1 // SPDX-License-Identifier: GPL-2.0-only
4 * Local APIC virtualization
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2007 Novell
8 * Copyright (C) 2007 Intel
9 * Copyright 2009 Red Hat, Inc. and/or its affiliates.
12 * Dor Laor <dor.laor@qumranet.com>
13 * Gregory Haskins <ghaskins@novell.com>
14 * Yaozu (Eddie) Dong <eddie.dong@intel.com>
16 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
19 #include <linux/kvm_host.h>
20 #include <linux/kvm.h>
22 #include <linux/highmem.h>
23 #include <linux/smp.h>
24 #include <linux/hrtimer.h>
26 #include <linux/export.h>
27 #include <linux/math64.h>
28 #include <linux/slab.h>
29 #include <asm/processor.h>
32 #include <asm/current.h>
33 #include <asm/apicdef.h>
34 #include <asm/delay.h>
35 #include <linux/atomic.h>
36 #include <linux/jump_label.h>
37 #include "kvm_cache_regs.h"
45 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
47 #define mod_64(x, y) ((x) % (y))
55 /* 14 is the version for Xeon and Pentium 8.4.8*/
56 #define APIC_VERSION (0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
57 #define LAPIC_MMIO_LENGTH (1 << 12)
58 /* followed define is not in apicdef.h */
59 #define MAX_APIC_VECTOR 256
60 #define APIC_VECTORS_PER_REG 32
62 #define APIC_BROADCAST 0xFF
63 #define X2APIC_BROADCAST 0xFFFFFFFFul
65 static bool lapic_timer_advance_dynamic __read_mostly;
66 #define LAPIC_TIMER_ADVANCE_ADJUST_MIN 100 /* clock cycles */
67 #define LAPIC_TIMER_ADVANCE_ADJUST_MAX 10000 /* clock cycles */
68 #define LAPIC_TIMER_ADVANCE_NS_INIT 1000
69 #define LAPIC_TIMER_ADVANCE_NS_MAX 5000
70 /* step-by-step approximation to mitigate fluctuation */
71 #define LAPIC_TIMER_ADVANCE_ADJUST_STEP 8
73 static inline int apic_test_vector(int vec, void *bitmap)
75 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
78 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
80 struct kvm_lapic *apic = vcpu->arch.apic;
82 return apic_test_vector(vector, apic->regs + APIC_ISR) ||
83 apic_test_vector(vector, apic->regs + APIC_IRR);
86 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
88 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
91 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
93 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
96 struct static_key_deferred apic_hw_disabled __read_mostly;
97 struct static_key_deferred apic_sw_disabled __read_mostly;
99 static inline int apic_enabled(struct kvm_lapic *apic)
101 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
105 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
108 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
109 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
111 static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
113 return apic->vcpu->vcpu_id;
116 bool kvm_can_post_timer_interrupt(struct kvm_vcpu *vcpu)
118 return pi_inject_timer && kvm_vcpu_apicv_active(vcpu);
120 EXPORT_SYMBOL_GPL(kvm_can_post_timer_interrupt);
122 static bool kvm_use_posted_timer_interrupt(struct kvm_vcpu *vcpu)
124 return kvm_can_post_timer_interrupt(vcpu) && vcpu->mode == IN_GUEST_MODE;
127 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
128 u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
130 case KVM_APIC_MODE_X2APIC: {
131 u32 offset = (dest_id >> 16) * 16;
132 u32 max_apic_id = map->max_apic_id;
134 if (offset <= max_apic_id) {
135 u8 cluster_size = min(max_apic_id - offset + 1, 16U);
137 offset = array_index_nospec(offset, map->max_apic_id + 1);
138 *cluster = &map->phys_map[offset];
139 *mask = dest_id & (0xffff >> (16 - cluster_size));
146 case KVM_APIC_MODE_XAPIC_FLAT:
147 *cluster = map->xapic_flat_map;
148 *mask = dest_id & 0xff;
150 case KVM_APIC_MODE_XAPIC_CLUSTER:
151 *cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
152 *mask = dest_id & 0xf;
160 static void kvm_apic_map_free(struct rcu_head *rcu)
162 struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
167 static void recalculate_apic_map(struct kvm *kvm)
169 struct kvm_apic_map *new, *old = NULL;
170 struct kvm_vcpu *vcpu;
172 u32 max_id = 255; /* enough space for any xAPIC ID */
174 mutex_lock(&kvm->arch.apic_map_lock);
176 kvm_for_each_vcpu(i, vcpu, kvm)
177 if (kvm_apic_present(vcpu))
178 max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
180 new = kvzalloc(sizeof(struct kvm_apic_map) +
181 sizeof(struct kvm_lapic *) * ((u64)max_id + 1),
187 new->max_apic_id = max_id;
189 kvm_for_each_vcpu(i, vcpu, kvm) {
190 struct kvm_lapic *apic = vcpu->arch.apic;
191 struct kvm_lapic **cluster;
197 if (!kvm_apic_present(vcpu))
200 xapic_id = kvm_xapic_id(apic);
201 x2apic_id = kvm_x2apic_id(apic);
203 /* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
204 if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
205 x2apic_id <= new->max_apic_id)
206 new->phys_map[x2apic_id] = apic;
208 * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
209 * prevent them from masking VCPUs with APIC ID <= 0xff.
211 if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
212 new->phys_map[xapic_id] = apic;
214 if (!kvm_apic_sw_enabled(apic))
217 ldr = kvm_lapic_get_reg(apic, APIC_LDR);
219 if (apic_x2apic_mode(apic)) {
220 new->mode |= KVM_APIC_MODE_X2APIC;
222 ldr = GET_APIC_LOGICAL_ID(ldr);
223 if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
224 new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
226 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
229 if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
233 cluster[ffs(mask) - 1] = apic;
236 old = rcu_dereference_protected(kvm->arch.apic_map,
237 lockdep_is_held(&kvm->arch.apic_map_lock));
238 rcu_assign_pointer(kvm->arch.apic_map, new);
239 mutex_unlock(&kvm->arch.apic_map_lock);
242 call_rcu(&old->rcu, kvm_apic_map_free);
244 kvm_make_scan_ioapic_request(kvm);
247 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
249 bool enabled = val & APIC_SPIV_APIC_ENABLED;
251 kvm_lapic_set_reg(apic, APIC_SPIV, val);
253 if (enabled != apic->sw_enabled) {
254 apic->sw_enabled = enabled;
256 static_key_slow_dec_deferred(&apic_sw_disabled);
258 static_key_slow_inc(&apic_sw_disabled.key);
260 recalculate_apic_map(apic->vcpu->kvm);
264 static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
266 kvm_lapic_set_reg(apic, APIC_ID, id << 24);
267 recalculate_apic_map(apic->vcpu->kvm);
270 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
272 kvm_lapic_set_reg(apic, APIC_LDR, id);
273 recalculate_apic_map(apic->vcpu->kvm);
276 static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
278 return ((id >> 4) << 16) | (1 << (id & 0xf));
281 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
283 u32 ldr = kvm_apic_calc_x2apic_ldr(id);
285 WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
287 kvm_lapic_set_reg(apic, APIC_ID, id);
288 kvm_lapic_set_reg(apic, APIC_LDR, ldr);
289 recalculate_apic_map(apic->vcpu->kvm);
292 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
294 return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
297 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
299 return kvm_lapic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
302 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
304 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
307 static inline int apic_lvtt_period(struct kvm_lapic *apic)
309 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
312 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
314 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
317 static inline int apic_lvt_nmi_mode(u32 lvt_val)
319 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
322 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
324 struct kvm_lapic *apic = vcpu->arch.apic;
325 struct kvm_cpuid_entry2 *feat;
326 u32 v = APIC_VERSION;
328 if (!lapic_in_kernel(vcpu))
332 * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
333 * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
334 * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
335 * version first and level-triggered interrupts never get EOIed in
338 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
339 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))) &&
340 !ioapic_in_kernel(vcpu->kvm))
341 v |= APIC_LVR_DIRECTED_EOI;
342 kvm_lapic_set_reg(apic, APIC_LVR, v);
345 static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
346 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
347 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
348 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
349 LINT_MASK, LINT_MASK, /* LVT0-1 */
350 LVT_MASK /* LVTERR */
353 static int find_highest_vector(void *bitmap)
358 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
359 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
360 reg = bitmap + REG_POS(vec);
362 return __fls(*reg) + vec;
368 static u8 count_vectors(void *bitmap)
374 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
375 reg = bitmap + REG_POS(vec);
376 count += hweight32(*reg);
382 bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
385 u32 pir_val, irr_val, prev_irr_val;
388 max_updated_irr = -1;
391 for (i = vec = 0; i <= 7; i++, vec += 32) {
392 pir_val = READ_ONCE(pir[i]);
393 irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
395 prev_irr_val = irr_val;
396 irr_val |= xchg(&pir[i], 0);
397 *((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
398 if (prev_irr_val != irr_val) {
400 __fls(irr_val ^ prev_irr_val) + vec;
404 *max_irr = __fls(irr_val) + vec;
407 return ((max_updated_irr != -1) &&
408 (max_updated_irr == *max_irr));
410 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
412 bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
414 struct kvm_lapic *apic = vcpu->arch.apic;
416 return __kvm_apic_update_irr(pir, apic->regs, max_irr);
418 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
420 static inline int apic_search_irr(struct kvm_lapic *apic)
422 return find_highest_vector(apic->regs + APIC_IRR);
425 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
430 * Note that irr_pending is just a hint. It will be always
431 * true with virtual interrupt delivery enabled.
433 if (!apic->irr_pending)
436 result = apic_search_irr(apic);
437 ASSERT(result == -1 || result >= 16);
442 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
444 struct kvm_vcpu *vcpu;
448 if (unlikely(vcpu->arch.apicv_active)) {
449 /* need to update RVI */
450 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
451 kvm_x86_ops->hwapic_irr_update(vcpu,
452 apic_find_highest_irr(apic));
454 apic->irr_pending = false;
455 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
456 if (apic_search_irr(apic) != -1)
457 apic->irr_pending = true;
461 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
463 struct kvm_vcpu *vcpu;
465 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
471 * With APIC virtualization enabled, all caching is disabled
472 * because the processor can modify ISR under the hood. Instead
475 if (unlikely(vcpu->arch.apicv_active))
476 kvm_x86_ops->hwapic_isr_update(vcpu, vec);
479 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
481 * ISR (in service register) bit is set when injecting an interrupt.
482 * The highest vector is injected. Thus the latest bit set matches
483 * the highest bit in ISR.
485 apic->highest_isr_cache = vec;
489 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
494 * Note that isr_count is always 1, and highest_isr_cache
495 * is always -1, with APIC virtualization enabled.
497 if (!apic->isr_count)
499 if (likely(apic->highest_isr_cache != -1))
500 return apic->highest_isr_cache;
502 result = find_highest_vector(apic->regs + APIC_ISR);
503 ASSERT(result == -1 || result >= 16);
508 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
510 struct kvm_vcpu *vcpu;
511 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
517 * We do get here for APIC virtualization enabled if the guest
518 * uses the Hyper-V APIC enlightenment. In this case we may need
519 * to trigger a new interrupt delivery by writing the SVI field;
520 * on the other hand isr_count and highest_isr_cache are unused
521 * and must be left alone.
523 if (unlikely(vcpu->arch.apicv_active))
524 kvm_x86_ops->hwapic_isr_update(vcpu,
525 apic_find_highest_isr(apic));
528 BUG_ON(apic->isr_count < 0);
529 apic->highest_isr_cache = -1;
533 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
535 /* This may race with setting of irr in __apic_accept_irq() and
536 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
537 * will cause vmexit immediately and the value will be recalculated
538 * on the next vmentry.
540 return apic_find_highest_irr(vcpu->arch.apic);
542 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
544 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
545 int vector, int level, int trig_mode,
546 struct dest_map *dest_map);
548 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
549 struct dest_map *dest_map)
551 struct kvm_lapic *apic = vcpu->arch.apic;
553 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
554 irq->level, irq->trig_mode, dest_map);
557 static int __pv_send_ipi(unsigned long *ipi_bitmap, struct kvm_apic_map *map,
558 struct kvm_lapic_irq *irq, u32 min)
561 struct kvm_vcpu *vcpu;
563 if (min > map->max_apic_id)
566 for_each_set_bit(i, ipi_bitmap,
567 min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
568 if (map->phys_map[min + i]) {
569 vcpu = map->phys_map[min + i]->vcpu;
570 count += kvm_apic_set_irq(vcpu, irq, NULL);
577 int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
578 unsigned long ipi_bitmap_high, u32 min,
579 unsigned long icr, int op_64_bit)
581 struct kvm_apic_map *map;
582 struct kvm_lapic_irq irq = {0};
583 int cluster_size = op_64_bit ? 64 : 32;
586 if (icr & (APIC_DEST_MASK | APIC_SHORT_MASK))
589 irq.vector = icr & APIC_VECTOR_MASK;
590 irq.delivery_mode = icr & APIC_MODE_MASK;
591 irq.level = (icr & APIC_INT_ASSERT) != 0;
592 irq.trig_mode = icr & APIC_INT_LEVELTRIG;
595 map = rcu_dereference(kvm->arch.apic_map);
599 count = __pv_send_ipi(&ipi_bitmap_low, map, &irq, min);
601 count += __pv_send_ipi(&ipi_bitmap_high, map, &irq, min);
608 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
611 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
615 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
618 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
622 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
624 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
627 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
630 if (pv_eoi_get_user(vcpu, &val) < 0)
631 printk(KERN_WARNING "Can't read EOI MSR value: 0x%llx\n",
632 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
636 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
638 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
639 printk(KERN_WARNING "Can't set EOI MSR value: 0x%llx\n",
640 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
643 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
646 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
648 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
649 printk(KERN_WARNING "Can't clear EOI MSR value: 0x%llx\n",
650 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
653 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
656 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
659 if (apic->vcpu->arch.apicv_active)
660 highest_irr = kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
662 highest_irr = apic_find_highest_irr(apic);
663 if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
668 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
670 u32 tpr, isrv, ppr, old_ppr;
673 old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
674 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
675 isr = apic_find_highest_isr(apic);
676 isrv = (isr != -1) ? isr : 0;
678 if ((tpr & 0xf0) >= (isrv & 0xf0))
685 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
687 return ppr < old_ppr;
690 static void apic_update_ppr(struct kvm_lapic *apic)
694 if (__apic_update_ppr(apic, &ppr) &&
695 apic_has_interrupt_for_ppr(apic, ppr) != -1)
696 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
699 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
701 apic_update_ppr(vcpu->arch.apic);
703 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
705 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
707 kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
708 apic_update_ppr(apic);
711 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
713 return mda == (apic_x2apic_mode(apic) ?
714 X2APIC_BROADCAST : APIC_BROADCAST);
717 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
719 if (kvm_apic_broadcast(apic, mda))
722 if (apic_x2apic_mode(apic))
723 return mda == kvm_x2apic_id(apic);
726 * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
727 * it were in x2APIC mode. Hotplugged VCPUs start in xAPIC mode and
728 * this allows unique addressing of VCPUs with APIC ID over 0xff.
729 * The 0xff condition is needed because writeable xAPIC ID.
731 if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
734 return mda == kvm_xapic_id(apic);
737 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
741 if (kvm_apic_broadcast(apic, mda))
744 logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
746 if (apic_x2apic_mode(apic))
747 return ((logical_id >> 16) == (mda >> 16))
748 && (logical_id & mda & 0xffff) != 0;
750 logical_id = GET_APIC_LOGICAL_ID(logical_id);
752 switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
754 return (logical_id & mda) != 0;
755 case APIC_DFR_CLUSTER:
756 return ((logical_id >> 4) == (mda >> 4))
757 && (logical_id & mda & 0xf) != 0;
763 /* The KVM local APIC implementation has two quirks:
765 * - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
766 * in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
767 * KVM doesn't do that aliasing.
769 * - in-kernel IOAPIC messages have to be delivered directly to
770 * x2APIC, because the kernel does not support interrupt remapping.
771 * In order to support broadcast without interrupt remapping, x2APIC
772 * rewrites the destination of non-IPI messages from APIC_BROADCAST
773 * to X2APIC_BROADCAST.
775 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API. This is
776 * important when userspace wants to use x2APIC-format MSIs, because
777 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
779 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
780 struct kvm_lapic *source, struct kvm_lapic *target)
782 bool ipi = source != NULL;
784 if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
785 !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
786 return X2APIC_BROADCAST;
791 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
792 int shorthand, unsigned int dest, int dest_mode)
794 struct kvm_lapic *target = vcpu->arch.apic;
795 u32 mda = kvm_apic_mda(vcpu, dest, source, target);
799 case APIC_DEST_NOSHORT:
800 if (dest_mode == APIC_DEST_PHYSICAL)
801 return kvm_apic_match_physical_addr(target, mda);
803 return kvm_apic_match_logical_addr(target, mda);
805 return target == source;
806 case APIC_DEST_ALLINC:
808 case APIC_DEST_ALLBUT:
809 return target != source;
814 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
816 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
817 const unsigned long *bitmap, u32 bitmap_size)
822 mod = vector % dest_vcpus;
824 for (i = 0; i <= mod; i++) {
825 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
826 BUG_ON(idx == bitmap_size);
832 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
834 if (!kvm->arch.disabled_lapic_found) {
835 kvm->arch.disabled_lapic_found = true;
837 "Disabled LAPIC found during irq injection\n");
841 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
842 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
844 if (kvm->arch.x2apic_broadcast_quirk_disabled) {
845 if ((irq->dest_id == APIC_BROADCAST &&
846 map->mode != KVM_APIC_MODE_X2APIC))
848 if (irq->dest_id == X2APIC_BROADCAST)
851 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
852 if (irq->dest_id == (x2apic_ipi ?
853 X2APIC_BROADCAST : APIC_BROADCAST))
860 /* Return true if the interrupt can be handled by using *bitmap as index mask
861 * for valid destinations in *dst array.
862 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
863 * Note: we may have zero kvm_lapic destinations when we return true, which
864 * means that the interrupt should be dropped. In this case, *bitmap would be
865 * zero and *dst undefined.
867 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
868 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
869 struct kvm_apic_map *map, struct kvm_lapic ***dst,
870 unsigned long *bitmap)
874 if (irq->shorthand == APIC_DEST_SELF && src) {
878 } else if (irq->shorthand)
881 if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
884 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
885 if (irq->dest_id > map->max_apic_id) {
888 u32 dest_id = array_index_nospec(irq->dest_id, map->max_apic_id + 1);
889 *dst = &map->phys_map[dest_id];
896 if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
900 if (!kvm_lowest_prio_delivery(irq))
903 if (!kvm_vector_hashing_enabled()) {
905 for_each_set_bit(i, bitmap, 16) {
910 else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
911 (*dst)[lowest]->vcpu) < 0)
918 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
921 if (!(*dst)[lowest]) {
922 kvm_apic_disabled_lapic_found(kvm);
928 *bitmap = (lowest >= 0) ? 1 << lowest : 0;
933 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
934 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
936 struct kvm_apic_map *map;
937 unsigned long bitmap;
938 struct kvm_lapic **dst = NULL;
944 if (irq->shorthand == APIC_DEST_SELF) {
945 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
950 map = rcu_dereference(kvm->arch.apic_map);
952 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
955 for_each_set_bit(i, &bitmap, 16) {
958 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
967 * This routine tries to handle interrupts in posted mode, here is how
968 * it deals with different cases:
969 * - For single-destination interrupts, handle it in posted mode
970 * - Else if vector hashing is enabled and it is a lowest-priority
971 * interrupt, handle it in posted mode and use the following mechanism
972 * to find the destination vCPU.
973 * 1. For lowest-priority interrupts, store all the possible
974 * destination vCPUs in an array.
975 * 2. Use "guest vector % max number of destination vCPUs" to find
976 * the right destination vCPU in the array for the lowest-priority
978 * - Otherwise, use remapped mode to inject the interrupt.
980 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
981 struct kvm_vcpu **dest_vcpu)
983 struct kvm_apic_map *map;
984 unsigned long bitmap;
985 struct kvm_lapic **dst = NULL;
992 map = rcu_dereference(kvm->arch.apic_map);
994 if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
995 hweight16(bitmap) == 1) {
996 unsigned long i = find_first_bit(&bitmap, 16);
999 *dest_vcpu = dst[i]->vcpu;
1009 * Add a pending IRQ into lapic.
1010 * Return 1 if successfully added and 0 if discarded.
1012 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
1013 int vector, int level, int trig_mode,
1014 struct dest_map *dest_map)
1017 struct kvm_vcpu *vcpu = apic->vcpu;
1019 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
1021 switch (delivery_mode) {
1022 case APIC_DM_LOWEST:
1023 vcpu->arch.apic_arb_prio++;
1026 if (unlikely(trig_mode && !level))
1029 /* FIXME add logic for vcpu on reset */
1030 if (unlikely(!apic_enabled(apic)))
1036 __set_bit(vcpu->vcpu_id, dest_map->map);
1037 dest_map->vectors[vcpu->vcpu_id] = vector;
1040 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
1042 kvm_lapic_set_vector(vector,
1043 apic->regs + APIC_TMR);
1045 kvm_lapic_clear_vector(vector,
1046 apic->regs + APIC_TMR);
1049 if (vcpu->arch.apicv_active)
1050 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
1052 kvm_lapic_set_irr(vector, apic);
1054 kvm_make_request(KVM_REQ_EVENT, vcpu);
1055 kvm_vcpu_kick(vcpu);
1061 vcpu->arch.pv.pv_unhalted = 1;
1062 kvm_make_request(KVM_REQ_EVENT, vcpu);
1063 kvm_vcpu_kick(vcpu);
1068 kvm_make_request(KVM_REQ_SMI, vcpu);
1069 kvm_vcpu_kick(vcpu);
1074 kvm_inject_nmi(vcpu);
1075 kvm_vcpu_kick(vcpu);
1079 if (!trig_mode || level) {
1081 /* assumes that there are only KVM_APIC_INIT/SIPI */
1082 apic->pending_events = (1UL << KVM_APIC_INIT);
1083 kvm_make_request(KVM_REQ_EVENT, vcpu);
1084 kvm_vcpu_kick(vcpu);
1088 case APIC_DM_STARTUP:
1090 apic->sipi_vector = vector;
1091 /* make sure sipi_vector is visible for the receiver */
1093 set_bit(KVM_APIC_SIPI, &apic->pending_events);
1094 kvm_make_request(KVM_REQ_EVENT, vcpu);
1095 kvm_vcpu_kick(vcpu);
1098 case APIC_DM_EXTINT:
1100 * Should only be called by kvm_apic_local_deliver() with LVT0,
1101 * before NMI watchdog was enabled. Already handled by
1102 * kvm_apic_accept_pic_intr().
1107 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1115 * This routine identifies the destination vcpus mask meant to receive the
1116 * IOAPIC interrupts. It either uses kvm_apic_map_get_dest_lapic() to find
1117 * out the destination vcpus array and set the bitmap or it traverses to
1118 * each available vcpu to identify the same.
1120 void kvm_bitmap_or_dest_vcpus(struct kvm *kvm, struct kvm_lapic_irq *irq,
1121 unsigned long *vcpu_bitmap)
1123 struct kvm_lapic **dest_vcpu = NULL;
1124 struct kvm_lapic *src = NULL;
1125 struct kvm_apic_map *map;
1126 struct kvm_vcpu *vcpu;
1127 unsigned long bitmap;
1132 map = rcu_dereference(kvm->arch.apic_map);
1134 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dest_vcpu,
1137 for_each_set_bit(i, &bitmap, 16) {
1140 vcpu_idx = dest_vcpu[i]->vcpu->vcpu_idx;
1141 __set_bit(vcpu_idx, vcpu_bitmap);
1144 kvm_for_each_vcpu(i, vcpu, kvm) {
1145 if (!kvm_apic_present(vcpu))
1147 if (!kvm_apic_match_dest(vcpu, NULL,
1152 __set_bit(i, vcpu_bitmap);
1158 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1160 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1163 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1165 return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1168 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1172 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1173 if (!kvm_ioapic_handles_vector(apic, vector))
1176 /* Request a KVM exit to inform the userspace IOAPIC. */
1177 if (irqchip_split(apic->vcpu->kvm)) {
1178 apic->vcpu->arch.pending_ioapic_eoi = vector;
1179 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1183 if (apic_test_vector(vector, apic->regs + APIC_TMR))
1184 trigger_mode = IOAPIC_LEVEL_TRIG;
1186 trigger_mode = IOAPIC_EDGE_TRIG;
1188 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1191 static int apic_set_eoi(struct kvm_lapic *apic)
1193 int vector = apic_find_highest_isr(apic);
1195 trace_kvm_eoi(apic, vector);
1198 * Not every write EOI will has corresponding ISR,
1199 * one example is when Kernel check timer on setup_IO_APIC
1204 apic_clear_isr(vector, apic);
1205 apic_update_ppr(apic);
1207 if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1208 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1210 kvm_ioapic_send_eoi(apic, vector);
1211 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1216 * this interface assumes a trap-like exit, which has already finished
1217 * desired side effect including vISR and vPPR update.
1219 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1221 struct kvm_lapic *apic = vcpu->arch.apic;
1223 trace_kvm_eoi(apic, vector);
1225 kvm_ioapic_send_eoi(apic, vector);
1226 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1228 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1230 static void apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
1232 struct kvm_lapic_irq irq;
1234 irq.vector = icr_low & APIC_VECTOR_MASK;
1235 irq.delivery_mode = icr_low & APIC_MODE_MASK;
1236 irq.dest_mode = icr_low & APIC_DEST_MASK;
1237 irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1238 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1239 irq.shorthand = icr_low & APIC_SHORT_MASK;
1240 irq.msi_redir_hint = false;
1241 if (apic_x2apic_mode(apic))
1242 irq.dest_id = icr_high;
1244 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1246 trace_kvm_apic_ipi(icr_low, irq.dest_id);
1248 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1251 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1253 ktime_t remaining, now;
1257 ASSERT(apic != NULL);
1259 /* if initial count is 0, current count should also be 0 */
1260 if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1261 apic->lapic_timer.period == 0)
1265 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1266 if (ktime_to_ns(remaining) < 0)
1269 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1270 tmcct = div64_u64(ns,
1271 (APIC_BUS_CYCLE_NS * apic->divide_count));
1276 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1278 struct kvm_vcpu *vcpu = apic->vcpu;
1279 struct kvm_run *run = vcpu->run;
1281 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1282 run->tpr_access.rip = kvm_rip_read(vcpu);
1283 run->tpr_access.is_write = write;
1286 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1288 if (apic->vcpu->arch.tpr_access_reporting)
1289 __report_tpr_access(apic, write);
1292 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1296 if (offset >= LAPIC_MMIO_LENGTH)
1303 case APIC_TMCCT: /* Timer CCR */
1304 if (apic_lvtt_tscdeadline(apic))
1307 val = apic_get_tmcct(apic);
1310 apic_update_ppr(apic);
1311 val = kvm_lapic_get_reg(apic, offset);
1314 report_tpr_access(apic, false);
1317 val = kvm_lapic_get_reg(apic, offset);
1324 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1326 return container_of(dev, struct kvm_lapic, dev);
1329 #define APIC_REG_MASK(reg) (1ull << ((reg) >> 4))
1330 #define APIC_REGS_MASK(first, count) \
1331 (APIC_REG_MASK(first) * ((1ull << (count)) - 1))
1333 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1336 unsigned char alignment = offset & 0xf;
1338 /* this bitmask has a bit cleared for each reserved register */
1339 u64 valid_reg_mask =
1340 APIC_REG_MASK(APIC_ID) |
1341 APIC_REG_MASK(APIC_LVR) |
1342 APIC_REG_MASK(APIC_TASKPRI) |
1343 APIC_REG_MASK(APIC_PROCPRI) |
1344 APIC_REG_MASK(APIC_LDR) |
1345 APIC_REG_MASK(APIC_DFR) |
1346 APIC_REG_MASK(APIC_SPIV) |
1347 APIC_REGS_MASK(APIC_ISR, APIC_ISR_NR) |
1348 APIC_REGS_MASK(APIC_TMR, APIC_ISR_NR) |
1349 APIC_REGS_MASK(APIC_IRR, APIC_ISR_NR) |
1350 APIC_REG_MASK(APIC_ESR) |
1351 APIC_REG_MASK(APIC_ICR) |
1352 APIC_REG_MASK(APIC_ICR2) |
1353 APIC_REG_MASK(APIC_LVTT) |
1354 APIC_REG_MASK(APIC_LVTTHMR) |
1355 APIC_REG_MASK(APIC_LVTPC) |
1356 APIC_REG_MASK(APIC_LVT0) |
1357 APIC_REG_MASK(APIC_LVT1) |
1358 APIC_REG_MASK(APIC_LVTERR) |
1359 APIC_REG_MASK(APIC_TMICT) |
1360 APIC_REG_MASK(APIC_TMCCT) |
1361 APIC_REG_MASK(APIC_TDCR);
1363 /* ARBPRI is not valid on x2APIC */
1364 if (!apic_x2apic_mode(apic))
1365 valid_reg_mask |= APIC_REG_MASK(APIC_ARBPRI);
1367 if (offset > 0x3f0 || !(valid_reg_mask & APIC_REG_MASK(offset)))
1370 result = __apic_read(apic, offset & ~0xf);
1372 trace_kvm_apic_read(offset, result);
1378 memcpy(data, (char *)&result + alignment, len);
1381 printk(KERN_ERR "Local APIC read with len = %x, "
1382 "should be 1,2, or 4 instead\n", len);
1387 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1389 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1391 return addr >= apic->base_address &&
1392 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1395 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1396 gpa_t address, int len, void *data)
1398 struct kvm_lapic *apic = to_lapic(this);
1399 u32 offset = address - apic->base_address;
1401 if (!apic_mmio_in_range(apic, address))
1404 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1405 if (!kvm_check_has_quirk(vcpu->kvm,
1406 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1409 memset(data, 0xff, len);
1413 kvm_lapic_reg_read(apic, offset, len, data);
1418 static void update_divide_count(struct kvm_lapic *apic)
1420 u32 tmp1, tmp2, tdcr;
1422 tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1424 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1425 apic->divide_count = 0x1 << (tmp2 & 0x7);
1428 static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1431 * Do not allow the guest to program periodic timers with small
1432 * interval, since the hrtimers are not throttled by the host
1435 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1436 s64 min_period = min_timer_period_us * 1000LL;
1438 if (apic->lapic_timer.period < min_period) {
1439 pr_info_ratelimited(
1440 "kvm: vcpu %i: requested %lld ns "
1441 "lapic timer period limited to %lld ns\n",
1442 apic->vcpu->vcpu_id,
1443 apic->lapic_timer.period, min_period);
1444 apic->lapic_timer.period = min_period;
1449 static void apic_update_lvtt(struct kvm_lapic *apic)
1451 u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1452 apic->lapic_timer.timer_mode_mask;
1454 if (apic->lapic_timer.timer_mode != timer_mode) {
1455 if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1456 APIC_LVT_TIMER_TSCDEADLINE)) {
1457 hrtimer_cancel(&apic->lapic_timer.timer);
1458 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1459 apic->lapic_timer.period = 0;
1460 apic->lapic_timer.tscdeadline = 0;
1462 apic->lapic_timer.timer_mode = timer_mode;
1463 limit_periodic_timer_frequency(apic);
1468 * On APICv, this test will cause a busy wait
1469 * during a higher-priority task.
1472 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1474 struct kvm_lapic *apic = vcpu->arch.apic;
1475 u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1477 if (kvm_apic_hw_enabled(apic)) {
1478 int vec = reg & APIC_VECTOR_MASK;
1479 void *bitmap = apic->regs + APIC_ISR;
1481 if (vcpu->arch.apicv_active)
1482 bitmap = apic->regs + APIC_IRR;
1484 if (apic_test_vector(vec, bitmap))
1490 static inline void __wait_lapic_expire(struct kvm_vcpu *vcpu, u64 guest_cycles)
1492 u64 timer_advance_ns = vcpu->arch.apic->lapic_timer.timer_advance_ns;
1495 * If the guest TSC is running at a different ratio than the host, then
1496 * convert the delay to nanoseconds to achieve an accurate delay. Note
1497 * that __delay() uses delay_tsc whenever the hardware has TSC, thus
1498 * always for VMX enabled hardware.
1500 if (vcpu->arch.tsc_scaling_ratio == kvm_default_tsc_scaling_ratio) {
1501 __delay(min(guest_cycles,
1502 nsec_to_cycles(vcpu, timer_advance_ns)));
1504 u64 delay_ns = guest_cycles * 1000000ULL;
1505 do_div(delay_ns, vcpu->arch.virtual_tsc_khz);
1506 ndelay(min_t(u32, delay_ns, timer_advance_ns));
1510 static inline void adjust_lapic_timer_advance(struct kvm_vcpu *vcpu,
1511 s64 advance_expire_delta)
1513 struct kvm_lapic *apic = vcpu->arch.apic;
1514 u32 timer_advance_ns = apic->lapic_timer.timer_advance_ns;
1517 /* Do not adjust for tiny fluctuations or large random spikes. */
1518 if (abs(advance_expire_delta) > LAPIC_TIMER_ADVANCE_ADJUST_MAX ||
1519 abs(advance_expire_delta) < LAPIC_TIMER_ADVANCE_ADJUST_MIN)
1523 if (advance_expire_delta < 0) {
1524 ns = -advance_expire_delta * 1000000ULL;
1525 do_div(ns, vcpu->arch.virtual_tsc_khz);
1526 timer_advance_ns -= ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1529 ns = advance_expire_delta * 1000000ULL;
1530 do_div(ns, vcpu->arch.virtual_tsc_khz);
1531 timer_advance_ns += ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1534 if (unlikely(timer_advance_ns > LAPIC_TIMER_ADVANCE_NS_MAX))
1535 timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
1536 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
1539 static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1541 struct kvm_lapic *apic = vcpu->arch.apic;
1542 u64 guest_tsc, tsc_deadline;
1544 if (apic->lapic_timer.expired_tscdeadline == 0)
1547 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1548 apic->lapic_timer.expired_tscdeadline = 0;
1549 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1550 apic->lapic_timer.advance_expire_delta = guest_tsc - tsc_deadline;
1552 if (guest_tsc < tsc_deadline)
1553 __wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
1555 if (lapic_timer_advance_dynamic)
1556 adjust_lapic_timer_advance(vcpu, apic->lapic_timer.advance_expire_delta);
1559 void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1561 if (lapic_timer_int_injected(vcpu))
1562 __kvm_wait_lapic_expire(vcpu);
1564 EXPORT_SYMBOL_GPL(kvm_wait_lapic_expire);
1566 static void kvm_apic_inject_pending_timer_irqs(struct kvm_lapic *apic)
1568 struct kvm_timer *ktimer = &apic->lapic_timer;
1570 kvm_apic_local_deliver(apic, APIC_LVTT);
1571 if (apic_lvtt_tscdeadline(apic)) {
1572 ktimer->tscdeadline = 0;
1573 } else if (apic_lvtt_oneshot(apic)) {
1574 ktimer->tscdeadline = 0;
1575 ktimer->target_expiration = 0;
1579 static void apic_timer_expired(struct kvm_lapic *apic)
1581 struct kvm_vcpu *vcpu = apic->vcpu;
1582 struct kvm_timer *ktimer = &apic->lapic_timer;
1584 if (atomic_read(&apic->lapic_timer.pending))
1587 if (apic_lvtt_tscdeadline(apic) || ktimer->hv_timer_in_use)
1588 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1590 if (kvm_use_posted_timer_interrupt(apic->vcpu)) {
1591 if (apic->lapic_timer.timer_advance_ns)
1592 __kvm_wait_lapic_expire(vcpu);
1593 kvm_apic_inject_pending_timer_irqs(apic);
1597 atomic_inc(&apic->lapic_timer.pending);
1598 kvm_set_pending_timer(vcpu);
1601 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1603 struct kvm_timer *ktimer = &apic->lapic_timer;
1604 u64 guest_tsc, tscdeadline = ktimer->tscdeadline;
1607 struct kvm_vcpu *vcpu = apic->vcpu;
1608 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1609 unsigned long flags;
1612 if (unlikely(!tscdeadline || !this_tsc_khz))
1615 local_irq_save(flags);
1618 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1620 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1621 do_div(ns, this_tsc_khz);
1623 if (likely(tscdeadline > guest_tsc) &&
1624 likely(ns > apic->lapic_timer.timer_advance_ns)) {
1625 expire = ktime_add_ns(now, ns);
1626 expire = ktime_sub_ns(expire, ktimer->timer_advance_ns);
1627 hrtimer_start(&ktimer->timer, expire, HRTIMER_MODE_ABS_HARD);
1629 apic_timer_expired(apic);
1631 local_irq_restore(flags);
1634 static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1636 ktime_t now, remaining;
1637 u64 ns_remaining_old, ns_remaining_new;
1639 apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1640 * APIC_BUS_CYCLE_NS * apic->divide_count;
1641 limit_periodic_timer_frequency(apic);
1644 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1645 if (ktime_to_ns(remaining) < 0)
1648 ns_remaining_old = ktime_to_ns(remaining);
1649 ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1650 apic->divide_count, old_divisor);
1652 apic->lapic_timer.tscdeadline +=
1653 nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1654 nsec_to_cycles(apic->vcpu, ns_remaining_old);
1655 apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1658 static bool set_target_expiration(struct kvm_lapic *apic)
1664 apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1665 * APIC_BUS_CYCLE_NS * apic->divide_count;
1667 if (!apic->lapic_timer.period) {
1668 apic->lapic_timer.tscdeadline = 0;
1672 limit_periodic_timer_frequency(apic);
1674 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1675 nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1676 apic->lapic_timer.target_expiration = ktime_add_ns(now, apic->lapic_timer.period);
1681 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1683 ktime_t now = ktime_get();
1688 * Synchronize both deadlines to the same time source or
1689 * differences in the periods (caused by differences in the
1690 * underlying clocks or numerical approximation errors) will
1691 * cause the two to drift apart over time as the errors
1694 apic->lapic_timer.target_expiration =
1695 ktime_add_ns(apic->lapic_timer.target_expiration,
1696 apic->lapic_timer.period);
1697 delta = ktime_sub(apic->lapic_timer.target_expiration, now);
1698 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1699 nsec_to_cycles(apic->vcpu, delta);
1702 static void start_sw_period(struct kvm_lapic *apic)
1704 if (!apic->lapic_timer.period)
1707 if (ktime_after(ktime_get(),
1708 apic->lapic_timer.target_expiration)) {
1709 apic_timer_expired(apic);
1711 if (apic_lvtt_oneshot(apic))
1714 advance_periodic_target_expiration(apic);
1717 hrtimer_start(&apic->lapic_timer.timer,
1718 apic->lapic_timer.target_expiration,
1722 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1724 if (!lapic_in_kernel(vcpu))
1727 return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1729 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1731 static void cancel_hv_timer(struct kvm_lapic *apic)
1733 WARN_ON(preemptible());
1734 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1735 kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1736 apic->lapic_timer.hv_timer_in_use = false;
1739 static bool start_hv_timer(struct kvm_lapic *apic)
1741 struct kvm_timer *ktimer = &apic->lapic_timer;
1742 struct kvm_vcpu *vcpu = apic->vcpu;
1745 WARN_ON(preemptible());
1746 if (!kvm_x86_ops->set_hv_timer)
1749 if (!ktimer->tscdeadline)
1752 if (kvm_x86_ops->set_hv_timer(vcpu, ktimer->tscdeadline, &expired))
1755 ktimer->hv_timer_in_use = true;
1756 hrtimer_cancel(&ktimer->timer);
1759 * To simplify handling the periodic timer, leave the hv timer running
1760 * even if the deadline timer has expired, i.e. rely on the resulting
1761 * VM-Exit to recompute the periodic timer's target expiration.
1763 if (!apic_lvtt_period(apic)) {
1765 * Cancel the hv timer if the sw timer fired while the hv timer
1766 * was being programmed, or if the hv timer itself expired.
1768 if (atomic_read(&ktimer->pending)) {
1769 cancel_hv_timer(apic);
1770 } else if (expired) {
1771 apic_timer_expired(apic);
1772 cancel_hv_timer(apic);
1776 trace_kvm_hv_timer_state(vcpu->vcpu_id, ktimer->hv_timer_in_use);
1781 static void start_sw_timer(struct kvm_lapic *apic)
1783 struct kvm_timer *ktimer = &apic->lapic_timer;
1785 WARN_ON(preemptible());
1786 if (apic->lapic_timer.hv_timer_in_use)
1787 cancel_hv_timer(apic);
1788 if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1791 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1792 start_sw_period(apic);
1793 else if (apic_lvtt_tscdeadline(apic))
1794 start_sw_tscdeadline(apic);
1795 trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1798 static void restart_apic_timer(struct kvm_lapic *apic)
1802 if (!apic_lvtt_period(apic) && atomic_read(&apic->lapic_timer.pending))
1805 if (!start_hv_timer(apic))
1806 start_sw_timer(apic);
1811 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1813 struct kvm_lapic *apic = vcpu->arch.apic;
1816 /* If the preempt notifier has already run, it also called apic_timer_expired */
1817 if (!apic->lapic_timer.hv_timer_in_use)
1819 WARN_ON(swait_active(&vcpu->wq));
1820 cancel_hv_timer(apic);
1821 apic_timer_expired(apic);
1823 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1824 advance_periodic_target_expiration(apic);
1825 restart_apic_timer(apic);
1830 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1832 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1834 restart_apic_timer(vcpu->arch.apic);
1836 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1838 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1840 struct kvm_lapic *apic = vcpu->arch.apic;
1843 /* Possibly the TSC deadline timer is not enabled yet */
1844 if (apic->lapic_timer.hv_timer_in_use)
1845 start_sw_timer(apic);
1848 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1850 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
1852 struct kvm_lapic *apic = vcpu->arch.apic;
1854 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1855 restart_apic_timer(apic);
1858 static void start_apic_timer(struct kvm_lapic *apic)
1860 atomic_set(&apic->lapic_timer.pending, 0);
1862 if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1863 && !set_target_expiration(apic))
1866 restart_apic_timer(apic);
1869 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1871 bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1873 if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1874 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1875 if (lvt0_in_nmi_mode) {
1876 atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1878 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1882 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1886 trace_kvm_apic_write(reg, val);
1889 case APIC_ID: /* Local APIC ID */
1890 if (!apic_x2apic_mode(apic))
1891 kvm_apic_set_xapic_id(apic, val >> 24);
1897 report_tpr_access(apic, true);
1898 apic_set_tpr(apic, val & 0xff);
1906 if (!apic_x2apic_mode(apic))
1907 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1913 if (!apic_x2apic_mode(apic)) {
1914 kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1915 recalculate_apic_map(apic->vcpu->kvm);
1922 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1923 mask |= APIC_SPIV_DIRECTED_EOI;
1924 apic_set_spiv(apic, val & mask);
1925 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1929 for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
1930 lvt_val = kvm_lapic_get_reg(apic,
1931 APIC_LVTT + 0x10 * i);
1932 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
1933 lvt_val | APIC_LVT_MASKED);
1935 apic_update_lvtt(apic);
1936 atomic_set(&apic->lapic_timer.pending, 0);
1942 /* No delay here, so we always clear the pending bit */
1944 apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
1945 kvm_lapic_set_reg(apic, APIC_ICR, val);
1949 if (!apic_x2apic_mode(apic))
1951 kvm_lapic_set_reg(apic, APIC_ICR2, val);
1955 apic_manage_nmi_watchdog(apic, val);
1961 /* TODO: Check vector */
1965 if (!kvm_apic_sw_enabled(apic))
1966 val |= APIC_LVT_MASKED;
1967 size = ARRAY_SIZE(apic_lvt_mask);
1968 index = array_index_nospec(
1969 (reg - APIC_LVTT) >> 4, size);
1970 val &= apic_lvt_mask[index];
1971 kvm_lapic_set_reg(apic, reg, val);
1976 if (!kvm_apic_sw_enabled(apic))
1977 val |= APIC_LVT_MASKED;
1978 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1979 kvm_lapic_set_reg(apic, APIC_LVTT, val);
1980 apic_update_lvtt(apic);
1984 if (apic_lvtt_tscdeadline(apic))
1987 hrtimer_cancel(&apic->lapic_timer.timer);
1988 kvm_lapic_set_reg(apic, APIC_TMICT, val);
1989 start_apic_timer(apic);
1993 uint32_t old_divisor = apic->divide_count;
1995 kvm_lapic_set_reg(apic, APIC_TDCR, val);
1996 update_divide_count(apic);
1997 if (apic->divide_count != old_divisor &&
1998 apic->lapic_timer.period) {
1999 hrtimer_cancel(&apic->lapic_timer.timer);
2000 update_target_expiration(apic, old_divisor);
2001 restart_apic_timer(apic);
2006 if (apic_x2apic_mode(apic) && val != 0)
2011 if (apic_x2apic_mode(apic)) {
2012 kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
2023 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
2025 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
2026 gpa_t address, int len, const void *data)
2028 struct kvm_lapic *apic = to_lapic(this);
2029 unsigned int offset = address - apic->base_address;
2032 if (!apic_mmio_in_range(apic, address))
2035 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
2036 if (!kvm_check_has_quirk(vcpu->kvm,
2037 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
2044 * APIC register must be aligned on 128-bits boundary.
2045 * 32/64/128 bits registers must be accessed thru 32 bits.
2048 if (len != 4 || (offset & 0xf))
2053 kvm_lapic_reg_write(apic, offset & 0xff0, val);
2058 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
2060 kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
2062 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
2064 /* emulate APIC access in a trap manner */
2065 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
2069 /* hw has done the conditional check and inst decode */
2072 kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
2074 /* TODO: optimize to just emulate side effect w/o one more write */
2075 kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
2077 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
2079 void kvm_free_lapic(struct kvm_vcpu *vcpu)
2081 struct kvm_lapic *apic = vcpu->arch.apic;
2083 if (!vcpu->arch.apic)
2086 hrtimer_cancel(&apic->lapic_timer.timer);
2088 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
2089 static_key_slow_dec_deferred(&apic_hw_disabled);
2091 if (!apic->sw_enabled)
2092 static_key_slow_dec_deferred(&apic_sw_disabled);
2095 free_page((unsigned long)apic->regs);
2101 *----------------------------------------------------------------------
2103 *----------------------------------------------------------------------
2105 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
2107 struct kvm_lapic *apic = vcpu->arch.apic;
2109 if (!lapic_in_kernel(vcpu) ||
2110 !apic_lvtt_tscdeadline(apic))
2113 return apic->lapic_timer.tscdeadline;
2116 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
2118 struct kvm_lapic *apic = vcpu->arch.apic;
2120 if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
2121 apic_lvtt_period(apic))
2124 hrtimer_cancel(&apic->lapic_timer.timer);
2125 apic->lapic_timer.tscdeadline = data;
2126 start_apic_timer(apic);
2129 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
2131 struct kvm_lapic *apic = vcpu->arch.apic;
2133 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
2134 | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
2137 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
2141 tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
2143 return (tpr & 0xf0) >> 4;
2146 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
2148 u64 old_value = vcpu->arch.apic_base;
2149 struct kvm_lapic *apic = vcpu->arch.apic;
2152 value |= MSR_IA32_APICBASE_BSP;
2154 vcpu->arch.apic_base = value;
2156 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
2157 kvm_update_cpuid(vcpu);
2162 /* update jump label if enable bit changes */
2163 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
2164 if (value & MSR_IA32_APICBASE_ENABLE) {
2165 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2166 static_key_slow_dec_deferred(&apic_hw_disabled);
2168 static_key_slow_inc(&apic_hw_disabled.key);
2169 recalculate_apic_map(vcpu->kvm);
2173 if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
2174 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
2176 if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE))
2177 kvm_x86_ops->set_virtual_apic_mode(vcpu);
2179 apic->base_address = apic->vcpu->arch.apic_base &
2180 MSR_IA32_APICBASE_BASE;
2182 if ((value & MSR_IA32_APICBASE_ENABLE) &&
2183 apic->base_address != APIC_DEFAULT_PHYS_BASE)
2184 pr_warn_once("APIC base relocation is unsupported by KVM");
2187 void kvm_apic_update_apicv(struct kvm_vcpu *vcpu)
2189 struct kvm_lapic *apic = vcpu->arch.apic;
2191 if (vcpu->arch.apicv_active) {
2192 /* irr_pending is always true when apicv is activated. */
2193 apic->irr_pending = true;
2194 apic->isr_count = 1;
2196 apic->irr_pending = (apic_search_irr(apic) != -1);
2197 apic->isr_count = count_vectors(apic->regs + APIC_ISR);
2200 EXPORT_SYMBOL_GPL(kvm_apic_update_apicv);
2202 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2204 struct kvm_lapic *apic = vcpu->arch.apic;
2210 /* Stop the timer in case it's a reset to an active apic */
2211 hrtimer_cancel(&apic->lapic_timer.timer);
2214 kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
2215 MSR_IA32_APICBASE_ENABLE);
2216 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2218 kvm_apic_set_version(apic->vcpu);
2220 for (i = 0; i < KVM_APIC_LVT_NUM; i++)
2221 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
2222 apic_update_lvtt(apic);
2223 if (kvm_vcpu_is_reset_bsp(vcpu) &&
2224 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2225 kvm_lapic_set_reg(apic, APIC_LVT0,
2226 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2227 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2229 kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
2230 apic_set_spiv(apic, 0xff);
2231 kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2232 if (!apic_x2apic_mode(apic))
2233 kvm_apic_set_ldr(apic, 0);
2234 kvm_lapic_set_reg(apic, APIC_ESR, 0);
2235 kvm_lapic_set_reg(apic, APIC_ICR, 0);
2236 kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2237 kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2238 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2239 for (i = 0; i < 8; i++) {
2240 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2241 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2242 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2244 kvm_apic_update_apicv(vcpu);
2245 apic->highest_isr_cache = -1;
2246 update_divide_count(apic);
2247 atomic_set(&apic->lapic_timer.pending, 0);
2248 if (kvm_vcpu_is_bsp(vcpu))
2249 kvm_lapic_set_base(vcpu,
2250 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
2251 vcpu->arch.pv_eoi.msr_val = 0;
2252 apic_update_ppr(apic);
2253 if (vcpu->arch.apicv_active) {
2254 kvm_x86_ops->apicv_post_state_restore(vcpu);
2255 kvm_x86_ops->hwapic_irr_update(vcpu, -1);
2256 kvm_x86_ops->hwapic_isr_update(vcpu, -1);
2259 vcpu->arch.apic_arb_prio = 0;
2260 vcpu->arch.apic_attention = 0;
2264 *----------------------------------------------------------------------
2266 *----------------------------------------------------------------------
2269 static bool lapic_is_periodic(struct kvm_lapic *apic)
2271 return apic_lvtt_period(apic);
2274 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2276 struct kvm_lapic *apic = vcpu->arch.apic;
2278 if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2279 return atomic_read(&apic->lapic_timer.pending);
2284 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2286 u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2287 int vector, mode, trig_mode;
2289 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2290 vector = reg & APIC_VECTOR_MASK;
2291 mode = reg & APIC_MODE_MASK;
2292 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2293 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2299 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2301 struct kvm_lapic *apic = vcpu->arch.apic;
2304 kvm_apic_local_deliver(apic, APIC_LVT0);
2307 static const struct kvm_io_device_ops apic_mmio_ops = {
2308 .read = apic_mmio_read,
2309 .write = apic_mmio_write,
2312 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2314 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2315 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2317 apic_timer_expired(apic);
2319 if (lapic_is_periodic(apic)) {
2320 advance_periodic_target_expiration(apic);
2321 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2322 return HRTIMER_RESTART;
2324 return HRTIMER_NORESTART;
2327 int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
2329 struct kvm_lapic *apic;
2331 ASSERT(vcpu != NULL);
2333 apic = kzalloc(sizeof(*apic), GFP_KERNEL_ACCOUNT);
2337 vcpu->arch.apic = apic;
2339 apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
2341 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2343 goto nomem_free_apic;
2347 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2348 HRTIMER_MODE_ABS_HARD);
2349 apic->lapic_timer.timer.function = apic_timer_fn;
2350 if (timer_advance_ns == -1) {
2351 apic->lapic_timer.timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
2352 lapic_timer_advance_dynamic = true;
2354 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
2355 lapic_timer_advance_dynamic = false;
2359 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2360 * thinking that APIC state has changed.
2362 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2363 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2364 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2369 vcpu->arch.apic = NULL;
2374 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2376 struct kvm_lapic *apic = vcpu->arch.apic;
2379 if (!kvm_apic_hw_enabled(apic))
2382 __apic_update_ppr(apic, &ppr);
2383 return apic_has_interrupt_for_ppr(apic, ppr);
2386 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2388 u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2390 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2392 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2393 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2398 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2400 struct kvm_lapic *apic = vcpu->arch.apic;
2402 if (atomic_read(&apic->lapic_timer.pending) > 0) {
2403 kvm_apic_inject_pending_timer_irqs(apic);
2404 atomic_set(&apic->lapic_timer.pending, 0);
2408 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2410 int vector = kvm_apic_has_interrupt(vcpu);
2411 struct kvm_lapic *apic = vcpu->arch.apic;
2418 * We get here even with APIC virtualization enabled, if doing
2419 * nested virtualization and L1 runs with the "acknowledge interrupt
2420 * on exit" mode. Then we cannot inject the interrupt via RVI,
2421 * because the process would deliver it through the IDT.
2424 apic_clear_irr(vector, apic);
2425 if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2427 * For auto-EOI interrupts, there might be another pending
2428 * interrupt above PPR, so check whether to raise another
2431 apic_update_ppr(apic);
2434 * For normal interrupts, PPR has been raised and there cannot
2435 * be a higher-priority pending interrupt---except if there was
2436 * a concurrent interrupt injection, but that would have
2437 * triggered KVM_REQ_EVENT already.
2439 apic_set_isr(vector, apic);
2440 __apic_update_ppr(apic, &ppr);
2446 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2447 struct kvm_lapic_state *s, bool set)
2449 if (apic_x2apic_mode(vcpu->arch.apic)) {
2450 u32 *id = (u32 *)(s->regs + APIC_ID);
2451 u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2453 if (vcpu->kvm->arch.x2apic_format) {
2454 if (*id != vcpu->vcpu_id)
2463 /* In x2APIC mode, the LDR is fixed and based on the id */
2465 *ldr = kvm_apic_calc_x2apic_ldr(*id);
2471 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2473 memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2474 return kvm_apic_state_fixup(vcpu, s, false);
2477 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2479 struct kvm_lapic *apic = vcpu->arch.apic;
2483 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2484 /* set SPIV separately to get count of SW disabled APICs right */
2485 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2487 r = kvm_apic_state_fixup(vcpu, s, true);
2490 memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
2492 recalculate_apic_map(vcpu->kvm);
2493 kvm_apic_set_version(vcpu);
2495 apic_update_ppr(apic);
2496 hrtimer_cancel(&apic->lapic_timer.timer);
2497 apic_update_lvtt(apic);
2498 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2499 update_divide_count(apic);
2500 start_apic_timer(apic);
2501 kvm_apic_update_apicv(vcpu);
2502 apic->highest_isr_cache = -1;
2503 if (vcpu->arch.apicv_active) {
2504 kvm_x86_ops->apicv_post_state_restore(vcpu);
2505 kvm_x86_ops->hwapic_irr_update(vcpu,
2506 apic_find_highest_irr(apic));
2507 kvm_x86_ops->hwapic_isr_update(vcpu,
2508 apic_find_highest_isr(apic));
2510 kvm_make_request(KVM_REQ_EVENT, vcpu);
2511 if (ioapic_in_kernel(vcpu->kvm))
2512 kvm_rtc_eoi_tracking_restore_one(vcpu);
2514 vcpu->arch.apic_arb_prio = 0;
2519 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2521 struct hrtimer *timer;
2523 if (!lapic_in_kernel(vcpu) ||
2524 kvm_can_post_timer_interrupt(vcpu))
2527 timer = &vcpu->arch.apic->lapic_timer.timer;
2528 if (hrtimer_cancel(timer))
2529 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_HARD);
2533 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2535 * Detect whether guest triggered PV EOI since the
2536 * last entry. If yes, set EOI on guests's behalf.
2537 * Clear PV EOI in guest memory in any case.
2539 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2540 struct kvm_lapic *apic)
2545 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2546 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2548 * KVM_APIC_PV_EOI_PENDING is unset:
2549 * -> host disabled PV EOI.
2550 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2551 * -> host enabled PV EOI, guest did not execute EOI yet.
2552 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2553 * -> host enabled PV EOI, guest executed EOI.
2555 BUG_ON(!pv_eoi_enabled(vcpu));
2556 pending = pv_eoi_get_pending(vcpu);
2558 * Clear pending bit in any case: it will be set again on vmentry.
2559 * While this might not be ideal from performance point of view,
2560 * this makes sure pv eoi is only enabled when we know it's safe.
2562 pv_eoi_clr_pending(vcpu);
2565 vector = apic_set_eoi(apic);
2566 trace_kvm_pv_eoi(apic, vector);
2569 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2573 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2574 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2576 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2579 if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2583 apic_set_tpr(vcpu->arch.apic, data & 0xff);
2587 * apic_sync_pv_eoi_to_guest - called before vmentry
2589 * Detect whether it's safe to enable PV EOI and
2592 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2593 struct kvm_lapic *apic)
2595 if (!pv_eoi_enabled(vcpu) ||
2596 /* IRR set or many bits in ISR: could be nested. */
2597 apic->irr_pending ||
2598 /* Cache not set: could be safe but we don't bother. */
2599 apic->highest_isr_cache == -1 ||
2600 /* Need EOI to update ioapic. */
2601 kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2603 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2604 * so we need not do anything here.
2609 pv_eoi_set_pending(apic->vcpu);
2612 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2615 int max_irr, max_isr;
2616 struct kvm_lapic *apic = vcpu->arch.apic;
2618 apic_sync_pv_eoi_to_guest(vcpu, apic);
2620 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2623 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2624 max_irr = apic_find_highest_irr(apic);
2627 max_isr = apic_find_highest_isr(apic);
2630 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2632 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2636 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2639 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2640 &vcpu->arch.apic->vapic_cache,
2641 vapic_addr, sizeof(u32)))
2643 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2645 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2648 vcpu->arch.apic->vapic_addr = vapic_addr;
2652 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2654 struct kvm_lapic *apic = vcpu->arch.apic;
2655 u32 reg = (msr - APIC_BASE_MSR) << 4;
2657 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2660 if (reg == APIC_ICR2)
2663 /* if this is ICR write vector before command */
2664 if (reg == APIC_ICR)
2665 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2666 return kvm_lapic_reg_write(apic, reg, (u32)data);
2669 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2671 struct kvm_lapic *apic = vcpu->arch.apic;
2672 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2674 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2677 if (reg == APIC_DFR || reg == APIC_ICR2)
2680 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2682 if (reg == APIC_ICR)
2683 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2685 *data = (((u64)high) << 32) | low;
2690 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2692 struct kvm_lapic *apic = vcpu->arch.apic;
2694 if (!lapic_in_kernel(vcpu))
2697 /* if this is ICR write vector before command */
2698 if (reg == APIC_ICR)
2699 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2700 return kvm_lapic_reg_write(apic, reg, (u32)data);
2703 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2705 struct kvm_lapic *apic = vcpu->arch.apic;
2708 if (!lapic_in_kernel(vcpu))
2711 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2713 if (reg == APIC_ICR)
2714 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2716 *data = (((u64)high) << 32) | low;
2721 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data, unsigned long len)
2723 u64 addr = data & ~KVM_MSR_ENABLED;
2724 struct gfn_to_hva_cache *ghc = &vcpu->arch.pv_eoi.data;
2725 unsigned long new_len;
2727 if (!IS_ALIGNED(addr, 4))
2730 vcpu->arch.pv_eoi.msr_val = data;
2731 if (!pv_eoi_enabled(vcpu))
2734 if (addr == ghc->gpa && len <= ghc->len)
2739 return kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, addr, new_len);
2742 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2744 struct kvm_lapic *apic = vcpu->arch.apic;
2748 if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2752 * INITs are latched while CPU is in specific states
2753 * (SMM, VMX non-root mode, SVM with GIF=0).
2754 * Because a CPU cannot be in these states immediately
2755 * after it has processed an INIT signal (and thus in
2756 * KVM_MP_STATE_INIT_RECEIVED state), just eat SIPIs
2757 * and leave the INIT pending.
2759 if (kvm_vcpu_latch_init(vcpu)) {
2760 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2761 if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2762 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2766 pe = xchg(&apic->pending_events, 0);
2767 if (test_bit(KVM_APIC_INIT, &pe)) {
2768 kvm_vcpu_reset(vcpu, true);
2769 if (kvm_vcpu_is_bsp(apic->vcpu))
2770 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2772 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2774 if (test_bit(KVM_APIC_SIPI, &pe) &&
2775 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2776 /* evaluate pending_events before reading the vector */
2778 sipi_vector = apic->sipi_vector;
2779 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2780 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2784 void kvm_lapic_init(void)
2786 /* do not patch jump label more than once per second */
2787 jump_label_rate_limit(&apic_hw_disabled, HZ);
2788 jump_label_rate_limit(&apic_sw_disabled, HZ);
2791 void kvm_lapic_exit(void)
2793 static_key_deferred_flush(&apic_hw_disabled);
2794 static_key_deferred_flush(&apic_sw_disabled);