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>
33 #include <asm/current.h>
34 #include <asm/apicdef.h>
35 #include <asm/delay.h>
36 #include <linux/atomic.h>
37 #include <linux/jump_label.h>
38 #include "kvm_cache_regs.h"
47 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
49 #define mod_64(x, y) ((x) % (y))
57 /* 14 is the version for Xeon and Pentium 8.4.8*/
58 #define APIC_VERSION 0x14UL
59 #define LAPIC_MMIO_LENGTH (1 << 12)
60 /* followed define is not in apicdef.h */
61 #define MAX_APIC_VECTOR 256
62 #define APIC_VECTORS_PER_REG 32
64 static bool lapic_timer_advance_dynamic __read_mostly;
65 #define LAPIC_TIMER_ADVANCE_ADJUST_MIN 100 /* clock cycles */
66 #define LAPIC_TIMER_ADVANCE_ADJUST_MAX 10000 /* clock cycles */
67 #define LAPIC_TIMER_ADVANCE_NS_INIT 1000
68 #define LAPIC_TIMER_ADVANCE_NS_MAX 5000
69 /* step-by-step approximation to mitigate fluctuation */
70 #define LAPIC_TIMER_ADVANCE_ADJUST_STEP 8
71 static int kvm_lapic_msr_read(struct kvm_lapic *apic, u32 reg, u64 *data);
72 static int kvm_lapic_msr_write(struct kvm_lapic *apic, u32 reg, u64 data);
74 static inline void __kvm_lapic_set_reg(char *regs, int reg_off, u32 val)
76 *((u32 *) (regs + reg_off)) = val;
79 static inline void kvm_lapic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
81 __kvm_lapic_set_reg(apic->regs, reg_off, val);
84 static __always_inline u64 __kvm_lapic_get_reg64(char *regs, int reg)
86 BUILD_BUG_ON(reg != APIC_ICR);
87 return *((u64 *) (regs + reg));
90 static __always_inline u64 kvm_lapic_get_reg64(struct kvm_lapic *apic, int reg)
92 return __kvm_lapic_get_reg64(apic->regs, reg);
95 static __always_inline void __kvm_lapic_set_reg64(char *regs, int reg, u64 val)
97 BUILD_BUG_ON(reg != APIC_ICR);
98 *((u64 *) (regs + reg)) = val;
101 static __always_inline void kvm_lapic_set_reg64(struct kvm_lapic *apic,
104 __kvm_lapic_set_reg64(apic->regs, reg, val);
107 static inline int apic_test_vector(int vec, void *bitmap)
109 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
112 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
114 struct kvm_lapic *apic = vcpu->arch.apic;
116 return apic_test_vector(vector, apic->regs + APIC_ISR) ||
117 apic_test_vector(vector, apic->regs + APIC_IRR);
120 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
122 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
125 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
127 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
130 __read_mostly DEFINE_STATIC_KEY_DEFERRED_FALSE(apic_hw_disabled, HZ);
131 __read_mostly DEFINE_STATIC_KEY_DEFERRED_FALSE(apic_sw_disabled, HZ);
133 static inline int apic_enabled(struct kvm_lapic *apic)
135 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
139 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
142 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
143 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
145 static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
147 return apic->vcpu->vcpu_id;
150 static bool kvm_can_post_timer_interrupt(struct kvm_vcpu *vcpu)
152 return pi_inject_timer && kvm_vcpu_apicv_active(vcpu) &&
153 (kvm_mwait_in_guest(vcpu->kvm) || kvm_hlt_in_guest(vcpu->kvm));
156 bool kvm_can_use_hv_timer(struct kvm_vcpu *vcpu)
158 return kvm_x86_ops.set_hv_timer
159 && !(kvm_mwait_in_guest(vcpu->kvm) ||
160 kvm_can_post_timer_interrupt(vcpu));
162 EXPORT_SYMBOL_GPL(kvm_can_use_hv_timer);
164 static bool kvm_use_posted_timer_interrupt(struct kvm_vcpu *vcpu)
166 return kvm_can_post_timer_interrupt(vcpu) && vcpu->mode == IN_GUEST_MODE;
169 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
170 u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
172 case KVM_APIC_MODE_X2APIC: {
173 u32 offset = (dest_id >> 16) * 16;
174 u32 max_apic_id = map->max_apic_id;
176 if (offset <= max_apic_id) {
177 u8 cluster_size = min(max_apic_id - offset + 1, 16U);
179 offset = array_index_nospec(offset, map->max_apic_id + 1);
180 *cluster = &map->phys_map[offset];
181 *mask = dest_id & (0xffff >> (16 - cluster_size));
188 case KVM_APIC_MODE_XAPIC_FLAT:
189 *cluster = map->xapic_flat_map;
190 *mask = dest_id & 0xff;
192 case KVM_APIC_MODE_XAPIC_CLUSTER:
193 *cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
194 *mask = dest_id & 0xf;
202 static void kvm_apic_map_free(struct rcu_head *rcu)
204 struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
210 * CLEAN -> DIRTY and UPDATE_IN_PROGRESS -> DIRTY changes happen without a lock.
212 * DIRTY -> UPDATE_IN_PROGRESS and UPDATE_IN_PROGRESS -> CLEAN happen with
213 * apic_map_lock_held.
221 void kvm_recalculate_apic_map(struct kvm *kvm)
223 struct kvm_apic_map *new, *old = NULL;
224 struct kvm_vcpu *vcpu;
226 u32 max_id = 255; /* enough space for any xAPIC ID */
228 /* Read kvm->arch.apic_map_dirty before kvm->arch.apic_map. */
229 if (atomic_read_acquire(&kvm->arch.apic_map_dirty) == CLEAN)
232 WARN_ONCE(!irqchip_in_kernel(kvm),
233 "Dirty APIC map without an in-kernel local APIC");
235 mutex_lock(&kvm->arch.apic_map_lock);
237 * Read kvm->arch.apic_map_dirty before kvm->arch.apic_map
238 * (if clean) or the APIC registers (if dirty).
240 if (atomic_cmpxchg_acquire(&kvm->arch.apic_map_dirty,
241 DIRTY, UPDATE_IN_PROGRESS) == CLEAN) {
242 /* Someone else has updated the map. */
243 mutex_unlock(&kvm->arch.apic_map_lock);
247 kvm_for_each_vcpu(i, vcpu, kvm)
248 if (kvm_apic_present(vcpu))
249 max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
251 new = kvzalloc(sizeof(struct kvm_apic_map) +
252 sizeof(struct kvm_lapic *) * ((u64)max_id + 1),
258 new->max_apic_id = max_id;
260 kvm_for_each_vcpu(i, vcpu, kvm) {
261 struct kvm_lapic *apic = vcpu->arch.apic;
262 struct kvm_lapic **cluster;
268 if (!kvm_apic_present(vcpu))
271 xapic_id = kvm_xapic_id(apic);
272 x2apic_id = kvm_x2apic_id(apic);
274 /* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
275 if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
276 x2apic_id <= new->max_apic_id)
277 new->phys_map[x2apic_id] = apic;
279 * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
280 * prevent them from masking VCPUs with APIC ID <= 0xff.
282 if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
283 new->phys_map[xapic_id] = apic;
285 if (!kvm_apic_sw_enabled(apic))
288 ldr = kvm_lapic_get_reg(apic, APIC_LDR);
290 if (apic_x2apic_mode(apic)) {
291 new->mode |= KVM_APIC_MODE_X2APIC;
293 ldr = GET_APIC_LOGICAL_ID(ldr);
294 if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
295 new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
297 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
300 if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
304 cluster[ffs(mask) - 1] = apic;
307 old = rcu_dereference_protected(kvm->arch.apic_map,
308 lockdep_is_held(&kvm->arch.apic_map_lock));
309 rcu_assign_pointer(kvm->arch.apic_map, new);
311 * Write kvm->arch.apic_map before clearing apic->apic_map_dirty.
312 * If another update has come in, leave it DIRTY.
314 atomic_cmpxchg_release(&kvm->arch.apic_map_dirty,
315 UPDATE_IN_PROGRESS, CLEAN);
316 mutex_unlock(&kvm->arch.apic_map_lock);
319 call_rcu(&old->rcu, kvm_apic_map_free);
321 kvm_make_scan_ioapic_request(kvm);
324 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
326 bool enabled = val & APIC_SPIV_APIC_ENABLED;
328 kvm_lapic_set_reg(apic, APIC_SPIV, val);
330 if (enabled != apic->sw_enabled) {
331 apic->sw_enabled = enabled;
333 static_branch_slow_dec_deferred(&apic_sw_disabled);
335 static_branch_inc(&apic_sw_disabled.key);
337 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
340 /* Check if there are APF page ready requests pending */
342 kvm_make_request(KVM_REQ_APF_READY, apic->vcpu);
345 static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
347 kvm_lapic_set_reg(apic, APIC_ID, id << 24);
348 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
351 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
353 kvm_lapic_set_reg(apic, APIC_LDR, id);
354 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
357 static inline void kvm_apic_set_dfr(struct kvm_lapic *apic, u32 val)
359 kvm_lapic_set_reg(apic, APIC_DFR, val);
360 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
363 static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
365 return ((id >> 4) << 16) | (1 << (id & 0xf));
368 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
370 u32 ldr = kvm_apic_calc_x2apic_ldr(id);
372 WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
374 kvm_lapic_set_reg(apic, APIC_ID, id);
375 kvm_lapic_set_reg(apic, APIC_LDR, ldr);
376 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
379 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
381 return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
384 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
386 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
389 static inline int apic_lvtt_period(struct kvm_lapic *apic)
391 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
394 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
396 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
399 static inline int apic_lvt_nmi_mode(u32 lvt_val)
401 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
404 static inline bool kvm_lapic_lvt_supported(struct kvm_lapic *apic, int lvt_index)
406 return apic->nr_lvt_entries > lvt_index;
409 static inline int kvm_apic_calc_nr_lvt_entries(struct kvm_vcpu *vcpu)
411 return KVM_APIC_MAX_NR_LVT_ENTRIES - !(vcpu->arch.mcg_cap & MCG_CMCI_P);
414 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
416 struct kvm_lapic *apic = vcpu->arch.apic;
419 if (!lapic_in_kernel(vcpu))
422 v = APIC_VERSION | ((apic->nr_lvt_entries - 1) << 16);
425 * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
426 * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
427 * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
428 * version first and level-triggered interrupts never get EOIed in
431 if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) &&
432 !ioapic_in_kernel(vcpu->kvm))
433 v |= APIC_LVR_DIRECTED_EOI;
434 kvm_lapic_set_reg(apic, APIC_LVR, v);
437 void kvm_apic_after_set_mcg_cap(struct kvm_vcpu *vcpu)
439 int nr_lvt_entries = kvm_apic_calc_nr_lvt_entries(vcpu);
440 struct kvm_lapic *apic = vcpu->arch.apic;
443 if (!lapic_in_kernel(vcpu) || nr_lvt_entries == apic->nr_lvt_entries)
446 /* Initialize/mask any "new" LVT entries. */
447 for (i = apic->nr_lvt_entries; i < nr_lvt_entries; i++)
448 kvm_lapic_set_reg(apic, APIC_LVTx(i), APIC_LVT_MASKED);
450 apic->nr_lvt_entries = nr_lvt_entries;
452 /* The number of LVT entries is reflected in the version register. */
453 kvm_apic_set_version(vcpu);
456 static const unsigned int apic_lvt_mask[KVM_APIC_MAX_NR_LVT_ENTRIES] = {
457 [LVT_TIMER] = LVT_MASK, /* timer mode mask added at runtime */
458 [LVT_THERMAL_MONITOR] = LVT_MASK | APIC_MODE_MASK,
459 [LVT_PERFORMANCE_COUNTER] = LVT_MASK | APIC_MODE_MASK,
460 [LVT_LINT0] = LINT_MASK,
461 [LVT_LINT1] = LINT_MASK,
462 [LVT_ERROR] = LVT_MASK,
463 [LVT_CMCI] = LVT_MASK | APIC_MODE_MASK
466 static int find_highest_vector(void *bitmap)
471 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
472 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
473 reg = bitmap + REG_POS(vec);
475 return __fls(*reg) + vec;
481 static u8 count_vectors(void *bitmap)
487 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
488 reg = bitmap + REG_POS(vec);
489 count += hweight32(*reg);
495 bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
498 u32 pir_val, irr_val, prev_irr_val;
501 max_updated_irr = -1;
504 for (i = vec = 0; i <= 7; i++, vec += 32) {
505 pir_val = READ_ONCE(pir[i]);
506 irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
508 prev_irr_val = irr_val;
509 irr_val |= xchg(&pir[i], 0);
510 *((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
511 if (prev_irr_val != irr_val) {
513 __fls(irr_val ^ prev_irr_val) + vec;
517 *max_irr = __fls(irr_val) + vec;
520 return ((max_updated_irr != -1) &&
521 (max_updated_irr == *max_irr));
523 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
525 bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
527 struct kvm_lapic *apic = vcpu->arch.apic;
529 return __kvm_apic_update_irr(pir, apic->regs, max_irr);
531 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
533 static inline int apic_search_irr(struct kvm_lapic *apic)
535 return find_highest_vector(apic->regs + APIC_IRR);
538 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
543 * Note that irr_pending is just a hint. It will be always
544 * true with virtual interrupt delivery enabled.
546 if (!apic->irr_pending)
549 result = apic_search_irr(apic);
550 ASSERT(result == -1 || result >= 16);
555 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
557 if (unlikely(apic->apicv_active)) {
558 /* need to update RVI */
559 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
560 static_call_cond(kvm_x86_hwapic_irr_update)(apic->vcpu,
561 apic_find_highest_irr(apic));
563 apic->irr_pending = false;
564 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
565 if (apic_search_irr(apic) != -1)
566 apic->irr_pending = true;
570 void kvm_apic_clear_irr(struct kvm_vcpu *vcpu, int vec)
572 apic_clear_irr(vec, vcpu->arch.apic);
574 EXPORT_SYMBOL_GPL(kvm_apic_clear_irr);
576 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
578 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
582 * With APIC virtualization enabled, all caching is disabled
583 * because the processor can modify ISR under the hood. Instead
586 if (unlikely(apic->apicv_active))
587 static_call_cond(kvm_x86_hwapic_isr_update)(vec);
590 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
592 * ISR (in service register) bit is set when injecting an interrupt.
593 * The highest vector is injected. Thus the latest bit set matches
594 * the highest bit in ISR.
596 apic->highest_isr_cache = vec;
600 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
605 * Note that isr_count is always 1, and highest_isr_cache
606 * is always -1, with APIC virtualization enabled.
608 if (!apic->isr_count)
610 if (likely(apic->highest_isr_cache != -1))
611 return apic->highest_isr_cache;
613 result = find_highest_vector(apic->regs + APIC_ISR);
614 ASSERT(result == -1 || result >= 16);
619 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
621 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
625 * We do get here for APIC virtualization enabled if the guest
626 * uses the Hyper-V APIC enlightenment. In this case we may need
627 * to trigger a new interrupt delivery by writing the SVI field;
628 * on the other hand isr_count and highest_isr_cache are unused
629 * and must be left alone.
631 if (unlikely(apic->apicv_active))
632 static_call_cond(kvm_x86_hwapic_isr_update)(apic_find_highest_isr(apic));
635 BUG_ON(apic->isr_count < 0);
636 apic->highest_isr_cache = -1;
640 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
642 /* This may race with setting of irr in __apic_accept_irq() and
643 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
644 * will cause vmexit immediately and the value will be recalculated
645 * on the next vmentry.
647 return apic_find_highest_irr(vcpu->arch.apic);
649 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
651 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
652 int vector, int level, int trig_mode,
653 struct dest_map *dest_map);
655 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
656 struct dest_map *dest_map)
658 struct kvm_lapic *apic = vcpu->arch.apic;
660 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
661 irq->level, irq->trig_mode, dest_map);
664 static int __pv_send_ipi(unsigned long *ipi_bitmap, struct kvm_apic_map *map,
665 struct kvm_lapic_irq *irq, u32 min)
668 struct kvm_vcpu *vcpu;
670 if (min > map->max_apic_id)
673 for_each_set_bit(i, ipi_bitmap,
674 min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
675 if (map->phys_map[min + i]) {
676 vcpu = map->phys_map[min + i]->vcpu;
677 count += kvm_apic_set_irq(vcpu, irq, NULL);
684 int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
685 unsigned long ipi_bitmap_high, u32 min,
686 unsigned long icr, int op_64_bit)
688 struct kvm_apic_map *map;
689 struct kvm_lapic_irq irq = {0};
690 int cluster_size = op_64_bit ? 64 : 32;
693 if (icr & (APIC_DEST_MASK | APIC_SHORT_MASK))
696 irq.vector = icr & APIC_VECTOR_MASK;
697 irq.delivery_mode = icr & APIC_MODE_MASK;
698 irq.level = (icr & APIC_INT_ASSERT) != 0;
699 irq.trig_mode = icr & APIC_INT_LEVELTRIG;
702 map = rcu_dereference(kvm->arch.apic_map);
706 count = __pv_send_ipi(&ipi_bitmap_low, map, &irq, min);
708 count += __pv_send_ipi(&ipi_bitmap_high, map, &irq, min);
715 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
718 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
722 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
725 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
729 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
731 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
734 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
736 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0)
739 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
742 static bool pv_eoi_test_and_clr_pending(struct kvm_vcpu *vcpu)
746 if (pv_eoi_get_user(vcpu, &val) < 0)
749 val &= KVM_PV_EOI_ENABLED;
751 if (val && pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0)
755 * Clear pending bit in any case: it will be set again on vmentry.
756 * While this might not be ideal from performance point of view,
757 * this makes sure pv eoi is only enabled when we know it's safe.
759 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
764 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
767 if (kvm_x86_ops.sync_pir_to_irr)
768 highest_irr = static_call(kvm_x86_sync_pir_to_irr)(apic->vcpu);
770 highest_irr = apic_find_highest_irr(apic);
771 if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
776 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
778 u32 tpr, isrv, ppr, old_ppr;
781 old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
782 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
783 isr = apic_find_highest_isr(apic);
784 isrv = (isr != -1) ? isr : 0;
786 if ((tpr & 0xf0) >= (isrv & 0xf0))
793 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
795 return ppr < old_ppr;
798 static void apic_update_ppr(struct kvm_lapic *apic)
802 if (__apic_update_ppr(apic, &ppr) &&
803 apic_has_interrupt_for_ppr(apic, ppr) != -1)
804 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
807 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
809 apic_update_ppr(vcpu->arch.apic);
811 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
813 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
815 kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
816 apic_update_ppr(apic);
819 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
821 return mda == (apic_x2apic_mode(apic) ?
822 X2APIC_BROADCAST : APIC_BROADCAST);
825 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
827 if (kvm_apic_broadcast(apic, mda))
831 * Hotplug hack: Accept interrupts for vCPUs in xAPIC mode as if they
832 * were in x2APIC mode if the target APIC ID can't be encoded as an
833 * xAPIC ID. This allows unique addressing of hotplugged vCPUs (which
834 * start in xAPIC mode) with an APIC ID that is unaddressable in xAPIC
835 * mode. Match the x2APIC ID if and only if the target APIC ID can't
836 * be encoded in xAPIC to avoid spurious matches against a vCPU that
837 * changed its (addressable) xAPIC ID (which is writable).
839 if (apic_x2apic_mode(apic) || mda > 0xff)
840 return mda == kvm_x2apic_id(apic);
842 return mda == kvm_xapic_id(apic);
845 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
849 if (kvm_apic_broadcast(apic, mda))
852 logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
854 if (apic_x2apic_mode(apic))
855 return ((logical_id >> 16) == (mda >> 16))
856 && (logical_id & mda & 0xffff) != 0;
858 logical_id = GET_APIC_LOGICAL_ID(logical_id);
860 switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
862 return (logical_id & mda) != 0;
863 case APIC_DFR_CLUSTER:
864 return ((logical_id >> 4) == (mda >> 4))
865 && (logical_id & mda & 0xf) != 0;
871 /* The KVM local APIC implementation has two quirks:
873 * - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
874 * in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
875 * KVM doesn't do that aliasing.
877 * - in-kernel IOAPIC messages have to be delivered directly to
878 * x2APIC, because the kernel does not support interrupt remapping.
879 * In order to support broadcast without interrupt remapping, x2APIC
880 * rewrites the destination of non-IPI messages from APIC_BROADCAST
881 * to X2APIC_BROADCAST.
883 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API. This is
884 * important when userspace wants to use x2APIC-format MSIs, because
885 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
887 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
888 struct kvm_lapic *source, struct kvm_lapic *target)
890 bool ipi = source != NULL;
892 if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
893 !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
894 return X2APIC_BROADCAST;
899 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
900 int shorthand, unsigned int dest, int dest_mode)
902 struct kvm_lapic *target = vcpu->arch.apic;
903 u32 mda = kvm_apic_mda(vcpu, dest, source, target);
907 case APIC_DEST_NOSHORT:
908 if (dest_mode == APIC_DEST_PHYSICAL)
909 return kvm_apic_match_physical_addr(target, mda);
911 return kvm_apic_match_logical_addr(target, mda);
913 return target == source;
914 case APIC_DEST_ALLINC:
916 case APIC_DEST_ALLBUT:
917 return target != source;
922 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
924 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
925 const unsigned long *bitmap, u32 bitmap_size)
930 mod = vector % dest_vcpus;
932 for (i = 0; i <= mod; i++) {
933 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
934 BUG_ON(idx == bitmap_size);
940 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
942 if (!kvm->arch.disabled_lapic_found) {
943 kvm->arch.disabled_lapic_found = true;
945 "Disabled LAPIC found during irq injection\n");
949 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
950 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
952 if (kvm->arch.x2apic_broadcast_quirk_disabled) {
953 if ((irq->dest_id == APIC_BROADCAST &&
954 map->mode != KVM_APIC_MODE_X2APIC))
956 if (irq->dest_id == X2APIC_BROADCAST)
959 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
960 if (irq->dest_id == (x2apic_ipi ?
961 X2APIC_BROADCAST : APIC_BROADCAST))
968 /* Return true if the interrupt can be handled by using *bitmap as index mask
969 * for valid destinations in *dst array.
970 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
971 * Note: we may have zero kvm_lapic destinations when we return true, which
972 * means that the interrupt should be dropped. In this case, *bitmap would be
973 * zero and *dst undefined.
975 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
976 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
977 struct kvm_apic_map *map, struct kvm_lapic ***dst,
978 unsigned long *bitmap)
982 if (irq->shorthand == APIC_DEST_SELF && src) {
986 } else if (irq->shorthand)
989 if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
992 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
993 if (irq->dest_id > map->max_apic_id) {
996 u32 dest_id = array_index_nospec(irq->dest_id, map->max_apic_id + 1);
997 *dst = &map->phys_map[dest_id];
1004 if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
1008 if (!kvm_lowest_prio_delivery(irq))
1011 if (!kvm_vector_hashing_enabled()) {
1013 for_each_set_bit(i, bitmap, 16) {
1018 else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
1019 (*dst)[lowest]->vcpu) < 0)
1026 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
1029 if (!(*dst)[lowest]) {
1030 kvm_apic_disabled_lapic_found(kvm);
1036 *bitmap = (lowest >= 0) ? 1 << lowest : 0;
1041 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
1042 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
1044 struct kvm_apic_map *map;
1045 unsigned long bitmap;
1046 struct kvm_lapic **dst = NULL;
1052 if (irq->shorthand == APIC_DEST_SELF) {
1053 if (KVM_BUG_ON(!src, kvm)) {
1057 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
1062 map = rcu_dereference(kvm->arch.apic_map);
1064 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
1067 for_each_set_bit(i, &bitmap, 16) {
1070 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
1079 * This routine tries to handle interrupts in posted mode, here is how
1080 * it deals with different cases:
1081 * - For single-destination interrupts, handle it in posted mode
1082 * - Else if vector hashing is enabled and it is a lowest-priority
1083 * interrupt, handle it in posted mode and use the following mechanism
1084 * to find the destination vCPU.
1085 * 1. For lowest-priority interrupts, store all the possible
1086 * destination vCPUs in an array.
1087 * 2. Use "guest vector % max number of destination vCPUs" to find
1088 * the right destination vCPU in the array for the lowest-priority
1090 * - Otherwise, use remapped mode to inject the interrupt.
1092 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
1093 struct kvm_vcpu **dest_vcpu)
1095 struct kvm_apic_map *map;
1096 unsigned long bitmap;
1097 struct kvm_lapic **dst = NULL;
1104 map = rcu_dereference(kvm->arch.apic_map);
1106 if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
1107 hweight16(bitmap) == 1) {
1108 unsigned long i = find_first_bit(&bitmap, 16);
1111 *dest_vcpu = dst[i]->vcpu;
1121 * Add a pending IRQ into lapic.
1122 * Return 1 if successfully added and 0 if discarded.
1124 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
1125 int vector, int level, int trig_mode,
1126 struct dest_map *dest_map)
1129 struct kvm_vcpu *vcpu = apic->vcpu;
1131 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
1133 switch (delivery_mode) {
1134 case APIC_DM_LOWEST:
1135 vcpu->arch.apic_arb_prio++;
1138 if (unlikely(trig_mode && !level))
1141 /* FIXME add logic for vcpu on reset */
1142 if (unlikely(!apic_enabled(apic)))
1148 __set_bit(vcpu->vcpu_id, dest_map->map);
1149 dest_map->vectors[vcpu->vcpu_id] = vector;
1152 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
1154 kvm_lapic_set_vector(vector,
1155 apic->regs + APIC_TMR);
1157 kvm_lapic_clear_vector(vector,
1158 apic->regs + APIC_TMR);
1161 static_call(kvm_x86_deliver_interrupt)(apic, delivery_mode,
1167 vcpu->arch.pv.pv_unhalted = 1;
1168 kvm_make_request(KVM_REQ_EVENT, vcpu);
1169 kvm_vcpu_kick(vcpu);
1174 kvm_make_request(KVM_REQ_SMI, vcpu);
1175 kvm_vcpu_kick(vcpu);
1180 kvm_inject_nmi(vcpu);
1181 kvm_vcpu_kick(vcpu);
1185 if (!trig_mode || level) {
1187 /* assumes that there are only KVM_APIC_INIT/SIPI */
1188 apic->pending_events = (1UL << KVM_APIC_INIT);
1189 kvm_make_request(KVM_REQ_EVENT, vcpu);
1190 kvm_vcpu_kick(vcpu);
1194 case APIC_DM_STARTUP:
1196 apic->sipi_vector = vector;
1197 /* make sure sipi_vector is visible for the receiver */
1199 set_bit(KVM_APIC_SIPI, &apic->pending_events);
1200 kvm_make_request(KVM_REQ_EVENT, vcpu);
1201 kvm_vcpu_kick(vcpu);
1204 case APIC_DM_EXTINT:
1206 * Should only be called by kvm_apic_local_deliver() with LVT0,
1207 * before NMI watchdog was enabled. Already handled by
1208 * kvm_apic_accept_pic_intr().
1213 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1221 * This routine identifies the destination vcpus mask meant to receive the
1222 * IOAPIC interrupts. It either uses kvm_apic_map_get_dest_lapic() to find
1223 * out the destination vcpus array and set the bitmap or it traverses to
1224 * each available vcpu to identify the same.
1226 void kvm_bitmap_or_dest_vcpus(struct kvm *kvm, struct kvm_lapic_irq *irq,
1227 unsigned long *vcpu_bitmap)
1229 struct kvm_lapic **dest_vcpu = NULL;
1230 struct kvm_lapic *src = NULL;
1231 struct kvm_apic_map *map;
1232 struct kvm_vcpu *vcpu;
1233 unsigned long bitmap, i;
1238 map = rcu_dereference(kvm->arch.apic_map);
1240 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dest_vcpu,
1243 for_each_set_bit(i, &bitmap, 16) {
1246 vcpu_idx = dest_vcpu[i]->vcpu->vcpu_idx;
1247 __set_bit(vcpu_idx, vcpu_bitmap);
1250 kvm_for_each_vcpu(i, vcpu, kvm) {
1251 if (!kvm_apic_present(vcpu))
1253 if (!kvm_apic_match_dest(vcpu, NULL,
1258 __set_bit(i, vcpu_bitmap);
1264 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1266 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1269 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1271 return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1274 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1278 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1279 if (!kvm_ioapic_handles_vector(apic, vector))
1282 /* Request a KVM exit to inform the userspace IOAPIC. */
1283 if (irqchip_split(apic->vcpu->kvm)) {
1284 apic->vcpu->arch.pending_ioapic_eoi = vector;
1285 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1289 if (apic_test_vector(vector, apic->regs + APIC_TMR))
1290 trigger_mode = IOAPIC_LEVEL_TRIG;
1292 trigger_mode = IOAPIC_EDGE_TRIG;
1294 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1297 static int apic_set_eoi(struct kvm_lapic *apic)
1299 int vector = apic_find_highest_isr(apic);
1301 trace_kvm_eoi(apic, vector);
1304 * Not every write EOI will has corresponding ISR,
1305 * one example is when Kernel check timer on setup_IO_APIC
1310 apic_clear_isr(vector, apic);
1311 apic_update_ppr(apic);
1313 if (to_hv_vcpu(apic->vcpu) &&
1314 test_bit(vector, to_hv_synic(apic->vcpu)->vec_bitmap))
1315 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1317 kvm_ioapic_send_eoi(apic, vector);
1318 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1323 * this interface assumes a trap-like exit, which has already finished
1324 * desired side effect including vISR and vPPR update.
1326 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1328 struct kvm_lapic *apic = vcpu->arch.apic;
1330 trace_kvm_eoi(apic, vector);
1332 kvm_ioapic_send_eoi(apic, vector);
1333 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1335 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1337 void kvm_apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
1339 struct kvm_lapic_irq irq;
1341 /* KVM has no delay and should always clear the BUSY/PENDING flag. */
1342 WARN_ON_ONCE(icr_low & APIC_ICR_BUSY);
1344 irq.vector = icr_low & APIC_VECTOR_MASK;
1345 irq.delivery_mode = icr_low & APIC_MODE_MASK;
1346 irq.dest_mode = icr_low & APIC_DEST_MASK;
1347 irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1348 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1349 irq.shorthand = icr_low & APIC_SHORT_MASK;
1350 irq.msi_redir_hint = false;
1351 if (apic_x2apic_mode(apic))
1352 irq.dest_id = icr_high;
1354 irq.dest_id = GET_XAPIC_DEST_FIELD(icr_high);
1356 trace_kvm_apic_ipi(icr_low, irq.dest_id);
1358 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1360 EXPORT_SYMBOL_GPL(kvm_apic_send_ipi);
1362 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1364 ktime_t remaining, now;
1368 ASSERT(apic != NULL);
1370 /* if initial count is 0, current count should also be 0 */
1371 if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1372 apic->lapic_timer.period == 0)
1376 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1377 if (ktime_to_ns(remaining) < 0)
1380 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1381 tmcct = div64_u64(ns,
1382 (APIC_BUS_CYCLE_NS * apic->divide_count));
1387 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1389 struct kvm_vcpu *vcpu = apic->vcpu;
1390 struct kvm_run *run = vcpu->run;
1392 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1393 run->tpr_access.rip = kvm_rip_read(vcpu);
1394 run->tpr_access.is_write = write;
1397 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1399 if (apic->vcpu->arch.tpr_access_reporting)
1400 __report_tpr_access(apic, write);
1403 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1407 if (offset >= LAPIC_MMIO_LENGTH)
1414 case APIC_TMCCT: /* Timer CCR */
1415 if (apic_lvtt_tscdeadline(apic))
1418 val = apic_get_tmcct(apic);
1421 apic_update_ppr(apic);
1422 val = kvm_lapic_get_reg(apic, offset);
1425 report_tpr_access(apic, false);
1428 val = kvm_lapic_get_reg(apic, offset);
1435 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1437 return container_of(dev, struct kvm_lapic, dev);
1440 #define APIC_REG_MASK(reg) (1ull << ((reg) >> 4))
1441 #define APIC_REGS_MASK(first, count) \
1442 (APIC_REG_MASK(first) * ((1ull << (count)) - 1))
1444 static int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1447 unsigned char alignment = offset & 0xf;
1449 /* this bitmask has a bit cleared for each reserved register */
1450 u64 valid_reg_mask =
1451 APIC_REG_MASK(APIC_ID) |
1452 APIC_REG_MASK(APIC_LVR) |
1453 APIC_REG_MASK(APIC_TASKPRI) |
1454 APIC_REG_MASK(APIC_PROCPRI) |
1455 APIC_REG_MASK(APIC_LDR) |
1456 APIC_REG_MASK(APIC_DFR) |
1457 APIC_REG_MASK(APIC_SPIV) |
1458 APIC_REGS_MASK(APIC_ISR, APIC_ISR_NR) |
1459 APIC_REGS_MASK(APIC_TMR, APIC_ISR_NR) |
1460 APIC_REGS_MASK(APIC_IRR, APIC_ISR_NR) |
1461 APIC_REG_MASK(APIC_ESR) |
1462 APIC_REG_MASK(APIC_ICR) |
1463 APIC_REG_MASK(APIC_LVTT) |
1464 APIC_REG_MASK(APIC_LVTTHMR) |
1465 APIC_REG_MASK(APIC_LVTPC) |
1466 APIC_REG_MASK(APIC_LVT0) |
1467 APIC_REG_MASK(APIC_LVT1) |
1468 APIC_REG_MASK(APIC_LVTERR) |
1469 APIC_REG_MASK(APIC_TMICT) |
1470 APIC_REG_MASK(APIC_TMCCT) |
1471 APIC_REG_MASK(APIC_TDCR);
1473 if (kvm_lapic_lvt_supported(apic, LVT_CMCI))
1474 valid_reg_mask |= APIC_REG_MASK(APIC_LVTCMCI);
1477 * ARBPRI and ICR2 are not valid in x2APIC mode. WARN if KVM reads ICR
1478 * in x2APIC mode as it's an 8-byte register in x2APIC and needs to be
1479 * manually handled by the caller.
1481 if (!apic_x2apic_mode(apic))
1482 valid_reg_mask |= APIC_REG_MASK(APIC_ARBPRI) |
1483 APIC_REG_MASK(APIC_ICR2);
1485 WARN_ON_ONCE(offset == APIC_ICR);
1487 if (alignment + len > 4)
1490 if (offset > 0x3f0 || !(valid_reg_mask & APIC_REG_MASK(offset)))
1493 result = __apic_read(apic, offset & ~0xf);
1495 trace_kvm_apic_read(offset, result);
1501 memcpy(data, (char *)&result + alignment, len);
1504 printk(KERN_ERR "Local APIC read with len = %x, "
1505 "should be 1,2, or 4 instead\n", len);
1511 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1513 return addr >= apic->base_address &&
1514 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1517 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1518 gpa_t address, int len, void *data)
1520 struct kvm_lapic *apic = to_lapic(this);
1521 u32 offset = address - apic->base_address;
1523 if (!apic_mmio_in_range(apic, address))
1526 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1527 if (!kvm_check_has_quirk(vcpu->kvm,
1528 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1531 memset(data, 0xff, len);
1535 kvm_lapic_reg_read(apic, offset, len, data);
1540 static void update_divide_count(struct kvm_lapic *apic)
1542 u32 tmp1, tmp2, tdcr;
1544 tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1546 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1547 apic->divide_count = 0x1 << (tmp2 & 0x7);
1550 static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1553 * Do not allow the guest to program periodic timers with small
1554 * interval, since the hrtimers are not throttled by the host
1557 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1558 s64 min_period = min_timer_period_us * 1000LL;
1560 if (apic->lapic_timer.period < min_period) {
1561 pr_info_ratelimited(
1562 "kvm: vcpu %i: requested %lld ns "
1563 "lapic timer period limited to %lld ns\n",
1564 apic->vcpu->vcpu_id,
1565 apic->lapic_timer.period, min_period);
1566 apic->lapic_timer.period = min_period;
1571 static void cancel_hv_timer(struct kvm_lapic *apic);
1573 static void cancel_apic_timer(struct kvm_lapic *apic)
1575 hrtimer_cancel(&apic->lapic_timer.timer);
1577 if (apic->lapic_timer.hv_timer_in_use)
1578 cancel_hv_timer(apic);
1580 atomic_set(&apic->lapic_timer.pending, 0);
1583 static void apic_update_lvtt(struct kvm_lapic *apic)
1585 u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1586 apic->lapic_timer.timer_mode_mask;
1588 if (apic->lapic_timer.timer_mode != timer_mode) {
1589 if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1590 APIC_LVT_TIMER_TSCDEADLINE)) {
1591 cancel_apic_timer(apic);
1592 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1593 apic->lapic_timer.period = 0;
1594 apic->lapic_timer.tscdeadline = 0;
1596 apic->lapic_timer.timer_mode = timer_mode;
1597 limit_periodic_timer_frequency(apic);
1602 * On APICv, this test will cause a busy wait
1603 * during a higher-priority task.
1606 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1608 struct kvm_lapic *apic = vcpu->arch.apic;
1609 u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1611 if (kvm_apic_hw_enabled(apic)) {
1612 int vec = reg & APIC_VECTOR_MASK;
1613 void *bitmap = apic->regs + APIC_ISR;
1615 if (apic->apicv_active)
1616 bitmap = apic->regs + APIC_IRR;
1618 if (apic_test_vector(vec, bitmap))
1624 static inline void __wait_lapic_expire(struct kvm_vcpu *vcpu, u64 guest_cycles)
1626 u64 timer_advance_ns = vcpu->arch.apic->lapic_timer.timer_advance_ns;
1629 * If the guest TSC is running at a different ratio than the host, then
1630 * convert the delay to nanoseconds to achieve an accurate delay. Note
1631 * that __delay() uses delay_tsc whenever the hardware has TSC, thus
1632 * always for VMX enabled hardware.
1634 if (vcpu->arch.tsc_scaling_ratio == kvm_caps.default_tsc_scaling_ratio) {
1635 __delay(min(guest_cycles,
1636 nsec_to_cycles(vcpu, timer_advance_ns)));
1638 u64 delay_ns = guest_cycles * 1000000ULL;
1639 do_div(delay_ns, vcpu->arch.virtual_tsc_khz);
1640 ndelay(min_t(u32, delay_ns, timer_advance_ns));
1644 static inline void adjust_lapic_timer_advance(struct kvm_vcpu *vcpu,
1645 s64 advance_expire_delta)
1647 struct kvm_lapic *apic = vcpu->arch.apic;
1648 u32 timer_advance_ns = apic->lapic_timer.timer_advance_ns;
1651 /* Do not adjust for tiny fluctuations or large random spikes. */
1652 if (abs(advance_expire_delta) > LAPIC_TIMER_ADVANCE_ADJUST_MAX ||
1653 abs(advance_expire_delta) < LAPIC_TIMER_ADVANCE_ADJUST_MIN)
1657 if (advance_expire_delta < 0) {
1658 ns = -advance_expire_delta * 1000000ULL;
1659 do_div(ns, vcpu->arch.virtual_tsc_khz);
1660 timer_advance_ns -= ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1663 ns = advance_expire_delta * 1000000ULL;
1664 do_div(ns, vcpu->arch.virtual_tsc_khz);
1665 timer_advance_ns += ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1668 if (unlikely(timer_advance_ns > LAPIC_TIMER_ADVANCE_NS_MAX))
1669 timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
1670 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
1673 static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1675 struct kvm_lapic *apic = vcpu->arch.apic;
1676 u64 guest_tsc, tsc_deadline;
1678 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1679 apic->lapic_timer.expired_tscdeadline = 0;
1680 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1681 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1683 if (lapic_timer_advance_dynamic) {
1684 adjust_lapic_timer_advance(vcpu, guest_tsc - tsc_deadline);
1686 * If the timer fired early, reread the TSC to account for the
1687 * overhead of the above adjustment to avoid waiting longer
1688 * than is necessary.
1690 if (guest_tsc < tsc_deadline)
1691 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1694 if (guest_tsc < tsc_deadline)
1695 __wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
1698 void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1700 if (lapic_in_kernel(vcpu) &&
1701 vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1702 vcpu->arch.apic->lapic_timer.timer_advance_ns &&
1703 lapic_timer_int_injected(vcpu))
1704 __kvm_wait_lapic_expire(vcpu);
1706 EXPORT_SYMBOL_GPL(kvm_wait_lapic_expire);
1708 static void kvm_apic_inject_pending_timer_irqs(struct kvm_lapic *apic)
1710 struct kvm_timer *ktimer = &apic->lapic_timer;
1712 kvm_apic_local_deliver(apic, APIC_LVTT);
1713 if (apic_lvtt_tscdeadline(apic)) {
1714 ktimer->tscdeadline = 0;
1715 } else if (apic_lvtt_oneshot(apic)) {
1716 ktimer->tscdeadline = 0;
1717 ktimer->target_expiration = 0;
1721 static void apic_timer_expired(struct kvm_lapic *apic, bool from_timer_fn)
1723 struct kvm_vcpu *vcpu = apic->vcpu;
1724 struct kvm_timer *ktimer = &apic->lapic_timer;
1726 if (atomic_read(&apic->lapic_timer.pending))
1729 if (apic_lvtt_tscdeadline(apic) || ktimer->hv_timer_in_use)
1730 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1732 if (!from_timer_fn && apic->apicv_active) {
1733 WARN_ON(kvm_get_running_vcpu() != vcpu);
1734 kvm_apic_inject_pending_timer_irqs(apic);
1738 if (kvm_use_posted_timer_interrupt(apic->vcpu)) {
1740 * Ensure the guest's timer has truly expired before posting an
1741 * interrupt. Open code the relevant checks to avoid querying
1742 * lapic_timer_int_injected(), which will be false since the
1743 * interrupt isn't yet injected. Waiting until after injecting
1744 * is not an option since that won't help a posted interrupt.
1746 if (vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1747 vcpu->arch.apic->lapic_timer.timer_advance_ns)
1748 __kvm_wait_lapic_expire(vcpu);
1749 kvm_apic_inject_pending_timer_irqs(apic);
1753 atomic_inc(&apic->lapic_timer.pending);
1754 kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
1756 kvm_vcpu_kick(vcpu);
1759 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1761 struct kvm_timer *ktimer = &apic->lapic_timer;
1762 u64 guest_tsc, tscdeadline = ktimer->tscdeadline;
1765 struct kvm_vcpu *vcpu = apic->vcpu;
1766 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1767 unsigned long flags;
1770 if (unlikely(!tscdeadline || !this_tsc_khz))
1773 local_irq_save(flags);
1776 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1778 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1779 do_div(ns, this_tsc_khz);
1781 if (likely(tscdeadline > guest_tsc) &&
1782 likely(ns > apic->lapic_timer.timer_advance_ns)) {
1783 expire = ktime_add_ns(now, ns);
1784 expire = ktime_sub_ns(expire, ktimer->timer_advance_ns);
1785 hrtimer_start(&ktimer->timer, expire, HRTIMER_MODE_ABS_HARD);
1787 apic_timer_expired(apic, false);
1789 local_irq_restore(flags);
1792 static inline u64 tmict_to_ns(struct kvm_lapic *apic, u32 tmict)
1794 return (u64)tmict * APIC_BUS_CYCLE_NS * (u64)apic->divide_count;
1797 static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1799 ktime_t now, remaining;
1800 u64 ns_remaining_old, ns_remaining_new;
1802 apic->lapic_timer.period =
1803 tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1804 limit_periodic_timer_frequency(apic);
1807 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1808 if (ktime_to_ns(remaining) < 0)
1811 ns_remaining_old = ktime_to_ns(remaining);
1812 ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1813 apic->divide_count, old_divisor);
1815 apic->lapic_timer.tscdeadline +=
1816 nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1817 nsec_to_cycles(apic->vcpu, ns_remaining_old);
1818 apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1821 static bool set_target_expiration(struct kvm_lapic *apic, u32 count_reg)
1828 apic->lapic_timer.period =
1829 tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1831 if (!apic->lapic_timer.period) {
1832 apic->lapic_timer.tscdeadline = 0;
1836 limit_periodic_timer_frequency(apic);
1837 deadline = apic->lapic_timer.period;
1839 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1840 if (unlikely(count_reg != APIC_TMICT)) {
1841 deadline = tmict_to_ns(apic,
1842 kvm_lapic_get_reg(apic, count_reg));
1843 if (unlikely(deadline <= 0))
1844 deadline = apic->lapic_timer.period;
1845 else if (unlikely(deadline > apic->lapic_timer.period)) {
1846 pr_info_ratelimited(
1847 "kvm: vcpu %i: requested lapic timer restore with "
1848 "starting count register %#x=%u (%lld ns) > initial count (%lld ns). "
1849 "Using initial count to start timer.\n",
1850 apic->vcpu->vcpu_id,
1852 kvm_lapic_get_reg(apic, count_reg),
1853 deadline, apic->lapic_timer.period);
1854 kvm_lapic_set_reg(apic, count_reg, 0);
1855 deadline = apic->lapic_timer.period;
1860 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1861 nsec_to_cycles(apic->vcpu, deadline);
1862 apic->lapic_timer.target_expiration = ktime_add_ns(now, deadline);
1867 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1869 ktime_t now = ktime_get();
1874 * Synchronize both deadlines to the same time source or
1875 * differences in the periods (caused by differences in the
1876 * underlying clocks or numerical approximation errors) will
1877 * cause the two to drift apart over time as the errors
1880 apic->lapic_timer.target_expiration =
1881 ktime_add_ns(apic->lapic_timer.target_expiration,
1882 apic->lapic_timer.period);
1883 delta = ktime_sub(apic->lapic_timer.target_expiration, now);
1884 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1885 nsec_to_cycles(apic->vcpu, delta);
1888 static void start_sw_period(struct kvm_lapic *apic)
1890 if (!apic->lapic_timer.period)
1893 if (ktime_after(ktime_get(),
1894 apic->lapic_timer.target_expiration)) {
1895 apic_timer_expired(apic, false);
1897 if (apic_lvtt_oneshot(apic))
1900 advance_periodic_target_expiration(apic);
1903 hrtimer_start(&apic->lapic_timer.timer,
1904 apic->lapic_timer.target_expiration,
1905 HRTIMER_MODE_ABS_HARD);
1908 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1910 if (!lapic_in_kernel(vcpu))
1913 return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1915 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1917 static void cancel_hv_timer(struct kvm_lapic *apic)
1919 WARN_ON(preemptible());
1920 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1921 static_call(kvm_x86_cancel_hv_timer)(apic->vcpu);
1922 apic->lapic_timer.hv_timer_in_use = false;
1925 static bool start_hv_timer(struct kvm_lapic *apic)
1927 struct kvm_timer *ktimer = &apic->lapic_timer;
1928 struct kvm_vcpu *vcpu = apic->vcpu;
1931 WARN_ON(preemptible());
1932 if (!kvm_can_use_hv_timer(vcpu))
1935 if (!ktimer->tscdeadline)
1938 if (static_call(kvm_x86_set_hv_timer)(vcpu, ktimer->tscdeadline, &expired))
1941 ktimer->hv_timer_in_use = true;
1942 hrtimer_cancel(&ktimer->timer);
1945 * To simplify handling the periodic timer, leave the hv timer running
1946 * even if the deadline timer has expired, i.e. rely on the resulting
1947 * VM-Exit to recompute the periodic timer's target expiration.
1949 if (!apic_lvtt_period(apic)) {
1951 * Cancel the hv timer if the sw timer fired while the hv timer
1952 * was being programmed, or if the hv timer itself expired.
1954 if (atomic_read(&ktimer->pending)) {
1955 cancel_hv_timer(apic);
1956 } else if (expired) {
1957 apic_timer_expired(apic, false);
1958 cancel_hv_timer(apic);
1962 trace_kvm_hv_timer_state(vcpu->vcpu_id, ktimer->hv_timer_in_use);
1967 static void start_sw_timer(struct kvm_lapic *apic)
1969 struct kvm_timer *ktimer = &apic->lapic_timer;
1971 WARN_ON(preemptible());
1972 if (apic->lapic_timer.hv_timer_in_use)
1973 cancel_hv_timer(apic);
1974 if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1977 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1978 start_sw_period(apic);
1979 else if (apic_lvtt_tscdeadline(apic))
1980 start_sw_tscdeadline(apic);
1981 trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1984 static void restart_apic_timer(struct kvm_lapic *apic)
1988 if (!apic_lvtt_period(apic) && atomic_read(&apic->lapic_timer.pending))
1991 if (!start_hv_timer(apic))
1992 start_sw_timer(apic);
1997 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1999 struct kvm_lapic *apic = vcpu->arch.apic;
2002 /* If the preempt notifier has already run, it also called apic_timer_expired */
2003 if (!apic->lapic_timer.hv_timer_in_use)
2005 WARN_ON(kvm_vcpu_is_blocking(vcpu));
2006 apic_timer_expired(apic, false);
2007 cancel_hv_timer(apic);
2009 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
2010 advance_periodic_target_expiration(apic);
2011 restart_apic_timer(apic);
2016 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
2018 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
2020 restart_apic_timer(vcpu->arch.apic);
2023 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
2025 struct kvm_lapic *apic = vcpu->arch.apic;
2028 /* Possibly the TSC deadline timer is not enabled yet */
2029 if (apic->lapic_timer.hv_timer_in_use)
2030 start_sw_timer(apic);
2034 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
2036 struct kvm_lapic *apic = vcpu->arch.apic;
2038 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
2039 restart_apic_timer(apic);
2042 static void __start_apic_timer(struct kvm_lapic *apic, u32 count_reg)
2044 atomic_set(&apic->lapic_timer.pending, 0);
2046 if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
2047 && !set_target_expiration(apic, count_reg))
2050 restart_apic_timer(apic);
2053 static void start_apic_timer(struct kvm_lapic *apic)
2055 __start_apic_timer(apic, APIC_TMICT);
2058 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
2060 bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
2062 if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
2063 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
2064 if (lvt0_in_nmi_mode) {
2065 atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
2067 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
2071 static void kvm_lapic_xapic_id_updated(struct kvm_lapic *apic)
2073 struct kvm *kvm = apic->vcpu->kvm;
2075 if (KVM_BUG_ON(apic_x2apic_mode(apic), kvm))
2078 if (kvm_xapic_id(apic) == apic->vcpu->vcpu_id)
2081 kvm_set_apicv_inhibit(apic->vcpu->kvm, APICV_INHIBIT_REASON_APIC_ID_MODIFIED);
2084 static int get_lvt_index(u32 reg)
2086 if (reg == APIC_LVTCMCI)
2088 if (reg < APIC_LVTT || reg > APIC_LVTERR)
2090 return array_index_nospec(
2091 (reg - APIC_LVTT) >> 4, KVM_APIC_MAX_NR_LVT_ENTRIES);
2094 static int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
2098 trace_kvm_apic_write(reg, val);
2101 case APIC_ID: /* Local APIC ID */
2102 if (!apic_x2apic_mode(apic)) {
2103 kvm_apic_set_xapic_id(apic, val >> 24);
2104 kvm_lapic_xapic_id_updated(apic);
2111 report_tpr_access(apic, true);
2112 apic_set_tpr(apic, val & 0xff);
2120 if (!apic_x2apic_mode(apic))
2121 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
2127 if (!apic_x2apic_mode(apic))
2128 kvm_apic_set_dfr(apic, val | 0x0FFFFFFF);
2135 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
2136 mask |= APIC_SPIV_DIRECTED_EOI;
2137 apic_set_spiv(apic, val & mask);
2138 if (!(val & APIC_SPIV_APIC_ENABLED)) {
2141 for (i = 0; i < apic->nr_lvt_entries; i++) {
2142 kvm_lapic_set_reg(apic, APIC_LVTx(i),
2143 kvm_lapic_get_reg(apic, APIC_LVTx(i)) | APIC_LVT_MASKED);
2145 apic_update_lvtt(apic);
2146 atomic_set(&apic->lapic_timer.pending, 0);
2152 WARN_ON_ONCE(apic_x2apic_mode(apic));
2154 /* No delay here, so we always clear the pending bit */
2155 val &= ~APIC_ICR_BUSY;
2156 kvm_apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
2157 kvm_lapic_set_reg(apic, APIC_ICR, val);
2160 if (apic_x2apic_mode(apic))
2163 kvm_lapic_set_reg(apic, APIC_ICR2, val & 0xff000000);
2167 apic_manage_nmi_watchdog(apic, val);
2173 case APIC_LVTCMCI: {
2174 u32 index = get_lvt_index(reg);
2175 if (!kvm_lapic_lvt_supported(apic, index)) {
2179 if (!kvm_apic_sw_enabled(apic))
2180 val |= APIC_LVT_MASKED;
2181 val &= apic_lvt_mask[index];
2182 kvm_lapic_set_reg(apic, reg, val);
2187 if (!kvm_apic_sw_enabled(apic))
2188 val |= APIC_LVT_MASKED;
2189 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
2190 kvm_lapic_set_reg(apic, APIC_LVTT, val);
2191 apic_update_lvtt(apic);
2195 if (apic_lvtt_tscdeadline(apic))
2198 cancel_apic_timer(apic);
2199 kvm_lapic_set_reg(apic, APIC_TMICT, val);
2200 start_apic_timer(apic);
2204 uint32_t old_divisor = apic->divide_count;
2206 kvm_lapic_set_reg(apic, APIC_TDCR, val & 0xb);
2207 update_divide_count(apic);
2208 if (apic->divide_count != old_divisor &&
2209 apic->lapic_timer.period) {
2210 hrtimer_cancel(&apic->lapic_timer.timer);
2211 update_target_expiration(apic, old_divisor);
2212 restart_apic_timer(apic);
2217 if (apic_x2apic_mode(apic) && val != 0)
2222 if (apic_x2apic_mode(apic))
2223 kvm_apic_send_ipi(apic, APIC_DEST_SELF | (val & APIC_VECTOR_MASK), 0);
2233 * Recalculate APIC maps if necessary, e.g. if the software enable bit
2234 * was toggled, the APIC ID changed, etc... The maps are marked dirty
2235 * on relevant changes, i.e. this is a nop for most writes.
2237 kvm_recalculate_apic_map(apic->vcpu->kvm);
2242 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
2243 gpa_t address, int len, const void *data)
2245 struct kvm_lapic *apic = to_lapic(this);
2246 unsigned int offset = address - apic->base_address;
2249 if (!apic_mmio_in_range(apic, address))
2252 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
2253 if (!kvm_check_has_quirk(vcpu->kvm,
2254 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
2261 * APIC register must be aligned on 128-bits boundary.
2262 * 32/64/128 bits registers must be accessed thru 32 bits.
2265 if (len != 4 || (offset & 0xf))
2270 kvm_lapic_reg_write(apic, offset & 0xff0, val);
2275 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
2277 kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
2279 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
2281 /* emulate APIC access in a trap manner */
2282 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
2284 struct kvm_lapic *apic = vcpu->arch.apic;
2287 if (apic_x2apic_mode(apic))
2288 kvm_lapic_msr_read(apic, offset, &val);
2290 val = kvm_lapic_get_reg(apic, offset);
2293 * ICR is a single 64-bit register when x2APIC is enabled. For legacy
2294 * xAPIC, ICR writes need to go down the common (slightly slower) path
2295 * to get the upper half from ICR2.
2297 if (apic_x2apic_mode(apic) && offset == APIC_ICR) {
2298 kvm_apic_send_ipi(apic, (u32)val, (u32)(val >> 32));
2299 trace_kvm_apic_write(APIC_ICR, val);
2301 /* TODO: optimize to just emulate side effect w/o one more write */
2302 kvm_lapic_reg_write(apic, offset, (u32)val);
2305 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
2307 void kvm_free_lapic(struct kvm_vcpu *vcpu)
2309 struct kvm_lapic *apic = vcpu->arch.apic;
2311 if (!vcpu->arch.apic)
2314 hrtimer_cancel(&apic->lapic_timer.timer);
2316 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
2317 static_branch_slow_dec_deferred(&apic_hw_disabled);
2319 if (!apic->sw_enabled)
2320 static_branch_slow_dec_deferred(&apic_sw_disabled);
2323 free_page((unsigned long)apic->regs);
2329 *----------------------------------------------------------------------
2331 *----------------------------------------------------------------------
2333 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
2335 struct kvm_lapic *apic = vcpu->arch.apic;
2337 if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2340 return apic->lapic_timer.tscdeadline;
2343 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
2345 struct kvm_lapic *apic = vcpu->arch.apic;
2347 if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2350 hrtimer_cancel(&apic->lapic_timer.timer);
2351 apic->lapic_timer.tscdeadline = data;
2352 start_apic_timer(apic);
2355 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
2357 apic_set_tpr(vcpu->arch.apic, (cr8 & 0x0f) << 4);
2360 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
2364 tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
2366 return (tpr & 0xf0) >> 4;
2369 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
2371 u64 old_value = vcpu->arch.apic_base;
2372 struct kvm_lapic *apic = vcpu->arch.apic;
2374 vcpu->arch.apic_base = value;
2376 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
2377 kvm_update_cpuid_runtime(vcpu);
2382 /* update jump label if enable bit changes */
2383 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
2384 if (value & MSR_IA32_APICBASE_ENABLE) {
2385 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2386 static_branch_slow_dec_deferred(&apic_hw_disabled);
2387 /* Check if there are APF page ready requests pending */
2388 kvm_make_request(KVM_REQ_APF_READY, vcpu);
2390 static_branch_inc(&apic_hw_disabled.key);
2391 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2395 if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
2396 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
2398 if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) {
2399 kvm_vcpu_update_apicv(vcpu);
2400 static_call_cond(kvm_x86_set_virtual_apic_mode)(vcpu);
2403 apic->base_address = apic->vcpu->arch.apic_base &
2404 MSR_IA32_APICBASE_BASE;
2406 if ((value & MSR_IA32_APICBASE_ENABLE) &&
2407 apic->base_address != APIC_DEFAULT_PHYS_BASE) {
2408 kvm_set_apicv_inhibit(apic->vcpu->kvm,
2409 APICV_INHIBIT_REASON_APIC_BASE_MODIFIED);
2413 void kvm_apic_update_apicv(struct kvm_vcpu *vcpu)
2415 struct kvm_lapic *apic = vcpu->arch.apic;
2417 if (apic->apicv_active) {
2418 /* irr_pending is always true when apicv is activated. */
2419 apic->irr_pending = true;
2420 apic->isr_count = 1;
2423 * Don't clear irr_pending, searching the IRR can race with
2424 * updates from the CPU as APICv is still active from hardware's
2425 * perspective. The flag will be cleared as appropriate when
2426 * KVM injects the interrupt.
2428 apic->isr_count = count_vectors(apic->regs + APIC_ISR);
2431 EXPORT_SYMBOL_GPL(kvm_apic_update_apicv);
2433 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2435 struct kvm_lapic *apic = vcpu->arch.apic;
2440 msr_val = APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE;
2441 if (kvm_vcpu_is_reset_bsp(vcpu))
2442 msr_val |= MSR_IA32_APICBASE_BSP;
2443 kvm_lapic_set_base(vcpu, msr_val);
2449 /* Stop the timer in case it's a reset to an active apic */
2450 hrtimer_cancel(&apic->lapic_timer.timer);
2452 /* The xAPIC ID is set at RESET even if the APIC was already enabled. */
2454 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2455 kvm_apic_set_version(apic->vcpu);
2457 for (i = 0; i < apic->nr_lvt_entries; i++)
2458 kvm_lapic_set_reg(apic, APIC_LVTx(i), APIC_LVT_MASKED);
2459 apic_update_lvtt(apic);
2460 if (kvm_vcpu_is_reset_bsp(vcpu) &&
2461 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2462 kvm_lapic_set_reg(apic, APIC_LVT0,
2463 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2464 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2466 kvm_apic_set_dfr(apic, 0xffffffffU);
2467 apic_set_spiv(apic, 0xff);
2468 kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2469 if (!apic_x2apic_mode(apic))
2470 kvm_apic_set_ldr(apic, 0);
2471 kvm_lapic_set_reg(apic, APIC_ESR, 0);
2472 if (!apic_x2apic_mode(apic)) {
2473 kvm_lapic_set_reg(apic, APIC_ICR, 0);
2474 kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2476 kvm_lapic_set_reg64(apic, APIC_ICR, 0);
2478 kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2479 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2480 for (i = 0; i < 8; i++) {
2481 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2482 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2483 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2485 kvm_apic_update_apicv(vcpu);
2486 apic->highest_isr_cache = -1;
2487 update_divide_count(apic);
2488 atomic_set(&apic->lapic_timer.pending, 0);
2490 vcpu->arch.pv_eoi.msr_val = 0;
2491 apic_update_ppr(apic);
2492 if (apic->apicv_active) {
2493 static_call_cond(kvm_x86_apicv_post_state_restore)(vcpu);
2494 static_call_cond(kvm_x86_hwapic_irr_update)(vcpu, -1);
2495 static_call_cond(kvm_x86_hwapic_isr_update)(-1);
2498 vcpu->arch.apic_arb_prio = 0;
2499 vcpu->arch.apic_attention = 0;
2501 kvm_recalculate_apic_map(vcpu->kvm);
2505 *----------------------------------------------------------------------
2507 *----------------------------------------------------------------------
2510 static bool lapic_is_periodic(struct kvm_lapic *apic)
2512 return apic_lvtt_period(apic);
2515 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2517 struct kvm_lapic *apic = vcpu->arch.apic;
2519 if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2520 return atomic_read(&apic->lapic_timer.pending);
2525 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2527 u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2528 int vector, mode, trig_mode;
2530 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2531 vector = reg & APIC_VECTOR_MASK;
2532 mode = reg & APIC_MODE_MASK;
2533 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2534 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2540 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2542 struct kvm_lapic *apic = vcpu->arch.apic;
2545 kvm_apic_local_deliver(apic, APIC_LVT0);
2548 static const struct kvm_io_device_ops apic_mmio_ops = {
2549 .read = apic_mmio_read,
2550 .write = apic_mmio_write,
2553 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2555 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2556 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2558 apic_timer_expired(apic, true);
2560 if (lapic_is_periodic(apic)) {
2561 advance_periodic_target_expiration(apic);
2562 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2563 return HRTIMER_RESTART;
2565 return HRTIMER_NORESTART;
2568 int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
2570 struct kvm_lapic *apic;
2572 ASSERT(vcpu != NULL);
2574 apic = kzalloc(sizeof(*apic), GFP_KERNEL_ACCOUNT);
2578 vcpu->arch.apic = apic;
2580 apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
2582 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2584 goto nomem_free_apic;
2588 apic->nr_lvt_entries = kvm_apic_calc_nr_lvt_entries(vcpu);
2590 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2591 HRTIMER_MODE_ABS_HARD);
2592 apic->lapic_timer.timer.function = apic_timer_fn;
2593 if (timer_advance_ns == -1) {
2594 apic->lapic_timer.timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
2595 lapic_timer_advance_dynamic = true;
2597 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
2598 lapic_timer_advance_dynamic = false;
2602 * Stuff the APIC ENABLE bit in lieu of temporarily incrementing
2603 * apic_hw_disabled; the full RESET value is set by kvm_lapic_reset().
2605 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2606 static_branch_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2607 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2612 vcpu->arch.apic = NULL;
2617 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2619 struct kvm_lapic *apic = vcpu->arch.apic;
2622 if (!kvm_apic_present(vcpu))
2625 __apic_update_ppr(apic, &ppr);
2626 return apic_has_interrupt_for_ppr(apic, ppr);
2628 EXPORT_SYMBOL_GPL(kvm_apic_has_interrupt);
2630 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2632 u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2634 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2636 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2637 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2642 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2644 struct kvm_lapic *apic = vcpu->arch.apic;
2646 if (atomic_read(&apic->lapic_timer.pending) > 0) {
2647 kvm_apic_inject_pending_timer_irqs(apic);
2648 atomic_set(&apic->lapic_timer.pending, 0);
2652 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2654 int vector = kvm_apic_has_interrupt(vcpu);
2655 struct kvm_lapic *apic = vcpu->arch.apic;
2662 * We get here even with APIC virtualization enabled, if doing
2663 * nested virtualization and L1 runs with the "acknowledge interrupt
2664 * on exit" mode. Then we cannot inject the interrupt via RVI,
2665 * because the process would deliver it through the IDT.
2668 apic_clear_irr(vector, apic);
2669 if (to_hv_vcpu(vcpu) && test_bit(vector, to_hv_synic(vcpu)->auto_eoi_bitmap)) {
2671 * For auto-EOI interrupts, there might be another pending
2672 * interrupt above PPR, so check whether to raise another
2675 apic_update_ppr(apic);
2678 * For normal interrupts, PPR has been raised and there cannot
2679 * be a higher-priority pending interrupt---except if there was
2680 * a concurrent interrupt injection, but that would have
2681 * triggered KVM_REQ_EVENT already.
2683 apic_set_isr(vector, apic);
2684 __apic_update_ppr(apic, &ppr);
2690 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2691 struct kvm_lapic_state *s, bool set)
2693 if (apic_x2apic_mode(vcpu->arch.apic)) {
2694 u32 *id = (u32 *)(s->regs + APIC_ID);
2695 u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2698 if (vcpu->kvm->arch.x2apic_format) {
2699 if (*id != vcpu->vcpu_id)
2709 * In x2APIC mode, the LDR is fixed and based on the id. And
2710 * ICR is internally a single 64-bit register, but needs to be
2711 * split to ICR+ICR2 in userspace for backwards compatibility.
2714 *ldr = kvm_apic_calc_x2apic_ldr(*id);
2716 icr = __kvm_lapic_get_reg(s->regs, APIC_ICR) |
2717 (u64)__kvm_lapic_get_reg(s->regs, APIC_ICR2) << 32;
2718 __kvm_lapic_set_reg64(s->regs, APIC_ICR, icr);
2720 icr = __kvm_lapic_get_reg64(s->regs, APIC_ICR);
2721 __kvm_lapic_set_reg(s->regs, APIC_ICR2, icr >> 32);
2724 kvm_lapic_xapic_id_updated(vcpu->arch.apic);
2730 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2732 memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2735 * Get calculated timer current count for remaining timer period (if
2736 * any) and store it in the returned register set.
2738 __kvm_lapic_set_reg(s->regs, APIC_TMCCT,
2739 __apic_read(vcpu->arch.apic, APIC_TMCCT));
2741 return kvm_apic_state_fixup(vcpu, s, false);
2744 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2746 struct kvm_lapic *apic = vcpu->arch.apic;
2749 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2750 /* set SPIV separately to get count of SW disabled APICs right */
2751 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2753 r = kvm_apic_state_fixup(vcpu, s, true);
2755 kvm_recalculate_apic_map(vcpu->kvm);
2758 memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
2760 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2761 kvm_recalculate_apic_map(vcpu->kvm);
2762 kvm_apic_set_version(vcpu);
2764 apic_update_ppr(apic);
2765 cancel_apic_timer(apic);
2766 apic->lapic_timer.expired_tscdeadline = 0;
2767 apic_update_lvtt(apic);
2768 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2769 update_divide_count(apic);
2770 __start_apic_timer(apic, APIC_TMCCT);
2771 kvm_lapic_set_reg(apic, APIC_TMCCT, 0);
2772 kvm_apic_update_apicv(vcpu);
2773 apic->highest_isr_cache = -1;
2774 if (apic->apicv_active) {
2775 static_call_cond(kvm_x86_apicv_post_state_restore)(vcpu);
2776 static_call_cond(kvm_x86_hwapic_irr_update)(vcpu, apic_find_highest_irr(apic));
2777 static_call_cond(kvm_x86_hwapic_isr_update)(apic_find_highest_isr(apic));
2779 kvm_make_request(KVM_REQ_EVENT, vcpu);
2780 if (ioapic_in_kernel(vcpu->kvm))
2781 kvm_rtc_eoi_tracking_restore_one(vcpu);
2783 vcpu->arch.apic_arb_prio = 0;
2788 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2790 struct hrtimer *timer;
2792 if (!lapic_in_kernel(vcpu) ||
2793 kvm_can_post_timer_interrupt(vcpu))
2796 timer = &vcpu->arch.apic->lapic_timer.timer;
2797 if (hrtimer_cancel(timer))
2798 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_HARD);
2802 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2804 * Detect whether guest triggered PV EOI since the
2805 * last entry. If yes, set EOI on guests's behalf.
2806 * Clear PV EOI in guest memory in any case.
2808 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2809 struct kvm_lapic *apic)
2813 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2814 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2816 * KVM_APIC_PV_EOI_PENDING is unset:
2817 * -> host disabled PV EOI.
2818 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2819 * -> host enabled PV EOI, guest did not execute EOI yet.
2820 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2821 * -> host enabled PV EOI, guest executed EOI.
2823 BUG_ON(!pv_eoi_enabled(vcpu));
2825 if (pv_eoi_test_and_clr_pending(vcpu))
2827 vector = apic_set_eoi(apic);
2828 trace_kvm_pv_eoi(apic, vector);
2831 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2835 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2836 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2838 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2841 if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2845 apic_set_tpr(vcpu->arch.apic, data & 0xff);
2849 * apic_sync_pv_eoi_to_guest - called before vmentry
2851 * Detect whether it's safe to enable PV EOI and
2854 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2855 struct kvm_lapic *apic)
2857 if (!pv_eoi_enabled(vcpu) ||
2858 /* IRR set or many bits in ISR: could be nested. */
2859 apic->irr_pending ||
2860 /* Cache not set: could be safe but we don't bother. */
2861 apic->highest_isr_cache == -1 ||
2862 /* Need EOI to update ioapic. */
2863 kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2865 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2866 * so we need not do anything here.
2871 pv_eoi_set_pending(apic->vcpu);
2874 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2877 int max_irr, max_isr;
2878 struct kvm_lapic *apic = vcpu->arch.apic;
2880 apic_sync_pv_eoi_to_guest(vcpu, apic);
2882 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2885 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2886 max_irr = apic_find_highest_irr(apic);
2889 max_isr = apic_find_highest_isr(apic);
2892 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2894 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2898 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2901 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2902 &vcpu->arch.apic->vapic_cache,
2903 vapic_addr, sizeof(u32)))
2905 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2907 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2910 vcpu->arch.apic->vapic_addr = vapic_addr;
2914 int kvm_x2apic_icr_write(struct kvm_lapic *apic, u64 data)
2916 data &= ~APIC_ICR_BUSY;
2918 kvm_apic_send_ipi(apic, (u32)data, (u32)(data >> 32));
2919 kvm_lapic_set_reg64(apic, APIC_ICR, data);
2920 trace_kvm_apic_write(APIC_ICR, data);
2924 static int kvm_lapic_msr_read(struct kvm_lapic *apic, u32 reg, u64 *data)
2928 if (reg == APIC_ICR) {
2929 *data = kvm_lapic_get_reg64(apic, APIC_ICR);
2933 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2941 static int kvm_lapic_msr_write(struct kvm_lapic *apic, u32 reg, u64 data)
2944 * ICR is a 64-bit register in x2APIC mode (and Hyper'v PV vAPIC) and
2945 * can be written as such, all other registers remain accessible only
2946 * through 32-bit reads/writes.
2948 if (reg == APIC_ICR)
2949 return kvm_x2apic_icr_write(apic, data);
2951 return kvm_lapic_reg_write(apic, reg, (u32)data);
2954 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2956 struct kvm_lapic *apic = vcpu->arch.apic;
2957 u32 reg = (msr - APIC_BASE_MSR) << 4;
2959 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2962 return kvm_lapic_msr_write(apic, reg, data);
2965 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2967 struct kvm_lapic *apic = vcpu->arch.apic;
2968 u32 reg = (msr - APIC_BASE_MSR) << 4;
2970 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2973 if (reg == APIC_DFR)
2976 return kvm_lapic_msr_read(apic, reg, data);
2979 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2981 if (!lapic_in_kernel(vcpu))
2984 return kvm_lapic_msr_write(vcpu->arch.apic, reg, data);
2987 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2989 if (!lapic_in_kernel(vcpu))
2992 return kvm_lapic_msr_read(vcpu->arch.apic, reg, data);
2995 int kvm_lapic_set_pv_eoi(struct kvm_vcpu *vcpu, u64 data, unsigned long len)
2997 u64 addr = data & ~KVM_MSR_ENABLED;
2998 struct gfn_to_hva_cache *ghc = &vcpu->arch.pv_eoi.data;
2999 unsigned long new_len;
3002 if (!IS_ALIGNED(addr, 4))
3005 if (data & KVM_MSR_ENABLED) {
3006 if (addr == ghc->gpa && len <= ghc->len)
3011 ret = kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, addr, new_len);
3016 vcpu->arch.pv_eoi.msr_val = data;
3021 int kvm_apic_accept_events(struct kvm_vcpu *vcpu)
3023 struct kvm_lapic *apic = vcpu->arch.apic;
3028 if (!lapic_in_kernel(vcpu))
3032 * Read pending events before calling the check_events
3035 pe = smp_load_acquire(&apic->pending_events);
3039 if (is_guest_mode(vcpu)) {
3040 r = kvm_check_nested_events(vcpu);
3042 return r == -EBUSY ? 0 : r;
3044 * If an event has happened and caused a vmexit,
3045 * we know INITs are latched and therefore
3046 * we will not incorrectly deliver an APIC
3047 * event instead of a vmexit.
3052 * INITs are latched while CPU is in specific states
3053 * (SMM, VMX root mode, SVM with GIF=0).
3054 * Because a CPU cannot be in these states immediately
3055 * after it has processed an INIT signal (and thus in
3056 * KVM_MP_STATE_INIT_RECEIVED state), just eat SIPIs
3057 * and leave the INIT pending.
3059 if (kvm_vcpu_latch_init(vcpu)) {
3060 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
3061 if (test_bit(KVM_APIC_SIPI, &pe))
3062 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
3066 if (test_bit(KVM_APIC_INIT, &pe)) {
3067 clear_bit(KVM_APIC_INIT, &apic->pending_events);
3068 kvm_vcpu_reset(vcpu, true);
3069 if (kvm_vcpu_is_bsp(apic->vcpu))
3070 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3072 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3074 if (test_bit(KVM_APIC_SIPI, &pe)) {
3075 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
3076 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
3077 /* evaluate pending_events before reading the vector */
3079 sipi_vector = apic->sipi_vector;
3080 static_call(kvm_x86_vcpu_deliver_sipi_vector)(vcpu, sipi_vector);
3081 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3087 void kvm_lapic_exit(void)
3089 static_key_deferred_flush(&apic_hw_disabled);
3090 WARN_ON(static_branch_unlikely(&apic_hw_disabled.key));
3091 static_key_deferred_flush(&apic_sw_disabled);
3092 WARN_ON(static_branch_unlikely(&apic_sw_disabled.key));