI */
1117 apic->pending_events = (1UL << KVM_APIC_INIT);
1118 kvm_make_request(KVM_REQ_EVENT, vcpu);
1119 kvm_vcpu_kick(vcpu);
1123 case APIC_DM_STARTUP:
1125 apic->sipi_vector = vector;
1126 /* make sure sipi_vector is visible for the receiver */
1128 set_bit(KVM_APIC_SIPI, &apic->pending_events);
1129 kvm_make_request(KVM_REQ_EVENT, vcpu);
1130 kvm_vcpu_kick(vcpu);
1133 case APIC_DM_EXTINT:
1135 * Should only be called by kvm_apic_local_deliver() with LVT0,
1136 * before NMI watchdog was enabled. Already handled by
1137 * kvm_apic_accept_pic_intr().
1142 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1150 * This routine identifies the destination vcpus mask meant to receive the
1151 * IOAPIC interrupts. It either uses kvm_apic_map_get_dest_lapic() to find
1152 * out the destination vcpus array and set the bitmap or it traverses to
1153 * each available vcpu to identify the same.
1155 void kvm_bitmap_or_dest_vcpus(struct kvm *kvm, struct kvm_lapic_irq *irq,
1156 unsigned long *vcpu_bitmap)
1158 struct kvm_lapic **dest_vcpu = NULL;
1159 struct kvm_lapic *src = NULL;
1160 struct kvm_apic_map *map;
1161 struct kvm_vcpu *vcpu;
1162 unsigned long bitmap;
1167 map = rcu_dereference(kvm->arch.apic_map);
1169 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dest_vcpu,
1172 for_each_set_bit(i, &bitmap, 16) {
1175 vcpu_idx = dest_vcpu[i]->vcpu->vcpu_idx;
1176 __set_bit(vcpu_idx, vcpu_bitmap);
1179 kvm_for_each_vcpu(i, vcpu, kvm) {
1180 if (!kvm_apic_present(vcpu))
1182 if (!kvm_apic_match_dest(vcpu, NULL,
1187 __set_bit(i, vcpu_bitmap);
1193 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1195 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1198 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1200 return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1203 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1207 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1208 if (!kvm_ioapic_handles_vector(apic, vector))
1211 /* Request a KVM exit to inform the userspace IOAPIC. */
1212 if (irqchip_split(apic->vcpu->kvm)) {
1213 apic->vcpu->arch.pending_ioapic_eoi = vector;
1214 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1218 if (apic_test_vector(vector, apic->regs + APIC_TMR))
1219 trigger_mode = IOAPIC_LEVEL_TRIG;
1221 trigger_mode = IOAPIC_EDGE_TRIG;
1223 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1226 static int apic_set_eoi(struct kvm_lapic *apic)
1228 int vector = apic_find_highest_isr(apic);
1230 trace_kvm_eoi(apic, vector);
1233 * Not every write EOI will has corresponding ISR,
1234 * one example is when Kernel check timer on setup_IO_APIC
1239 apic_clear_isr(vector, apic);
1240 apic_update_ppr(apic);
1242 if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1243 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1245 kvm_ioapic_send_eoi(apic, vector);
1246 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1251 * this interface assumes a trap-like exit, which has already finished
1252 * desired side effect including vISR and vPPR update.
1254 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1256 struct kvm_lapic *apic = vcpu->arch.apic;
1258 trace_kvm_eoi(apic, vector);
1260 kvm_ioapic_send_eoi(apic, vector);
1261 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1263 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1265 void kvm_apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
1267 struct kvm_lapic_irq irq;
1269 irq.vector = icr_low & APIC_VECTOR_MASK;
1270 irq.delivery_mode = icr_low & APIC_MODE_MASK;
1271 irq.dest_mode = icr_low & APIC_DEST_MASK;
1272 irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1273 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1274 irq.shorthand = icr_low & APIC_SHORT_MASK;
1275 irq.msi_redir_hint = false;
1276 if (apic_x2apic_mode(apic))
1277 irq.dest_id = icr_high;
1279 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1281 trace_kvm_apic_ipi(icr_low, irq.dest_id);
1283 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1286 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1288 ktime_t remaining, now;
1292 ASSERT(apic != NULL);
1294 /* if initial count is 0, current count should also be 0 */
1295 if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1296 apic->lapic_timer.period == 0)
1300 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1301 if (ktime_to_ns(remaining) < 0)
1304 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1305 tmcct = div64_u64(ns,
1306 (APIC_BUS_CYCLE_NS * apic->divide_count));
1311 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1313 struct kvm_vcpu *vcpu = apic->vcpu;
1314 struct kvm_run *run = vcpu->run;
1316 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1317 run->tpr_access.rip = kvm_rip_read(vcpu);
1318 run->tpr_access.is_write = write;
1321 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1323 if (apic->vcpu->arch.tpr_access_reporting)
1324 __report_tpr_access(apic, write);
1327 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1331 if (offset >= LAPIC_MMIO_LENGTH)
1338 case APIC_TMCCT: /* Timer CCR */
1339 if (apic_lvtt_tscdeadline(apic))
1342 val = apic_get_tmcct(apic);
1345 apic_update_ppr(apic);
1346 val = kvm_lapic_get_reg(apic, offset);
1349 report_tpr_access(apic, false);
1352 val = kvm_lapic_get_reg(apic, offset);
1359 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1361 return container_of(dev, struct kvm_lapic, dev);
1364 #define APIC_REG_MASK(reg) (1ull << ((reg) >> 4))
1365 #define APIC_REGS_MASK(first, count) \
1366 (APIC_REG_MASK(first) * ((1ull << (count)) - 1))
1368 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1371 unsigned char alignment = offset & 0xf;
1373 /* this bitmask has a bit cleared for each reserved register */
1374 u64 valid_reg_mask =
1375 APIC_REG_MASK(APIC_ID) |
1376 APIC_REG_MASK(APIC_LVR) |
1377 APIC_REG_MASK(APIC_TASKPRI) |
1378 APIC_REG_MASK(APIC_PROCPRI) |
1379 APIC_REG_MASK(APIC_LDR) |
1380 APIC_REG_MASK(APIC_DFR) |
1381 APIC_REG_MASK(APIC_SPIV) |
1382 APIC_REGS_MASK(APIC_ISR, APIC_ISR_NR) |
1383 APIC_REGS_MASK(APIC_TMR, APIC_ISR_NR) |
1384 APIC_REGS_MASK(APIC_IRR, APIC_ISR_NR) |
1385 APIC_REG_MASK(APIC_ESR) |
1386 APIC_REG_MASK(APIC_ICR) |
1387 APIC_REG_MASK(APIC_ICR2) |
1388 APIC_REG_MASK(APIC_LVTT) |
1389 APIC_REG_MASK(APIC_LVTTHMR) |
1390 APIC_REG_MASK(APIC_LVTPC) |
1391 APIC_REG_MASK(APIC_LVT0) |
1392 APIC_REG_MASK(APIC_LVT1) |
1393 APIC_REG_MASK(APIC_LVTERR) |
1394 APIC_REG_MASK(APIC_TMICT) |
1395 APIC_REG_MASK(APIC_TMCCT) |
1396 APIC_REG_MASK(APIC_TDCR);
1398 /* ARBPRI is not valid on x2APIC */
1399 if (!apic_x2apic_mode(apic))
1400 valid_reg_mask |= APIC_REG_MASK(APIC_ARBPRI);
1402 if (offset > 0x3f0 || !(valid_reg_mask & APIC_REG_MASK(offset)))
1405 result = __apic_read(apic, offset & ~0xf);
1407 trace_kvm_apic_read(offset, result);
1413 memcpy(data, (char *)&result + alignment, len);
1416 printk(KERN_ERR "Local APIC read with len = %x, "
1417 "should be 1,2, or 4 instead\n", len);
1422 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1424 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1426 return addr >= apic->base_address &&
1427 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1430 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1431 gpa_t address, int len, void *data)
1433 struct kvm_lapic *apic = to_lapic(this);
1434 u32 offset = address - apic->base_address;
1436 if (!apic_mmio_in_range(apic, address))
1439 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1440 if (!kvm_check_has_quirk(vcpu->kvm,
1441 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1444 memset(data, 0xff, len);
1448 kvm_lapic_reg_read(apic, offset, len, data);
1453 static void update_divide_count(struct kvm_lapic *apic)
1455 u32 tmp1, tmp2, tdcr;
1457 tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1459 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1460 apic->divide_count = 0x1 << (tmp2 & 0x7);
1463 static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1466 * Do not allow the guest to program periodic timers with small
1467 * interval, since the hrtimers are not throttled by the host
1470 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1471 s64 min_period = min_timer_period_us * 1000LL;
1473 if (apic->lapic_timer.period < min_period) {
1474 pr_info_ratelimited(
1475 "kvm: vcpu %i: requested %lld ns "
1476 "lapic timer period limited to %lld ns\n",
1477 apic->vcpu->vcpu_id,
1478 apic->lapic_timer.period, min_period);
1479 apic->lapic_timer.period = min_period;
1484 static void cancel_hv_timer(struct kvm_lapic *apic);
1486 static void apic_update_lvtt(struct kvm_lapic *apic)
1488 u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1489 apic->lapic_timer.timer_mode_mask;
1491 if (apic->lapic_timer.timer_mode != timer_mode) {
1492 if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1493 APIC_LVT_TIMER_TSCDEADLINE)) {
1494 hrtimer_cancel(&apic->lapic_timer.timer);
1496 if (apic->lapic_timer.hv_timer_in_use)
1497 cancel_hv_timer(apic);
1499 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1500 apic->lapic_timer.period = 0;
1501 apic->lapic_timer.tscdeadline = 0;
1503 apic->lapic_timer.timer_mode = timer_mode;
1504 limit_periodic_timer_frequency(apic);
1509 * On APICv, this test will cause a busy wait
1510 * during a higher-priority task.
1513 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1515 struct kvm_lapic *apic = vcpu->arch.apic;
1516 u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1518 if (kvm_apic_hw_enabled(apic)) {
1519 int vec = reg & APIC_VECTOR_MASK;
1520 void *bitmap = apic->regs + APIC_ISR;
1522 if (vcpu->arch.apicv_active)
1523 bitmap = apic->regs + APIC_IRR;
1525 if (apic_test_vector(vec, bitmap))
1531 static inline void __wait_lapic_expire(struct kvm_vcpu *vcpu, u64 guest_cycles)
1533 u64 timer_advance_ns = vcpu->arch.apic->lapic_timer.timer_advance_ns;
1536 * If the guest TSC is running at a different ratio than the host, then
1537 * convert the delay to nanoseconds to achieve an accurate delay. Note
1538 * that __delay() uses delay_tsc whenever the hardware has TSC, thus
1539 * always for VMX enabled hardware.
1541 if (vcpu->arch.tsc_scaling_ratio == kvm_default_tsc_scaling_ratio) {
1542 __delay(min(guest_cycles,
1543 nsec_to_cycles(vcpu, timer_advance_ns)));
1545 u64 delay_ns = guest_cycles * 1000000ULL;
1546 do_div(delay_ns, vcpu->arch.virtual_tsc_khz);
1547 ndelay(min_t(u32, delay_ns, timer_advance_ns));
1551 static inline void adjust_lapic_timer_advance(struct kvm_vcpu *vcpu,
1552 s64 advance_expire_delta)
1554 struct kvm_lapic *apic = vcpu->arch.apic;
1555 u32 timer_advance_ns = apic->lapic_timer.timer_advance_ns;
1558 /* Do not adjust for tiny fluctuations or large random spikes. */
1559 if (abs(advance_expire_delta) > LAPIC_TIMER_ADVANCE_ADJUST_MAX ||
1560 abs(advance_expire_delta) < LAPIC_TIMER_ADVANCE_ADJUST_MIN)
1564 if (advance_expire_delta < 0) {
1565 ns = -advance_expire_delta * 1000000ULL;
1566 do_div(ns, vcpu->arch.virtual_tsc_khz);
1567 timer_advance_ns -= ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1570 ns = advance_expire_delta * 1000000ULL;
1571 do_div(ns, vcpu->arch.virtual_tsc_khz);
1572 timer_advance_ns += ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1575 if (unlikely(timer_advance_ns > LAPIC_TIMER_ADVANCE_NS_MAX))
1576 timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
1577 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
1580 static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1582 struct kvm_lapic *apic = vcpu->arch.apic;
1583 u64 guest_tsc, tsc_deadline;
1585 if (apic->lapic_timer.expired_tscdeadline == 0)
1588 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1589 apic->lapic_timer.expired_tscdeadline = 0;
1590 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1591 apic->lapic_timer.advance_expire_delta = guest_tsc - tsc_deadline;
1593 if (guest_tsc < tsc_deadline)
1594 __wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
1596 if (lapic_timer_advance_dynamic)
1597 adjust_lapic_timer_advance(vcpu, apic->lapic_timer.advance_expire_delta);
1600 void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1602 if (lapic_timer_int_injected(vcpu))
1603 __kvm_wait_lapic_expire(vcpu);
1605 EXPORT_SYMBOL_GPL(kvm_wait_lapic_expire);
1607 static void kvm_apic_inject_pending_timer_irqs(struct kvm_lapic *apic)
1609 struct kvm_timer *ktimer = &apic->lapic_timer;
1611 kvm_apic_local_deliver(apic, APIC_LVTT);
1612 if (apic_lvtt_tscdeadline(apic)) {
1613 ktimer->tscdeadline = 0;
1614 } else if (apic_lvtt_oneshot(apic)) {
1615 ktimer->tscdeadline = 0;
1616 ktimer->target_expiration = 0;
1620 static void apic_timer_expired(struct kvm_lapic *apic, bool from_timer_fn)
1622 struct kvm_vcpu *vcpu = apic->vcpu;
1623 struct kvm_timer *ktimer = &apic->lapic_timer;
1625 if (atomic_read(&apic->lapic_timer.pending))
1628 if (apic_lvtt_tscdeadline(apic) || ktimer->hv_timer_in_use)
1629 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1631 if (!from_timer_fn && vcpu->arch.apicv_active) {
1632 WARN_ON(kvm_get_running_vcpu() != vcpu);
1633 kvm_apic_inject_pending_timer_irqs(apic);
1637 if (kvm_use_posted_timer_interrupt(apic->vcpu)) {
1638 if (apic->lapic_timer.timer_advance_ns)
1639 __kvm_wait_lapic_expire(vcpu);
1640 kvm_apic_inject_pending_timer_irqs(apic);
1644 atomic_inc(&apic->lapic_timer.pending);
1645 kvm_set_pending_timer(vcpu);
1648 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1650 struct kvm_timer *ktimer = &apic->lapic_timer;
1651 u64 guest_tsc, tscdeadline = ktimer->tscdeadline;
1654 struct kvm_vcpu *vcpu = apic->vcpu;
1655 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1656 unsigned long flags;
1659 if (unlikely(!tscdeadline || !this_tsc_khz))
1662 local_irq_save(flags);
1665 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1667 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1668 do_div(ns, this_tsc_khz);
1670 if (likely(tscdeadline > guest_tsc) &&
1671 likely(ns > apic->lapic_timer.timer_advance_ns)) {
1672 expire = ktime_add_ns(now, ns);
1673 expire = ktime_sub_ns(expire, ktimer->timer_advance_ns);
1674 hrtimer_start(&ktimer->timer, expire, HRTIMER_MODE_ABS_HARD);
1676 apic_timer_expired(apic, false);
1678 local_irq_restore(flags);
1681 static inline u64 tmict_to_ns(struct kvm_lapic *apic, u32 tmict)
1683 return (u64)tmict * APIC_BUS_CYCLE_NS * (u64)apic->divide_count;
1686 static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1688 ktime_t now, remaining;
1689 u64 ns_remaining_old, ns_remaining_new;
1691 apic->lapic_timer.period =
1692 tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1693 limit_periodic_timer_frequency(apic);
1696 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1697 if (ktime_to_ns(remaining) < 0)
1700 ns_remaining_old = ktime_to_ns(remaining);
1701 ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1702 apic->divide_count, old_divisor);
1704 apic->lapic_timer.tscdeadline +=
1705 nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1706 nsec_to_cycles(apic->vcpu, ns_remaining_old);
1707 apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1710 static bool set_target_expiration(struct kvm_lapic *apic, u32 count_reg)
1717 apic->lapic_timer.period =
1718 tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1720 if (!apic->lapic_timer.period) {
1721 apic->lapic_timer.tscdeadline = 0;
1725 limit_periodic_timer_frequency(apic);
1726 deadline = apic->lapic_timer.period;
1728 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1729 if (unlikely(count_reg != APIC_TMICT)) {
1730 deadline = tmict_to_ns(apic,
1731 kvm_lapic_get_reg(apic, count_reg));
1732 if (unlikely(deadline <= 0))
1733 deadline = apic->lapic_timer.period;
1734 else if (unlikely(deadline > apic->lapic_timer.period)) {
1735 pr_info_ratelimited(
1736 "kvm: vcpu %i: requested lapic timer restore with "
1737 "starting count register %#x=%u (%lld ns) > initial count (%lld ns). "
1738 "Using initial count to start timer.\n",
1739 apic->vcpu->vcpu_id,
1741 kvm_lapic_get_reg(apic, count_reg),
1742 deadline, apic->lapic_timer.period);
1743 kvm_lapic_set_reg(apic, count_reg, 0);
1744 deadline = apic->lapic_timer.period;
1749 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1750 nsec_to_cycles(apic->vcpu, deadline);
1751 apic->lapic_timer.target_expiration = ktime_add_ns(now, deadline);
1756 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1758 ktime_t now = ktime_get();
1763 * Synchronize both deadlines to the same time source or
1764 * differences in the periods (caused by differences in the
1765 * underlying clocks or numerical approximation errors) will
1766 * cause the two to drift apart over time as the errors
1769 apic->lapic_timer.target_expiration =
1770 ktime_add_ns(apic->lapic_timer.target_expiration,
1771 apic->lapic_timer.period);
1772 delta = ktime_sub(apic->lapic_timer.target_expiration, now);
1773 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1774 nsec_to_cycles(apic->vcpu, delta);
1777 static void start_sw_period(struct kvm_lapic *apic)
1779 if (!apic->lapic_timer.period)
1782 if (ktime_after(ktime_get(),
1783 apic->lapic_timer.target_expiration)) {
1784 apic_timer_expired(apic, false);
1786 if (apic_lvtt_oneshot(apic))
1789 advance_periodic_target_expiration(apic);
1792 hrtimer_start(&apic->lapic_timer.timer,
1793 apic->lapic_timer.target_expiration,
1794 HRTIMER_MODE_ABS_HARD);
1797 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1799 if (!lapic_in_kernel(vcpu))
1802 return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1804 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1806 static void cancel_hv_timer(struct kvm_lapic *apic)
1808 WARN_ON(preemptible());
1809 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1810 kvm_x86_ops.cancel_hv_timer(apic->vcpu);
1811 apic->lapic_timer.hv_timer_in_use = false;
1814 static bool start_hv_timer(struct kvm_lapic *apic)
1816 struct kvm_timer *ktimer = &apic->lapic_timer;
1817 struct kvm_vcpu *vcpu = apic->vcpu;
1820 WARN_ON(preemptible());
1821 if (!kvm_can_use_hv_timer(vcpu))
1824 if (!ktimer->tscdeadline)
1827 if (kvm_x86_ops.set_hv_timer(vcpu, ktimer->tscdeadline, &expired))
1830 ktimer->hv_timer_in_use = true;
1831 hrtimer_cancel(&ktimer->timer);
1834 * To simplify handling the periodic timer, leave the hv timer running
1835 * even if the deadline timer has expired, i.e. rely on the resulting
1836 * VM-Exit to recompute the periodic timer's target expiration.
1838 if (!apic_lvtt_period(apic)) {
1840 * Cancel the hv timer if the sw timer fired while the hv timer
1841 * was being programmed, or if the hv timer itself expired.
1843 if (atomic_read(&ktimer->pending)) {
1844 cancel_hv_timer(apic);
1845 } else if (expired) {
1846 apic_timer_expired(apic, false);
1847 cancel_hv_timer(apic);
1851 trace_kvm_hv_timer_state(vcpu->vcpu_id, ktimer->hv_timer_in_use);
1856 static void start_sw_timer(struct kvm_lapic *apic)
1858 struct kvm_timer *ktimer = &apic->lapic_timer;
1860 WARN_ON(preemptible());
1861 if (apic->lapic_timer.hv_timer_in_use)
1862 cancel_hv_timer(apic);
1863 if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1866 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1867 start_sw_period(apic);
1868 else if (apic_lvtt_tscdeadline(apic))
1869 start_sw_tscdeadline(apic);
1870 trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1873 static void restart_apic_timer(struct kvm_lapic *apic)
1877 if (!apic_lvtt_period(apic) && atomic_read(&apic->lapic_timer.pending))
1880 if (!start_hv_timer(apic))
1881 start_sw_timer(apic);
1886 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1888 struct kvm_lapic *apic = vcpu->arch.apic;
1891 /* If the preempt notifier has already run, it also called apic_timer_expired */
1892 if (!apic->lapic_timer.hv_timer_in_use)
1894 WARN_ON(rcuwait_active(&vcpu->wait));
1895 cancel_hv_timer(apic);
1896 apic_timer_expired(apic, false);
1898 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1899 advance_periodic_target_expiration(apic);
1900 restart_apic_timer(apic);
1905 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1907 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1909 restart_apic_timer(vcpu->arch.apic);
1911 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1913 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1915 struct kvm_lapic *apic = vcpu->arch.apic;
1918 /* Possibly the TSC deadline timer is not enabled yet */
1919 if (apic->lapic_timer.hv_timer_in_use)
1920 start_sw_timer(apic);
1923 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1925 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
1927 struct kvm_lapic *apic = vcpu->arch.apic;
1929 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1930 restart_apic_timer(apic);
1933 static void __start_apic_timer(struct kvm_lapic *apic, u32 count_reg)
1935 atomic_set(&apic->lapic_timer.pending, 0);
1937 if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1938 && !set_target_expiration(apic, count_reg))
1941 restart_apic_timer(apic);
1944 static void start_apic_timer(struct kvm_lapic *apic)
1946 __start_apic_timer(apic, APIC_TMICT);
1949 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1951 bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1953 if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1954 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1955 if (lvt0_in_nmi_mode) {
1956 atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1958 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1962 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1966 trace_kvm_apic_write(reg, val);
1969 case APIC_ID: /* Local APIC ID */
1970 if (!apic_x2apic_mode(apic))
1971 kvm_apic_set_xapic_id(apic, val >> 24);
1977 report_tpr_access(apic, true);
1978 apic_set_tpr(apic, val & 0xff);
1986 if (!apic_x2apic_mode(apic))
1987 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1993 if (!apic_x2apic_mode(apic))
1994 kvm_apic_set_dfr(apic, val | 0x0FFFFFFF);
2001 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
2002 mask |= APIC_SPIV_DIRECTED_EOI;
2003 apic_set_spiv(apic, val & mask);
2004 if (!(val & APIC_SPIV_APIC_ENABLED)) {
2008 for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
2009 lvt_val = kvm_lapic_get_reg(apic,
2010 APIC_LVTT + 0x10 * i);
2011 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
2012 lvt_val | APIC_LVT_MASKED);
2014 apic_update_lvtt(apic);
2015 atomic_set(&apic->lapic_timer.pending, 0);
2021 /* No delay here, so we always clear the pending bit */
2023 kvm_apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
2024 kvm_lapic_set_reg(apic, APIC_ICR, val);
2028 if (!apic_x2apic_mode(apic))
2030 kvm_lapic_set_reg(apic, APIC_ICR2, val);
2034 apic_manage_nmi_watchdog(apic, val);
2040 /* TODO: Check vector */
2044 if (!kvm_apic_sw_enabled(apic))
2045 val |= APIC_LVT_MASKED;
2046 size = ARRAY_SIZE(apic_lvt_mask);
2047 index = array_index_nospec(
2048 (reg - APIC_LVTT) >> 4, size);
2049 val &= apic_lvt_mask[index];
2050 kvm_lapic_set_reg(apic, reg, val);
2055 if (!kvm_apic_sw_enabled(apic))
2056 val |= APIC_LVT_MASKED;
2057 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
2058 kvm_lapic_set_reg(apic, APIC_LVTT, val);
2059 apic_update_lvtt(apic);
2063 if (apic_lvtt_tscdeadline(apic))
2066 hrtimer_cancel(&apic->lapic_timer.timer);
2067 kvm_lapic_set_reg(apic, APIC_TMICT, val);
2068 start_apic_timer(apic);
2072 uint32_t old_divisor = apic->divide_count;
2074 kvm_lapic_set_reg(apic, APIC_TDCR, val & 0xb);
2075 update_divide_count(apic);
2076 if (apic->divide_count != old_divisor &&
2077 apic->lapic_timer.period) {
2078 hrtimer_cancel(&apic->lapic_timer.timer);
2079 update_target_expiration(apic, old_divisor);
2080 restart_apic_timer(apic);
2085 if (apic_x2apic_mode(apic) && val != 0)
2090 if (apic_x2apic_mode(apic)) {
2091 kvm_lapic_reg_write(apic, APIC_ICR,
2092 APIC_DEST_SELF | (val & APIC_VECTOR_MASK));
2101 kvm_recalculate_apic_map(apic->vcpu->kvm);
2105 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
2107 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
2108 gpa_t address, int len, const void *data)
2110 struct kvm_lapic *apic = to_lapic(this);
2111 unsigned int offset = address - apic->base_address;
2114 if (!apic_mmio_in_range(apic, address))
2117 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
2118 if (!kvm_check_has_quirk(vcpu->kvm,
2119 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
2126 * APIC register must be aligned on 128-bits boundary.
2127 * 32/64/128 bits registers must be accessed thru 32 bits.
2130 if (len != 4 || (offset & 0xf))
2135 kvm_lapic_reg_write(apic, offset & 0xff0, val);
2140 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
2142 kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
2144 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
2146 /* emulate APIC access in a trap manner */
2147 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
2151 /* hw has done the conditional check and inst decode */
2154 kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
2156 /* TODO: optimize to just emulate side effect w/o one more write */
2157 kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
2159 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
2161 void kvm_free_lapic(struct kvm_vcpu *vcpu)
2163 struct kvm_lapic *apic = vcpu->arch.apic;
2165 if (!vcpu->arch.apic)
2168 hrtimer_cancel(&apic->lapic_timer.timer);
2170 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
2171 static_key_slow_dec_deferred(&apic_hw_disabled);
2173 if (!apic->sw_enabled)
2174 static_key_slow_dec_deferred(&apic_sw_disabled);
2177 free_page((unsigned long)apic->regs);
2183 *----------------------------------------------------------------------
2185 *----------------------------------------------------------------------
2187 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
2189 struct kvm_lapic *apic = vcpu->arch.apic;
2191 if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2194 return apic->lapic_timer.tscdeadline;
2197 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
2199 struct kvm_lapic *apic = vcpu->arch.apic;
2201 if (!kvm_apic_present(vcpu) || apic_lvtt_oneshot(apic) ||
2202 apic_lvtt_period(apic))
2205 hrtimer_cancel(&apic->lapic_timer.timer);
2206 apic->lapic_timer.tscdeadline = data;
2207 start_apic_timer(apic);
2210 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
2212 struct kvm_lapic *apic = vcpu->arch.apic;
2214 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
2215 | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
2218 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
2222 tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
2224 return (tpr & 0xf0) >> 4;
2227 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
2229 u64 old_value = vcpu->arch.apic_base;
2230 struct kvm_lapic *apic = vcpu->arch.apic;
2233 value |= MSR_IA32_APICBASE_BSP;
2235 vcpu->arch.apic_base = value;
2237 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
2238 kvm_update_cpuid_runtime(vcpu);
2243 /* update jump label if enable bit changes */
2244 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
2245 if (value & MSR_IA32_APICBASE_ENABLE) {
2246 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2247 static_key_slow_dec_deferred(&apic_hw_disabled);
2249 static_key_slow_inc(&apic_hw_disabled.key);
2250 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2254 if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
2255 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
2257 if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE))
2258 kvm_x86_ops.set_virtual_apic_mode(vcpu);
2260 apic->base_address = apic->vcpu->arch.apic_base &
2261 MSR_IA32_APICBASE_BASE;
2263 if ((value & MSR_IA32_APICBASE_ENABLE) &&
2264 apic->base_address != APIC_DEFAULT_PHYS_BASE)
2265 pr_warn_once("APIC base relocation is unsupported by KVM");
2268 void kvm_apic_update_apicv(struct kvm_vcpu *vcpu)
2270 struct kvm_lapic *apic = vcpu->arch.apic;
2272 if (vcpu->arch.apicv_active) {
2273 /* irr_pending is always true when apicv is activated. */
2274 apic->irr_pending = true;
2275 apic->isr_count = 1;
2277 apic->irr_pending = (apic_search_irr(apic) != -1);
2278 apic->isr_count = count_vectors(apic->regs + APIC_ISR);
2281 EXPORT_SYMBOL_GPL(kvm_apic_update_apicv);
2283 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2285 struct kvm_lapic *apic = vcpu->arch.apic;
2291 /* Stop the timer in case it's a reset to an active apic */
2292 hrtimer_cancel(&apic->lapic_timer.timer);
2295 kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
2296 MSR_IA32_APICBASE_ENABLE);
2297 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2299 kvm_apic_set_version(apic->vcpu);
2301 for (i = 0; i < KVM_APIC_LVT_NUM; i++)
2302 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
2303 apic_update_lvtt(apic);
2304 if (kvm_vcpu_is_reset_bsp(vcpu) &&
2305 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2306 kvm_lapic_set_reg(apic, APIC_LVT0,
2307 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2308 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2310 kvm_apic_set_dfr(apic, 0xffffffffU);
2311 apic_set_spiv(apic, 0xff);
2312 kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2313 if (!apic_x2apic_mode(apic))
2314 kvm_apic_set_ldr(apic, 0);
2315 kvm_lapic_set_reg(apic, APIC_ESR, 0);
2316 kvm_lapic_set_reg(apic, APIC_ICR, 0);
2317 kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2318 kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2319 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2320 for (i = 0; i < 8; i++) {
2321 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2322 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2323 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2325 kvm_apic_update_apicv(vcpu);
2326 apic->highest_isr_cache = -1;
2327 update_divide_count(apic);
2328 atomic_set(&apic->lapic_timer.pending, 0);
2329 if (kvm_vcpu_is_bsp(vcpu))
2330 kvm_lapic_set_base(vcpu,
2331 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
2332 vcpu->arch.pv_eoi.msr_val = 0;
2333 apic_update_ppr(apic);
2334 if (vcpu->arch.apicv_active) {
2335 kvm_x86_ops.apicv_post_state_restore(vcpu);
2336 kvm_x86_ops.hwapic_irr_update(vcpu, -1);
2337 kvm_x86_ops.hwapic_isr_update(vcpu, -1);
2340 vcpu->arch.apic_arb_prio = 0;
2341 vcpu->arch.apic_attention = 0;
2343 kvm_recalculate_apic_map(vcpu->kvm);
2347 *----------------------------------------------------------------------
2349 *----------------------------------------------------------------------
2352 static bool lapic_is_periodic(struct kvm_lapic *apic)
2354 return apic_lvtt_period(apic);
2357 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2359 struct kvm_lapic *apic = vcpu->arch.apic;
2361 if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2362 return atomic_read(&apic->lapic_timer.pending);
2367 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2369 u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2370 int vector, mode, trig_mode;
2372 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2373 vector = reg & APIC_VECTOR_MASK;
2374 mode = reg & APIC_MODE_MASK;
2375 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2376 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2382 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2384 struct kvm_lapic *apic = vcpu->arch.apic;
2387 kvm_apic_local_deliver(apic, APIC_LVT0);
2390 static const struct kvm_io_device_ops apic_mmio_ops = {
2391 .read = apic_mmio_read,
2392 .write = apic_mmio_write,
2395 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2397 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2398 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2400 apic_timer_expired(apic, true);
2402 if (lapic_is_periodic(apic)) {
2403 advance_periodic_target_expiration(apic);
2404 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2405 return HRTIMER_RESTART;
2407 return HRTIMER_NORESTART;
2410 int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
2412 struct kvm_lapic *apic;
2414 ASSERT(vcpu != NULL);
2416 apic = kzalloc(sizeof(*apic), GFP_KERNEL_ACCOUNT);
2420 vcpu->arch.apic = apic;
2422 apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
2424 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2426 goto nomem_free_apic;
2430 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2431 HRTIMER_MODE_ABS_HARD);
2432 apic->lapic_timer.timer.function = apic_timer_fn;
2433 if (timer_advance_ns == -1) {
2434 apic->lapic_timer.timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
2435 lapic_timer_advance_dynamic = true;
2437 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
2438 lapic_timer_advance_dynamic = false;
2442 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2443 * thinking that APIC state has changed.
2445 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2446 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2447 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2452 vcpu->arch.apic = NULL;
2457 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2459 struct kvm_lapic *apic = vcpu->arch.apic;
2462 if (!kvm_apic_hw_enabled(apic))
2465 __apic_update_ppr(apic, &ppr);
2466 return apic_has_interrupt_for_ppr(apic, ppr);
2469 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2471 u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2473 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2475 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2476 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2481 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2483 struct kvm_lapic *apic = vcpu->arch.apic;
2485 if (atomic_read(&apic->lapic_timer.pending) > 0) {
2486 kvm_apic_inject_pending_timer_irqs(apic);
2487 atomic_set(&apic->lapic_timer.pending, 0);
2491 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2493 int vector = kvm_apic_has_interrupt(vcpu);
2494 struct kvm_lapic *apic = vcpu->arch.apic;
2501 * We get here even with APIC virtualization enabled, if doing
2502 * nested virtualization and L1 runs with the "acknowledge interrupt
2503 * on exit" mode. Then we cannot inject the interrupt via RVI,
2504 * because the process would deliver it through the IDT.
2507 apic_clear_irr(vector, apic);
2508 if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2510 * For auto-EOI interrupts, there might be another pending
2511 * interrupt above PPR, so check whether to raise another
2514 apic_update_ppr(apic);
2517 * For normal interrupts, PPR has been raised and there cannot
2518 * be a higher-priority pending interrupt---except if there was
2519 * a concurrent interrupt injection, but that would have
2520 * triggered KVM_REQ_EVENT already.
2522 apic_set_isr(vector, apic);
2523 __apic_update_ppr(apic, &ppr);
2529 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2530 struct kvm_lapic_state *s, bool set)
2532 if (apic_x2apic_mode(vcpu->arch.apic)) {
2533 u32 *id = (u32 *)(s->regs + APIC_ID);
2534 u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2536 if (vcpu->kvm->arch.x2apic_format) {
2537 if (*id != vcpu->vcpu_id)
2546 /* In x2APIC mode, the LDR is fixed and based on the id */
2548 *ldr = kvm_apic_calc_x2apic_ldr(*id);
2554 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2556 memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2559 * Get calculated timer current count for remaining timer period (if
2560 * any) and store it in the returned register set.
2562 __kvm_lapic_set_reg(s->regs, APIC_TMCCT,
2563 __apic_read(vcpu->arch.apic, APIC_TMCCT));
2565 return kvm_apic_state_fixup(vcpu, s, false);
2568 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2570 struct kvm_lapic *apic = vcpu->arch.apic;
2573 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2574 /* set SPIV separately to get count of SW disabled APICs right */
2575 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2577 r = kvm_apic_state_fixup(vcpu, s, true);
2579 kvm_recalculate_apic_map(vcpu->kvm);
2582 memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
2584 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2585 kvm_recalculate_apic_map(vcpu->kvm);
2586 kvm_apic_set_version(vcpu);
2588 apic_update_ppr(apic);
2589 hrtimer_cancel(&apic->lapic_timer.timer);
2590 apic_update_lvtt(apic);
2591 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2592 update_divide_count(apic);
2593 __start_apic_timer(apic, APIC_TMCCT);
2594 kvm_apic_update_apicv(vcpu);
2595 apic->highest_isr_cache = -1;
2596 if (vcpu->arch.apicv_active) {
2597 kvm_x86_ops.apicv_post_state_restore(vcpu);
2598 kvm_x86_ops.hwapic_irr_update(vcpu,
2599 apic_find_highest_irr(apic));
2600 kvm_x86_ops.hwapic_isr_update(vcpu,
2601 apic_find_highest_isr(apic));
2603 kvm_make_request(KVM_REQ_EVENT, vcpu);
2604 if (ioapic_in_kernel(vcpu->kvm))
2605 kvm_rtc_eoi_tracking_restore_one(vcpu);
2607 vcpu->arch.apic_arb_prio = 0;
2612 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2614 struct hrtimer *timer;
2616 if (!lapic_in_kernel(vcpu) ||
2617 kvm_can_post_timer_interrupt(vcpu))
2620 timer = &vcpu->arch.apic->lapic_timer.timer;
2621 if (hrtimer_cancel(timer))
2622 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_HARD);
2626 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2628 * Detect whether guest triggered PV EOI since the
2629 * last entry. If yes, set EOI on guests's behalf.
2630 * Clear PV EOI in guest memory in any case.
2632 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2633 struct kvm_lapic *apic)
2638 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2639 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2641 * KVM_APIC_PV_EOI_PENDING is unset:
2642 * -> host disabled PV EOI.
2643 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2644 * -> host enabled PV EOI, guest did not execute EOI yet.
2645 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2646 * -> host enabled PV EOI, guest executed EOI.
2648 BUG_ON(!pv_eoi_enabled(vcpu));
2649 pending = pv_eoi_get_pending(vcpu);
2651 * Clear pending bit in any case: it will be set again on vmentry.
2652 * While this might not be ideal from performance point of view,
2653 * this makes sure pv eoi is only enabled when we know it's safe.
2655 pv_eoi_clr_pending(vcpu);
2658 vector = apic_set_eoi(apic);
2659 trace_kvm_pv_eoi(apic, vector);
2662 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2666 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2667 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2669 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2672 if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2676 apic_set_tpr(vcpu->arch.apic, data & 0xff);
2680 * apic_sync_pv_eoi_to_guest - called before vmentry
2682 * Detect whether it's safe to enable PV EOI and
2685 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2686 struct kvm_lapic *apic)
2688 if (!pv_eoi_enabled(vcpu) ||
2689 /* IRR set or many bits in ISR: could be nested. */
2690 apic->irr_pending ||
2691 /* Cache not set: could be safe but we don't bother. */
2692 apic->highest_isr_cache == -1 ||
2693 /* Need EOI to update ioapic. */
2694 kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2696 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2697 * so we need not do anything here.
2702 pv_eoi_set_pending(apic->vcpu);
2705 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2708 int max_irr, max_isr;
2709 struct kvm_lapic *apic = vcpu->arch.apic;
2711 apic_sync_pv_eoi_to_guest(vcpu, apic);
2713 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2716 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2717 max_irr = apic_find_highest_irr(apic);
2720 max_isr = apic_find_highest_isr(apic);
2723 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2725 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2729 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2732 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2733 &vcpu->arch.apic->vapic_cache,
2734 vapic_addr, sizeof(u32)))
2736 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2738 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2741 vcpu->arch.apic->vapic_addr = vapic_addr;
2745 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2747 struct kvm_lapic *apic = vcpu->arch.apic;
2748 u32 reg = (msr - APIC_BASE_MSR) << 4;
2750 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2753 if (reg == APIC_ICR2)
2756 /* if this is ICR write vector before command */
2757 if (reg == APIC_ICR)
2758 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2759 return kvm_lapic_reg_write(apic, reg, (u32)data);
2762 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2764 struct kvm_lapic *apic = vcpu->arch.apic;
2765 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2767 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2770 if (reg == APIC_DFR || reg == APIC_ICR2)
2773 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2775 if (reg == APIC_ICR)
2776 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2778 *data = (((u64)high) << 32) | low;
2783 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2785 struct kvm_lapic *apic = vcpu->arch.apic;
2787 if (!lapic_in_kernel(vcpu))
2790 /* if this is ICR write vector before command */
2791 if (reg == APIC_ICR)
2792 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2793 return kvm_lapic_reg_write(apic, reg, (u32)data);
2796 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2798 struct kvm_lapic *apic = vcpu->arch.apic;
2801 if (!lapic_in_kernel(vcpu))
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_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data, unsigned long len)
2816 u64 addr = data & ~KVM_MSR_ENABLED;
2817 struct gfn_to_hva_cache *ghc = &vcpu->arch.pv_eoi.data;
2818 unsigned long new_len;
2820 if (!IS_ALIGNED(addr, 4))
2823 vcpu->arch.pv_eoi.msr_val = data;
2824 if (!pv_eoi_enabled(vcpu))
2827 if (addr == ghc->gpa && len <= ghc->len)
2832 return kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, addr, new_len);
2835 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2837 struct kvm_lapic *apic = vcpu->arch.apic;
2841 if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2845 * INITs are latched while CPU is in specific states
2846 * (SMM, VMX non-root mode, SVM with GIF=0).
2847 * Because a CPU cannot be in these states immediately
2848 * after it has processed an INIT signal (and thus in
2849 * KVM_MP_STATE_INIT_RECEIVED state), just eat SIPIs
2850 * and leave the INIT pending.
2852 if (kvm_vcpu_latch_init(vcpu)) {
2853 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2854 if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2855 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2859 pe = xchg(&apic->pending_events, 0);
2860 if (test_bit(KVM_APIC_INIT, &pe)) {
2861 kvm_vcpu_reset(vcpu, true);
2862 if (kvm_vcpu_is_bsp(apic->vcpu))
2863 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2865 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2867 if (test_bit(KVM_APIC_SIPI, &pe) &&
2868 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2869 /* evaluate pending_events before reading the vector */
2871 sipi_vector = apic->sipi_vector;
2872 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2873 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2877 void kvm_lapic_init(void)
2879 /* do not patch jump label more than once per second */
2880 jump_label_rate_limit(&apic_hw_disabled, HZ);
2881 jump_label_rate_limit(&apic_sw_disabled, HZ);
2884 void kvm_lapic_exit(void)
2886 static_key_deferred_flush(&apic_hw_disabled);
2887 static_key_deferred_flush(&apic_sw_disabled);