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/module.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 <linux/atomic.h>
37 #include <linux/jump_label.h>
38 #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 #define APIC_BUS_CYCLE_NS 1
57 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
58 #define apic_debug(fmt, arg...)
60 #define APIC_LVT_NUM 6
61 /* 14 is the version for Xeon and Pentium 8.4.8*/
62 #define APIC_VERSION (0x14UL | ((APIC_LVT_NUM - 1) << 16))
63 #define LAPIC_MMIO_LENGTH (1 << 12)
64 /* followed define is not in apicdef.h */
65 #define APIC_SHORT_MASK 0xc0000
66 #define APIC_DEST_NOSHORT 0x0
67 #define APIC_DEST_MASK 0x800
68 #define MAX_APIC_VECTOR 256
69 #define APIC_VECTORS_PER_REG 32
71 #define APIC_BROADCAST 0xFF
72 #define X2APIC_BROADCAST 0xFFFFFFFFul
74 #define VEC_POS(v) ((v) & (32 - 1))
75 #define REG_POS(v) (((v) >> 5) << 4)
77 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
79 *((u32 *) (apic->regs + reg_off)) = val;
82 static inline int apic_test_vector(int vec, void *bitmap)
84 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
87 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
89 struct kvm_lapic *apic = vcpu->arch.apic;
91 return apic_test_vector(vector, apic->regs + APIC_ISR) ||
92 apic_test_vector(vector, apic->regs + APIC_IRR);
95 static inline void apic_set_vector(int vec, void *bitmap)
97 set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
100 static inline void apic_clear_vector(int vec, void *bitmap)
102 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
105 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
107 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
110 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
112 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
115 struct static_key_deferred apic_hw_disabled __read_mostly;
116 struct static_key_deferred apic_sw_disabled __read_mostly;
118 static inline int apic_enabled(struct kvm_lapic *apic)
120 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
124 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
127 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
128 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
130 static inline int kvm_apic_id(struct kvm_lapic *apic)
132 return (kvm_apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
135 #define KVM_X2APIC_CID_BITS 0
137 static void recalculate_apic_map(struct kvm *kvm)
139 struct kvm_apic_map *new, *old = NULL;
140 struct kvm_vcpu *vcpu;
143 new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
145 mutex_lock(&kvm->arch.apic_map_lock);
151 /* flat mode is default */
154 new->lid_mask = 0xff;
155 new->broadcast = APIC_BROADCAST;
157 kvm_for_each_vcpu(i, vcpu, kvm) {
158 struct kvm_lapic *apic = vcpu->arch.apic;
162 if (!kvm_apic_present(vcpu))
166 * All APICs have to be configured in the same mode by an OS.
167 * We take advatage of this while building logical id loockup
168 * table. After reset APICs are in xapic/flat mode, so if we
169 * find apic with different setting we assume this is the mode
170 * OS wants all apics to be in; build lookup table accordingly.
172 if (apic_x2apic_mode(apic)) {
175 new->cid_mask = (1 << KVM_X2APIC_CID_BITS) - 1;
176 new->lid_mask = 0xffff;
177 new->broadcast = X2APIC_BROADCAST;
178 } else if (kvm_apic_sw_enabled(apic) &&
179 !new->cid_mask /* flat mode */ &&
180 kvm_apic_get_reg(apic, APIC_DFR) == APIC_DFR_CLUSTER) {
186 new->phys_map[kvm_apic_id(apic)] = apic;
188 ldr = kvm_apic_get_reg(apic, APIC_LDR);
189 cid = apic_cluster_id(new, ldr);
190 lid = apic_logical_id(new, ldr);
193 new->logical_map[cid][ffs(lid) - 1] = apic;
196 old = rcu_dereference_protected(kvm->arch.apic_map,
197 lockdep_is_held(&kvm->arch.apic_map_lock));
198 rcu_assign_pointer(kvm->arch.apic_map, new);
199 mutex_unlock(&kvm->arch.apic_map_lock);
204 kvm_vcpu_request_scan_ioapic(kvm);
207 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
209 u32 prev = kvm_apic_get_reg(apic, APIC_SPIV);
211 apic_set_reg(apic, APIC_SPIV, val);
212 if ((prev ^ val) & APIC_SPIV_APIC_ENABLED) {
213 if (val & APIC_SPIV_APIC_ENABLED) {
214 static_key_slow_dec_deferred(&apic_sw_disabled);
215 recalculate_apic_map(apic->vcpu->kvm);
217 static_key_slow_inc(&apic_sw_disabled.key);
221 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
223 apic_set_reg(apic, APIC_ID, id << 24);
224 recalculate_apic_map(apic->vcpu->kvm);
227 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
229 apic_set_reg(apic, APIC_LDR, id);
230 recalculate_apic_map(apic->vcpu->kvm);
233 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
235 return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
238 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
240 return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
243 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
245 return ((kvm_apic_get_reg(apic, APIC_LVTT) &
246 apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_ONESHOT);
249 static inline int apic_lvtt_period(struct kvm_lapic *apic)
251 return ((kvm_apic_get_reg(apic, APIC_LVTT) &
252 apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_PERIODIC);
255 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
257 return ((kvm_apic_get_reg(apic, APIC_LVTT) &
258 apic->lapic_timer.timer_mode_mask) ==
259 APIC_LVT_TIMER_TSCDEADLINE);
262 static inline int apic_lvt_nmi_mode(u32 lvt_val)
264 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
267 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
269 struct kvm_lapic *apic = vcpu->arch.apic;
270 struct kvm_cpuid_entry2 *feat;
271 u32 v = APIC_VERSION;
273 if (!kvm_vcpu_has_lapic(vcpu))
276 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
277 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
278 v |= APIC_LVR_DIRECTED_EOI;
279 apic_set_reg(apic, APIC_LVR, v);
282 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
283 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
284 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
285 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
286 LINT_MASK, LINT_MASK, /* LVT0-1 */
287 LVT_MASK /* LVTERR */
290 static int find_highest_vector(void *bitmap)
295 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
296 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
297 reg = bitmap + REG_POS(vec);
299 return fls(*reg) - 1 + vec;
305 static u8 count_vectors(void *bitmap)
311 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
312 reg = bitmap + REG_POS(vec);
313 count += hweight32(*reg);
319 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
322 struct kvm_lapic *apic = vcpu->arch.apic;
324 for (i = 0; i <= 7; i++) {
325 pir_val = xchg(&pir[i], 0);
327 *((u32 *)(apic->regs + APIC_IRR + i * 0x10)) |= pir_val;
330 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
332 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
334 apic->irr_pending = true;
335 apic_set_vector(vec, apic->regs + APIC_IRR);
338 static inline int apic_search_irr(struct kvm_lapic *apic)
340 return find_highest_vector(apic->regs + APIC_IRR);
343 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
348 * Note that irr_pending is just a hint. It will be always
349 * true with virtual interrupt delivery enabled.
351 if (!apic->irr_pending)
354 kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
355 result = apic_search_irr(apic);
356 ASSERT(result == -1 || result >= 16);
361 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
363 struct kvm_vcpu *vcpu;
367 apic_clear_vector(vec, apic->regs + APIC_IRR);
368 if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
369 /* try to update RVI */
370 kvm_make_request(KVM_REQ_EVENT, vcpu);
372 vec = apic_search_irr(apic);
373 apic->irr_pending = (vec != -1);
377 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
379 struct kvm_vcpu *vcpu;
381 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
387 * With APIC virtualization enabled, all caching is disabled
388 * because the processor can modify ISR under the hood. Instead
391 if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
392 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
395 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
397 * ISR (in service register) bit is set when injecting an interrupt.
398 * The highest vector is injected. Thus the latest bit set matches
399 * the highest bit in ISR.
401 apic->highest_isr_cache = vec;
405 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
410 * Note that isr_count is always 1, and highest_isr_cache
411 * is always -1, with APIC virtualization enabled.
413 if (!apic->isr_count)
415 if (likely(apic->highest_isr_cache != -1))
416 return apic->highest_isr_cache;
418 result = find_highest_vector(apic->regs + APIC_ISR);
419 ASSERT(result == -1 || result >= 16);
424 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
426 struct kvm_vcpu *vcpu;
427 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
433 * We do get here for APIC virtualization enabled if the guest
434 * uses the Hyper-V APIC enlightenment. In this case we may need
435 * to trigger a new interrupt delivery by writing the SVI field;
436 * on the other hand isr_count and highest_isr_cache are unused
437 * and must be left alone.
439 if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
440 kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
441 apic_find_highest_isr(apic));
444 BUG_ON(apic->isr_count < 0);
445 apic->highest_isr_cache = -1;
449 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
453 /* This may race with setting of irr in __apic_accept_irq() and
454 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
455 * will cause vmexit immediately and the value will be recalculated
456 * on the next vmentry.
458 if (!kvm_vcpu_has_lapic(vcpu))
460 highest_irr = apic_find_highest_irr(vcpu->arch.apic);
465 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
466 int vector, int level, int trig_mode,
467 unsigned long *dest_map);
469 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
470 unsigned long *dest_map)
472 struct kvm_lapic *apic = vcpu->arch.apic;
474 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
475 irq->level, irq->trig_mode, dest_map);
478 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
481 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
485 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
488 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
492 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
494 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
497 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
500 if (pv_eoi_get_user(vcpu, &val) < 0)
501 apic_debug("Can't read EOI MSR value: 0x%llx\n",
502 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
506 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
508 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
509 apic_debug("Can't set EOI MSR value: 0x%llx\n",
510 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
513 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
516 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
518 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
519 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
520 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
523 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
526 void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
528 struct kvm_lapic *apic = vcpu->arch.apic;
531 for (i = 0; i < 8; i++)
532 apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
535 static void apic_update_ppr(struct kvm_lapic *apic)
537 u32 tpr, isrv, ppr, old_ppr;
540 old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
541 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
542 isr = apic_find_highest_isr(apic);
543 isrv = (isr != -1) ? isr : 0;
545 if ((tpr & 0xf0) >= (isrv & 0xf0))
550 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
551 apic, ppr, isr, isrv);
553 if (old_ppr != ppr) {
554 apic_set_reg(apic, APIC_PROCPRI, ppr);
556 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
560 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
562 apic_set_reg(apic, APIC_TASKPRI, tpr);
563 apic_update_ppr(apic);
566 static int kvm_apic_broadcast(struct kvm_lapic *apic, u32 dest)
568 return dest == (apic_x2apic_mode(apic) ?
569 X2APIC_BROADCAST : APIC_BROADCAST);
572 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 dest)
574 return kvm_apic_id(apic) == dest || kvm_apic_broadcast(apic, dest);
577 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
582 if (kvm_apic_broadcast(apic, mda))
585 if (apic_x2apic_mode(apic)) {
586 logical_id = kvm_apic_get_reg(apic, APIC_LDR);
587 return logical_id & mda;
590 logical_id = GET_APIC_LOGICAL_ID(kvm_apic_get_reg(apic, APIC_LDR));
592 switch (kvm_apic_get_reg(apic, APIC_DFR)) {
594 if (logical_id & mda)
597 case APIC_DFR_CLUSTER:
598 if (((logical_id >> 4) == (mda >> 0x4))
599 && (logical_id & mda & 0xf))
603 apic_debug("Bad DFR vcpu %d: %08x\n",
604 apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
611 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
612 int short_hand, unsigned int dest, int dest_mode)
615 struct kvm_lapic *target = vcpu->arch.apic;
617 apic_debug("target %p, source %p, dest 0x%x, "
618 "dest_mode 0x%x, short_hand 0x%x\n",
619 target, source, dest, dest_mode, short_hand);
622 switch (short_hand) {
623 case APIC_DEST_NOSHORT:
626 result = kvm_apic_match_physical_addr(target, dest);
629 result = kvm_apic_match_logical_addr(target, dest);
632 result = (target == source);
634 case APIC_DEST_ALLINC:
637 case APIC_DEST_ALLBUT:
638 result = (target != source);
641 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
649 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
650 struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
652 struct kvm_apic_map *map;
653 unsigned long bitmap = 1;
654 struct kvm_lapic **dst;
660 if (irq->shorthand == APIC_DEST_SELF) {
661 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
669 map = rcu_dereference(kvm->arch.apic_map);
674 if (irq->dest_id == map->broadcast)
677 if (irq->dest_mode == 0) { /* physical mode */
678 if (irq->delivery_mode == APIC_DM_LOWEST)
680 dst = &map->phys_map[irq->dest_id & 0xff];
682 u32 mda = irq->dest_id << (32 - map->ldr_bits);
684 dst = map->logical_map[apic_cluster_id(map, mda)];
686 bitmap = apic_logical_id(map, mda);
688 if (irq->delivery_mode == APIC_DM_LOWEST) {
690 for_each_set_bit(i, &bitmap, 16) {
695 else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
699 bitmap = (l >= 0) ? 1 << l : 0;
703 for_each_set_bit(i, &bitmap, 16) {
708 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
718 * Add a pending IRQ into lapic.
719 * Return 1 if successfully added and 0 if discarded.
721 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
722 int vector, int level, int trig_mode,
723 unsigned long *dest_map)
726 struct kvm_vcpu *vcpu = apic->vcpu;
728 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
730 switch (delivery_mode) {
732 vcpu->arch.apic_arb_prio++;
734 /* FIXME add logic for vcpu on reset */
735 if (unlikely(!apic_enabled(apic)))
741 __set_bit(vcpu->vcpu_id, dest_map);
743 if (kvm_x86_ops->deliver_posted_interrupt)
744 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
746 apic_set_irr(vector, apic);
748 kvm_make_request(KVM_REQ_EVENT, vcpu);
755 vcpu->arch.pv.pv_unhalted = 1;
756 kvm_make_request(KVM_REQ_EVENT, vcpu);
761 apic_debug("Ignoring guest SMI\n");
766 kvm_inject_nmi(vcpu);
771 if (!trig_mode || level) {
773 /* assumes that there are only KVM_APIC_INIT/SIPI */
774 apic->pending_events = (1UL << KVM_APIC_INIT);
775 /* make sure pending_events is visible before sending
778 kvm_make_request(KVM_REQ_EVENT, vcpu);
781 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
786 case APIC_DM_STARTUP:
787 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
788 vcpu->vcpu_id, vector);
790 apic->sipi_vector = vector;
791 /* make sure sipi_vector is visible for the receiver */
793 set_bit(KVM_APIC_SIPI, &apic->pending_events);
794 kvm_make_request(KVM_REQ_EVENT, vcpu);
800 * Should only be called by kvm_apic_local_deliver() with LVT0,
801 * before NMI watchdog was enabled. Already handled by
802 * kvm_apic_accept_pic_intr().
807 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
814 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
816 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
819 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
821 if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
822 kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
824 if (apic_test_vector(vector, apic->regs + APIC_TMR))
825 trigger_mode = IOAPIC_LEVEL_TRIG;
827 trigger_mode = IOAPIC_EDGE_TRIG;
828 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
832 static int apic_set_eoi(struct kvm_lapic *apic)
834 int vector = apic_find_highest_isr(apic);
836 trace_kvm_eoi(apic, vector);
839 * Not every write EOI will has corresponding ISR,
840 * one example is when Kernel check timer on setup_IO_APIC
845 apic_clear_isr(vector, apic);
846 apic_update_ppr(apic);
848 kvm_ioapic_send_eoi(apic, vector);
849 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
854 * this interface assumes a trap-like exit, which has already finished
855 * desired side effect including vISR and vPPR update.
857 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
859 struct kvm_lapic *apic = vcpu->arch.apic;
861 trace_kvm_eoi(apic, vector);
863 kvm_ioapic_send_eoi(apic, vector);
864 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
866 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
868 static void apic_send_ipi(struct kvm_lapic *apic)
870 u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
871 u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
872 struct kvm_lapic_irq irq;
874 irq.vector = icr_low & APIC_VECTOR_MASK;
875 irq.delivery_mode = icr_low & APIC_MODE_MASK;
876 irq.dest_mode = icr_low & APIC_DEST_MASK;
877 irq.level = icr_low & APIC_INT_ASSERT;
878 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
879 irq.shorthand = icr_low & APIC_SHORT_MASK;
880 if (apic_x2apic_mode(apic))
881 irq.dest_id = icr_high;
883 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
885 trace_kvm_apic_ipi(icr_low, irq.dest_id);
887 apic_debug("icr_high 0x%x, icr_low 0x%x, "
888 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
889 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
890 icr_high, icr_low, irq.shorthand, irq.dest_id,
891 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
894 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
897 static u32 apic_get_tmcct(struct kvm_lapic *apic)
903 ASSERT(apic != NULL);
905 /* if initial count is 0, current count should also be 0 */
906 if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
907 apic->lapic_timer.period == 0)
910 remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
911 if (ktime_to_ns(remaining) < 0)
912 remaining = ktime_set(0, 0);
914 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
915 tmcct = div64_u64(ns,
916 (APIC_BUS_CYCLE_NS * apic->divide_count));
921 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
923 struct kvm_vcpu *vcpu = apic->vcpu;
924 struct kvm_run *run = vcpu->run;
926 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
927 run->tpr_access.rip = kvm_rip_read(vcpu);
928 run->tpr_access.is_write = write;
931 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
933 if (apic->vcpu->arch.tpr_access_reporting)
934 __report_tpr_access(apic, write);
937 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
941 if (offset >= LAPIC_MMIO_LENGTH)
946 if (apic_x2apic_mode(apic))
947 val = kvm_apic_id(apic);
949 val = kvm_apic_id(apic) << 24;
952 apic_debug("Access APIC ARBPRI register which is for P6\n");
955 case APIC_TMCCT: /* Timer CCR */
956 if (apic_lvtt_tscdeadline(apic))
959 val = apic_get_tmcct(apic);
962 apic_update_ppr(apic);
963 val = kvm_apic_get_reg(apic, offset);
966 report_tpr_access(apic, false);
969 val = kvm_apic_get_reg(apic, offset);
976 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
978 return container_of(dev, struct kvm_lapic, dev);
981 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
984 unsigned char alignment = offset & 0xf;
986 /* this bitmask has a bit cleared for each reserved register */
987 static const u64 rmask = 0x43ff01ffffffe70cULL;
989 if ((alignment + len) > 4) {
990 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
995 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
996 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1001 result = __apic_read(apic, offset & ~0xf);
1003 trace_kvm_apic_read(offset, result);
1009 memcpy(data, (char *)&result + alignment, len);
1012 printk(KERN_ERR "Local APIC read with len = %x, "
1013 "should be 1,2, or 4 instead\n", len);
1019 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1021 return kvm_apic_hw_enabled(apic) &&
1022 addr >= apic->base_address &&
1023 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1026 static int apic_mmio_read(struct kvm_io_device *this,
1027 gpa_t address, int len, void *data)
1029 struct kvm_lapic *apic = to_lapic(this);
1030 u32 offset = address - apic->base_address;
1032 if (!apic_mmio_in_range(apic, address))
1035 apic_reg_read(apic, offset, len, data);
1040 static void update_divide_count(struct kvm_lapic *apic)
1042 u32 tmp1, tmp2, tdcr;
1044 tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1046 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1047 apic->divide_count = 0x1 << (tmp2 & 0x7);
1049 apic_debug("timer divide count is 0x%x\n",
1050 apic->divide_count);
1053 static void start_apic_timer(struct kvm_lapic *apic)
1056 atomic_set(&apic->lapic_timer.pending, 0);
1058 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1059 /* lapic timer in oneshot or periodic mode */
1060 now = apic->lapic_timer.timer.base->get_time();
1061 apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1062 * APIC_BUS_CYCLE_NS * apic->divide_count;
1064 if (!apic->lapic_timer.period)
1067 * Do not allow the guest to program periodic timers with small
1068 * interval, since the hrtimers are not throttled by the host
1071 if (apic_lvtt_period(apic)) {
1072 s64 min_period = min_timer_period_us * 1000LL;
1074 if (apic->lapic_timer.period < min_period) {
1075 pr_info_ratelimited(
1076 "kvm: vcpu %i: requested %lld ns "
1077 "lapic timer period limited to %lld ns\n",
1078 apic->vcpu->vcpu_id,
1079 apic->lapic_timer.period, min_period);
1080 apic->lapic_timer.period = min_period;
1084 hrtimer_start(&apic->lapic_timer.timer,
1085 ktime_add_ns(now, apic->lapic_timer.period),
1088 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1090 "timer initial count 0x%x, period %lldns, "
1091 "expire @ 0x%016" PRIx64 ".\n", __func__,
1092 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1093 kvm_apic_get_reg(apic, APIC_TMICT),
1094 apic->lapic_timer.period,
1095 ktime_to_ns(ktime_add_ns(now,
1096 apic->lapic_timer.period)));
1097 } else if (apic_lvtt_tscdeadline(apic)) {
1098 /* lapic timer in tsc deadline mode */
1099 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1101 struct kvm_vcpu *vcpu = apic->vcpu;
1102 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1103 unsigned long flags;
1105 if (unlikely(!tscdeadline || !this_tsc_khz))
1108 local_irq_save(flags);
1110 now = apic->lapic_timer.timer.base->get_time();
1111 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1112 if (likely(tscdeadline > guest_tsc)) {
1113 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1114 do_div(ns, this_tsc_khz);
1116 hrtimer_start(&apic->lapic_timer.timer,
1117 ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
1119 local_irq_restore(flags);
1123 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1125 int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1127 if (apic_lvt_nmi_mode(lvt0_val)) {
1128 if (!nmi_wd_enabled) {
1129 apic_debug("Receive NMI setting on APIC_LVT0 "
1130 "for cpu %d\n", apic->vcpu->vcpu_id);
1131 apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1133 } else if (nmi_wd_enabled)
1134 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1137 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1141 trace_kvm_apic_write(reg, val);
1144 case APIC_ID: /* Local APIC ID */
1145 if (!apic_x2apic_mode(apic))
1146 kvm_apic_set_id(apic, val >> 24);
1152 report_tpr_access(apic, true);
1153 apic_set_tpr(apic, val & 0xff);
1161 if (!apic_x2apic_mode(apic))
1162 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1168 if (!apic_x2apic_mode(apic)) {
1169 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1170 recalculate_apic_map(apic->vcpu->kvm);
1177 if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1178 mask |= APIC_SPIV_DIRECTED_EOI;
1179 apic_set_spiv(apic, val & mask);
1180 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1184 for (i = 0; i < APIC_LVT_NUM; i++) {
1185 lvt_val = kvm_apic_get_reg(apic,
1186 APIC_LVTT + 0x10 * i);
1187 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1188 lvt_val | APIC_LVT_MASKED);
1190 atomic_set(&apic->lapic_timer.pending, 0);
1196 /* No delay here, so we always clear the pending bit */
1197 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1198 apic_send_ipi(apic);
1202 if (!apic_x2apic_mode(apic))
1204 apic_set_reg(apic, APIC_ICR2, val);
1208 apic_manage_nmi_watchdog(apic, val);
1213 /* TODO: Check vector */
1214 if (!kvm_apic_sw_enabled(apic))
1215 val |= APIC_LVT_MASKED;
1217 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1218 apic_set_reg(apic, reg, val);
1223 if ((kvm_apic_get_reg(apic, APIC_LVTT) &
1224 apic->lapic_timer.timer_mode_mask) !=
1225 (val & apic->lapic_timer.timer_mode_mask))
1226 hrtimer_cancel(&apic->lapic_timer.timer);
1228 if (!kvm_apic_sw_enabled(apic))
1229 val |= APIC_LVT_MASKED;
1230 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1231 apic_set_reg(apic, APIC_LVTT, val);
1235 if (apic_lvtt_tscdeadline(apic))
1238 hrtimer_cancel(&apic->lapic_timer.timer);
1239 apic_set_reg(apic, APIC_TMICT, val);
1240 start_apic_timer(apic);
1245 apic_debug("KVM_WRITE:TDCR %x\n", val);
1246 apic_set_reg(apic, APIC_TDCR, val);
1247 update_divide_count(apic);
1251 if (apic_x2apic_mode(apic) && val != 0) {
1252 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1258 if (apic_x2apic_mode(apic)) {
1259 apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1268 apic_debug("Local APIC Write to read-only register %x\n", reg);
1272 static int apic_mmio_write(struct kvm_io_device *this,
1273 gpa_t address, int len, const void *data)
1275 struct kvm_lapic *apic = to_lapic(this);
1276 unsigned int offset = address - apic->base_address;
1279 if (!apic_mmio_in_range(apic, address))
1283 * APIC register must be aligned on 128-bits boundary.
1284 * 32/64/128 bits registers must be accessed thru 32 bits.
1287 if (len != 4 || (offset & 0xf)) {
1288 /* Don't shout loud, $infamous_os would cause only noise. */
1289 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1295 /* too common printing */
1296 if (offset != APIC_EOI)
1297 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1298 "0x%x\n", __func__, offset, len, val);
1300 apic_reg_write(apic, offset & 0xff0, val);
1305 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1307 if (kvm_vcpu_has_lapic(vcpu))
1308 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1310 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1312 /* emulate APIC access in a trap manner */
1313 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1317 /* hw has done the conditional check and inst decode */
1320 apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1322 /* TODO: optimize to just emulate side effect w/o one more write */
1323 apic_reg_write(vcpu->arch.apic, offset, val);
1325 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1327 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1329 struct kvm_lapic *apic = vcpu->arch.apic;
1331 if (!vcpu->arch.apic)
1334 hrtimer_cancel(&apic->lapic_timer.timer);
1336 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1337 static_key_slow_dec_deferred(&apic_hw_disabled);
1339 if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED))
1340 static_key_slow_dec_deferred(&apic_sw_disabled);
1343 free_page((unsigned long)apic->regs);
1349 *----------------------------------------------------------------------
1351 *----------------------------------------------------------------------
1354 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1356 struct kvm_lapic *apic = vcpu->arch.apic;
1358 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1359 apic_lvtt_period(apic))
1362 return apic->lapic_timer.tscdeadline;
1365 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1367 struct kvm_lapic *apic = vcpu->arch.apic;
1369 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1370 apic_lvtt_period(apic))
1373 hrtimer_cancel(&apic->lapic_timer.timer);
1374 /* Inject here so clearing tscdeadline won't override new value */
1375 if (apic_has_pending_timer(vcpu))
1376 kvm_inject_apic_timer_irqs(vcpu);
1377 apic->lapic_timer.tscdeadline = data;
1378 start_apic_timer(apic);
1381 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1383 struct kvm_lapic *apic = vcpu->arch.apic;
1385 if (!kvm_vcpu_has_lapic(vcpu))
1388 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1389 | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1392 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1396 if (!kvm_vcpu_has_lapic(vcpu))
1399 tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1401 return (tpr & 0xf0) >> 4;
1404 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1406 u64 old_value = vcpu->arch.apic_base;
1407 struct kvm_lapic *apic = vcpu->arch.apic;
1410 value |= MSR_IA32_APICBASE_BSP;
1411 vcpu->arch.apic_base = value;
1415 if (!kvm_vcpu_is_bsp(apic->vcpu))
1416 value &= ~MSR_IA32_APICBASE_BSP;
1417 vcpu->arch.apic_base = value;
1419 /* update jump label if enable bit changes */
1420 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1421 if (value & MSR_IA32_APICBASE_ENABLE)
1422 static_key_slow_dec_deferred(&apic_hw_disabled);
1424 static_key_slow_inc(&apic_hw_disabled.key);
1425 recalculate_apic_map(vcpu->kvm);
1428 if ((old_value ^ value) & X2APIC_ENABLE) {
1429 if (value & X2APIC_ENABLE) {
1430 u32 id = kvm_apic_id(apic);
1431 u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1432 kvm_apic_set_ldr(apic, ldr);
1433 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1435 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1438 apic->base_address = apic->vcpu->arch.apic_base &
1439 MSR_IA32_APICBASE_BASE;
1441 /* with FSB delivery interrupt, we can restart APIC functionality */
1442 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1443 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1447 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1449 struct kvm_lapic *apic;
1452 apic_debug("%s\n", __func__);
1455 apic = vcpu->arch.apic;
1456 ASSERT(apic != NULL);
1458 /* Stop the timer in case it's a reset to an active apic */
1459 hrtimer_cancel(&apic->lapic_timer.timer);
1461 kvm_apic_set_id(apic, vcpu->vcpu_id);
1462 kvm_apic_set_version(apic->vcpu);
1464 for (i = 0; i < APIC_LVT_NUM; i++)
1465 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1466 apic_set_reg(apic, APIC_LVT0,
1467 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1469 apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1470 apic_set_spiv(apic, 0xff);
1471 apic_set_reg(apic, APIC_TASKPRI, 0);
1472 kvm_apic_set_ldr(apic, 0);
1473 apic_set_reg(apic, APIC_ESR, 0);
1474 apic_set_reg(apic, APIC_ICR, 0);
1475 apic_set_reg(apic, APIC_ICR2, 0);
1476 apic_set_reg(apic, APIC_TDCR, 0);
1477 apic_set_reg(apic, APIC_TMICT, 0);
1478 for (i = 0; i < 8; i++) {
1479 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1480 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1481 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1483 apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1484 apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm);
1485 apic->highest_isr_cache = -1;
1486 update_divide_count(apic);
1487 atomic_set(&apic->lapic_timer.pending, 0);
1488 if (kvm_vcpu_is_bsp(vcpu))
1489 kvm_lapic_set_base(vcpu,
1490 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1491 vcpu->arch.pv_eoi.msr_val = 0;
1492 apic_update_ppr(apic);
1494 vcpu->arch.apic_arb_prio = 0;
1495 vcpu->arch.apic_attention = 0;
1497 apic_debug("%s: vcpu=%p, id=%d, base_msr="
1498 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1499 vcpu, kvm_apic_id(apic),
1500 vcpu->arch.apic_base, apic->base_address);
1504 *----------------------------------------------------------------------
1506 *----------------------------------------------------------------------
1509 static bool lapic_is_periodic(struct kvm_lapic *apic)
1511 return apic_lvtt_period(apic);
1514 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1516 struct kvm_lapic *apic = vcpu->arch.apic;
1518 if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1519 apic_lvt_enabled(apic, APIC_LVTT))
1520 return atomic_read(&apic->lapic_timer.pending);
1525 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1527 u32 reg = kvm_apic_get_reg(apic, lvt_type);
1528 int vector, mode, trig_mode;
1530 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1531 vector = reg & APIC_VECTOR_MASK;
1532 mode = reg & APIC_MODE_MASK;
1533 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1534 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1540 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1542 struct kvm_lapic *apic = vcpu->arch.apic;
1545 kvm_apic_local_deliver(apic, APIC_LVT0);
1548 static const struct kvm_io_device_ops apic_mmio_ops = {
1549 .read = apic_mmio_read,
1550 .write = apic_mmio_write,
1553 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1555 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1556 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1557 struct kvm_vcpu *vcpu = apic->vcpu;
1558 wait_queue_head_t *q = &vcpu->wq;
1561 * There is a race window between reading and incrementing, but we do
1562 * not care about potentially losing timer events in the !reinject
1563 * case anyway. Note: KVM_REQ_PENDING_TIMER is implicitly checked
1564 * in vcpu_enter_guest.
1566 if (!atomic_read(&ktimer->pending)) {
1567 atomic_inc(&ktimer->pending);
1568 /* FIXME: this code should not know anything about vcpus */
1569 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1572 if (waitqueue_active(q))
1573 wake_up_interruptible(q);
1575 if (lapic_is_periodic(apic)) {
1576 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1577 return HRTIMER_RESTART;
1579 return HRTIMER_NORESTART;
1582 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1584 struct kvm_lapic *apic;
1586 ASSERT(vcpu != NULL);
1587 apic_debug("apic_init %d\n", vcpu->vcpu_id);
1589 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1593 vcpu->arch.apic = apic;
1595 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1597 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1599 goto nomem_free_apic;
1603 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1605 apic->lapic_timer.timer.function = apic_timer_fn;
1608 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1609 * thinking that APIC satet has changed.
1611 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1612 kvm_lapic_set_base(vcpu,
1613 APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1615 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1616 kvm_lapic_reset(vcpu);
1617 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1626 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1628 struct kvm_lapic *apic = vcpu->arch.apic;
1631 if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1634 apic_update_ppr(apic);
1635 highest_irr = apic_find_highest_irr(apic);
1636 if ((highest_irr == -1) ||
1637 ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1642 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1644 u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1647 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1649 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1650 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1655 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1657 struct kvm_lapic *apic = vcpu->arch.apic;
1659 if (!kvm_vcpu_has_lapic(vcpu))
1662 if (atomic_read(&apic->lapic_timer.pending) > 0) {
1663 kvm_apic_local_deliver(apic, APIC_LVTT);
1664 if (apic_lvtt_tscdeadline(apic))
1665 apic->lapic_timer.tscdeadline = 0;
1666 atomic_set(&apic->lapic_timer.pending, 0);
1670 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1672 int vector = kvm_apic_has_interrupt(vcpu);
1673 struct kvm_lapic *apic = vcpu->arch.apic;
1679 * We get here even with APIC virtualization enabled, if doing
1680 * nested virtualization and L1 runs with the "acknowledge interrupt
1681 * on exit" mode. Then we cannot inject the interrupt via RVI,
1682 * because the process would deliver it through the IDT.
1685 apic_set_isr(vector, apic);
1686 apic_update_ppr(apic);
1687 apic_clear_irr(vector, apic);
1691 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1692 struct kvm_lapic_state *s)
1694 struct kvm_lapic *apic = vcpu->arch.apic;
1696 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1697 /* set SPIV separately to get count of SW disabled APICs right */
1698 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1699 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1700 /* call kvm_apic_set_id() to put apic into apic_map */
1701 kvm_apic_set_id(apic, kvm_apic_id(apic));
1702 kvm_apic_set_version(vcpu);
1704 apic_update_ppr(apic);
1705 hrtimer_cancel(&apic->lapic_timer.timer);
1706 update_divide_count(apic);
1707 start_apic_timer(apic);
1708 apic->irr_pending = true;
1709 apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm) ?
1710 1 : count_vectors(apic->regs + APIC_ISR);
1711 apic->highest_isr_cache = -1;
1712 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, apic_find_highest_isr(apic));
1713 kvm_make_request(KVM_REQ_EVENT, vcpu);
1714 kvm_rtc_eoi_tracking_restore_one(vcpu);
1717 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1719 struct hrtimer *timer;
1721 if (!kvm_vcpu_has_lapic(vcpu))
1724 timer = &vcpu->arch.apic->lapic_timer.timer;
1725 if (hrtimer_cancel(timer))
1726 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1730 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1732 * Detect whether guest triggered PV EOI since the
1733 * last entry. If yes, set EOI on guests's behalf.
1734 * Clear PV EOI in guest memory in any case.
1736 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1737 struct kvm_lapic *apic)
1742 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1743 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1745 * KVM_APIC_PV_EOI_PENDING is unset:
1746 * -> host disabled PV EOI.
1747 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1748 * -> host enabled PV EOI, guest did not execute EOI yet.
1749 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1750 * -> host enabled PV EOI, guest executed EOI.
1752 BUG_ON(!pv_eoi_enabled(vcpu));
1753 pending = pv_eoi_get_pending(vcpu);
1755 * Clear pending bit in any case: it will be set again on vmentry.
1756 * While this might not be ideal from performance point of view,
1757 * this makes sure pv eoi is only enabled when we know it's safe.
1759 pv_eoi_clr_pending(vcpu);
1762 vector = apic_set_eoi(apic);
1763 trace_kvm_pv_eoi(apic, vector);
1766 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1770 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1771 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1773 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1776 kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1779 apic_set_tpr(vcpu->arch.apic, data & 0xff);
1783 * apic_sync_pv_eoi_to_guest - called before vmentry
1785 * Detect whether it's safe to enable PV EOI and
1788 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1789 struct kvm_lapic *apic)
1791 if (!pv_eoi_enabled(vcpu) ||
1792 /* IRR set or many bits in ISR: could be nested. */
1793 apic->irr_pending ||
1794 /* Cache not set: could be safe but we don't bother. */
1795 apic->highest_isr_cache == -1 ||
1796 /* Need EOI to update ioapic. */
1797 kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1799 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1800 * so we need not do anything here.
1805 pv_eoi_set_pending(apic->vcpu);
1808 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1811 int max_irr, max_isr;
1812 struct kvm_lapic *apic = vcpu->arch.apic;
1814 apic_sync_pv_eoi_to_guest(vcpu, apic);
1816 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1819 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1820 max_irr = apic_find_highest_irr(apic);
1823 max_isr = apic_find_highest_isr(apic);
1826 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1828 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1832 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1835 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1836 &vcpu->arch.apic->vapic_cache,
1837 vapic_addr, sizeof(u32)))
1839 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1841 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1844 vcpu->arch.apic->vapic_addr = vapic_addr;
1848 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1850 struct kvm_lapic *apic = vcpu->arch.apic;
1851 u32 reg = (msr - APIC_BASE_MSR) << 4;
1853 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1856 /* if this is ICR write vector before command */
1858 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1859 return apic_reg_write(apic, reg, (u32)data);
1862 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1864 struct kvm_lapic *apic = vcpu->arch.apic;
1865 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1867 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1870 if (apic_reg_read(apic, reg, 4, &low))
1873 apic_reg_read(apic, APIC_ICR2, 4, &high);
1875 *data = (((u64)high) << 32) | low;
1880 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1882 struct kvm_lapic *apic = vcpu->arch.apic;
1884 if (!kvm_vcpu_has_lapic(vcpu))
1887 /* if this is ICR write vector before command */
1888 if (reg == APIC_ICR)
1889 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1890 return apic_reg_write(apic, reg, (u32)data);
1893 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1895 struct kvm_lapic *apic = vcpu->arch.apic;
1898 if (!kvm_vcpu_has_lapic(vcpu))
1901 if (apic_reg_read(apic, reg, 4, &low))
1903 if (reg == APIC_ICR)
1904 apic_reg_read(apic, APIC_ICR2, 4, &high);
1906 *data = (((u64)high) << 32) | low;
1911 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
1913 u64 addr = data & ~KVM_MSR_ENABLED;
1914 if (!IS_ALIGNED(addr, 4))
1917 vcpu->arch.pv_eoi.msr_val = data;
1918 if (!pv_eoi_enabled(vcpu))
1920 return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
1924 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
1926 struct kvm_lapic *apic = vcpu->arch.apic;
1927 unsigned int sipi_vector;
1930 if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
1933 pe = xchg(&apic->pending_events, 0);
1935 if (test_bit(KVM_APIC_INIT, &pe)) {
1936 kvm_lapic_reset(vcpu);
1937 kvm_vcpu_reset(vcpu);
1938 if (kvm_vcpu_is_bsp(apic->vcpu))
1939 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1941 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
1943 if (test_bit(KVM_APIC_SIPI, &pe) &&
1944 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
1945 /* evaluate pending_events before reading the vector */
1947 sipi_vector = apic->sipi_vector;
1948 apic_debug("vcpu %d received sipi with vector # %x\n",
1949 vcpu->vcpu_id, sipi_vector);
1950 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
1951 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1955 void kvm_lapic_init(void)
1957 /* do not patch jump label more than once per second */
1958 jump_label_rate_limit(&apic_hw_disabled, HZ);
1959 jump_label_rate_limit(&apic_sw_disabled, HZ);