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 <asm/delay.h>
37 #include <linux/atomic.h>
38 #include <linux/jump_label.h>
39 #include "kvm_cache_regs.h"
46 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
48 #define mod_64(x, y) ((x) % (y))
56 #define APIC_BUS_CYCLE_NS 1
58 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
59 #define apic_debug(fmt, arg...)
61 #define APIC_LVT_NUM 6
62 /* 14 is the version for Xeon and Pentium 8.4.8*/
63 #define APIC_VERSION (0x14UL | ((APIC_LVT_NUM - 1) << 16))
64 #define LAPIC_MMIO_LENGTH (1 << 12)
65 /* followed define is not in apicdef.h */
66 #define APIC_SHORT_MASK 0xc0000
67 #define APIC_DEST_NOSHORT 0x0
68 #define APIC_DEST_MASK 0x800
69 #define MAX_APIC_VECTOR 256
70 #define APIC_VECTORS_PER_REG 32
72 #define APIC_BROADCAST 0xFF
73 #define X2APIC_BROADCAST 0xFFFFFFFFul
75 #define VEC_POS(v) ((v) & (32 - 1))
76 #define REG_POS(v) (((v) >> 5) << 4)
78 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
80 *((u32 *) (apic->regs + reg_off)) = val;
83 static inline int apic_test_vector(int vec, void *bitmap)
85 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
88 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
90 struct kvm_lapic *apic = vcpu->arch.apic;
92 return apic_test_vector(vector, apic->regs + APIC_ISR) ||
93 apic_test_vector(vector, apic->regs + APIC_IRR);
96 static inline void apic_set_vector(int vec, void *bitmap)
98 set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
101 static inline void apic_clear_vector(int vec, void *bitmap)
103 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
106 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
108 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
111 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
113 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
116 struct static_key_deferred apic_hw_disabled __read_mostly;
117 struct static_key_deferred apic_sw_disabled __read_mostly;
119 static inline int apic_enabled(struct kvm_lapic *apic)
121 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
125 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
128 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
129 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
131 static inline int kvm_apic_id(struct kvm_lapic *apic)
133 return (kvm_apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
136 static void recalculate_apic_map(struct kvm *kvm)
138 struct kvm_apic_map *new, *old = NULL;
139 struct kvm_vcpu *vcpu;
142 new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
144 mutex_lock(&kvm->arch.apic_map_lock);
150 /* flat mode is default */
153 new->lid_mask = 0xff;
155 kvm_for_each_vcpu(i, vcpu, kvm) {
156 struct kvm_lapic *apic = vcpu->arch.apic;
158 if (!kvm_apic_present(vcpu))
161 if (apic_x2apic_mode(apic)) {
164 new->cid_mask = new->lid_mask = 0xffff;
165 } else if (kvm_apic_get_reg(apic, APIC_LDR)) {
166 if (kvm_apic_get_reg(apic, APIC_DFR) ==
174 new->lid_mask = 0xff;
179 * All APICs have to be configured in the same mode by an OS.
180 * We take advatage of this while building logical id loockup
181 * table. After reset APICs are in software disabled mode, so if
182 * we find apic with different setting we assume this is the mode
183 * OS wants all apics to be in; build lookup table accordingly.
185 if (kvm_apic_sw_enabled(apic))
189 kvm_for_each_vcpu(i, vcpu, kvm) {
190 struct kvm_lapic *apic = vcpu->arch.apic;
194 if (!kvm_apic_present(vcpu))
197 aid = kvm_apic_id(apic);
198 ldr = kvm_apic_get_reg(apic, APIC_LDR);
199 cid = apic_cluster_id(new, ldr);
200 lid = apic_logical_id(new, ldr);
202 if (aid < ARRAY_SIZE(new->phys_map))
203 new->phys_map[aid] = apic;
204 if (lid && cid < ARRAY_SIZE(new->logical_map))
205 new->logical_map[cid][ffs(lid) - 1] = apic;
208 old = rcu_dereference_protected(kvm->arch.apic_map,
209 lockdep_is_held(&kvm->arch.apic_map_lock));
210 rcu_assign_pointer(kvm->arch.apic_map, new);
211 mutex_unlock(&kvm->arch.apic_map_lock);
216 kvm_vcpu_request_scan_ioapic(kvm);
219 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
221 bool enabled = val & APIC_SPIV_APIC_ENABLED;
223 apic_set_reg(apic, APIC_SPIV, val);
225 if (enabled != apic->sw_enabled) {
226 apic->sw_enabled = enabled;
228 static_key_slow_dec_deferred(&apic_sw_disabled);
229 recalculate_apic_map(apic->vcpu->kvm);
231 static_key_slow_inc(&apic_sw_disabled.key);
235 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
237 apic_set_reg(apic, APIC_ID, id << 24);
238 recalculate_apic_map(apic->vcpu->kvm);
241 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
243 apic_set_reg(apic, APIC_LDR, id);
244 recalculate_apic_map(apic->vcpu->kvm);
247 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
249 return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
252 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
254 return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
257 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
259 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
262 static inline int apic_lvtt_period(struct kvm_lapic *apic)
264 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
267 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
269 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
272 static inline int apic_lvt_nmi_mode(u32 lvt_val)
274 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
277 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
279 struct kvm_lapic *apic = vcpu->arch.apic;
280 struct kvm_cpuid_entry2 *feat;
281 u32 v = APIC_VERSION;
283 if (!kvm_vcpu_has_lapic(vcpu))
286 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
287 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
288 v |= APIC_LVR_DIRECTED_EOI;
289 apic_set_reg(apic, APIC_LVR, v);
292 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
293 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
294 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
295 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
296 LINT_MASK, LINT_MASK, /* LVT0-1 */
297 LVT_MASK /* LVTERR */
300 static int find_highest_vector(void *bitmap)
305 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
306 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
307 reg = bitmap + REG_POS(vec);
309 return fls(*reg) - 1 + vec;
315 static u8 count_vectors(void *bitmap)
321 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
322 reg = bitmap + REG_POS(vec);
323 count += hweight32(*reg);
329 void __kvm_apic_update_irr(u32 *pir, void *regs)
333 for (i = 0; i <= 7; i++) {
334 pir_val = xchg(&pir[i], 0);
336 *((u32 *)(regs + APIC_IRR + i * 0x10)) |= pir_val;
339 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
341 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
343 struct kvm_lapic *apic = vcpu->arch.apic;
345 __kvm_apic_update_irr(pir, apic->regs);
347 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
349 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
351 apic_set_vector(vec, apic->regs + APIC_IRR);
353 * irr_pending must be true if any interrupt is pending; set it after
354 * APIC_IRR to avoid race with apic_clear_irr
356 apic->irr_pending = true;
359 static inline int apic_search_irr(struct kvm_lapic *apic)
361 return find_highest_vector(apic->regs + APIC_IRR);
364 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
369 * Note that irr_pending is just a hint. It will be always
370 * true with virtual interrupt delivery enabled.
372 if (!apic->irr_pending)
375 kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
376 result = apic_search_irr(apic);
377 ASSERT(result == -1 || result >= 16);
382 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
384 struct kvm_vcpu *vcpu;
388 if (unlikely(kvm_apic_vid_enabled(vcpu->kvm))) {
389 /* try to update RVI */
390 apic_clear_vector(vec, apic->regs + APIC_IRR);
391 kvm_make_request(KVM_REQ_EVENT, vcpu);
393 apic->irr_pending = false;
394 apic_clear_vector(vec, apic->regs + APIC_IRR);
395 if (apic_search_irr(apic) != -1)
396 apic->irr_pending = true;
400 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
402 struct kvm_vcpu *vcpu;
404 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
410 * With APIC virtualization enabled, all caching is disabled
411 * because the processor can modify ISR under the hood. Instead
414 if (unlikely(kvm_x86_ops->hwapic_isr_update))
415 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
418 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
420 * ISR (in service register) bit is set when injecting an interrupt.
421 * The highest vector is injected. Thus the latest bit set matches
422 * the highest bit in ISR.
424 apic->highest_isr_cache = vec;
428 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
433 * Note that isr_count is always 1, and highest_isr_cache
434 * is always -1, with APIC virtualization enabled.
436 if (!apic->isr_count)
438 if (likely(apic->highest_isr_cache != -1))
439 return apic->highest_isr_cache;
441 result = find_highest_vector(apic->regs + APIC_ISR);
442 ASSERT(result == -1 || result >= 16);
447 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
449 struct kvm_vcpu *vcpu;
450 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
456 * We do get here for APIC virtualization enabled if the guest
457 * uses the Hyper-V APIC enlightenment. In this case we may need
458 * to trigger a new interrupt delivery by writing the SVI field;
459 * on the other hand isr_count and highest_isr_cache are unused
460 * and must be left alone.
462 if (unlikely(kvm_x86_ops->hwapic_isr_update))
463 kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
464 apic_find_highest_isr(apic));
467 BUG_ON(apic->isr_count < 0);
468 apic->highest_isr_cache = -1;
472 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
476 /* This may race with setting of irr in __apic_accept_irq() and
477 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
478 * will cause vmexit immediately and the value will be recalculated
479 * on the next vmentry.
481 if (!kvm_vcpu_has_lapic(vcpu))
483 highest_irr = apic_find_highest_irr(vcpu->arch.apic);
488 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
489 int vector, int level, int trig_mode,
490 unsigned long *dest_map);
492 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
493 unsigned long *dest_map)
495 struct kvm_lapic *apic = vcpu->arch.apic;
497 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
498 irq->level, irq->trig_mode, dest_map);
501 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
504 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
508 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
511 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
515 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
517 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
520 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
523 if (pv_eoi_get_user(vcpu, &val) < 0)
524 apic_debug("Can't read EOI MSR value: 0x%llx\n",
525 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
529 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
531 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
532 apic_debug("Can't set EOI MSR value: 0x%llx\n",
533 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
536 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
539 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
541 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
542 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
543 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
546 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
549 void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
551 struct kvm_lapic *apic = vcpu->arch.apic;
554 for (i = 0; i < 8; i++)
555 apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
558 static void apic_update_ppr(struct kvm_lapic *apic)
560 u32 tpr, isrv, ppr, old_ppr;
563 old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
564 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
565 isr = apic_find_highest_isr(apic);
566 isrv = (isr != -1) ? isr : 0;
568 if ((tpr & 0xf0) >= (isrv & 0xf0))
573 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
574 apic, ppr, isr, isrv);
576 if (old_ppr != ppr) {
577 apic_set_reg(apic, APIC_PROCPRI, ppr);
579 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
583 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
585 apic_set_reg(apic, APIC_TASKPRI, tpr);
586 apic_update_ppr(apic);
589 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
591 if (apic_x2apic_mode(apic))
592 return mda == X2APIC_BROADCAST;
594 return GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST;
597 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
599 if (kvm_apic_broadcast(apic, mda))
602 if (apic_x2apic_mode(apic))
603 return mda == kvm_apic_id(apic);
605 return mda == SET_APIC_DEST_FIELD(kvm_apic_id(apic));
608 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
612 if (kvm_apic_broadcast(apic, mda))
615 logical_id = kvm_apic_get_reg(apic, APIC_LDR);
617 if (apic_x2apic_mode(apic))
618 return ((logical_id >> 16) == (mda >> 16))
619 && (logical_id & mda & 0xffff) != 0;
621 logical_id = GET_APIC_LOGICAL_ID(logical_id);
622 mda = GET_APIC_DEST_FIELD(mda);
624 switch (kvm_apic_get_reg(apic, APIC_DFR)) {
626 return (logical_id & mda) != 0;
627 case APIC_DFR_CLUSTER:
628 return ((logical_id >> 4) == (mda >> 4))
629 && (logical_id & mda & 0xf) != 0;
631 apic_debug("Bad DFR vcpu %d: %08x\n",
632 apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
637 /* KVM APIC implementation has two quirks
638 * - dest always begins at 0 while xAPIC MDA has offset 24,
639 * - IOxAPIC messages have to be delivered (directly) to x2APIC.
641 static u32 kvm_apic_mda(unsigned int dest_id, struct kvm_lapic *source,
642 struct kvm_lapic *target)
644 bool ipi = source != NULL;
645 bool x2apic_mda = apic_x2apic_mode(ipi ? source : target);
647 if (!ipi && dest_id == APIC_BROADCAST && x2apic_mda)
648 return X2APIC_BROADCAST;
650 return x2apic_mda ? dest_id : SET_APIC_DEST_FIELD(dest_id);
653 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
654 int short_hand, unsigned int dest, int dest_mode)
656 struct kvm_lapic *target = vcpu->arch.apic;
657 u32 mda = kvm_apic_mda(dest, source, target);
659 apic_debug("target %p, source %p, dest 0x%x, "
660 "dest_mode 0x%x, short_hand 0x%x\n",
661 target, source, dest, dest_mode, short_hand);
664 switch (short_hand) {
665 case APIC_DEST_NOSHORT:
666 if (dest_mode == APIC_DEST_PHYSICAL)
667 return kvm_apic_match_physical_addr(target, mda);
669 return kvm_apic_match_logical_addr(target, mda);
671 return target == source;
672 case APIC_DEST_ALLINC:
674 case APIC_DEST_ALLBUT:
675 return target != source;
677 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
683 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
684 struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
686 struct kvm_apic_map *map;
687 unsigned long bitmap = 1;
688 struct kvm_lapic **dst;
691 bool x2apic_ipi = src && apic_x2apic_mode(src);
695 if (irq->shorthand == APIC_DEST_SELF) {
696 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
703 if (irq->dest_id == (x2apic_ipi ? X2APIC_BROADCAST : APIC_BROADCAST))
707 map = rcu_dereference(kvm->arch.apic_map);
714 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
715 if (irq->dest_id >= ARRAY_SIZE(map->phys_map))
718 dst = &map->phys_map[irq->dest_id];
720 u32 mda = irq->dest_id << (32 - map->ldr_bits);
721 u16 cid = apic_cluster_id(map, mda);
723 if (cid >= ARRAY_SIZE(map->logical_map))
726 dst = map->logical_map[cid];
728 bitmap = apic_logical_id(map, mda);
730 if (irq->delivery_mode == APIC_DM_LOWEST) {
732 for_each_set_bit(i, &bitmap, 16) {
737 else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
741 bitmap = (l >= 0) ? 1 << l : 0;
745 for_each_set_bit(i, &bitmap, 16) {
750 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
758 * Add a pending IRQ into lapic.
759 * Return 1 if successfully added and 0 if discarded.
761 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
762 int vector, int level, int trig_mode,
763 unsigned long *dest_map)
766 struct kvm_vcpu *vcpu = apic->vcpu;
768 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
770 switch (delivery_mode) {
772 vcpu->arch.apic_arb_prio++;
774 /* FIXME add logic for vcpu on reset */
775 if (unlikely(!apic_enabled(apic)))
781 __set_bit(vcpu->vcpu_id, dest_map);
783 if (kvm_x86_ops->deliver_posted_interrupt)
784 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
786 apic_set_irr(vector, apic);
788 kvm_make_request(KVM_REQ_EVENT, vcpu);
795 vcpu->arch.pv.pv_unhalted = 1;
796 kvm_make_request(KVM_REQ_EVENT, vcpu);
801 apic_debug("Ignoring guest SMI\n");
806 kvm_inject_nmi(vcpu);
811 if (!trig_mode || level) {
813 /* assumes that there are only KVM_APIC_INIT/SIPI */
814 apic->pending_events = (1UL << KVM_APIC_INIT);
815 /* make sure pending_events is visible before sending
818 kvm_make_request(KVM_REQ_EVENT, vcpu);
821 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
826 case APIC_DM_STARTUP:
827 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
828 vcpu->vcpu_id, vector);
830 apic->sipi_vector = vector;
831 /* make sure sipi_vector is visible for the receiver */
833 set_bit(KVM_APIC_SIPI, &apic->pending_events);
834 kvm_make_request(KVM_REQ_EVENT, vcpu);
840 * Should only be called by kvm_apic_local_deliver() with LVT0,
841 * before NMI watchdog was enabled. Already handled by
842 * kvm_apic_accept_pic_intr().
847 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
854 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
856 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
859 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
861 if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
862 kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
864 if (apic_test_vector(vector, apic->regs + APIC_TMR))
865 trigger_mode = IOAPIC_LEVEL_TRIG;
867 trigger_mode = IOAPIC_EDGE_TRIG;
868 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
872 static int apic_set_eoi(struct kvm_lapic *apic)
874 int vector = apic_find_highest_isr(apic);
876 trace_kvm_eoi(apic, vector);
879 * Not every write EOI will has corresponding ISR,
880 * one example is when Kernel check timer on setup_IO_APIC
885 apic_clear_isr(vector, apic);
886 apic_update_ppr(apic);
888 kvm_ioapic_send_eoi(apic, vector);
889 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
894 * this interface assumes a trap-like exit, which has already finished
895 * desired side effect including vISR and vPPR update.
897 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
899 struct kvm_lapic *apic = vcpu->arch.apic;
901 trace_kvm_eoi(apic, vector);
903 kvm_ioapic_send_eoi(apic, vector);
904 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
906 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
908 static void apic_send_ipi(struct kvm_lapic *apic)
910 u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
911 u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
912 struct kvm_lapic_irq irq;
914 irq.vector = icr_low & APIC_VECTOR_MASK;
915 irq.delivery_mode = icr_low & APIC_MODE_MASK;
916 irq.dest_mode = icr_low & APIC_DEST_MASK;
917 irq.level = icr_low & APIC_INT_ASSERT;
918 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
919 irq.shorthand = icr_low & APIC_SHORT_MASK;
920 if (apic_x2apic_mode(apic))
921 irq.dest_id = icr_high;
923 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
925 trace_kvm_apic_ipi(icr_low, irq.dest_id);
927 apic_debug("icr_high 0x%x, icr_low 0x%x, "
928 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
929 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
930 icr_high, icr_low, irq.shorthand, irq.dest_id,
931 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
934 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
937 static u32 apic_get_tmcct(struct kvm_lapic *apic)
943 ASSERT(apic != NULL);
945 /* if initial count is 0, current count should also be 0 */
946 if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
947 apic->lapic_timer.period == 0)
950 remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
951 if (ktime_to_ns(remaining) < 0)
952 remaining = ktime_set(0, 0);
954 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
955 tmcct = div64_u64(ns,
956 (APIC_BUS_CYCLE_NS * apic->divide_count));
961 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
963 struct kvm_vcpu *vcpu = apic->vcpu;
964 struct kvm_run *run = vcpu->run;
966 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
967 run->tpr_access.rip = kvm_rip_read(vcpu);
968 run->tpr_access.is_write = write;
971 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
973 if (apic->vcpu->arch.tpr_access_reporting)
974 __report_tpr_access(apic, write);
977 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
981 if (offset >= LAPIC_MMIO_LENGTH)
986 if (apic_x2apic_mode(apic))
987 val = kvm_apic_id(apic);
989 val = kvm_apic_id(apic) << 24;
992 apic_debug("Access APIC ARBPRI register which is for P6\n");
995 case APIC_TMCCT: /* Timer CCR */
996 if (apic_lvtt_tscdeadline(apic))
999 val = apic_get_tmcct(apic);
1002 apic_update_ppr(apic);
1003 val = kvm_apic_get_reg(apic, offset);
1006 report_tpr_access(apic, false);
1009 val = kvm_apic_get_reg(apic, offset);
1016 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1018 return container_of(dev, struct kvm_lapic, dev);
1021 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1024 unsigned char alignment = offset & 0xf;
1026 /* this bitmask has a bit cleared for each reserved register */
1027 static const u64 rmask = 0x43ff01ffffffe70cULL;
1029 if ((alignment + len) > 4) {
1030 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1035 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1036 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1041 result = __apic_read(apic, offset & ~0xf);
1043 trace_kvm_apic_read(offset, result);
1049 memcpy(data, (char *)&result + alignment, len);
1052 printk(KERN_ERR "Local APIC read with len = %x, "
1053 "should be 1,2, or 4 instead\n", len);
1059 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1061 return kvm_apic_hw_enabled(apic) &&
1062 addr >= apic->base_address &&
1063 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1066 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1067 gpa_t address, int len, void *data)
1069 struct kvm_lapic *apic = to_lapic(this);
1070 u32 offset = address - apic->base_address;
1072 if (!apic_mmio_in_range(apic, address))
1075 apic_reg_read(apic, offset, len, data);
1080 static void update_divide_count(struct kvm_lapic *apic)
1082 u32 tmp1, tmp2, tdcr;
1084 tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1086 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1087 apic->divide_count = 0x1 << (tmp2 & 0x7);
1089 apic_debug("timer divide count is 0x%x\n",
1090 apic->divide_count);
1093 static void apic_timer_expired(struct kvm_lapic *apic)
1095 struct kvm_vcpu *vcpu = apic->vcpu;
1096 wait_queue_head_t *q = &vcpu->wq;
1097 struct kvm_timer *ktimer = &apic->lapic_timer;
1099 if (atomic_read(&apic->lapic_timer.pending))
1102 atomic_inc(&apic->lapic_timer.pending);
1103 kvm_set_pending_timer(vcpu);
1105 if (waitqueue_active(q))
1106 wake_up_interruptible(q);
1108 if (apic_lvtt_tscdeadline(apic))
1109 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1113 * On APICv, this test will cause a busy wait
1114 * during a higher-priority task.
1117 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1119 struct kvm_lapic *apic = vcpu->arch.apic;
1120 u32 reg = kvm_apic_get_reg(apic, APIC_LVTT);
1122 if (kvm_apic_hw_enabled(apic)) {
1123 int vec = reg & APIC_VECTOR_MASK;
1124 void *bitmap = apic->regs + APIC_ISR;
1126 if (kvm_x86_ops->deliver_posted_interrupt)
1127 bitmap = apic->regs + APIC_IRR;
1129 if (apic_test_vector(vec, bitmap))
1135 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1137 struct kvm_lapic *apic = vcpu->arch.apic;
1138 u64 guest_tsc, tsc_deadline;
1140 if (!kvm_vcpu_has_lapic(vcpu))
1143 if (apic->lapic_timer.expired_tscdeadline == 0)
1146 if (!lapic_timer_int_injected(vcpu))
1149 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1150 apic->lapic_timer.expired_tscdeadline = 0;
1151 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1152 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1154 /* __delay is delay_tsc whenever the hardware has TSC, thus always. */
1155 if (guest_tsc < tsc_deadline)
1156 __delay(tsc_deadline - guest_tsc);
1159 static void start_apic_timer(struct kvm_lapic *apic)
1163 atomic_set(&apic->lapic_timer.pending, 0);
1165 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1166 /* lapic timer in oneshot or periodic mode */
1167 now = apic->lapic_timer.timer.base->get_time();
1168 apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1169 * APIC_BUS_CYCLE_NS * apic->divide_count;
1171 if (!apic->lapic_timer.period)
1174 * Do not allow the guest to program periodic timers with small
1175 * interval, since the hrtimers are not throttled by the host
1178 if (apic_lvtt_period(apic)) {
1179 s64 min_period = min_timer_period_us * 1000LL;
1181 if (apic->lapic_timer.period < min_period) {
1182 pr_info_ratelimited(
1183 "kvm: vcpu %i: requested %lld ns "
1184 "lapic timer period limited to %lld ns\n",
1185 apic->vcpu->vcpu_id,
1186 apic->lapic_timer.period, min_period);
1187 apic->lapic_timer.period = min_period;
1191 hrtimer_start(&apic->lapic_timer.timer,
1192 ktime_add_ns(now, apic->lapic_timer.period),
1195 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1197 "timer initial count 0x%x, period %lldns, "
1198 "expire @ 0x%016" PRIx64 ".\n", __func__,
1199 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1200 kvm_apic_get_reg(apic, APIC_TMICT),
1201 apic->lapic_timer.period,
1202 ktime_to_ns(ktime_add_ns(now,
1203 apic->lapic_timer.period)));
1204 } else if (apic_lvtt_tscdeadline(apic)) {
1205 /* lapic timer in tsc deadline mode */
1206 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1209 struct kvm_vcpu *vcpu = apic->vcpu;
1210 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1211 unsigned long flags;
1213 if (unlikely(!tscdeadline || !this_tsc_khz))
1216 local_irq_save(flags);
1218 now = apic->lapic_timer.timer.base->get_time();
1219 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1220 if (likely(tscdeadline > guest_tsc)) {
1221 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1222 do_div(ns, this_tsc_khz);
1223 expire = ktime_add_ns(now, ns);
1224 expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1225 hrtimer_start(&apic->lapic_timer.timer,
1226 expire, HRTIMER_MODE_ABS);
1228 apic_timer_expired(apic);
1230 local_irq_restore(flags);
1234 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1236 int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1238 if (apic_lvt_nmi_mode(lvt0_val)) {
1239 if (!nmi_wd_enabled) {
1240 apic_debug("Receive NMI setting on APIC_LVT0 "
1241 "for cpu %d\n", apic->vcpu->vcpu_id);
1242 apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1244 } else if (nmi_wd_enabled)
1245 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1248 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1252 trace_kvm_apic_write(reg, val);
1255 case APIC_ID: /* Local APIC ID */
1256 if (!apic_x2apic_mode(apic))
1257 kvm_apic_set_id(apic, val >> 24);
1263 report_tpr_access(apic, true);
1264 apic_set_tpr(apic, val & 0xff);
1272 if (!apic_x2apic_mode(apic))
1273 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1279 if (!apic_x2apic_mode(apic)) {
1280 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1281 recalculate_apic_map(apic->vcpu->kvm);
1288 if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1289 mask |= APIC_SPIV_DIRECTED_EOI;
1290 apic_set_spiv(apic, val & mask);
1291 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1295 for (i = 0; i < APIC_LVT_NUM; i++) {
1296 lvt_val = kvm_apic_get_reg(apic,
1297 APIC_LVTT + 0x10 * i);
1298 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1299 lvt_val | APIC_LVT_MASKED);
1301 atomic_set(&apic->lapic_timer.pending, 0);
1307 /* No delay here, so we always clear the pending bit */
1308 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1309 apic_send_ipi(apic);
1313 if (!apic_x2apic_mode(apic))
1315 apic_set_reg(apic, APIC_ICR2, val);
1319 apic_manage_nmi_watchdog(apic, val);
1324 /* TODO: Check vector */
1325 if (!kvm_apic_sw_enabled(apic))
1326 val |= APIC_LVT_MASKED;
1328 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1329 apic_set_reg(apic, reg, val);
1334 u32 timer_mode = val & apic->lapic_timer.timer_mode_mask;
1336 if (apic->lapic_timer.timer_mode != timer_mode) {
1337 apic->lapic_timer.timer_mode = timer_mode;
1338 hrtimer_cancel(&apic->lapic_timer.timer);
1341 if (!kvm_apic_sw_enabled(apic))
1342 val |= APIC_LVT_MASKED;
1343 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1344 apic_set_reg(apic, APIC_LVTT, val);
1349 if (apic_lvtt_tscdeadline(apic))
1352 hrtimer_cancel(&apic->lapic_timer.timer);
1353 apic_set_reg(apic, APIC_TMICT, val);
1354 start_apic_timer(apic);
1359 apic_debug("KVM_WRITE:TDCR %x\n", val);
1360 apic_set_reg(apic, APIC_TDCR, val);
1361 update_divide_count(apic);
1365 if (apic_x2apic_mode(apic) && val != 0) {
1366 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1372 if (apic_x2apic_mode(apic)) {
1373 apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1382 apic_debug("Local APIC Write to read-only register %x\n", reg);
1386 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1387 gpa_t address, int len, const void *data)
1389 struct kvm_lapic *apic = to_lapic(this);
1390 unsigned int offset = address - apic->base_address;
1393 if (!apic_mmio_in_range(apic, address))
1397 * APIC register must be aligned on 128-bits boundary.
1398 * 32/64/128 bits registers must be accessed thru 32 bits.
1401 if (len != 4 || (offset & 0xf)) {
1402 /* Don't shout loud, $infamous_os would cause only noise. */
1403 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1409 /* too common printing */
1410 if (offset != APIC_EOI)
1411 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1412 "0x%x\n", __func__, offset, len, val);
1414 apic_reg_write(apic, offset & 0xff0, val);
1419 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1421 if (kvm_vcpu_has_lapic(vcpu))
1422 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1424 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1426 /* emulate APIC access in a trap manner */
1427 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1431 /* hw has done the conditional check and inst decode */
1434 apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1436 /* TODO: optimize to just emulate side effect w/o one more write */
1437 apic_reg_write(vcpu->arch.apic, offset, val);
1439 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1441 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1443 struct kvm_lapic *apic = vcpu->arch.apic;
1445 if (!vcpu->arch.apic)
1448 hrtimer_cancel(&apic->lapic_timer.timer);
1450 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1451 static_key_slow_dec_deferred(&apic_hw_disabled);
1453 if (!apic->sw_enabled)
1454 static_key_slow_dec_deferred(&apic_sw_disabled);
1457 free_page((unsigned long)apic->regs);
1463 *----------------------------------------------------------------------
1465 *----------------------------------------------------------------------
1468 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1470 struct kvm_lapic *apic = vcpu->arch.apic;
1472 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1473 apic_lvtt_period(apic))
1476 return apic->lapic_timer.tscdeadline;
1479 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1481 struct kvm_lapic *apic = vcpu->arch.apic;
1483 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1484 apic_lvtt_period(apic))
1487 hrtimer_cancel(&apic->lapic_timer.timer);
1488 apic->lapic_timer.tscdeadline = data;
1489 start_apic_timer(apic);
1492 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1494 struct kvm_lapic *apic = vcpu->arch.apic;
1496 if (!kvm_vcpu_has_lapic(vcpu))
1499 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1500 | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1503 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1507 if (!kvm_vcpu_has_lapic(vcpu))
1510 tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1512 return (tpr & 0xf0) >> 4;
1515 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1517 u64 old_value = vcpu->arch.apic_base;
1518 struct kvm_lapic *apic = vcpu->arch.apic;
1521 value |= MSR_IA32_APICBASE_BSP;
1522 vcpu->arch.apic_base = value;
1526 if (!kvm_vcpu_is_bsp(apic->vcpu))
1527 value &= ~MSR_IA32_APICBASE_BSP;
1528 vcpu->arch.apic_base = value;
1530 /* update jump label if enable bit changes */
1531 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1532 if (value & MSR_IA32_APICBASE_ENABLE)
1533 static_key_slow_dec_deferred(&apic_hw_disabled);
1535 static_key_slow_inc(&apic_hw_disabled.key);
1536 recalculate_apic_map(vcpu->kvm);
1539 if ((old_value ^ value) & X2APIC_ENABLE) {
1540 if (value & X2APIC_ENABLE) {
1541 u32 id = kvm_apic_id(apic);
1542 u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1543 kvm_apic_set_ldr(apic, ldr);
1544 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1546 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1549 apic->base_address = apic->vcpu->arch.apic_base &
1550 MSR_IA32_APICBASE_BASE;
1552 if ((value & MSR_IA32_APICBASE_ENABLE) &&
1553 apic->base_address != APIC_DEFAULT_PHYS_BASE)
1554 pr_warn_once("APIC base relocation is unsupported by KVM");
1556 /* with FSB delivery interrupt, we can restart APIC functionality */
1557 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1558 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1562 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1564 struct kvm_lapic *apic;
1567 apic_debug("%s\n", __func__);
1570 apic = vcpu->arch.apic;
1571 ASSERT(apic != NULL);
1573 /* Stop the timer in case it's a reset to an active apic */
1574 hrtimer_cancel(&apic->lapic_timer.timer);
1576 kvm_apic_set_id(apic, vcpu->vcpu_id);
1577 kvm_apic_set_version(apic->vcpu);
1579 for (i = 0; i < APIC_LVT_NUM; i++)
1580 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1581 apic->lapic_timer.timer_mode = 0;
1582 apic_set_reg(apic, APIC_LVT0,
1583 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1585 apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1586 apic_set_spiv(apic, 0xff);
1587 apic_set_reg(apic, APIC_TASKPRI, 0);
1588 kvm_apic_set_ldr(apic, 0);
1589 apic_set_reg(apic, APIC_ESR, 0);
1590 apic_set_reg(apic, APIC_ICR, 0);
1591 apic_set_reg(apic, APIC_ICR2, 0);
1592 apic_set_reg(apic, APIC_TDCR, 0);
1593 apic_set_reg(apic, APIC_TMICT, 0);
1594 for (i = 0; i < 8; i++) {
1595 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1596 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1597 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1599 apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1600 apic->isr_count = kvm_x86_ops->hwapic_isr_update ? 1 : 0;
1601 apic->highest_isr_cache = -1;
1602 update_divide_count(apic);
1603 atomic_set(&apic->lapic_timer.pending, 0);
1604 if (kvm_vcpu_is_bsp(vcpu))
1605 kvm_lapic_set_base(vcpu,
1606 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1607 vcpu->arch.pv_eoi.msr_val = 0;
1608 apic_update_ppr(apic);
1610 vcpu->arch.apic_arb_prio = 0;
1611 vcpu->arch.apic_attention = 0;
1613 apic_debug("%s: vcpu=%p, id=%d, base_msr="
1614 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1615 vcpu, kvm_apic_id(apic),
1616 vcpu->arch.apic_base, apic->base_address);
1620 *----------------------------------------------------------------------
1622 *----------------------------------------------------------------------
1625 static bool lapic_is_periodic(struct kvm_lapic *apic)
1627 return apic_lvtt_period(apic);
1630 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1632 struct kvm_lapic *apic = vcpu->arch.apic;
1634 if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1635 apic_lvt_enabled(apic, APIC_LVTT))
1636 return atomic_read(&apic->lapic_timer.pending);
1641 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1643 u32 reg = kvm_apic_get_reg(apic, lvt_type);
1644 int vector, mode, trig_mode;
1646 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1647 vector = reg & APIC_VECTOR_MASK;
1648 mode = reg & APIC_MODE_MASK;
1649 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1650 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1656 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1658 struct kvm_lapic *apic = vcpu->arch.apic;
1661 kvm_apic_local_deliver(apic, APIC_LVT0);
1664 static const struct kvm_io_device_ops apic_mmio_ops = {
1665 .read = apic_mmio_read,
1666 .write = apic_mmio_write,
1669 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1671 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1672 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1674 apic_timer_expired(apic);
1676 if (lapic_is_periodic(apic)) {
1677 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1678 return HRTIMER_RESTART;
1680 return HRTIMER_NORESTART;
1683 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1685 struct kvm_lapic *apic;
1687 ASSERT(vcpu != NULL);
1688 apic_debug("apic_init %d\n", vcpu->vcpu_id);
1690 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1694 vcpu->arch.apic = apic;
1696 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1698 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1700 goto nomem_free_apic;
1704 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1706 apic->lapic_timer.timer.function = apic_timer_fn;
1709 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1710 * thinking that APIC satet has changed.
1712 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1713 kvm_lapic_set_base(vcpu,
1714 APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1716 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1717 kvm_lapic_reset(vcpu);
1718 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1727 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1729 struct kvm_lapic *apic = vcpu->arch.apic;
1732 if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1735 apic_update_ppr(apic);
1736 highest_irr = apic_find_highest_irr(apic);
1737 if ((highest_irr == -1) ||
1738 ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1743 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1745 u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1748 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1750 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1751 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1756 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1758 struct kvm_lapic *apic = vcpu->arch.apic;
1760 if (!kvm_vcpu_has_lapic(vcpu))
1763 if (atomic_read(&apic->lapic_timer.pending) > 0) {
1764 kvm_apic_local_deliver(apic, APIC_LVTT);
1765 if (apic_lvtt_tscdeadline(apic))
1766 apic->lapic_timer.tscdeadline = 0;
1767 atomic_set(&apic->lapic_timer.pending, 0);
1771 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1773 int vector = kvm_apic_has_interrupt(vcpu);
1774 struct kvm_lapic *apic = vcpu->arch.apic;
1780 * We get here even with APIC virtualization enabled, if doing
1781 * nested virtualization and L1 runs with the "acknowledge interrupt
1782 * on exit" mode. Then we cannot inject the interrupt via RVI,
1783 * because the process would deliver it through the IDT.
1786 apic_set_isr(vector, apic);
1787 apic_update_ppr(apic);
1788 apic_clear_irr(vector, apic);
1792 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1793 struct kvm_lapic_state *s)
1795 struct kvm_lapic *apic = vcpu->arch.apic;
1797 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1798 /* set SPIV separately to get count of SW disabled APICs right */
1799 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1800 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1801 /* call kvm_apic_set_id() to put apic into apic_map */
1802 kvm_apic_set_id(apic, kvm_apic_id(apic));
1803 kvm_apic_set_version(vcpu);
1805 apic_update_ppr(apic);
1806 hrtimer_cancel(&apic->lapic_timer.timer);
1807 update_divide_count(apic);
1808 start_apic_timer(apic);
1809 apic->irr_pending = true;
1810 apic->isr_count = kvm_x86_ops->hwapic_isr_update ?
1811 1 : count_vectors(apic->regs + APIC_ISR);
1812 apic->highest_isr_cache = -1;
1813 if (kvm_x86_ops->hwapic_irr_update)
1814 kvm_x86_ops->hwapic_irr_update(vcpu,
1815 apic_find_highest_irr(apic));
1816 if (unlikely(kvm_x86_ops->hwapic_isr_update))
1817 kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
1818 apic_find_highest_isr(apic));
1819 kvm_make_request(KVM_REQ_EVENT, vcpu);
1820 kvm_rtc_eoi_tracking_restore_one(vcpu);
1823 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1825 struct hrtimer *timer;
1827 if (!kvm_vcpu_has_lapic(vcpu))
1830 timer = &vcpu->arch.apic->lapic_timer.timer;
1831 if (hrtimer_cancel(timer))
1832 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1836 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1838 * Detect whether guest triggered PV EOI since the
1839 * last entry. If yes, set EOI on guests's behalf.
1840 * Clear PV EOI in guest memory in any case.
1842 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1843 struct kvm_lapic *apic)
1848 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1849 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1851 * KVM_APIC_PV_EOI_PENDING is unset:
1852 * -> host disabled PV EOI.
1853 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1854 * -> host enabled PV EOI, guest did not execute EOI yet.
1855 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1856 * -> host enabled PV EOI, guest executed EOI.
1858 BUG_ON(!pv_eoi_enabled(vcpu));
1859 pending = pv_eoi_get_pending(vcpu);
1861 * Clear pending bit in any case: it will be set again on vmentry.
1862 * While this might not be ideal from performance point of view,
1863 * this makes sure pv eoi is only enabled when we know it's safe.
1865 pv_eoi_clr_pending(vcpu);
1868 vector = apic_set_eoi(apic);
1869 trace_kvm_pv_eoi(apic, vector);
1872 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1876 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1877 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1879 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1882 kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1885 apic_set_tpr(vcpu->arch.apic, data & 0xff);
1889 * apic_sync_pv_eoi_to_guest - called before vmentry
1891 * Detect whether it's safe to enable PV EOI and
1894 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1895 struct kvm_lapic *apic)
1897 if (!pv_eoi_enabled(vcpu) ||
1898 /* IRR set or many bits in ISR: could be nested. */
1899 apic->irr_pending ||
1900 /* Cache not set: could be safe but we don't bother. */
1901 apic->highest_isr_cache == -1 ||
1902 /* Need EOI to update ioapic. */
1903 kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1905 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1906 * so we need not do anything here.
1911 pv_eoi_set_pending(apic->vcpu);
1914 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1917 int max_irr, max_isr;
1918 struct kvm_lapic *apic = vcpu->arch.apic;
1920 apic_sync_pv_eoi_to_guest(vcpu, apic);
1922 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1925 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1926 max_irr = apic_find_highest_irr(apic);
1929 max_isr = apic_find_highest_isr(apic);
1932 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1934 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1938 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1941 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1942 &vcpu->arch.apic->vapic_cache,
1943 vapic_addr, sizeof(u32)))
1945 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1947 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1950 vcpu->arch.apic->vapic_addr = vapic_addr;
1954 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1956 struct kvm_lapic *apic = vcpu->arch.apic;
1957 u32 reg = (msr - APIC_BASE_MSR) << 4;
1959 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1962 if (reg == APIC_ICR2)
1965 /* if this is ICR write vector before command */
1966 if (reg == APIC_ICR)
1967 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1968 return apic_reg_write(apic, reg, (u32)data);
1971 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1973 struct kvm_lapic *apic = vcpu->arch.apic;
1974 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1976 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1979 if (reg == APIC_DFR || reg == APIC_ICR2) {
1980 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
1985 if (apic_reg_read(apic, reg, 4, &low))
1987 if (reg == APIC_ICR)
1988 apic_reg_read(apic, APIC_ICR2, 4, &high);
1990 *data = (((u64)high) << 32) | low;
1995 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1997 struct kvm_lapic *apic = vcpu->arch.apic;
1999 if (!kvm_vcpu_has_lapic(vcpu))
2002 /* if this is ICR write vector before command */
2003 if (reg == APIC_ICR)
2004 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2005 return apic_reg_write(apic, reg, (u32)data);
2008 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2010 struct kvm_lapic *apic = vcpu->arch.apic;
2013 if (!kvm_vcpu_has_lapic(vcpu))
2016 if (apic_reg_read(apic, reg, 4, &low))
2018 if (reg == APIC_ICR)
2019 apic_reg_read(apic, APIC_ICR2, 4, &high);
2021 *data = (((u64)high) << 32) | low;
2026 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2028 u64 addr = data & ~KVM_MSR_ENABLED;
2029 if (!IS_ALIGNED(addr, 4))
2032 vcpu->arch.pv_eoi.msr_val = data;
2033 if (!pv_eoi_enabled(vcpu))
2035 return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2039 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2041 struct kvm_lapic *apic = vcpu->arch.apic;
2045 if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
2048 pe = xchg(&apic->pending_events, 0);
2050 if (test_bit(KVM_APIC_INIT, &pe)) {
2051 kvm_lapic_reset(vcpu);
2052 kvm_vcpu_reset(vcpu);
2053 if (kvm_vcpu_is_bsp(apic->vcpu))
2054 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2056 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2058 if (test_bit(KVM_APIC_SIPI, &pe) &&
2059 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2060 /* evaluate pending_events before reading the vector */
2062 sipi_vector = apic->sipi_vector;
2063 apic_debug("vcpu %d received sipi with vector # %x\n",
2064 vcpu->vcpu_id, sipi_vector);
2065 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2066 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2070 void kvm_lapic_init(void)
2072 /* do not patch jump label more than once per second */
2073 jump_label_rate_limit(&apic_hw_disabled, HZ);
2074 jump_label_rate_limit(&apic_sw_disabled, HZ);