3 * Local APIC virtualization
5 * Copyright (C) 2006 Qumranet, Inc.
6 * Copyright (C) 2007 Novell
7 * Copyright (C) 2007 Intel
8 * Copyright 2009 Red Hat, Inc. and/or its affiliates.
11 * Dor Laor <dor.laor@qumranet.com>
12 * Gregory Haskins <ghaskins@novell.com>
13 * Yaozu (Eddie) Dong <eddie.dong@intel.com>
15 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
24 #include <linux/highmem.h>
25 #include <linux/smp.h>
26 #include <linux/hrtimer.h>
28 #include <linux/export.h>
29 #include <linux/math64.h>
30 #include <linux/slab.h>
31 #include <asm/processor.h>
34 #include <asm/current.h>
35 #include <asm/apicdef.h>
36 #include <asm/delay.h>
37 #include <linux/atomic.h>
38 #include <linux/jump_label.h>
39 #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 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
58 #define apic_debug(fmt, arg...) do {} while (0)
60 /* 14 is the version for Xeon and Pentium 8.4.8*/
61 #define APIC_VERSION (0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
62 #define LAPIC_MMIO_LENGTH (1 << 12)
63 /* followed define is not in apicdef.h */
64 #define APIC_SHORT_MASK 0xc0000
65 #define APIC_DEST_NOSHORT 0x0
66 #define APIC_DEST_MASK 0x800
67 #define MAX_APIC_VECTOR 256
68 #define APIC_VECTORS_PER_REG 32
70 #define APIC_BROADCAST 0xFF
71 #define X2APIC_BROADCAST 0xFFFFFFFFul
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 void apic_clear_vector(int vec, void *bitmap)
88 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
91 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
93 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
96 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
98 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
101 struct static_key_deferred apic_hw_disabled __read_mostly;
102 struct static_key_deferred apic_sw_disabled __read_mostly;
104 static inline int apic_enabled(struct kvm_lapic *apic)
106 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
110 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
113 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
114 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
116 static inline u8 kvm_xapic_id(struct kvm_lapic *apic)
118 return kvm_lapic_get_reg(apic, APIC_ID) >> 24;
121 static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
123 return apic->vcpu->vcpu_id;
126 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
127 u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
129 case KVM_APIC_MODE_X2APIC: {
130 u32 offset = (dest_id >> 16) * 16;
131 u32 max_apic_id = map->max_apic_id;
133 if (offset <= max_apic_id) {
134 u8 cluster_size = min(max_apic_id - offset + 1, 16U);
136 offset = array_index_nospec(offset, map->max_apic_id + 1);
137 *cluster = &map->phys_map[offset];
138 *mask = dest_id & (0xffff >> (16 - cluster_size));
145 case KVM_APIC_MODE_XAPIC_FLAT:
146 *cluster = map->xapic_flat_map;
147 *mask = dest_id & 0xff;
149 case KVM_APIC_MODE_XAPIC_CLUSTER:
150 *cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
151 *mask = dest_id & 0xf;
159 static void kvm_apic_map_free(struct rcu_head *rcu)
161 struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
166 static void recalculate_apic_map(struct kvm *kvm)
168 struct kvm_apic_map *new, *old = NULL;
169 struct kvm_vcpu *vcpu;
171 u32 max_id = 255; /* enough space for any xAPIC ID */
173 mutex_lock(&kvm->arch.apic_map_lock);
175 kvm_for_each_vcpu(i, vcpu, kvm)
176 if (kvm_apic_present(vcpu))
177 max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
179 new = kvzalloc(sizeof(struct kvm_apic_map) +
180 sizeof(struct kvm_lapic *) * ((u64)max_id + 1), GFP_KERNEL);
185 new->max_apic_id = max_id;
187 kvm_for_each_vcpu(i, vcpu, kvm) {
188 struct kvm_lapic *apic = vcpu->arch.apic;
189 struct kvm_lapic **cluster;
195 if (!kvm_apic_present(vcpu))
198 xapic_id = kvm_xapic_id(apic);
199 x2apic_id = kvm_x2apic_id(apic);
201 /* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
202 if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
203 x2apic_id <= new->max_apic_id)
204 new->phys_map[x2apic_id] = apic;
206 * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
207 * prevent them from masking VCPUs with APIC ID <= 0xff.
209 if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
210 new->phys_map[xapic_id] = apic;
212 if (!kvm_apic_sw_enabled(apic))
215 ldr = kvm_lapic_get_reg(apic, APIC_LDR);
217 if (apic_x2apic_mode(apic)) {
218 new->mode |= KVM_APIC_MODE_X2APIC;
220 ldr = GET_APIC_LOGICAL_ID(ldr);
221 if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
222 new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
224 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
227 if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
231 cluster[ffs(mask) - 1] = apic;
234 old = rcu_dereference_protected(kvm->arch.apic_map,
235 lockdep_is_held(&kvm->arch.apic_map_lock));
236 rcu_assign_pointer(kvm->arch.apic_map, new);
237 mutex_unlock(&kvm->arch.apic_map_lock);
240 call_rcu(&old->rcu, kvm_apic_map_free);
242 kvm_make_scan_ioapic_request(kvm);
245 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
247 bool enabled = val & APIC_SPIV_APIC_ENABLED;
249 kvm_lapic_set_reg(apic, APIC_SPIV, val);
251 if (enabled != apic->sw_enabled) {
252 apic->sw_enabled = enabled;
254 static_key_slow_dec_deferred(&apic_sw_disabled);
255 recalculate_apic_map(apic->vcpu->kvm);
257 static_key_slow_inc(&apic_sw_disabled.key);
259 recalculate_apic_map(apic->vcpu->kvm);
263 static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
265 kvm_lapic_set_reg(apic, APIC_ID, id << 24);
266 recalculate_apic_map(apic->vcpu->kvm);
269 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
271 kvm_lapic_set_reg(apic, APIC_LDR, id);
272 recalculate_apic_map(apic->vcpu->kvm);
275 static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
277 return ((id >> 4) << 16) | (1 << (id & 0xf));
280 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
282 u32 ldr = kvm_apic_calc_x2apic_ldr(id);
284 WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
286 kvm_lapic_set_reg(apic, APIC_ID, id);
287 kvm_lapic_set_reg(apic, APIC_LDR, ldr);
288 recalculate_apic_map(apic->vcpu->kvm);
291 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
293 return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
296 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
298 return kvm_lapic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
301 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
303 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
306 static inline int apic_lvtt_period(struct kvm_lapic *apic)
308 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
311 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
313 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
316 static inline int apic_lvt_nmi_mode(u32 lvt_val)
318 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
321 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
323 struct kvm_lapic *apic = vcpu->arch.apic;
324 struct kvm_cpuid_entry2 *feat;
325 u32 v = APIC_VERSION;
327 if (!lapic_in_kernel(vcpu))
331 * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
332 * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
333 * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
334 * version first and level-triggered interrupts never get EOIed in
337 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
338 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))) &&
339 !ioapic_in_kernel(vcpu->kvm))
340 v |= APIC_LVR_DIRECTED_EOI;
341 kvm_lapic_set_reg(apic, APIC_LVR, v);
344 static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
345 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
346 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
347 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
348 LINT_MASK, LINT_MASK, /* LVT0-1 */
349 LVT_MASK /* LVTERR */
352 static int find_highest_vector(void *bitmap)
357 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
358 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
359 reg = bitmap + REG_POS(vec);
361 return __fls(*reg) + vec;
367 static u8 count_vectors(void *bitmap)
373 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
374 reg = bitmap + REG_POS(vec);
375 count += hweight32(*reg);
381 bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
384 u32 pir_val, irr_val, prev_irr_val;
387 max_updated_irr = -1;
390 for (i = vec = 0; i <= 7; i++, vec += 32) {
391 pir_val = READ_ONCE(pir[i]);
392 irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
394 prev_irr_val = irr_val;
395 irr_val |= xchg(&pir[i], 0);
396 *((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
397 if (prev_irr_val != irr_val) {
399 __fls(irr_val ^ prev_irr_val) + vec;
403 *max_irr = __fls(irr_val) + vec;
406 return ((max_updated_irr != -1) &&
407 (max_updated_irr == *max_irr));
409 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
411 bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
413 struct kvm_lapic *apic = vcpu->arch.apic;
415 return __kvm_apic_update_irr(pir, apic->regs, max_irr);
417 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
419 static inline int apic_search_irr(struct kvm_lapic *apic)
421 return find_highest_vector(apic->regs + APIC_IRR);
424 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
429 * Note that irr_pending is just a hint. It will be always
430 * true with virtual interrupt delivery enabled.
432 if (!apic->irr_pending)
435 result = apic_search_irr(apic);
436 ASSERT(result == -1 || result >= 16);
441 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
443 struct kvm_vcpu *vcpu;
447 if (unlikely(vcpu->arch.apicv_active)) {
448 /* need to update RVI */
449 apic_clear_vector(vec, apic->regs + APIC_IRR);
450 kvm_x86_ops->hwapic_irr_update(vcpu,
451 apic_find_highest_irr(apic));
453 apic->irr_pending = false;
454 apic_clear_vector(vec, apic->regs + APIC_IRR);
455 if (apic_search_irr(apic) != -1)
456 apic->irr_pending = true;
460 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
462 struct kvm_vcpu *vcpu;
464 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
470 * With APIC virtualization enabled, all caching is disabled
471 * because the processor can modify ISR under the hood. Instead
474 if (unlikely(vcpu->arch.apicv_active))
475 kvm_x86_ops->hwapic_isr_update(vcpu, vec);
478 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
480 * ISR (in service register) bit is set when injecting an interrupt.
481 * The highest vector is injected. Thus the latest bit set matches
482 * the highest bit in ISR.
484 apic->highest_isr_cache = vec;
488 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
493 * Note that isr_count is always 1, and highest_isr_cache
494 * is always -1, with APIC virtualization enabled.
496 if (!apic->isr_count)
498 if (likely(apic->highest_isr_cache != -1))
499 return apic->highest_isr_cache;
501 result = find_highest_vector(apic->regs + APIC_ISR);
502 ASSERT(result == -1 || result >= 16);
507 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
509 struct kvm_vcpu *vcpu;
510 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
516 * We do get here for APIC virtualization enabled if the guest
517 * uses the Hyper-V APIC enlightenment. In this case we may need
518 * to trigger a new interrupt delivery by writing the SVI field;
519 * on the other hand isr_count and highest_isr_cache are unused
520 * and must be left alone.
522 if (unlikely(vcpu->arch.apicv_active))
523 kvm_x86_ops->hwapic_isr_update(vcpu,
524 apic_find_highest_isr(apic));
527 BUG_ON(apic->isr_count < 0);
528 apic->highest_isr_cache = -1;
532 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
534 /* This may race with setting of irr in __apic_accept_irq() and
535 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
536 * will cause vmexit immediately and the value will be recalculated
537 * on the next vmentry.
539 return apic_find_highest_irr(vcpu->arch.apic);
541 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
543 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
544 int vector, int level, int trig_mode,
545 struct dest_map *dest_map);
547 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
548 struct dest_map *dest_map)
550 struct kvm_lapic *apic = vcpu->arch.apic;
552 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
553 irq->level, irq->trig_mode, dest_map);
556 int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
557 unsigned long ipi_bitmap_high, u32 min,
558 unsigned long icr, int op_64_bit)
561 struct kvm_apic_map *map;
562 struct kvm_vcpu *vcpu;
563 struct kvm_lapic_irq irq = {0};
564 int cluster_size = op_64_bit ? 64 : 32;
567 irq.vector = icr & APIC_VECTOR_MASK;
568 irq.delivery_mode = icr & APIC_MODE_MASK;
569 irq.level = (icr & APIC_INT_ASSERT) != 0;
570 irq.trig_mode = icr & APIC_INT_LEVELTRIG;
572 if (icr & APIC_DEST_MASK)
574 if (icr & APIC_SHORT_MASK)
578 map = rcu_dereference(kvm->arch.apic_map);
580 if (unlikely(!map)) {
585 if (min > map->max_apic_id)
587 /* Bits above cluster_size are masked in the caller. */
588 for_each_set_bit(i, &ipi_bitmap_low,
589 min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
590 if (map->phys_map[min + i]) {
591 vcpu = map->phys_map[min + i]->vcpu;
592 count += kvm_apic_set_irq(vcpu, &irq, NULL);
598 if (min > map->max_apic_id)
601 for_each_set_bit(i, &ipi_bitmap_high,
602 min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
603 if (map->phys_map[min + i]) {
604 vcpu = map->phys_map[min + i]->vcpu;
605 count += kvm_apic_set_irq(vcpu, &irq, NULL);
614 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
617 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
621 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
624 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
628 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
630 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
633 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
636 if (pv_eoi_get_user(vcpu, &val) < 0) {
637 apic_debug("Can't read EOI MSR value: 0x%llx\n",
638 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
644 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
646 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
647 apic_debug("Can't set EOI MSR value: 0x%llx\n",
648 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
651 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
654 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
656 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
657 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
658 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
661 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
664 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
667 if (apic->vcpu->arch.apicv_active)
668 highest_irr = kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
670 highest_irr = apic_find_highest_irr(apic);
671 if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
676 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
678 u32 tpr, isrv, ppr, old_ppr;
681 old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
682 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
683 isr = apic_find_highest_isr(apic);
684 isrv = (isr != -1) ? isr : 0;
686 if ((tpr & 0xf0) >= (isrv & 0xf0))
691 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
692 apic, ppr, isr, isrv);
696 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
698 return ppr < old_ppr;
701 static void apic_update_ppr(struct kvm_lapic *apic)
705 if (__apic_update_ppr(apic, &ppr) &&
706 apic_has_interrupt_for_ppr(apic, ppr) != -1)
707 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
710 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
712 apic_update_ppr(vcpu->arch.apic);
714 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
716 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
718 kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
719 apic_update_ppr(apic);
722 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
724 return mda == (apic_x2apic_mode(apic) ?
725 X2APIC_BROADCAST : APIC_BROADCAST);
728 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
730 if (kvm_apic_broadcast(apic, mda))
733 if (apic_x2apic_mode(apic))
734 return mda == kvm_x2apic_id(apic);
737 * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
738 * it were in x2APIC mode. Hotplugged VCPUs start in xAPIC mode and
739 * this allows unique addressing of VCPUs with APIC ID over 0xff.
740 * The 0xff condition is needed because writeable xAPIC ID.
742 if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
745 return mda == kvm_xapic_id(apic);
748 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
752 if (kvm_apic_broadcast(apic, mda))
755 logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
757 if (apic_x2apic_mode(apic))
758 return ((logical_id >> 16) == (mda >> 16))
759 && (logical_id & mda & 0xffff) != 0;
761 logical_id = GET_APIC_LOGICAL_ID(logical_id);
763 switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
765 return (logical_id & mda) != 0;
766 case APIC_DFR_CLUSTER:
767 return ((logical_id >> 4) == (mda >> 4))
768 && (logical_id & mda & 0xf) != 0;
770 apic_debug("Bad DFR vcpu %d: %08x\n",
771 apic->vcpu->vcpu_id, kvm_lapic_get_reg(apic, APIC_DFR));
776 /* The KVM local APIC implementation has two quirks:
778 * - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
779 * in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
780 * KVM doesn't do that aliasing.
782 * - in-kernel IOAPIC messages have to be delivered directly to
783 * x2APIC, because the kernel does not support interrupt remapping.
784 * In order to support broadcast without interrupt remapping, x2APIC
785 * rewrites the destination of non-IPI messages from APIC_BROADCAST
786 * to X2APIC_BROADCAST.
788 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API. This is
789 * important when userspace wants to use x2APIC-format MSIs, because
790 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
792 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
793 struct kvm_lapic *source, struct kvm_lapic *target)
795 bool ipi = source != NULL;
797 if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
798 !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
799 return X2APIC_BROADCAST;
804 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
805 int short_hand, unsigned int dest, int dest_mode)
807 struct kvm_lapic *target = vcpu->arch.apic;
808 u32 mda = kvm_apic_mda(vcpu, dest, source, target);
810 apic_debug("target %p, source %p, dest 0x%x, "
811 "dest_mode 0x%x, short_hand 0x%x\n",
812 target, source, dest, dest_mode, short_hand);
815 switch (short_hand) {
816 case APIC_DEST_NOSHORT:
817 if (dest_mode == APIC_DEST_PHYSICAL)
818 return kvm_apic_match_physical_addr(target, mda);
820 return kvm_apic_match_logical_addr(target, mda);
822 return target == source;
823 case APIC_DEST_ALLINC:
825 case APIC_DEST_ALLBUT:
826 return target != source;
828 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
833 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
835 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
836 const unsigned long *bitmap, u32 bitmap_size)
841 mod = vector % dest_vcpus;
843 for (i = 0; i <= mod; i++) {
844 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
845 BUG_ON(idx == bitmap_size);
851 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
853 if (!kvm->arch.disabled_lapic_found) {
854 kvm->arch.disabled_lapic_found = true;
856 "Disabled LAPIC found during irq injection\n");
860 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
861 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
863 if (kvm->arch.x2apic_broadcast_quirk_disabled) {
864 if ((irq->dest_id == APIC_BROADCAST &&
865 map->mode != KVM_APIC_MODE_X2APIC))
867 if (irq->dest_id == X2APIC_BROADCAST)
870 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
871 if (irq->dest_id == (x2apic_ipi ?
872 X2APIC_BROADCAST : APIC_BROADCAST))
879 /* Return true if the interrupt can be handled by using *bitmap as index mask
880 * for valid destinations in *dst array.
881 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
882 * Note: we may have zero kvm_lapic destinations when we return true, which
883 * means that the interrupt should be dropped. In this case, *bitmap would be
884 * zero and *dst undefined.
886 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
887 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
888 struct kvm_apic_map *map, struct kvm_lapic ***dst,
889 unsigned long *bitmap)
893 if (irq->shorthand == APIC_DEST_SELF && src) {
897 } else if (irq->shorthand)
900 if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
903 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
904 if (irq->dest_id > map->max_apic_id) {
907 u32 dest_id = array_index_nospec(irq->dest_id, map->max_apic_id + 1);
908 *dst = &map->phys_map[dest_id];
915 if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
919 if (!kvm_lowest_prio_delivery(irq))
922 if (!kvm_vector_hashing_enabled()) {
924 for_each_set_bit(i, bitmap, 16) {
929 else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
930 (*dst)[lowest]->vcpu) < 0)
937 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
940 if (!(*dst)[lowest]) {
941 kvm_apic_disabled_lapic_found(kvm);
947 *bitmap = (lowest >= 0) ? 1 << lowest : 0;
952 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
953 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
955 struct kvm_apic_map *map;
956 unsigned long bitmap;
957 struct kvm_lapic **dst = NULL;
963 if (irq->shorthand == APIC_DEST_SELF) {
964 if (KVM_BUG_ON(!src, kvm)) {
968 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
973 map = rcu_dereference(kvm->arch.apic_map);
975 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
977 for_each_set_bit(i, &bitmap, 16) {
982 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
990 * This routine tries to handler interrupts in posted mode, here is how
991 * it deals with different cases:
992 * - For single-destination interrupts, handle it in posted mode
993 * - Else if vector hashing is enabled and it is a lowest-priority
994 * interrupt, handle it in posted mode and use the following mechanism
995 * to find the destinaiton vCPU.
996 * 1. For lowest-priority interrupts, store all the possible
997 * destination vCPUs in an array.
998 * 2. Use "guest vector % max number of destination vCPUs" to find
999 * the right destination vCPU in the array for the lowest-priority
1001 * - Otherwise, use remapped mode to inject the interrupt.
1003 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
1004 struct kvm_vcpu **dest_vcpu)
1006 struct kvm_apic_map *map;
1007 unsigned long bitmap;
1008 struct kvm_lapic **dst = NULL;
1015 map = rcu_dereference(kvm->arch.apic_map);
1017 if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
1018 hweight16(bitmap) == 1) {
1019 unsigned long i = find_first_bit(&bitmap, 16);
1022 *dest_vcpu = dst[i]->vcpu;
1032 * Add a pending IRQ into lapic.
1033 * Return 1 if successfully added and 0 if discarded.
1035 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
1036 int vector, int level, int trig_mode,
1037 struct dest_map *dest_map)
1040 struct kvm_vcpu *vcpu = apic->vcpu;
1042 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
1044 switch (delivery_mode) {
1045 case APIC_DM_LOWEST:
1046 vcpu->arch.apic_arb_prio++;
1048 if (unlikely(trig_mode && !level))
1051 /* FIXME add logic for vcpu on reset */
1052 if (unlikely(!apic_enabled(apic)))
1058 __set_bit(vcpu->vcpu_id, dest_map->map);
1059 dest_map->vectors[vcpu->vcpu_id] = vector;
1062 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
1064 kvm_lapic_set_vector(vector, apic->regs + APIC_TMR);
1066 apic_clear_vector(vector, apic->regs + APIC_TMR);
1069 if (kvm_x86_ops->deliver_posted_interrupt(vcpu, vector)) {
1070 kvm_lapic_set_irr(vector, apic);
1071 kvm_make_request(KVM_REQ_EVENT, vcpu);
1072 kvm_vcpu_kick(vcpu);
1078 vcpu->arch.pv.pv_unhalted = 1;
1079 kvm_make_request(KVM_REQ_EVENT, vcpu);
1080 kvm_vcpu_kick(vcpu);
1085 kvm_make_request(KVM_REQ_SMI, vcpu);
1086 kvm_vcpu_kick(vcpu);
1091 kvm_inject_nmi(vcpu);
1092 kvm_vcpu_kick(vcpu);
1096 if (!trig_mode || level) {
1098 /* assumes that there are only KVM_APIC_INIT/SIPI */
1099 apic->pending_events = (1UL << KVM_APIC_INIT);
1100 /* make sure pending_events is visible before sending
1103 kvm_make_request(KVM_REQ_EVENT, vcpu);
1104 kvm_vcpu_kick(vcpu);
1106 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
1111 case APIC_DM_STARTUP:
1112 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
1113 vcpu->vcpu_id, vector);
1115 apic->sipi_vector = vector;
1116 /* make sure sipi_vector is visible for the receiver */
1118 set_bit(KVM_APIC_SIPI, &apic->pending_events);
1119 kvm_make_request(KVM_REQ_EVENT, vcpu);
1120 kvm_vcpu_kick(vcpu);
1123 case APIC_DM_EXTINT:
1125 * Should only be called by kvm_apic_local_deliver() with LVT0,
1126 * before NMI watchdog was enabled. Already handled by
1127 * kvm_apic_accept_pic_intr().
1132 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1139 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1141 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1144 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1146 return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1149 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1153 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1154 if (!kvm_ioapic_handles_vector(apic, vector))
1157 /* Request a KVM exit to inform the userspace IOAPIC. */
1158 if (irqchip_split(apic->vcpu->kvm)) {
1159 apic->vcpu->arch.pending_ioapic_eoi = vector;
1160 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1164 if (apic_test_vector(vector, apic->regs + APIC_TMR))
1165 trigger_mode = IOAPIC_LEVEL_TRIG;
1167 trigger_mode = IOAPIC_EDGE_TRIG;
1169 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1172 static int apic_set_eoi(struct kvm_lapic *apic)
1174 int vector = apic_find_highest_isr(apic);
1176 trace_kvm_eoi(apic, vector);
1179 * Not every write EOI will has corresponding ISR,
1180 * one example is when Kernel check timer on setup_IO_APIC
1185 apic_clear_isr(vector, apic);
1186 apic_update_ppr(apic);
1188 if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1189 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1191 kvm_ioapic_send_eoi(apic, vector);
1192 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1197 * this interface assumes a trap-like exit, which has already finished
1198 * desired side effect including vISR and vPPR update.
1200 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1202 struct kvm_lapic *apic = vcpu->arch.apic;
1204 trace_kvm_eoi(apic, vector);
1206 kvm_ioapic_send_eoi(apic, vector);
1207 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1209 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1211 static void apic_send_ipi(struct kvm_lapic *apic)
1213 u32 icr_low = kvm_lapic_get_reg(apic, APIC_ICR);
1214 u32 icr_high = kvm_lapic_get_reg(apic, APIC_ICR2);
1215 struct kvm_lapic_irq irq;
1217 irq.vector = icr_low & APIC_VECTOR_MASK;
1218 irq.delivery_mode = icr_low & APIC_MODE_MASK;
1219 irq.dest_mode = icr_low & APIC_DEST_MASK;
1220 irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1221 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1222 irq.shorthand = icr_low & APIC_SHORT_MASK;
1223 irq.msi_redir_hint = false;
1224 if (apic_x2apic_mode(apic))
1225 irq.dest_id = icr_high;
1227 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1229 trace_kvm_apic_ipi(icr_low, irq.dest_id);
1231 apic_debug("icr_high 0x%x, icr_low 0x%x, "
1232 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1233 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1234 "msi_redir_hint 0x%x\n",
1235 icr_high, icr_low, irq.shorthand, irq.dest_id,
1236 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
1237 irq.vector, irq.msi_redir_hint);
1239 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1242 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1244 ktime_t remaining, now;
1248 ASSERT(apic != NULL);
1250 /* if initial count is 0, current count should also be 0 */
1251 if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1252 apic->lapic_timer.period == 0)
1256 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1257 if (ktime_to_ns(remaining) < 0)
1260 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1261 tmcct = div64_u64(ns,
1262 (APIC_BUS_CYCLE_NS * apic->divide_count));
1267 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1269 struct kvm_vcpu *vcpu = apic->vcpu;
1270 struct kvm_run *run = vcpu->run;
1272 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1273 run->tpr_access.rip = kvm_rip_read(vcpu);
1274 run->tpr_access.is_write = write;
1277 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1279 if (apic->vcpu->arch.tpr_access_reporting)
1280 __report_tpr_access(apic, write);
1283 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1287 if (offset >= LAPIC_MMIO_LENGTH)
1292 apic_debug("Access APIC ARBPRI register which is for P6\n");
1295 case APIC_TMCCT: /* Timer CCR */
1296 if (apic_lvtt_tscdeadline(apic))
1299 val = apic_get_tmcct(apic);
1302 apic_update_ppr(apic);
1303 val = kvm_lapic_get_reg(apic, offset);
1306 report_tpr_access(apic, false);
1309 val = kvm_lapic_get_reg(apic, offset);
1316 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1318 return container_of(dev, struct kvm_lapic, dev);
1321 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1324 unsigned char alignment = offset & 0xf;
1326 /* this bitmask has a bit cleared for each reserved register */
1327 static const u64 rmask = 0x43ff01ffffffe70cULL;
1329 if ((alignment + len) > 4) {
1330 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1335 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1336 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1341 result = __apic_read(apic, offset & ~0xf);
1343 trace_kvm_apic_read(offset, result);
1349 memcpy(data, (char *)&result + alignment, len);
1352 printk(KERN_ERR "Local APIC read with len = %x, "
1353 "should be 1,2, or 4 instead\n", len);
1358 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1360 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1362 return addr >= apic->base_address &&
1363 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1366 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1367 gpa_t address, int len, void *data)
1369 struct kvm_lapic *apic = to_lapic(this);
1370 u32 offset = address - apic->base_address;
1372 if (!apic_mmio_in_range(apic, address))
1375 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1376 if (!kvm_check_has_quirk(vcpu->kvm,
1377 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1380 memset(data, 0xff, len);
1384 kvm_lapic_reg_read(apic, offset, len, data);
1389 static void update_divide_count(struct kvm_lapic *apic)
1391 u32 tmp1, tmp2, tdcr;
1393 tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1395 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1396 apic->divide_count = 0x1 << (tmp2 & 0x7);
1398 apic_debug("timer divide count is 0x%x\n",
1399 apic->divide_count);
1402 static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1405 * Do not allow the guest to program periodic timers with small
1406 * interval, since the hrtimers are not throttled by the host
1409 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1410 s64 min_period = min_timer_period_us * 1000LL;
1412 if (apic->lapic_timer.period < min_period) {
1413 pr_info_ratelimited(
1414 "kvm: vcpu %i: requested %lld ns "
1415 "lapic timer period limited to %lld ns\n",
1416 apic->vcpu->vcpu_id,
1417 apic->lapic_timer.period, min_period);
1418 apic->lapic_timer.period = min_period;
1423 static void apic_update_lvtt(struct kvm_lapic *apic)
1425 u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1426 apic->lapic_timer.timer_mode_mask;
1428 if (apic->lapic_timer.timer_mode != timer_mode) {
1429 if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1430 APIC_LVT_TIMER_TSCDEADLINE)) {
1431 hrtimer_cancel(&apic->lapic_timer.timer);
1432 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1433 apic->lapic_timer.period = 0;
1434 apic->lapic_timer.tscdeadline = 0;
1436 apic->lapic_timer.timer_mode = timer_mode;
1437 limit_periodic_timer_frequency(apic);
1441 static void apic_timer_expired(struct kvm_lapic *apic)
1443 struct kvm_vcpu *vcpu = apic->vcpu;
1444 struct swait_queue_head *q = &vcpu->wq;
1445 struct kvm_timer *ktimer = &apic->lapic_timer;
1447 if (atomic_read(&apic->lapic_timer.pending))
1450 atomic_inc(&apic->lapic_timer.pending);
1451 kvm_set_pending_timer(vcpu);
1454 * For x86, the atomic_inc() is serialized, thus
1455 * using swait_active() is safe.
1457 if (swait_active(q))
1460 if (apic_lvtt_tscdeadline(apic) || ktimer->hv_timer_in_use)
1461 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1465 * On APICv, this test will cause a busy wait
1466 * during a higher-priority task.
1469 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1471 struct kvm_lapic *apic = vcpu->arch.apic;
1472 u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1474 if (kvm_apic_hw_enabled(apic)) {
1475 int vec = reg & APIC_VECTOR_MASK;
1476 void *bitmap = apic->regs + APIC_ISR;
1478 if (vcpu->arch.apicv_active)
1479 bitmap = apic->regs + APIC_IRR;
1481 if (apic_test_vector(vec, bitmap))
1487 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1489 struct kvm_lapic *apic = vcpu->arch.apic;
1490 u64 guest_tsc, tsc_deadline;
1492 if (!lapic_in_kernel(vcpu))
1495 if (apic->lapic_timer.expired_tscdeadline == 0)
1498 if (!lapic_timer_int_injected(vcpu))
1501 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1502 apic->lapic_timer.expired_tscdeadline = 0;
1503 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1504 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1506 /* __delay is delay_tsc whenever the hardware has TSC, thus always. */
1507 if (guest_tsc < tsc_deadline)
1508 __delay(min(tsc_deadline - guest_tsc,
1509 nsec_to_cycles(vcpu, lapic_timer_advance_ns)));
1512 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1514 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1517 struct kvm_vcpu *vcpu = apic->vcpu;
1518 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1519 unsigned long flags;
1522 if (unlikely(!tscdeadline || !this_tsc_khz))
1525 local_irq_save(flags);
1528 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1529 if (likely(tscdeadline > guest_tsc)) {
1530 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1531 do_div(ns, this_tsc_khz);
1532 expire = ktime_add_ns(now, ns);
1533 expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1534 hrtimer_start(&apic->lapic_timer.timer,
1535 expire, HRTIMER_MODE_ABS_PINNED);
1537 apic_timer_expired(apic);
1539 local_irq_restore(flags);
1542 static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1544 ktime_t now, remaining;
1545 u64 ns_remaining_old, ns_remaining_new;
1547 apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1548 * APIC_BUS_CYCLE_NS * apic->divide_count;
1549 limit_periodic_timer_frequency(apic);
1552 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1553 if (ktime_to_ns(remaining) < 0)
1556 ns_remaining_old = ktime_to_ns(remaining);
1557 ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1558 apic->divide_count, old_divisor);
1560 apic->lapic_timer.tscdeadline +=
1561 nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1562 nsec_to_cycles(apic->vcpu, ns_remaining_old);
1563 apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1566 static bool set_target_expiration(struct kvm_lapic *apic)
1572 apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1573 * APIC_BUS_CYCLE_NS * apic->divide_count;
1575 if (!apic->lapic_timer.period) {
1576 apic->lapic_timer.tscdeadline = 0;
1580 limit_periodic_timer_frequency(apic);
1582 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1584 "timer initial count 0x%x, period %lldns, "
1585 "expire @ 0x%016" PRIx64 ".\n", __func__,
1586 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1587 kvm_lapic_get_reg(apic, APIC_TMICT),
1588 apic->lapic_timer.period,
1589 ktime_to_ns(ktime_add_ns(now,
1590 apic->lapic_timer.period)));
1592 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1593 nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1594 apic->lapic_timer.target_expiration = ktime_add_ns(now, apic->lapic_timer.period);
1599 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1601 ktime_t now = ktime_get();
1606 * Synchronize both deadlines to the same time source or
1607 * differences in the periods (caused by differences in the
1608 * underlying clocks or numerical approximation errors) will
1609 * cause the two to drift apart over time as the errors
1612 apic->lapic_timer.target_expiration =
1613 ktime_add_ns(apic->lapic_timer.target_expiration,
1614 apic->lapic_timer.period);
1615 delta = ktime_sub(apic->lapic_timer.target_expiration, now);
1616 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1617 nsec_to_cycles(apic->vcpu, delta);
1620 static void start_sw_period(struct kvm_lapic *apic)
1622 if (!apic->lapic_timer.period)
1625 if (ktime_after(ktime_get(),
1626 apic->lapic_timer.target_expiration)) {
1627 apic_timer_expired(apic);
1629 if (apic_lvtt_oneshot(apic))
1632 advance_periodic_target_expiration(apic);
1635 hrtimer_start(&apic->lapic_timer.timer,
1636 apic->lapic_timer.target_expiration,
1637 HRTIMER_MODE_ABS_PINNED);
1640 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1642 if (!lapic_in_kernel(vcpu))
1645 return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1647 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1649 static void cancel_hv_timer(struct kvm_lapic *apic)
1651 WARN_ON(preemptible());
1652 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1653 kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1654 apic->lapic_timer.hv_timer_in_use = false;
1657 static bool start_hv_timer(struct kvm_lapic *apic)
1659 struct kvm_timer *ktimer = &apic->lapic_timer;
1662 WARN_ON(preemptible());
1663 if (!kvm_x86_ops->set_hv_timer)
1666 if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1669 if (!ktimer->tscdeadline)
1672 r = kvm_x86_ops->set_hv_timer(apic->vcpu, ktimer->tscdeadline);
1676 ktimer->hv_timer_in_use = true;
1677 hrtimer_cancel(&ktimer->timer);
1680 * Also recheck ktimer->pending, in case the sw timer triggered in
1681 * the window. For periodic timer, leave the hv timer running for
1682 * simplicity, and the deadline will be recomputed on the next vmexit.
1684 if (!apic_lvtt_period(apic) && (r || atomic_read(&ktimer->pending))) {
1686 apic_timer_expired(apic);
1690 trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, true);
1694 static void start_sw_timer(struct kvm_lapic *apic)
1696 struct kvm_timer *ktimer = &apic->lapic_timer;
1698 WARN_ON(preemptible());
1699 if (apic->lapic_timer.hv_timer_in_use)
1700 cancel_hv_timer(apic);
1701 if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1704 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1705 start_sw_period(apic);
1706 else if (apic_lvtt_tscdeadline(apic))
1707 start_sw_tscdeadline(apic);
1708 trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1711 static void restart_apic_timer(struct kvm_lapic *apic)
1714 if (!start_hv_timer(apic))
1715 start_sw_timer(apic);
1719 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1721 struct kvm_lapic *apic = vcpu->arch.apic;
1724 /* If the preempt notifier has already run, it also called apic_timer_expired */
1725 if (!apic->lapic_timer.hv_timer_in_use)
1727 WARN_ON(swait_active(&vcpu->wq));
1728 cancel_hv_timer(apic);
1729 apic_timer_expired(apic);
1731 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1732 advance_periodic_target_expiration(apic);
1733 restart_apic_timer(apic);
1738 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1740 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1742 restart_apic_timer(vcpu->arch.apic);
1744 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1746 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1748 struct kvm_lapic *apic = vcpu->arch.apic;
1751 /* Possibly the TSC deadline timer is not enabled yet */
1752 if (apic->lapic_timer.hv_timer_in_use)
1753 start_sw_timer(apic);
1756 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1758 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
1760 struct kvm_lapic *apic = vcpu->arch.apic;
1762 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1763 restart_apic_timer(apic);
1766 static void start_apic_timer(struct kvm_lapic *apic)
1768 atomic_set(&apic->lapic_timer.pending, 0);
1770 if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1771 && !set_target_expiration(apic))
1774 restart_apic_timer(apic);
1777 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1779 bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1781 if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1782 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1783 if (lvt0_in_nmi_mode) {
1784 apic_debug("Receive NMI setting on APIC_LVT0 "
1785 "for cpu %d\n", apic->vcpu->vcpu_id);
1786 atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1788 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1792 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1796 trace_kvm_apic_write(reg, val);
1799 case APIC_ID: /* Local APIC ID */
1800 if (!apic_x2apic_mode(apic))
1801 kvm_apic_set_xapic_id(apic, val >> 24);
1807 report_tpr_access(apic, true);
1808 apic_set_tpr(apic, val & 0xff);
1816 if (!apic_x2apic_mode(apic))
1817 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1823 if (!apic_x2apic_mode(apic)) {
1824 kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1825 recalculate_apic_map(apic->vcpu->kvm);
1832 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1833 mask |= APIC_SPIV_DIRECTED_EOI;
1834 apic_set_spiv(apic, val & mask);
1835 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1839 for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
1840 lvt_val = kvm_lapic_get_reg(apic,
1841 APIC_LVTT + 0x10 * i);
1842 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
1843 lvt_val | APIC_LVT_MASKED);
1845 apic_update_lvtt(apic);
1846 atomic_set(&apic->lapic_timer.pending, 0);
1852 /* No delay here, so we always clear the pending bit */
1853 kvm_lapic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1854 apic_send_ipi(apic);
1858 if (!apic_x2apic_mode(apic))
1860 kvm_lapic_set_reg(apic, APIC_ICR2, val);
1864 apic_manage_nmi_watchdog(apic, val);
1869 /* TODO: Check vector */
1873 if (!kvm_apic_sw_enabled(apic))
1874 val |= APIC_LVT_MASKED;
1875 size = ARRAY_SIZE(apic_lvt_mask);
1876 index = array_index_nospec(
1877 (reg - APIC_LVTT) >> 4, size);
1878 val &= apic_lvt_mask[index];
1879 kvm_lapic_set_reg(apic, reg, val);
1884 if (!kvm_apic_sw_enabled(apic))
1885 val |= APIC_LVT_MASKED;
1886 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1887 kvm_lapic_set_reg(apic, APIC_LVTT, val);
1888 apic_update_lvtt(apic);
1892 if (apic_lvtt_tscdeadline(apic))
1895 hrtimer_cancel(&apic->lapic_timer.timer);
1896 kvm_lapic_set_reg(apic, APIC_TMICT, val);
1897 start_apic_timer(apic);
1901 uint32_t old_divisor = apic->divide_count;
1904 apic_debug("KVM_WRITE:TDCR %x\n", val);
1905 kvm_lapic_set_reg(apic, APIC_TDCR, val);
1906 update_divide_count(apic);
1907 if (apic->divide_count != old_divisor &&
1908 apic->lapic_timer.period) {
1909 hrtimer_cancel(&apic->lapic_timer.timer);
1910 update_target_expiration(apic, old_divisor);
1911 restart_apic_timer(apic);
1916 if (apic_x2apic_mode(apic) && val != 0) {
1917 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1923 if (apic_x2apic_mode(apic)) {
1924 kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1933 apic_debug("Local APIC Write to read-only register %x\n", reg);
1936 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
1938 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1939 gpa_t address, int len, const void *data)
1941 struct kvm_lapic *apic = to_lapic(this);
1942 unsigned int offset = address - apic->base_address;
1945 if (!apic_mmio_in_range(apic, address))
1948 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1949 if (!kvm_check_has_quirk(vcpu->kvm,
1950 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1957 * APIC register must be aligned on 128-bits boundary.
1958 * 32/64/128 bits registers must be accessed thru 32 bits.
1961 if (len != 4 || (offset & 0xf)) {
1962 /* Don't shout loud, $infamous_os would cause only noise. */
1963 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1969 /* too common printing */
1970 if (offset != APIC_EOI)
1971 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1972 "0x%x\n", __func__, offset, len, val);
1974 kvm_lapic_reg_write(apic, offset & 0xff0, val);
1979 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1981 kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1983 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1985 /* emulate APIC access in a trap manner */
1986 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1990 /* hw has done the conditional check and inst decode */
1993 kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
1995 /* TODO: optimize to just emulate side effect w/o one more write */
1996 kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
1998 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
2000 void kvm_free_lapic(struct kvm_vcpu *vcpu)
2002 struct kvm_lapic *apic = vcpu->arch.apic;
2004 if (!vcpu->arch.apic)
2007 hrtimer_cancel(&apic->lapic_timer.timer);
2009 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
2010 static_key_slow_dec_deferred(&apic_hw_disabled);
2012 if (!apic->sw_enabled)
2013 static_key_slow_dec_deferred(&apic_sw_disabled);
2016 free_page((unsigned long)apic->regs);
2022 *----------------------------------------------------------------------
2024 *----------------------------------------------------------------------
2026 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
2028 struct kvm_lapic *apic = vcpu->arch.apic;
2030 if (!lapic_in_kernel(vcpu) ||
2031 !apic_lvtt_tscdeadline(apic))
2034 return apic->lapic_timer.tscdeadline;
2037 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
2039 struct kvm_lapic *apic = vcpu->arch.apic;
2041 if (!kvm_apic_present(vcpu) || apic_lvtt_oneshot(apic) ||
2042 apic_lvtt_period(apic))
2045 hrtimer_cancel(&apic->lapic_timer.timer);
2046 apic->lapic_timer.tscdeadline = data;
2047 start_apic_timer(apic);
2050 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
2052 apic_set_tpr(vcpu->arch.apic, (cr8 & 0x0f) << 4);
2055 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
2059 tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
2061 return (tpr & 0xf0) >> 4;
2064 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
2066 u64 old_value = vcpu->arch.apic_base;
2067 struct kvm_lapic *apic = vcpu->arch.apic;
2070 value |= MSR_IA32_APICBASE_BSP;
2072 vcpu->arch.apic_base = value;
2074 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
2075 kvm_update_cpuid(vcpu);
2080 /* update jump label if enable bit changes */
2081 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
2082 if (value & MSR_IA32_APICBASE_ENABLE) {
2083 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2084 static_key_slow_dec_deferred(&apic_hw_disabled);
2086 static_key_slow_inc(&apic_hw_disabled.key);
2087 recalculate_apic_map(vcpu->kvm);
2091 if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
2092 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
2094 if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE))
2095 kvm_x86_ops->set_virtual_apic_mode(vcpu);
2097 apic->base_address = apic->vcpu->arch.apic_base &
2098 MSR_IA32_APICBASE_BASE;
2100 if ((value & MSR_IA32_APICBASE_ENABLE) &&
2101 apic->base_address != APIC_DEFAULT_PHYS_BASE)
2102 pr_warn_once("APIC base relocation is unsupported by KVM");
2104 /* with FSB delivery interrupt, we can restart APIC functionality */
2105 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
2106 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
2110 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2112 struct kvm_lapic *apic = vcpu->arch.apic;
2118 apic_debug("%s\n", __func__);
2120 /* Stop the timer in case it's a reset to an active apic */
2121 hrtimer_cancel(&apic->lapic_timer.timer);
2124 kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
2125 MSR_IA32_APICBASE_ENABLE);
2126 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2128 kvm_apic_set_version(apic->vcpu);
2130 for (i = 0; i < KVM_APIC_LVT_NUM; i++)
2131 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
2132 apic_update_lvtt(apic);
2133 if (kvm_vcpu_is_reset_bsp(vcpu) &&
2134 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2135 kvm_lapic_set_reg(apic, APIC_LVT0,
2136 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2137 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2139 kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
2140 apic_set_spiv(apic, 0xff);
2141 kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2142 if (!apic_x2apic_mode(apic))
2143 kvm_apic_set_ldr(apic, 0);
2144 kvm_lapic_set_reg(apic, APIC_ESR, 0);
2145 kvm_lapic_set_reg(apic, APIC_ICR, 0);
2146 kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2147 kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2148 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2149 for (i = 0; i < 8; i++) {
2150 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2151 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2152 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2154 apic->irr_pending = vcpu->arch.apicv_active;
2155 apic->isr_count = vcpu->arch.apicv_active ? 1 : 0;
2156 apic->highest_isr_cache = -1;
2157 update_divide_count(apic);
2158 atomic_set(&apic->lapic_timer.pending, 0);
2159 if (kvm_vcpu_is_bsp(vcpu))
2160 kvm_lapic_set_base(vcpu,
2161 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
2162 vcpu->arch.pv_eoi.msr_val = 0;
2163 apic_update_ppr(apic);
2164 if (vcpu->arch.apicv_active) {
2165 kvm_x86_ops->apicv_post_state_restore(vcpu);
2166 kvm_x86_ops->hwapic_irr_update(vcpu, -1);
2167 kvm_x86_ops->hwapic_isr_update(vcpu, -1);
2170 vcpu->arch.apic_arb_prio = 0;
2171 vcpu->arch.apic_attention = 0;
2173 apic_debug("%s: vcpu=%p, id=0x%x, base_msr="
2174 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
2175 vcpu, kvm_lapic_get_reg(apic, APIC_ID),
2176 vcpu->arch.apic_base, apic->base_address);
2180 *----------------------------------------------------------------------
2182 *----------------------------------------------------------------------
2185 static bool lapic_is_periodic(struct kvm_lapic *apic)
2187 return apic_lvtt_period(apic);
2190 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2192 struct kvm_lapic *apic = vcpu->arch.apic;
2194 if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2195 return atomic_read(&apic->lapic_timer.pending);
2200 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2202 u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2203 int vector, mode, trig_mode;
2205 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2206 vector = reg & APIC_VECTOR_MASK;
2207 mode = reg & APIC_MODE_MASK;
2208 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2209 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2215 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2217 struct kvm_lapic *apic = vcpu->arch.apic;
2220 kvm_apic_local_deliver(apic, APIC_LVT0);
2223 static const struct kvm_io_device_ops apic_mmio_ops = {
2224 .read = apic_mmio_read,
2225 .write = apic_mmio_write,
2228 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2230 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2231 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2233 apic_timer_expired(apic);
2235 if (lapic_is_periodic(apic)) {
2236 advance_periodic_target_expiration(apic);
2237 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2238 return HRTIMER_RESTART;
2240 return HRTIMER_NORESTART;
2243 int kvm_create_lapic(struct kvm_vcpu *vcpu)
2245 struct kvm_lapic *apic;
2247 ASSERT(vcpu != NULL);
2248 apic_debug("apic_init %d\n", vcpu->vcpu_id);
2250 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
2254 vcpu->arch.apic = apic;
2256 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
2258 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2260 goto nomem_free_apic;
2264 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2265 HRTIMER_MODE_ABS_PINNED);
2266 apic->lapic_timer.timer.function = apic_timer_fn;
2269 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2270 * thinking that APIC satet has changed.
2272 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2273 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2274 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2283 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2285 struct kvm_lapic *apic = vcpu->arch.apic;
2288 if (!kvm_apic_present(vcpu))
2291 __apic_update_ppr(apic, &ppr);
2292 return apic_has_interrupt_for_ppr(apic, ppr);
2295 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2297 u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2300 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2302 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2303 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2308 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2310 struct kvm_lapic *apic = vcpu->arch.apic;
2312 if (atomic_read(&apic->lapic_timer.pending) > 0) {
2313 kvm_apic_local_deliver(apic, APIC_LVTT);
2314 if (apic_lvtt_tscdeadline(apic))
2315 apic->lapic_timer.tscdeadline = 0;
2316 if (apic_lvtt_oneshot(apic)) {
2317 apic->lapic_timer.tscdeadline = 0;
2318 apic->lapic_timer.target_expiration = 0;
2320 atomic_set(&apic->lapic_timer.pending, 0);
2324 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2326 int vector = kvm_apic_has_interrupt(vcpu);
2327 struct kvm_lapic *apic = vcpu->arch.apic;
2334 * We get here even with APIC virtualization enabled, if doing
2335 * nested virtualization and L1 runs with the "acknowledge interrupt
2336 * on exit" mode. Then we cannot inject the interrupt via RVI,
2337 * because the process would deliver it through the IDT.
2340 apic_clear_irr(vector, apic);
2341 if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2343 * For auto-EOI interrupts, there might be another pending
2344 * interrupt above PPR, so check whether to raise another
2347 apic_update_ppr(apic);
2350 * For normal interrupts, PPR has been raised and there cannot
2351 * be a higher-priority pending interrupt---except if there was
2352 * a concurrent interrupt injection, but that would have
2353 * triggered KVM_REQ_EVENT already.
2355 apic_set_isr(vector, apic);
2356 __apic_update_ppr(apic, &ppr);
2362 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2363 struct kvm_lapic_state *s, bool set)
2365 if (apic_x2apic_mode(vcpu->arch.apic)) {
2366 u32 *id = (u32 *)(s->regs + APIC_ID);
2367 u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2369 if (vcpu->kvm->arch.x2apic_format) {
2370 if (*id != vcpu->vcpu_id)
2379 /* In x2APIC mode, the LDR is fixed and based on the id */
2381 *ldr = kvm_apic_calc_x2apic_ldr(*id);
2387 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2389 memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2390 return kvm_apic_state_fixup(vcpu, s, false);
2393 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2395 struct kvm_lapic *apic = vcpu->arch.apic;
2399 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2400 /* set SPIV separately to get count of SW disabled APICs right */
2401 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2403 r = kvm_apic_state_fixup(vcpu, s, true);
2406 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2408 recalculate_apic_map(vcpu->kvm);
2409 kvm_apic_set_version(vcpu);
2411 apic_update_ppr(apic);
2412 hrtimer_cancel(&apic->lapic_timer.timer);
2413 apic_update_lvtt(apic);
2414 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2415 update_divide_count(apic);
2416 start_apic_timer(apic);
2417 apic->irr_pending = true;
2418 apic->isr_count = vcpu->arch.apicv_active ?
2419 1 : count_vectors(apic->regs + APIC_ISR);
2420 apic->highest_isr_cache = -1;
2421 if (vcpu->arch.apicv_active) {
2422 kvm_x86_ops->apicv_post_state_restore(vcpu);
2423 kvm_x86_ops->hwapic_irr_update(vcpu,
2424 apic_find_highest_irr(apic));
2425 kvm_x86_ops->hwapic_isr_update(vcpu,
2426 apic_find_highest_isr(apic));
2428 kvm_make_request(KVM_REQ_EVENT, vcpu);
2429 if (ioapic_in_kernel(vcpu->kvm))
2430 kvm_rtc_eoi_tracking_restore_one(vcpu);
2432 vcpu->arch.apic_arb_prio = 0;
2437 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2439 struct hrtimer *timer;
2441 if (!lapic_in_kernel(vcpu))
2444 timer = &vcpu->arch.apic->lapic_timer.timer;
2445 if (hrtimer_cancel(timer))
2446 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
2450 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2452 * Detect whether guest triggered PV EOI since the
2453 * last entry. If yes, set EOI on guests's behalf.
2454 * Clear PV EOI in guest memory in any case.
2456 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2457 struct kvm_lapic *apic)
2462 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2463 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2465 * KVM_APIC_PV_EOI_PENDING is unset:
2466 * -> host disabled PV EOI.
2467 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2468 * -> host enabled PV EOI, guest did not execute EOI yet.
2469 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2470 * -> host enabled PV EOI, guest executed EOI.
2472 BUG_ON(!pv_eoi_enabled(vcpu));
2473 pending = pv_eoi_get_pending(vcpu);
2475 * Clear pending bit in any case: it will be set again on vmentry.
2476 * While this might not be ideal from performance point of view,
2477 * this makes sure pv eoi is only enabled when we know it's safe.
2479 pv_eoi_clr_pending(vcpu);
2482 vector = apic_set_eoi(apic);
2483 trace_kvm_pv_eoi(apic, vector);
2486 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2490 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2491 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2493 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2496 if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2500 apic_set_tpr(vcpu->arch.apic, data & 0xff);
2504 * apic_sync_pv_eoi_to_guest - called before vmentry
2506 * Detect whether it's safe to enable PV EOI and
2509 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2510 struct kvm_lapic *apic)
2512 if (!pv_eoi_enabled(vcpu) ||
2513 /* IRR set or many bits in ISR: could be nested. */
2514 apic->irr_pending ||
2515 /* Cache not set: could be safe but we don't bother. */
2516 apic->highest_isr_cache == -1 ||
2517 /* Need EOI to update ioapic. */
2518 kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2520 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2521 * so we need not do anything here.
2526 pv_eoi_set_pending(apic->vcpu);
2529 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2532 int max_irr, max_isr;
2533 struct kvm_lapic *apic = vcpu->arch.apic;
2535 apic_sync_pv_eoi_to_guest(vcpu, apic);
2537 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2540 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2541 max_irr = apic_find_highest_irr(apic);
2544 max_isr = apic_find_highest_isr(apic);
2547 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2549 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2553 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2556 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2557 &vcpu->arch.apic->vapic_cache,
2558 vapic_addr, sizeof(u32)))
2560 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2562 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2565 vcpu->arch.apic->vapic_addr = vapic_addr;
2569 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2571 struct kvm_lapic *apic = vcpu->arch.apic;
2572 u32 reg = (msr - APIC_BASE_MSR) << 4;
2574 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2577 if (reg == APIC_ICR2)
2580 /* if this is ICR write vector before command */
2581 if (reg == APIC_ICR)
2582 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2583 return kvm_lapic_reg_write(apic, reg, (u32)data);
2586 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2588 struct kvm_lapic *apic = vcpu->arch.apic;
2589 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2591 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2594 if (reg == APIC_DFR || reg == APIC_ICR2) {
2595 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2600 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2602 if (reg == APIC_ICR)
2603 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2605 *data = (((u64)high) << 32) | low;
2610 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2612 struct kvm_lapic *apic = vcpu->arch.apic;
2614 if (!lapic_in_kernel(vcpu))
2617 /* if this is ICR write vector before command */
2618 if (reg == APIC_ICR)
2619 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2620 return kvm_lapic_reg_write(apic, reg, (u32)data);
2623 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2625 struct kvm_lapic *apic = vcpu->arch.apic;
2628 if (!lapic_in_kernel(vcpu))
2631 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2633 if (reg == APIC_ICR)
2634 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2636 *data = (((u64)high) << 32) | low;
2641 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data, unsigned long len)
2643 u64 addr = data & ~KVM_MSR_ENABLED;
2644 struct gfn_to_hva_cache *ghc = &vcpu->arch.pv_eoi.data;
2645 unsigned long new_len;
2647 if (!IS_ALIGNED(addr, 4))
2650 vcpu->arch.pv_eoi.msr_val = data;
2651 if (!pv_eoi_enabled(vcpu))
2654 if (addr == ghc->gpa && len <= ghc->len)
2659 return kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, addr, new_len);
2662 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2664 struct kvm_lapic *apic = vcpu->arch.apic;
2668 if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2672 * INITs are latched while in SMM. Because an SMM CPU cannot
2673 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2674 * and delay processing of INIT until the next RSM.
2677 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2678 if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2679 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2683 pe = xchg(&apic->pending_events, 0);
2684 if (test_bit(KVM_APIC_INIT, &pe)) {
2685 kvm_vcpu_reset(vcpu, true);
2686 if (kvm_vcpu_is_bsp(apic->vcpu))
2687 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2689 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2691 if (test_bit(KVM_APIC_SIPI, &pe) &&
2692 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2693 /* evaluate pending_events before reading the vector */
2695 sipi_vector = apic->sipi_vector;
2696 apic_debug("vcpu %d received sipi with vector # %x\n",
2697 vcpu->vcpu_id, sipi_vector);
2698 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2699 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2703 void kvm_lapic_init(void)
2705 /* do not patch jump label more than once per second */
2706 jump_label_rate_limit(&apic_hw_disabled, HZ);
2707 jump_label_rate_limit(&apic_sw_disabled, HZ);
2710 void kvm_lapic_exit(void)
2712 static_key_deferred_flush(&apic_hw_disabled);
2713 static_key_deferred_flush(&apic_sw_disabled);