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;
154 new->broadcast = APIC_BROADCAST;
156 kvm_for_each_vcpu(i, vcpu, kvm) {
157 struct kvm_lapic *apic = vcpu->arch.apic;
159 if (!kvm_apic_present(vcpu))
162 if (apic_x2apic_mode(apic)) {
165 new->cid_mask = new->lid_mask = 0xffff;
166 new->broadcast = X2APIC_BROADCAST;
167 } else if (kvm_apic_get_reg(apic, APIC_LDR)) {
168 if (kvm_apic_get_reg(apic, APIC_DFR) ==
176 new->lid_mask = 0xff;
181 * All APICs have to be configured in the same mode by an OS.
182 * We take advatage of this while building logical id loockup
183 * table. After reset APICs are in software disabled mode, so if
184 * we find apic with different setting we assume this is the mode
185 * OS wants all apics to be in; build lookup table accordingly.
187 if (kvm_apic_sw_enabled(apic))
191 kvm_for_each_vcpu(i, vcpu, kvm) {
192 struct kvm_lapic *apic = vcpu->arch.apic;
196 if (!kvm_apic_present(vcpu))
199 aid = kvm_apic_id(apic);
200 ldr = kvm_apic_get_reg(apic, APIC_LDR);
201 cid = apic_cluster_id(new, ldr);
202 lid = apic_logical_id(new, ldr);
204 if (aid < ARRAY_SIZE(new->phys_map))
205 new->phys_map[aid] = apic;
206 if (lid && cid < ARRAY_SIZE(new->logical_map))
207 new->logical_map[cid][ffs(lid) - 1] = apic;
210 old = rcu_dereference_protected(kvm->arch.apic_map,
211 lockdep_is_held(&kvm->arch.apic_map_lock));
212 rcu_assign_pointer(kvm->arch.apic_map, new);
213 mutex_unlock(&kvm->arch.apic_map_lock);
218 kvm_vcpu_request_scan_ioapic(kvm);
221 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
223 bool enabled = val & APIC_SPIV_APIC_ENABLED;
225 apic_set_reg(apic, APIC_SPIV, val);
227 if (enabled != apic->sw_enabled) {
228 apic->sw_enabled = enabled;
230 static_key_slow_dec_deferred(&apic_sw_disabled);
231 recalculate_apic_map(apic->vcpu->kvm);
233 static_key_slow_inc(&apic_sw_disabled.key);
237 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
239 apic_set_reg(apic, APIC_ID, id << 24);
240 recalculate_apic_map(apic->vcpu->kvm);
243 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
245 apic_set_reg(apic, APIC_LDR, id);
246 recalculate_apic_map(apic->vcpu->kvm);
249 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
251 return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
254 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
256 return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
259 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
261 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
264 static inline int apic_lvtt_period(struct kvm_lapic *apic)
266 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
269 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
271 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
274 static inline int apic_lvt_nmi_mode(u32 lvt_val)
276 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
279 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
281 struct kvm_lapic *apic = vcpu->arch.apic;
282 struct kvm_cpuid_entry2 *feat;
283 u32 v = APIC_VERSION;
285 if (!kvm_vcpu_has_lapic(vcpu))
288 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
289 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
290 v |= APIC_LVR_DIRECTED_EOI;
291 apic_set_reg(apic, APIC_LVR, v);
294 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
295 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
296 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
297 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
298 LINT_MASK, LINT_MASK, /* LVT0-1 */
299 LVT_MASK /* LVTERR */
302 static int find_highest_vector(void *bitmap)
307 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
308 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
309 reg = bitmap + REG_POS(vec);
311 return fls(*reg) - 1 + vec;
317 static u8 count_vectors(void *bitmap)
323 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
324 reg = bitmap + REG_POS(vec);
325 count += hweight32(*reg);
331 void __kvm_apic_update_irr(u32 *pir, void *regs)
335 for (i = 0; i <= 7; i++) {
336 pir_val = xchg(&pir[i], 0);
338 *((u32 *)(regs + APIC_IRR + i * 0x10)) |= pir_val;
341 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
343 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
345 struct kvm_lapic *apic = vcpu->arch.apic;
347 __kvm_apic_update_irr(pir, apic->regs);
349 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
351 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
353 apic_set_vector(vec, apic->regs + APIC_IRR);
355 * irr_pending must be true if any interrupt is pending; set it after
356 * APIC_IRR to avoid race with apic_clear_irr
358 apic->irr_pending = true;
361 static inline int apic_search_irr(struct kvm_lapic *apic)
363 return find_highest_vector(apic->regs + APIC_IRR);
366 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
371 * Note that irr_pending is just a hint. It will be always
372 * true with virtual interrupt delivery enabled.
374 if (!apic->irr_pending)
377 kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
378 result = apic_search_irr(apic);
379 ASSERT(result == -1 || result >= 16);
384 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
386 struct kvm_vcpu *vcpu;
390 if (unlikely(kvm_apic_vid_enabled(vcpu->kvm))) {
391 /* try to update RVI */
392 apic_clear_vector(vec, apic->regs + APIC_IRR);
393 kvm_make_request(KVM_REQ_EVENT, vcpu);
395 apic->irr_pending = false;
396 apic_clear_vector(vec, apic->regs + APIC_IRR);
397 if (apic_search_irr(apic) != -1)
398 apic->irr_pending = true;
402 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
404 struct kvm_vcpu *vcpu;
406 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
412 * With APIC virtualization enabled, all caching is disabled
413 * because the processor can modify ISR under the hood. Instead
416 if (unlikely(kvm_x86_ops->hwapic_isr_update))
417 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
420 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
422 * ISR (in service register) bit is set when injecting an interrupt.
423 * The highest vector is injected. Thus the latest bit set matches
424 * the highest bit in ISR.
426 apic->highest_isr_cache = vec;
430 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
435 * Note that isr_count is always 1, and highest_isr_cache
436 * is always -1, with APIC virtualization enabled.
438 if (!apic->isr_count)
440 if (likely(apic->highest_isr_cache != -1))
441 return apic->highest_isr_cache;
443 result = find_highest_vector(apic->regs + APIC_ISR);
444 ASSERT(result == -1 || result >= 16);
449 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
451 struct kvm_vcpu *vcpu;
452 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
458 * We do get here for APIC virtualization enabled if the guest
459 * uses the Hyper-V APIC enlightenment. In this case we may need
460 * to trigger a new interrupt delivery by writing the SVI field;
461 * on the other hand isr_count and highest_isr_cache are unused
462 * and must be left alone.
464 if (unlikely(kvm_x86_ops->hwapic_isr_update))
465 kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
466 apic_find_highest_isr(apic));
469 BUG_ON(apic->isr_count < 0);
470 apic->highest_isr_cache = -1;
474 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
478 /* This may race with setting of irr in __apic_accept_irq() and
479 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
480 * will cause vmexit immediately and the value will be recalculated
481 * on the next vmentry.
483 if (!kvm_vcpu_has_lapic(vcpu))
485 highest_irr = apic_find_highest_irr(vcpu->arch.apic);
490 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
491 int vector, int level, int trig_mode,
492 unsigned long *dest_map);
494 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
495 unsigned long *dest_map)
497 struct kvm_lapic *apic = vcpu->arch.apic;
499 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
500 irq->level, irq->trig_mode, dest_map);
503 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
506 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
510 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
513 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
517 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
519 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
522 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
525 if (pv_eoi_get_user(vcpu, &val) < 0)
526 apic_debug("Can't read EOI MSR value: 0x%llx\n",
527 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
531 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
533 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
534 apic_debug("Can't set EOI MSR value: 0x%llx\n",
535 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
538 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
541 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
543 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
544 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
545 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
548 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
551 void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
553 struct kvm_lapic *apic = vcpu->arch.apic;
556 for (i = 0; i < 8; i++)
557 apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
560 static void apic_update_ppr(struct kvm_lapic *apic)
562 u32 tpr, isrv, ppr, old_ppr;
565 old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
566 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
567 isr = apic_find_highest_isr(apic);
568 isrv = (isr != -1) ? isr : 0;
570 if ((tpr & 0xf0) >= (isrv & 0xf0))
575 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
576 apic, ppr, isr, isrv);
578 if (old_ppr != ppr) {
579 apic_set_reg(apic, APIC_PROCPRI, ppr);
581 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
585 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
587 apic_set_reg(apic, APIC_TASKPRI, tpr);
588 apic_update_ppr(apic);
591 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
593 if (apic_x2apic_mode(apic))
594 return mda == X2APIC_BROADCAST;
596 return GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST;
599 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
601 if (kvm_apic_broadcast(apic, mda))
604 if (apic_x2apic_mode(apic))
605 return mda == kvm_apic_id(apic);
607 return mda == SET_APIC_DEST_FIELD(kvm_apic_id(apic));
610 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
614 if (kvm_apic_broadcast(apic, mda))
617 logical_id = kvm_apic_get_reg(apic, APIC_LDR);
619 if (apic_x2apic_mode(apic))
620 return ((logical_id >> 16) == (mda >> 16))
621 && (logical_id & mda & 0xffff) != 0;
623 logical_id = GET_APIC_LOGICAL_ID(logical_id);
624 mda = GET_APIC_DEST_FIELD(mda);
626 switch (kvm_apic_get_reg(apic, APIC_DFR)) {
628 return (logical_id & mda) != 0;
629 case APIC_DFR_CLUSTER:
630 return ((logical_id >> 4) == (mda >> 4))
631 && (logical_id & mda & 0xf) != 0;
633 apic_debug("Bad DFR vcpu %d: %08x\n",
634 apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
639 /* KVM APIC implementation has two quirks
640 * - dest always begins at 0 while xAPIC MDA has offset 24,
641 * - IOxAPIC messages have to be delivered (directly) to x2APIC.
643 static u32 kvm_apic_mda(unsigned int dest_id, struct kvm_lapic *source,
644 struct kvm_lapic *target)
646 bool ipi = source != NULL;
647 bool x2apic_mda = apic_x2apic_mode(ipi ? source : target);
649 if (!ipi && dest_id == APIC_BROADCAST && x2apic_mda)
650 return X2APIC_BROADCAST;
652 return x2apic_mda ? dest_id : SET_APIC_DEST_FIELD(dest_id);
655 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
656 int short_hand, unsigned int dest, int dest_mode)
658 struct kvm_lapic *target = vcpu->arch.apic;
659 u32 mda = kvm_apic_mda(dest, source, target);
661 apic_debug("target %p, source %p, dest 0x%x, "
662 "dest_mode 0x%x, short_hand 0x%x\n",
663 target, source, dest, dest_mode, short_hand);
666 switch (short_hand) {
667 case APIC_DEST_NOSHORT:
668 if (dest_mode == APIC_DEST_PHYSICAL)
669 return kvm_apic_match_physical_addr(target, mda);
671 return kvm_apic_match_logical_addr(target, mda);
673 return target == source;
674 case APIC_DEST_ALLINC:
676 case APIC_DEST_ALLBUT:
677 return target != source;
679 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
685 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
686 struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
688 struct kvm_apic_map *map;
689 unsigned long bitmap = 1;
690 struct kvm_lapic **dst;
696 if (irq->shorthand == APIC_DEST_SELF) {
697 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
705 map = rcu_dereference(kvm->arch.apic_map);
710 if (irq->dest_id == map->broadcast)
715 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
716 if (irq->dest_id >= ARRAY_SIZE(map->phys_map))
719 dst = &map->phys_map[irq->dest_id];
721 u32 mda = irq->dest_id << (32 - map->ldr_bits);
722 u16 cid = apic_cluster_id(map, mda);
724 if (cid >= ARRAY_SIZE(map->logical_map))
727 dst = map->logical_map[cid];
729 bitmap = apic_logical_id(map, mda);
731 if (irq->delivery_mode == APIC_DM_LOWEST) {
733 for_each_set_bit(i, &bitmap, 16) {
738 else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
742 bitmap = (l >= 0) ? 1 << l : 0;
746 for_each_set_bit(i, &bitmap, 16) {
751 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
759 * Add a pending IRQ into lapic.
760 * Return 1 if successfully added and 0 if discarded.
762 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
763 int vector, int level, int trig_mode,
764 unsigned long *dest_map)
767 struct kvm_vcpu *vcpu = apic->vcpu;
769 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
771 switch (delivery_mode) {
773 vcpu->arch.apic_arb_prio++;
775 /* FIXME add logic for vcpu on reset */
776 if (unlikely(!apic_enabled(apic)))
782 __set_bit(vcpu->vcpu_id, dest_map);
784 if (kvm_x86_ops->deliver_posted_interrupt)
785 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
787 apic_set_irr(vector, apic);
789 kvm_make_request(KVM_REQ_EVENT, vcpu);
796 vcpu->arch.pv.pv_unhalted = 1;
797 kvm_make_request(KVM_REQ_EVENT, vcpu);
802 apic_debug("Ignoring guest SMI\n");
807 kvm_inject_nmi(vcpu);
812 if (!trig_mode || level) {
814 /* assumes that there are only KVM_APIC_INIT/SIPI */
815 apic->pending_events = (1UL << KVM_APIC_INIT);
816 /* make sure pending_events is visible before sending
819 kvm_make_request(KVM_REQ_EVENT, vcpu);
822 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
827 case APIC_DM_STARTUP:
828 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
829 vcpu->vcpu_id, vector);
831 apic->sipi_vector = vector;
832 /* make sure sipi_vector is visible for the receiver */
834 set_bit(KVM_APIC_SIPI, &apic->pending_events);
835 kvm_make_request(KVM_REQ_EVENT, vcpu);
841 * Should only be called by kvm_apic_local_deliver() with LVT0,
842 * before NMI watchdog was enabled. Already handled by
843 * kvm_apic_accept_pic_intr().
848 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
855 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
857 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
860 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
862 if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
863 kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
865 if (apic_test_vector(vector, apic->regs + APIC_TMR))
866 trigger_mode = IOAPIC_LEVEL_TRIG;
868 trigger_mode = IOAPIC_EDGE_TRIG;
869 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
873 static int apic_set_eoi(struct kvm_lapic *apic)
875 int vector = apic_find_highest_isr(apic);
877 trace_kvm_eoi(apic, vector);
880 * Not every write EOI will has corresponding ISR,
881 * one example is when Kernel check timer on setup_IO_APIC
886 apic_clear_isr(vector, apic);
887 apic_update_ppr(apic);
889 kvm_ioapic_send_eoi(apic, vector);
890 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
895 * this interface assumes a trap-like exit, which has already finished
896 * desired side effect including vISR and vPPR update.
898 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
900 struct kvm_lapic *apic = vcpu->arch.apic;
902 trace_kvm_eoi(apic, vector);
904 kvm_ioapic_send_eoi(apic, vector);
905 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
907 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
909 static void apic_send_ipi(struct kvm_lapic *apic)
911 u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
912 u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
913 struct kvm_lapic_irq irq;
915 irq.vector = icr_low & APIC_VECTOR_MASK;
916 irq.delivery_mode = icr_low & APIC_MODE_MASK;
917 irq.dest_mode = icr_low & APIC_DEST_MASK;
918 irq.level = icr_low & APIC_INT_ASSERT;
919 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
920 irq.shorthand = icr_low & APIC_SHORT_MASK;
921 if (apic_x2apic_mode(apic))
922 irq.dest_id = icr_high;
924 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
926 trace_kvm_apic_ipi(icr_low, irq.dest_id);
928 apic_debug("icr_high 0x%x, icr_low 0x%x, "
929 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
930 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
931 icr_high, icr_low, irq.shorthand, irq.dest_id,
932 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
935 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
938 static u32 apic_get_tmcct(struct kvm_lapic *apic)
944 ASSERT(apic != NULL);
946 /* if initial count is 0, current count should also be 0 */
947 if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
948 apic->lapic_timer.period == 0)
951 remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
952 if (ktime_to_ns(remaining) < 0)
953 remaining = ktime_set(0, 0);
955 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
956 tmcct = div64_u64(ns,
957 (APIC_BUS_CYCLE_NS * apic->divide_count));
962 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
964 struct kvm_vcpu *vcpu = apic->vcpu;
965 struct kvm_run *run = vcpu->run;
967 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
968 run->tpr_access.rip = kvm_rip_read(vcpu);
969 run->tpr_access.is_write = write;
972 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
974 if (apic->vcpu->arch.tpr_access_reporting)
975 __report_tpr_access(apic, write);
978 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
982 if (offset >= LAPIC_MMIO_LENGTH)
987 if (apic_x2apic_mode(apic))
988 val = kvm_apic_id(apic);
990 val = kvm_apic_id(apic) << 24;
993 apic_debug("Access APIC ARBPRI register which is for P6\n");
996 case APIC_TMCCT: /* Timer CCR */
997 if (apic_lvtt_tscdeadline(apic))
1000 val = apic_get_tmcct(apic);
1003 apic_update_ppr(apic);
1004 val = kvm_apic_get_reg(apic, offset);
1007 report_tpr_access(apic, false);
1010 val = kvm_apic_get_reg(apic, offset);
1017 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1019 return container_of(dev, struct kvm_lapic, dev);
1022 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1025 unsigned char alignment = offset & 0xf;
1027 /* this bitmask has a bit cleared for each reserved register */
1028 static const u64 rmask = 0x43ff01ffffffe70cULL;
1030 if ((alignment + len) > 4) {
1031 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1036 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1037 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1042 result = __apic_read(apic, offset & ~0xf);
1044 trace_kvm_apic_read(offset, result);
1050 memcpy(data, (char *)&result + alignment, len);
1053 printk(KERN_ERR "Local APIC read with len = %x, "
1054 "should be 1,2, or 4 instead\n", len);
1060 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1062 return kvm_apic_hw_enabled(apic) &&
1063 addr >= apic->base_address &&
1064 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1067 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1068 gpa_t address, int len, void *data)
1070 struct kvm_lapic *apic = to_lapic(this);
1071 u32 offset = address - apic->base_address;
1073 if (!apic_mmio_in_range(apic, address))
1076 apic_reg_read(apic, offset, len, data);
1081 static void update_divide_count(struct kvm_lapic *apic)
1083 u32 tmp1, tmp2, tdcr;
1085 tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1087 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1088 apic->divide_count = 0x1 << (tmp2 & 0x7);
1090 apic_debug("timer divide count is 0x%x\n",
1091 apic->divide_count);
1094 static void apic_timer_expired(struct kvm_lapic *apic)
1096 struct kvm_vcpu *vcpu = apic->vcpu;
1097 wait_queue_head_t *q = &vcpu->wq;
1098 struct kvm_timer *ktimer = &apic->lapic_timer;
1100 if (atomic_read(&apic->lapic_timer.pending))
1103 atomic_inc(&apic->lapic_timer.pending);
1104 kvm_set_pending_timer(vcpu);
1106 if (waitqueue_active(q))
1107 wake_up_interruptible(q);
1109 if (apic_lvtt_tscdeadline(apic))
1110 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1114 * On APICv, this test will cause a busy wait
1115 * during a higher-priority task.
1118 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1120 struct kvm_lapic *apic = vcpu->arch.apic;
1121 u32 reg = kvm_apic_get_reg(apic, APIC_LVTT);
1123 if (kvm_apic_hw_enabled(apic)) {
1124 int vec = reg & APIC_VECTOR_MASK;
1125 void *bitmap = apic->regs + APIC_ISR;
1127 if (kvm_x86_ops->deliver_posted_interrupt)
1128 bitmap = apic->regs + APIC_IRR;
1130 if (apic_test_vector(vec, bitmap))
1136 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1138 struct kvm_lapic *apic = vcpu->arch.apic;
1139 u64 guest_tsc, tsc_deadline;
1141 if (!kvm_vcpu_has_lapic(vcpu))
1144 if (apic->lapic_timer.expired_tscdeadline == 0)
1147 if (!lapic_timer_int_injected(vcpu))
1150 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1151 apic->lapic_timer.expired_tscdeadline = 0;
1152 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1153 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1155 /* __delay is delay_tsc whenever the hardware has TSC, thus always. */
1156 if (guest_tsc < tsc_deadline)
1157 __delay(tsc_deadline - guest_tsc);
1160 static void start_apic_timer(struct kvm_lapic *apic)
1164 atomic_set(&apic->lapic_timer.pending, 0);
1166 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1167 /* lapic timer in oneshot or periodic mode */
1168 now = apic->lapic_timer.timer.base->get_time();
1169 apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1170 * APIC_BUS_CYCLE_NS * apic->divide_count;
1172 if (!apic->lapic_timer.period)
1175 * Do not allow the guest to program periodic timers with small
1176 * interval, since the hrtimers are not throttled by the host
1179 if (apic_lvtt_period(apic)) {
1180 s64 min_period = min_timer_period_us * 1000LL;
1182 if (apic->lapic_timer.period < min_period) {
1183 pr_info_ratelimited(
1184 "kvm: vcpu %i: requested %lld ns "
1185 "lapic timer period limited to %lld ns\n",
1186 apic->vcpu->vcpu_id,
1187 apic->lapic_timer.period, min_period);
1188 apic->lapic_timer.period = min_period;
1192 hrtimer_start(&apic->lapic_timer.timer,
1193 ktime_add_ns(now, apic->lapic_timer.period),
1196 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1198 "timer initial count 0x%x, period %lldns, "
1199 "expire @ 0x%016" PRIx64 ".\n", __func__,
1200 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1201 kvm_apic_get_reg(apic, APIC_TMICT),
1202 apic->lapic_timer.period,
1203 ktime_to_ns(ktime_add_ns(now,
1204 apic->lapic_timer.period)));
1205 } else if (apic_lvtt_tscdeadline(apic)) {
1206 /* lapic timer in tsc deadline mode */
1207 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1210 struct kvm_vcpu *vcpu = apic->vcpu;
1211 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1212 unsigned long flags;
1214 if (unlikely(!tscdeadline || !this_tsc_khz))
1217 local_irq_save(flags);
1219 now = apic->lapic_timer.timer.base->get_time();
1220 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1221 if (likely(tscdeadline > guest_tsc)) {
1222 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1223 do_div(ns, this_tsc_khz);
1224 expire = ktime_add_ns(now, ns);
1225 expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1226 hrtimer_start(&apic->lapic_timer.timer,
1227 expire, HRTIMER_MODE_ABS);
1229 apic_timer_expired(apic);
1231 local_irq_restore(flags);
1235 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1237 int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1239 if (apic_lvt_nmi_mode(lvt0_val)) {
1240 if (!nmi_wd_enabled) {
1241 apic_debug("Receive NMI setting on APIC_LVT0 "
1242 "for cpu %d\n", apic->vcpu->vcpu_id);
1243 apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1245 } else if (nmi_wd_enabled)
1246 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1249 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1253 trace_kvm_apic_write(reg, val);
1256 case APIC_ID: /* Local APIC ID */
1257 if (!apic_x2apic_mode(apic))
1258 kvm_apic_set_id(apic, val >> 24);
1264 report_tpr_access(apic, true);
1265 apic_set_tpr(apic, val & 0xff);
1273 if (!apic_x2apic_mode(apic))
1274 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1280 if (!apic_x2apic_mode(apic)) {
1281 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1282 recalculate_apic_map(apic->vcpu->kvm);
1289 if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1290 mask |= APIC_SPIV_DIRECTED_EOI;
1291 apic_set_spiv(apic, val & mask);
1292 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1296 for (i = 0; i < APIC_LVT_NUM; i++) {
1297 lvt_val = kvm_apic_get_reg(apic,
1298 APIC_LVTT + 0x10 * i);
1299 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1300 lvt_val | APIC_LVT_MASKED);
1302 atomic_set(&apic->lapic_timer.pending, 0);
1308 /* No delay here, so we always clear the pending bit */
1309 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1310 apic_send_ipi(apic);
1314 if (!apic_x2apic_mode(apic))
1316 apic_set_reg(apic, APIC_ICR2, val);
1320 apic_manage_nmi_watchdog(apic, val);
1325 /* TODO: Check vector */
1326 if (!kvm_apic_sw_enabled(apic))
1327 val |= APIC_LVT_MASKED;
1329 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1330 apic_set_reg(apic, reg, val);
1335 u32 timer_mode = val & apic->lapic_timer.timer_mode_mask;
1337 if (apic->lapic_timer.timer_mode != timer_mode) {
1338 apic->lapic_timer.timer_mode = timer_mode;
1339 hrtimer_cancel(&apic->lapic_timer.timer);
1342 if (!kvm_apic_sw_enabled(apic))
1343 val |= APIC_LVT_MASKED;
1344 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1345 apic_set_reg(apic, APIC_LVTT, val);
1350 if (apic_lvtt_tscdeadline(apic))
1353 hrtimer_cancel(&apic->lapic_timer.timer);
1354 apic_set_reg(apic, APIC_TMICT, val);
1355 start_apic_timer(apic);
1360 apic_debug("KVM_WRITE:TDCR %x\n", val);
1361 apic_set_reg(apic, APIC_TDCR, val);
1362 update_divide_count(apic);
1366 if (apic_x2apic_mode(apic) && val != 0) {
1367 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1373 if (apic_x2apic_mode(apic)) {
1374 apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1383 apic_debug("Local APIC Write to read-only register %x\n", reg);
1387 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1388 gpa_t address, int len, const void *data)
1390 struct kvm_lapic *apic = to_lapic(this);
1391 unsigned int offset = address - apic->base_address;
1394 if (!apic_mmio_in_range(apic, address))
1398 * APIC register must be aligned on 128-bits boundary.
1399 * 32/64/128 bits registers must be accessed thru 32 bits.
1402 if (len != 4 || (offset & 0xf)) {
1403 /* Don't shout loud, $infamous_os would cause only noise. */
1404 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1410 /* too common printing */
1411 if (offset != APIC_EOI)
1412 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1413 "0x%x\n", __func__, offset, len, val);
1415 apic_reg_write(apic, offset & 0xff0, val);
1420 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1422 if (kvm_vcpu_has_lapic(vcpu))
1423 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1425 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1427 /* emulate APIC access in a trap manner */
1428 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1432 /* hw has done the conditional check and inst decode */
1435 apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1437 /* TODO: optimize to just emulate side effect w/o one more write */
1438 apic_reg_write(vcpu->arch.apic, offset, val);
1440 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1442 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1444 struct kvm_lapic *apic = vcpu->arch.apic;
1446 if (!vcpu->arch.apic)
1449 hrtimer_cancel(&apic->lapic_timer.timer);
1451 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1452 static_key_slow_dec_deferred(&apic_hw_disabled);
1454 if (!apic->sw_enabled)
1455 static_key_slow_dec_deferred(&apic_sw_disabled);
1458 free_page((unsigned long)apic->regs);
1464 *----------------------------------------------------------------------
1466 *----------------------------------------------------------------------
1469 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1471 struct kvm_lapic *apic = vcpu->arch.apic;
1473 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1474 apic_lvtt_period(apic))
1477 return apic->lapic_timer.tscdeadline;
1480 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1482 struct kvm_lapic *apic = vcpu->arch.apic;
1484 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1485 apic_lvtt_period(apic))
1488 hrtimer_cancel(&apic->lapic_timer.timer);
1489 apic->lapic_timer.tscdeadline = data;
1490 start_apic_timer(apic);
1493 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1495 struct kvm_lapic *apic = vcpu->arch.apic;
1497 if (!kvm_vcpu_has_lapic(vcpu))
1500 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1501 | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1504 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1508 if (!kvm_vcpu_has_lapic(vcpu))
1511 tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1513 return (tpr & 0xf0) >> 4;
1516 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1518 u64 old_value = vcpu->arch.apic_base;
1519 struct kvm_lapic *apic = vcpu->arch.apic;
1522 value |= MSR_IA32_APICBASE_BSP;
1523 vcpu->arch.apic_base = value;
1527 if (!kvm_vcpu_is_bsp(apic->vcpu))
1528 value &= ~MSR_IA32_APICBASE_BSP;
1529 vcpu->arch.apic_base = value;
1531 /* update jump label if enable bit changes */
1532 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1533 if (value & MSR_IA32_APICBASE_ENABLE)
1534 static_key_slow_dec_deferred(&apic_hw_disabled);
1536 static_key_slow_inc(&apic_hw_disabled.key);
1537 recalculate_apic_map(vcpu->kvm);
1540 if ((old_value ^ value) & X2APIC_ENABLE) {
1541 if (value & X2APIC_ENABLE) {
1542 u32 id = kvm_apic_id(apic);
1543 u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1544 kvm_apic_set_ldr(apic, ldr);
1545 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1547 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1550 apic->base_address = apic->vcpu->arch.apic_base &
1551 MSR_IA32_APICBASE_BASE;
1553 if ((value & MSR_IA32_APICBASE_ENABLE) &&
1554 apic->base_address != APIC_DEFAULT_PHYS_BASE)
1555 pr_warn_once("APIC base relocation is unsupported by KVM");
1557 /* with FSB delivery interrupt, we can restart APIC functionality */
1558 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1559 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1563 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1565 struct kvm_lapic *apic;
1568 apic_debug("%s\n", __func__);
1571 apic = vcpu->arch.apic;
1572 ASSERT(apic != NULL);
1574 /* Stop the timer in case it's a reset to an active apic */
1575 hrtimer_cancel(&apic->lapic_timer.timer);
1577 kvm_apic_set_id(apic, vcpu->vcpu_id);
1578 kvm_apic_set_version(apic->vcpu);
1580 for (i = 0; i < APIC_LVT_NUM; i++)
1581 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1582 apic->lapic_timer.timer_mode = 0;
1583 apic_set_reg(apic, APIC_LVT0,
1584 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1586 apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1587 apic_set_spiv(apic, 0xff);
1588 apic_set_reg(apic, APIC_TASKPRI, 0);
1589 kvm_apic_set_ldr(apic, 0);
1590 apic_set_reg(apic, APIC_ESR, 0);
1591 apic_set_reg(apic, APIC_ICR, 0);
1592 apic_set_reg(apic, APIC_ICR2, 0);
1593 apic_set_reg(apic, APIC_TDCR, 0);
1594 apic_set_reg(apic, APIC_TMICT, 0);
1595 for (i = 0; i < 8; i++) {
1596 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1597 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1598 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1600 apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1601 apic->isr_count = kvm_x86_ops->hwapic_isr_update ? 1 : 0;
1602 apic->highest_isr_cache = -1;
1603 update_divide_count(apic);
1604 atomic_set(&apic->lapic_timer.pending, 0);
1605 if (kvm_vcpu_is_bsp(vcpu))
1606 kvm_lapic_set_base(vcpu,
1607 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1608 vcpu->arch.pv_eoi.msr_val = 0;
1609 apic_update_ppr(apic);
1611 vcpu->arch.apic_arb_prio = 0;
1612 vcpu->arch.apic_attention = 0;
1614 apic_debug("%s: vcpu=%p, id=%d, base_msr="
1615 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1616 vcpu, kvm_apic_id(apic),
1617 vcpu->arch.apic_base, apic->base_address);
1621 *----------------------------------------------------------------------
1623 *----------------------------------------------------------------------
1626 static bool lapic_is_periodic(struct kvm_lapic *apic)
1628 return apic_lvtt_period(apic);
1631 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1633 struct kvm_lapic *apic = vcpu->arch.apic;
1635 if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1636 apic_lvt_enabled(apic, APIC_LVTT))
1637 return atomic_read(&apic->lapic_timer.pending);
1642 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1644 u32 reg = kvm_apic_get_reg(apic, lvt_type);
1645 int vector, mode, trig_mode;
1647 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1648 vector = reg & APIC_VECTOR_MASK;
1649 mode = reg & APIC_MODE_MASK;
1650 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1651 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1657 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1659 struct kvm_lapic *apic = vcpu->arch.apic;
1662 kvm_apic_local_deliver(apic, APIC_LVT0);
1665 static const struct kvm_io_device_ops apic_mmio_ops = {
1666 .read = apic_mmio_read,
1667 .write = apic_mmio_write,
1670 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1672 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1673 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1675 apic_timer_expired(apic);
1677 if (lapic_is_periodic(apic)) {
1678 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1679 return HRTIMER_RESTART;
1681 return HRTIMER_NORESTART;
1684 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1686 struct kvm_lapic *apic;
1688 ASSERT(vcpu != NULL);
1689 apic_debug("apic_init %d\n", vcpu->vcpu_id);
1691 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1695 vcpu->arch.apic = apic;
1697 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1699 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1701 goto nomem_free_apic;
1705 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1707 apic->lapic_timer.timer.function = apic_timer_fn;
1710 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1711 * thinking that APIC satet has changed.
1713 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1714 kvm_lapic_set_base(vcpu,
1715 APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1717 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1718 kvm_lapic_reset(vcpu);
1719 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1728 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1730 struct kvm_lapic *apic = vcpu->arch.apic;
1733 if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1736 apic_update_ppr(apic);
1737 highest_irr = apic_find_highest_irr(apic);
1738 if ((highest_irr == -1) ||
1739 ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1744 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1746 u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1749 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1751 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1752 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1757 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1759 struct kvm_lapic *apic = vcpu->arch.apic;
1761 if (!kvm_vcpu_has_lapic(vcpu))
1764 if (atomic_read(&apic->lapic_timer.pending) > 0) {
1765 kvm_apic_local_deliver(apic, APIC_LVTT);
1766 if (apic_lvtt_tscdeadline(apic))
1767 apic->lapic_timer.tscdeadline = 0;
1768 atomic_set(&apic->lapic_timer.pending, 0);
1772 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1774 int vector = kvm_apic_has_interrupt(vcpu);
1775 struct kvm_lapic *apic = vcpu->arch.apic;
1781 * We get here even with APIC virtualization enabled, if doing
1782 * nested virtualization and L1 runs with the "acknowledge interrupt
1783 * on exit" mode. Then we cannot inject the interrupt via RVI,
1784 * because the process would deliver it through the IDT.
1787 apic_set_isr(vector, apic);
1788 apic_update_ppr(apic);
1789 apic_clear_irr(vector, apic);
1793 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1794 struct kvm_lapic_state *s)
1796 struct kvm_lapic *apic = vcpu->arch.apic;
1798 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1799 /* set SPIV separately to get count of SW disabled APICs right */
1800 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1801 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1802 /* call kvm_apic_set_id() to put apic into apic_map */
1803 kvm_apic_set_id(apic, kvm_apic_id(apic));
1804 kvm_apic_set_version(vcpu);
1806 apic_update_ppr(apic);
1807 hrtimer_cancel(&apic->lapic_timer.timer);
1808 update_divide_count(apic);
1809 start_apic_timer(apic);
1810 apic->irr_pending = true;
1811 apic->isr_count = kvm_x86_ops->hwapic_isr_update ?
1812 1 : count_vectors(apic->regs + APIC_ISR);
1813 apic->highest_isr_cache = -1;
1814 if (kvm_x86_ops->hwapic_irr_update)
1815 kvm_x86_ops->hwapic_irr_update(vcpu,
1816 apic_find_highest_irr(apic));
1817 if (unlikely(kvm_x86_ops->hwapic_isr_update))
1818 kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
1819 apic_find_highest_isr(apic));
1820 kvm_make_request(KVM_REQ_EVENT, vcpu);
1821 kvm_rtc_eoi_tracking_restore_one(vcpu);
1824 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1826 struct hrtimer *timer;
1828 if (!kvm_vcpu_has_lapic(vcpu))
1831 timer = &vcpu->arch.apic->lapic_timer.timer;
1832 if (hrtimer_cancel(timer))
1833 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1837 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1839 * Detect whether guest triggered PV EOI since the
1840 * last entry. If yes, set EOI on guests's behalf.
1841 * Clear PV EOI in guest memory in any case.
1843 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1844 struct kvm_lapic *apic)
1849 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1850 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1852 * KVM_APIC_PV_EOI_PENDING is unset:
1853 * -> host disabled PV EOI.
1854 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1855 * -> host enabled PV EOI, guest did not execute EOI yet.
1856 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1857 * -> host enabled PV EOI, guest executed EOI.
1859 BUG_ON(!pv_eoi_enabled(vcpu));
1860 pending = pv_eoi_get_pending(vcpu);
1862 * Clear pending bit in any case: it will be set again on vmentry.
1863 * While this might not be ideal from performance point of view,
1864 * this makes sure pv eoi is only enabled when we know it's safe.
1866 pv_eoi_clr_pending(vcpu);
1869 vector = apic_set_eoi(apic);
1870 trace_kvm_pv_eoi(apic, vector);
1873 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1877 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1878 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1880 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1883 kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1886 apic_set_tpr(vcpu->arch.apic, data & 0xff);
1890 * apic_sync_pv_eoi_to_guest - called before vmentry
1892 * Detect whether it's safe to enable PV EOI and
1895 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1896 struct kvm_lapic *apic)
1898 if (!pv_eoi_enabled(vcpu) ||
1899 /* IRR set or many bits in ISR: could be nested. */
1900 apic->irr_pending ||
1901 /* Cache not set: could be safe but we don't bother. */
1902 apic->highest_isr_cache == -1 ||
1903 /* Need EOI to update ioapic. */
1904 kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1906 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1907 * so we need not do anything here.
1912 pv_eoi_set_pending(apic->vcpu);
1915 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1918 int max_irr, max_isr;
1919 struct kvm_lapic *apic = vcpu->arch.apic;
1921 apic_sync_pv_eoi_to_guest(vcpu, apic);
1923 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1926 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1927 max_irr = apic_find_highest_irr(apic);
1930 max_isr = apic_find_highest_isr(apic);
1933 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1935 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1939 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1942 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1943 &vcpu->arch.apic->vapic_cache,
1944 vapic_addr, sizeof(u32)))
1946 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1948 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1951 vcpu->arch.apic->vapic_addr = vapic_addr;
1955 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1957 struct kvm_lapic *apic = vcpu->arch.apic;
1958 u32 reg = (msr - APIC_BASE_MSR) << 4;
1960 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1963 if (reg == APIC_ICR2)
1966 /* if this is ICR write vector before command */
1967 if (reg == APIC_ICR)
1968 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1969 return apic_reg_write(apic, reg, (u32)data);
1972 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1974 struct kvm_lapic *apic = vcpu->arch.apic;
1975 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1977 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1980 if (reg == APIC_DFR || reg == APIC_ICR2) {
1981 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
1986 if (apic_reg_read(apic, reg, 4, &low))
1988 if (reg == APIC_ICR)
1989 apic_reg_read(apic, APIC_ICR2, 4, &high);
1991 *data = (((u64)high) << 32) | low;
1996 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1998 struct kvm_lapic *apic = vcpu->arch.apic;
2000 if (!kvm_vcpu_has_lapic(vcpu))
2003 /* if this is ICR write vector before command */
2004 if (reg == APIC_ICR)
2005 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2006 return apic_reg_write(apic, reg, (u32)data);
2009 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2011 struct kvm_lapic *apic = vcpu->arch.apic;
2014 if (!kvm_vcpu_has_lapic(vcpu))
2017 if (apic_reg_read(apic, reg, 4, &low))
2019 if (reg == APIC_ICR)
2020 apic_reg_read(apic, APIC_ICR2, 4, &high);
2022 *data = (((u64)high) << 32) | low;
2027 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2029 u64 addr = data & ~KVM_MSR_ENABLED;
2030 if (!IS_ALIGNED(addr, 4))
2033 vcpu->arch.pv_eoi.msr_val = data;
2034 if (!pv_eoi_enabled(vcpu))
2036 return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2040 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2042 struct kvm_lapic *apic = vcpu->arch.apic;
2046 if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
2049 pe = xchg(&apic->pending_events, 0);
2051 if (test_bit(KVM_APIC_INIT, &pe)) {
2052 kvm_lapic_reset(vcpu);
2053 kvm_vcpu_reset(vcpu);
2054 if (kvm_vcpu_is_bsp(apic->vcpu))
2055 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2057 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2059 if (test_bit(KVM_APIC_SIPI, &pe) &&
2060 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2061 /* evaluate pending_events before reading the vector */
2063 sipi_vector = apic->sipi_vector;
2064 apic_debug("vcpu %d received sipi with vector # %x\n",
2065 vcpu->vcpu_id, sipi_vector);
2066 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2067 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2071 void kvm_lapic_init(void)
2073 /* do not patch jump label more than once per second */
2074 jump_label_rate_limit(&apic_hw_disabled, HZ);
2075 jump_label_rate_limit(&apic_sw_disabled, HZ);