3 * Local APIC virtualization
5 * Copyright (C) 2006 Qumranet, Inc.
6 * Copyright (C) 2007 Novell
7 * Copyright (C) 2007 Intel
8 * Copyright 2009 Red Hat, Inc. and/or its affiliates.
11 * Dor Laor <dor.laor@qumranet.com>
12 * Gregory Haskins <ghaskins@novell.com>
13 * Yaozu (Eddie) Dong <eddie.dong@intel.com>
15 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
24 #include <linux/highmem.h>
25 #include <linux/smp.h>
26 #include <linux/hrtimer.h>
28 #include <linux/module.h>
29 #include <linux/math64.h>
30 #include <linux/slab.h>
31 #include <asm/processor.h>
34 #include <asm/current.h>
35 #include <asm/apicdef.h>
36 #include <linux/atomic.h>
37 #include <linux/jump_label.h>
38 #include "kvm_cache_regs.h"
45 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
47 #define mod_64(x, y) ((x) % (y))
55 #define APIC_BUS_CYCLE_NS 1
57 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
58 #define apic_debug(fmt, arg...)
60 #define APIC_LVT_NUM 6
61 /* 14 is the version for Xeon and Pentium 8.4.8*/
62 #define APIC_VERSION (0x14UL | ((APIC_LVT_NUM - 1) << 16))
63 #define LAPIC_MMIO_LENGTH (1 << 12)
64 /* followed define is not in apicdef.h */
65 #define APIC_SHORT_MASK 0xc0000
66 #define APIC_DEST_NOSHORT 0x0
67 #define APIC_DEST_MASK 0x800
68 #define MAX_APIC_VECTOR 256
69 #define APIC_VECTORS_PER_REG 32
71 #define APIC_BROADCAST 0xFF
72 #define X2APIC_BROADCAST 0xFFFFFFFFul
74 #define VEC_POS(v) ((v) & (32 - 1))
75 #define REG_POS(v) (((v) >> 5) << 4)
77 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
79 *((u32 *) (apic->regs + reg_off)) = val;
82 static inline int apic_test_vector(int vec, void *bitmap)
84 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
87 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
89 struct kvm_lapic *apic = vcpu->arch.apic;
91 return apic_test_vector(vector, apic->regs + APIC_ISR) ||
92 apic_test_vector(vector, apic->regs + APIC_IRR);
95 static inline void apic_set_vector(int vec, void *bitmap)
97 set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
100 static inline void apic_clear_vector(int vec, void *bitmap)
102 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
105 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
107 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
110 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
112 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
115 struct static_key_deferred apic_hw_disabled __read_mostly;
116 struct static_key_deferred apic_sw_disabled __read_mostly;
118 static inline int apic_enabled(struct kvm_lapic *apic)
120 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
124 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
127 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
128 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
130 static inline int kvm_apic_id(struct kvm_lapic *apic)
132 return (kvm_apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
135 #define KVM_X2APIC_CID_BITS 0
137 static void recalculate_apic_map(struct kvm *kvm)
139 struct kvm_apic_map *new, *old = NULL;
140 struct kvm_vcpu *vcpu;
143 new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
145 mutex_lock(&kvm->arch.apic_map_lock);
151 /* flat mode is default */
154 new->lid_mask = 0xff;
155 new->broadcast = APIC_BROADCAST;
157 kvm_for_each_vcpu(i, vcpu, kvm) {
158 struct kvm_lapic *apic = vcpu->arch.apic;
162 if (!kvm_apic_present(vcpu))
166 * All APICs have to be configured in the same mode by an OS.
167 * We take advatage of this while building logical id loockup
168 * table. After reset APICs are in xapic/flat mode, so if we
169 * find apic with different setting we assume this is the mode
170 * OS wants all apics to be in; build lookup table accordingly.
172 if (apic_x2apic_mode(apic)) {
175 new->cid_mask = (1 << KVM_X2APIC_CID_BITS) - 1;
176 new->lid_mask = 0xffff;
177 new->broadcast = X2APIC_BROADCAST;
178 } else if (kvm_apic_sw_enabled(apic) &&
179 !new->cid_mask /* flat mode */ &&
180 kvm_apic_get_reg(apic, APIC_DFR) == APIC_DFR_CLUSTER) {
186 new->phys_map[kvm_apic_id(apic)] = apic;
188 ldr = kvm_apic_get_reg(apic, APIC_LDR);
189 cid = apic_cluster_id(new, ldr);
190 lid = apic_logical_id(new, ldr);
193 new->logical_map[cid][ffs(lid) - 1] = apic;
196 old = rcu_dereference_protected(kvm->arch.apic_map,
197 lockdep_is_held(&kvm->arch.apic_map_lock));
198 rcu_assign_pointer(kvm->arch.apic_map, new);
199 mutex_unlock(&kvm->arch.apic_map_lock);
204 kvm_vcpu_request_scan_ioapic(kvm);
207 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
209 bool enabled = val & APIC_SPIV_APIC_ENABLED;
211 apic_set_reg(apic, APIC_SPIV, val);
213 if (enabled != apic->sw_enabled) {
214 apic->sw_enabled = enabled;
216 static_key_slow_dec_deferred(&apic_sw_disabled);
217 recalculate_apic_map(apic->vcpu->kvm);
219 static_key_slow_inc(&apic_sw_disabled.key);
223 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
225 apic_set_reg(apic, APIC_ID, id << 24);
226 recalculate_apic_map(apic->vcpu->kvm);
229 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
231 apic_set_reg(apic, APIC_LDR, id);
232 recalculate_apic_map(apic->vcpu->kvm);
235 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
237 return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
240 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
242 return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
245 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
247 return ((kvm_apic_get_reg(apic, APIC_LVTT) &
248 apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_ONESHOT);
251 static inline int apic_lvtt_period(struct kvm_lapic *apic)
253 return ((kvm_apic_get_reg(apic, APIC_LVTT) &
254 apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_PERIODIC);
257 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
259 return ((kvm_apic_get_reg(apic, APIC_LVTT) &
260 apic->lapic_timer.timer_mode_mask) ==
261 APIC_LVT_TIMER_TSCDEADLINE);
264 static inline int apic_lvt_nmi_mode(u32 lvt_val)
266 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
269 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
271 struct kvm_lapic *apic = vcpu->arch.apic;
272 struct kvm_cpuid_entry2 *feat;
273 u32 v = APIC_VERSION;
275 if (!kvm_vcpu_has_lapic(vcpu))
278 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
279 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
280 v |= APIC_LVR_DIRECTED_EOI;
281 apic_set_reg(apic, APIC_LVR, v);
284 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
285 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
286 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
287 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
288 LINT_MASK, LINT_MASK, /* LVT0-1 */
289 LVT_MASK /* LVTERR */
292 static int find_highest_vector(void *bitmap)
297 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
298 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
299 reg = bitmap + REG_POS(vec);
301 return fls(*reg) - 1 + vec;
307 static u8 count_vectors(void *bitmap)
313 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
314 reg = bitmap + REG_POS(vec);
315 count += hweight32(*reg);
321 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
324 struct kvm_lapic *apic = vcpu->arch.apic;
326 for (i = 0; i <= 7; i++) {
327 pir_val = xchg(&pir[i], 0);
329 *((u32 *)(apic->regs + APIC_IRR + i * 0x10)) |= pir_val;
332 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
334 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
336 apic->irr_pending = true;
337 apic_set_vector(vec, apic->regs + APIC_IRR);
340 static inline int apic_search_irr(struct kvm_lapic *apic)
342 return find_highest_vector(apic->regs + APIC_IRR);
345 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
350 * Note that irr_pending is just a hint. It will be always
351 * true with virtual interrupt delivery enabled.
353 if (!apic->irr_pending)
356 kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
357 result = apic_search_irr(apic);
358 ASSERT(result == -1 || result >= 16);
363 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
365 struct kvm_vcpu *vcpu;
369 apic_clear_vector(vec, apic->regs + APIC_IRR);
370 if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
371 /* try to update RVI */
372 kvm_make_request(KVM_REQ_EVENT, vcpu);
374 vec = apic_search_irr(apic);
375 apic->irr_pending = (vec != -1);
379 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
381 struct kvm_vcpu *vcpu;
383 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
389 * With APIC virtualization enabled, all caching is disabled
390 * because the processor can modify ISR under the hood. Instead
393 if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
394 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
397 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
399 * ISR (in service register) bit is set when injecting an interrupt.
400 * The highest vector is injected. Thus the latest bit set matches
401 * the highest bit in ISR.
403 apic->highest_isr_cache = vec;
407 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
412 * Note that isr_count is always 1, and highest_isr_cache
413 * is always -1, with APIC virtualization enabled.
415 if (!apic->isr_count)
417 if (likely(apic->highest_isr_cache != -1))
418 return apic->highest_isr_cache;
420 result = find_highest_vector(apic->regs + APIC_ISR);
421 ASSERT(result == -1 || result >= 16);
426 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
428 struct kvm_vcpu *vcpu;
429 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
435 * We do get here for APIC virtualization enabled if the guest
436 * uses the Hyper-V APIC enlightenment. In this case we may need
437 * to trigger a new interrupt delivery by writing the SVI field;
438 * on the other hand isr_count and highest_isr_cache are unused
439 * and must be left alone.
441 if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
442 kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
443 apic_find_highest_isr(apic));
446 BUG_ON(apic->isr_count < 0);
447 apic->highest_isr_cache = -1;
451 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
455 /* This may race with setting of irr in __apic_accept_irq() and
456 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
457 * will cause vmexit immediately and the value will be recalculated
458 * on the next vmentry.
460 if (!kvm_vcpu_has_lapic(vcpu))
462 highest_irr = apic_find_highest_irr(vcpu->arch.apic);
467 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
468 int vector, int level, int trig_mode,
469 unsigned long *dest_map);
471 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
472 unsigned long *dest_map)
474 struct kvm_lapic *apic = vcpu->arch.apic;
476 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
477 irq->level, irq->trig_mode, dest_map);
480 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
483 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
487 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
490 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
494 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
496 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
499 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
502 if (pv_eoi_get_user(vcpu, &val) < 0)
503 apic_debug("Can't read EOI MSR value: 0x%llx\n",
504 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
508 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
510 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
511 apic_debug("Can't set EOI MSR value: 0x%llx\n",
512 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
515 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
518 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
520 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
521 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
522 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
525 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
528 void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
530 struct kvm_lapic *apic = vcpu->arch.apic;
533 for (i = 0; i < 8; i++)
534 apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
537 static void apic_update_ppr(struct kvm_lapic *apic)
539 u32 tpr, isrv, ppr, old_ppr;
542 old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
543 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
544 isr = apic_find_highest_isr(apic);
545 isrv = (isr != -1) ? isr : 0;
547 if ((tpr & 0xf0) >= (isrv & 0xf0))
552 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
553 apic, ppr, isr, isrv);
555 if (old_ppr != ppr) {
556 apic_set_reg(apic, APIC_PROCPRI, ppr);
558 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
562 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
564 apic_set_reg(apic, APIC_TASKPRI, tpr);
565 apic_update_ppr(apic);
568 static int kvm_apic_broadcast(struct kvm_lapic *apic, u32 dest)
570 return dest == (apic_x2apic_mode(apic) ?
571 X2APIC_BROADCAST : APIC_BROADCAST);
574 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 dest)
576 return kvm_apic_id(apic) == dest || kvm_apic_broadcast(apic, dest);
579 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
584 if (kvm_apic_broadcast(apic, mda))
587 if (apic_x2apic_mode(apic)) {
588 logical_id = kvm_apic_get_reg(apic, APIC_LDR);
589 return logical_id & mda;
592 logical_id = GET_APIC_LOGICAL_ID(kvm_apic_get_reg(apic, APIC_LDR));
594 switch (kvm_apic_get_reg(apic, APIC_DFR)) {
596 if (logical_id & mda)
599 case APIC_DFR_CLUSTER:
600 if (((logical_id >> 4) == (mda >> 0x4))
601 && (logical_id & mda & 0xf))
605 apic_debug("Bad DFR vcpu %d: %08x\n",
606 apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
613 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
614 int short_hand, unsigned int dest, int dest_mode)
617 struct kvm_lapic *target = vcpu->arch.apic;
619 apic_debug("target %p, source %p, dest 0x%x, "
620 "dest_mode 0x%x, short_hand 0x%x\n",
621 target, source, dest, dest_mode, short_hand);
624 switch (short_hand) {
625 case APIC_DEST_NOSHORT:
628 result = kvm_apic_match_physical_addr(target, dest);
631 result = kvm_apic_match_logical_addr(target, dest);
634 result = (target == source);
636 case APIC_DEST_ALLINC:
639 case APIC_DEST_ALLBUT:
640 result = (target != source);
643 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
651 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
652 struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
654 struct kvm_apic_map *map;
655 unsigned long bitmap = 1;
656 struct kvm_lapic **dst;
662 if (irq->shorthand == APIC_DEST_SELF) {
663 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
671 map = rcu_dereference(kvm->arch.apic_map);
676 if (irq->dest_id == map->broadcast)
679 if (irq->dest_mode == 0) { /* physical mode */
680 if (irq->delivery_mode == APIC_DM_LOWEST)
682 dst = &map->phys_map[irq->dest_id & 0xff];
684 u32 mda = irq->dest_id << (32 - map->ldr_bits);
686 dst = map->logical_map[apic_cluster_id(map, mda)];
688 bitmap = apic_logical_id(map, mda);
690 if (irq->delivery_mode == APIC_DM_LOWEST) {
692 for_each_set_bit(i, &bitmap, 16) {
697 else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
701 bitmap = (l >= 0) ? 1 << l : 0;
705 for_each_set_bit(i, &bitmap, 16) {
710 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
720 * Add a pending IRQ into lapic.
721 * Return 1 if successfully added and 0 if discarded.
723 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
724 int vector, int level, int trig_mode,
725 unsigned long *dest_map)
728 struct kvm_vcpu *vcpu = apic->vcpu;
730 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
732 switch (delivery_mode) {
734 vcpu->arch.apic_arb_prio++;
736 /* FIXME add logic for vcpu on reset */
737 if (unlikely(!apic_enabled(apic)))
743 __set_bit(vcpu->vcpu_id, dest_map);
745 if (kvm_x86_ops->deliver_posted_interrupt)
746 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
748 apic_set_irr(vector, apic);
750 kvm_make_request(KVM_REQ_EVENT, vcpu);
757 vcpu->arch.pv.pv_unhalted = 1;
758 kvm_make_request(KVM_REQ_EVENT, vcpu);
763 apic_debug("Ignoring guest SMI\n");
768 kvm_inject_nmi(vcpu);
773 if (!trig_mode || level) {
775 /* assumes that there are only KVM_APIC_INIT/SIPI */
776 apic->pending_events = (1UL << KVM_APIC_INIT);
777 /* make sure pending_events is visible before sending
780 kvm_make_request(KVM_REQ_EVENT, vcpu);
783 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
788 case APIC_DM_STARTUP:
789 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
790 vcpu->vcpu_id, vector);
792 apic->sipi_vector = vector;
793 /* make sure sipi_vector is visible for the receiver */
795 set_bit(KVM_APIC_SIPI, &apic->pending_events);
796 kvm_make_request(KVM_REQ_EVENT, vcpu);
802 * Should only be called by kvm_apic_local_deliver() with LVT0,
803 * before NMI watchdog was enabled. Already handled by
804 * kvm_apic_accept_pic_intr().
809 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
816 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
818 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
821 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
823 if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
824 kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
826 if (apic_test_vector(vector, apic->regs + APIC_TMR))
827 trigger_mode = IOAPIC_LEVEL_TRIG;
829 trigger_mode = IOAPIC_EDGE_TRIG;
830 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
834 static int apic_set_eoi(struct kvm_lapic *apic)
836 int vector = apic_find_highest_isr(apic);
838 trace_kvm_eoi(apic, vector);
841 * Not every write EOI will has corresponding ISR,
842 * one example is when Kernel check timer on setup_IO_APIC
847 apic_clear_isr(vector, apic);
848 apic_update_ppr(apic);
850 kvm_ioapic_send_eoi(apic, vector);
851 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
856 * this interface assumes a trap-like exit, which has already finished
857 * desired side effect including vISR and vPPR update.
859 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
861 struct kvm_lapic *apic = vcpu->arch.apic;
863 trace_kvm_eoi(apic, vector);
865 kvm_ioapic_send_eoi(apic, vector);
866 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
868 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
870 static void apic_send_ipi(struct kvm_lapic *apic)
872 u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
873 u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
874 struct kvm_lapic_irq irq;
876 irq.vector = icr_low & APIC_VECTOR_MASK;
877 irq.delivery_mode = icr_low & APIC_MODE_MASK;
878 irq.dest_mode = icr_low & APIC_DEST_MASK;
879 irq.level = icr_low & APIC_INT_ASSERT;
880 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
881 irq.shorthand = icr_low & APIC_SHORT_MASK;
882 if (apic_x2apic_mode(apic))
883 irq.dest_id = icr_high;
885 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
887 trace_kvm_apic_ipi(icr_low, irq.dest_id);
889 apic_debug("icr_high 0x%x, icr_low 0x%x, "
890 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
891 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
892 icr_high, icr_low, irq.shorthand, irq.dest_id,
893 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
896 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
899 static u32 apic_get_tmcct(struct kvm_lapic *apic)
905 ASSERT(apic != NULL);
907 /* if initial count is 0, current count should also be 0 */
908 if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
909 apic->lapic_timer.period == 0)
912 remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
913 if (ktime_to_ns(remaining) < 0)
914 remaining = ktime_set(0, 0);
916 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
917 tmcct = div64_u64(ns,
918 (APIC_BUS_CYCLE_NS * apic->divide_count));
923 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
925 struct kvm_vcpu *vcpu = apic->vcpu;
926 struct kvm_run *run = vcpu->run;
928 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
929 run->tpr_access.rip = kvm_rip_read(vcpu);
930 run->tpr_access.is_write = write;
933 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
935 if (apic->vcpu->arch.tpr_access_reporting)
936 __report_tpr_access(apic, write);
939 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
943 if (offset >= LAPIC_MMIO_LENGTH)
948 if (apic_x2apic_mode(apic))
949 val = kvm_apic_id(apic);
951 val = kvm_apic_id(apic) << 24;
954 apic_debug("Access APIC ARBPRI register which is for P6\n");
957 case APIC_TMCCT: /* Timer CCR */
958 if (apic_lvtt_tscdeadline(apic))
961 val = apic_get_tmcct(apic);
964 apic_update_ppr(apic);
965 val = kvm_apic_get_reg(apic, offset);
968 report_tpr_access(apic, false);
971 val = kvm_apic_get_reg(apic, offset);
978 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
980 return container_of(dev, struct kvm_lapic, dev);
983 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
986 unsigned char alignment = offset & 0xf;
988 /* this bitmask has a bit cleared for each reserved register */
989 static const u64 rmask = 0x43ff01ffffffe70cULL;
991 if ((alignment + len) > 4) {
992 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
997 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
998 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1003 result = __apic_read(apic, offset & ~0xf);
1005 trace_kvm_apic_read(offset, result);
1011 memcpy(data, (char *)&result + alignment, len);
1014 printk(KERN_ERR "Local APIC read with len = %x, "
1015 "should be 1,2, or 4 instead\n", len);
1021 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1023 return kvm_apic_hw_enabled(apic) &&
1024 addr >= apic->base_address &&
1025 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1028 static int apic_mmio_read(struct kvm_io_device *this,
1029 gpa_t address, int len, void *data)
1031 struct kvm_lapic *apic = to_lapic(this);
1032 u32 offset = address - apic->base_address;
1034 if (!apic_mmio_in_range(apic, address))
1037 apic_reg_read(apic, offset, len, data);
1042 static void update_divide_count(struct kvm_lapic *apic)
1044 u32 tmp1, tmp2, tdcr;
1046 tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1048 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1049 apic->divide_count = 0x1 << (tmp2 & 0x7);
1051 apic_debug("timer divide count is 0x%x\n",
1052 apic->divide_count);
1055 static void apic_timer_expired(struct kvm_lapic *apic)
1057 struct kvm_vcpu *vcpu = apic->vcpu;
1058 wait_queue_head_t *q = &vcpu->wq;
1061 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1064 if (atomic_read(&apic->lapic_timer.pending))
1067 atomic_inc(&apic->lapic_timer.pending);
1068 /* FIXME: this code should not know anything about vcpus */
1069 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1071 if (waitqueue_active(q))
1072 wake_up_interruptible(q);
1075 static void start_apic_timer(struct kvm_lapic *apic)
1078 atomic_set(&apic->lapic_timer.pending, 0);
1080 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1081 /* lapic timer in oneshot or periodic mode */
1082 now = apic->lapic_timer.timer.base->get_time();
1083 apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1084 * APIC_BUS_CYCLE_NS * apic->divide_count;
1086 if (!apic->lapic_timer.period)
1089 * Do not allow the guest to program periodic timers with small
1090 * interval, since the hrtimers are not throttled by the host
1093 if (apic_lvtt_period(apic)) {
1094 s64 min_period = min_timer_period_us * 1000LL;
1096 if (apic->lapic_timer.period < min_period) {
1097 pr_info_ratelimited(
1098 "kvm: vcpu %i: requested %lld ns "
1099 "lapic timer period limited to %lld ns\n",
1100 apic->vcpu->vcpu_id,
1101 apic->lapic_timer.period, min_period);
1102 apic->lapic_timer.period = min_period;
1106 hrtimer_start(&apic->lapic_timer.timer,
1107 ktime_add_ns(now, apic->lapic_timer.period),
1110 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1112 "timer initial count 0x%x, period %lldns, "
1113 "expire @ 0x%016" PRIx64 ".\n", __func__,
1114 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1115 kvm_apic_get_reg(apic, APIC_TMICT),
1116 apic->lapic_timer.period,
1117 ktime_to_ns(ktime_add_ns(now,
1118 apic->lapic_timer.period)));
1119 } else if (apic_lvtt_tscdeadline(apic)) {
1120 /* lapic timer in tsc deadline mode */
1121 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1123 struct kvm_vcpu *vcpu = apic->vcpu;
1124 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1125 unsigned long flags;
1127 if (unlikely(!tscdeadline || !this_tsc_khz))
1130 local_irq_save(flags);
1132 now = apic->lapic_timer.timer.base->get_time();
1133 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1134 if (likely(tscdeadline > guest_tsc)) {
1135 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1136 do_div(ns, this_tsc_khz);
1137 hrtimer_start(&apic->lapic_timer.timer,
1138 ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
1140 apic_timer_expired(apic);
1142 local_irq_restore(flags);
1146 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1148 int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1150 if (apic_lvt_nmi_mode(lvt0_val)) {
1151 if (!nmi_wd_enabled) {
1152 apic_debug("Receive NMI setting on APIC_LVT0 "
1153 "for cpu %d\n", apic->vcpu->vcpu_id);
1154 apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1156 } else if (nmi_wd_enabled)
1157 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1160 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1164 trace_kvm_apic_write(reg, val);
1167 case APIC_ID: /* Local APIC ID */
1168 if (!apic_x2apic_mode(apic))
1169 kvm_apic_set_id(apic, val >> 24);
1175 report_tpr_access(apic, true);
1176 apic_set_tpr(apic, val & 0xff);
1184 if (!apic_x2apic_mode(apic))
1185 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1191 if (!apic_x2apic_mode(apic)) {
1192 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1193 recalculate_apic_map(apic->vcpu->kvm);
1200 if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1201 mask |= APIC_SPIV_DIRECTED_EOI;
1202 apic_set_spiv(apic, val & mask);
1203 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1207 for (i = 0; i < APIC_LVT_NUM; i++) {
1208 lvt_val = kvm_apic_get_reg(apic,
1209 APIC_LVTT + 0x10 * i);
1210 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1211 lvt_val | APIC_LVT_MASKED);
1213 atomic_set(&apic->lapic_timer.pending, 0);
1219 /* No delay here, so we always clear the pending bit */
1220 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1221 apic_send_ipi(apic);
1225 if (!apic_x2apic_mode(apic))
1227 apic_set_reg(apic, APIC_ICR2, val);
1231 apic_manage_nmi_watchdog(apic, val);
1236 /* TODO: Check vector */
1237 if (!kvm_apic_sw_enabled(apic))
1238 val |= APIC_LVT_MASKED;
1240 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1241 apic_set_reg(apic, reg, val);
1246 u32 timer_mode = val & apic->lapic_timer.timer_mode_mask;
1248 if (apic->lapic_timer.timer_mode != timer_mode) {
1249 apic->lapic_timer.timer_mode = timer_mode;
1250 hrtimer_cancel(&apic->lapic_timer.timer);
1253 if (!kvm_apic_sw_enabled(apic))
1254 val |= APIC_LVT_MASKED;
1255 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1256 apic_set_reg(apic, APIC_LVTT, val);
1261 if (apic_lvtt_tscdeadline(apic))
1264 hrtimer_cancel(&apic->lapic_timer.timer);
1265 apic_set_reg(apic, APIC_TMICT, val);
1266 start_apic_timer(apic);
1271 apic_debug("KVM_WRITE:TDCR %x\n", val);
1272 apic_set_reg(apic, APIC_TDCR, val);
1273 update_divide_count(apic);
1277 if (apic_x2apic_mode(apic) && val != 0) {
1278 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1284 if (apic_x2apic_mode(apic)) {
1285 apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1294 apic_debug("Local APIC Write to read-only register %x\n", reg);
1298 static int apic_mmio_write(struct kvm_io_device *this,
1299 gpa_t address, int len, const void *data)
1301 struct kvm_lapic *apic = to_lapic(this);
1302 unsigned int offset = address - apic->base_address;
1305 if (!apic_mmio_in_range(apic, address))
1309 * APIC register must be aligned on 128-bits boundary.
1310 * 32/64/128 bits registers must be accessed thru 32 bits.
1313 if (len != 4 || (offset & 0xf)) {
1314 /* Don't shout loud, $infamous_os would cause only noise. */
1315 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1321 /* too common printing */
1322 if (offset != APIC_EOI)
1323 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1324 "0x%x\n", __func__, offset, len, val);
1326 apic_reg_write(apic, offset & 0xff0, val);
1331 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1333 if (kvm_vcpu_has_lapic(vcpu))
1334 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1336 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1338 /* emulate APIC access in a trap manner */
1339 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1343 /* hw has done the conditional check and inst decode */
1346 apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1348 /* TODO: optimize to just emulate side effect w/o one more write */
1349 apic_reg_write(vcpu->arch.apic, offset, val);
1351 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1353 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1355 struct kvm_lapic *apic = vcpu->arch.apic;
1357 if (!vcpu->arch.apic)
1360 hrtimer_cancel(&apic->lapic_timer.timer);
1362 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1363 static_key_slow_dec_deferred(&apic_hw_disabled);
1365 if (!apic->sw_enabled)
1366 static_key_slow_dec_deferred(&apic_sw_disabled);
1369 free_page((unsigned long)apic->regs);
1375 *----------------------------------------------------------------------
1377 *----------------------------------------------------------------------
1380 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1382 struct kvm_lapic *apic = vcpu->arch.apic;
1384 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1385 apic_lvtt_period(apic))
1388 return apic->lapic_timer.tscdeadline;
1391 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1393 struct kvm_lapic *apic = vcpu->arch.apic;
1395 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1396 apic_lvtt_period(apic))
1399 hrtimer_cancel(&apic->lapic_timer.timer);
1400 apic->lapic_timer.tscdeadline = data;
1401 start_apic_timer(apic);
1404 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1406 struct kvm_lapic *apic = vcpu->arch.apic;
1408 if (!kvm_vcpu_has_lapic(vcpu))
1411 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1412 | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1415 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1419 if (!kvm_vcpu_has_lapic(vcpu))
1422 tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1424 return (tpr & 0xf0) >> 4;
1427 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1429 u64 old_value = vcpu->arch.apic_base;
1430 struct kvm_lapic *apic = vcpu->arch.apic;
1433 value |= MSR_IA32_APICBASE_BSP;
1434 vcpu->arch.apic_base = value;
1438 if (!kvm_vcpu_is_bsp(apic->vcpu))
1439 value &= ~MSR_IA32_APICBASE_BSP;
1440 vcpu->arch.apic_base = value;
1442 /* update jump label if enable bit changes */
1443 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1444 if (value & MSR_IA32_APICBASE_ENABLE)
1445 static_key_slow_dec_deferred(&apic_hw_disabled);
1447 static_key_slow_inc(&apic_hw_disabled.key);
1448 recalculate_apic_map(vcpu->kvm);
1451 if ((old_value ^ value) & X2APIC_ENABLE) {
1452 if (value & X2APIC_ENABLE) {
1453 u32 id = kvm_apic_id(apic);
1454 u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1455 kvm_apic_set_ldr(apic, ldr);
1456 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1458 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1461 apic->base_address = apic->vcpu->arch.apic_base &
1462 MSR_IA32_APICBASE_BASE;
1464 /* with FSB delivery interrupt, we can restart APIC functionality */
1465 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1466 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1470 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1472 struct kvm_lapic *apic;
1475 apic_debug("%s\n", __func__);
1478 apic = vcpu->arch.apic;
1479 ASSERT(apic != NULL);
1481 /* Stop the timer in case it's a reset to an active apic */
1482 hrtimer_cancel(&apic->lapic_timer.timer);
1484 kvm_apic_set_id(apic, vcpu->vcpu_id);
1485 kvm_apic_set_version(apic->vcpu);
1487 for (i = 0; i < APIC_LVT_NUM; i++)
1488 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1489 apic->lapic_timer.timer_mode = 0;
1490 apic_set_reg(apic, APIC_LVT0,
1491 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1493 apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1494 apic_set_spiv(apic, 0xff);
1495 apic_set_reg(apic, APIC_TASKPRI, 0);
1496 kvm_apic_set_ldr(apic, 0);
1497 apic_set_reg(apic, APIC_ESR, 0);
1498 apic_set_reg(apic, APIC_ICR, 0);
1499 apic_set_reg(apic, APIC_ICR2, 0);
1500 apic_set_reg(apic, APIC_TDCR, 0);
1501 apic_set_reg(apic, APIC_TMICT, 0);
1502 for (i = 0; i < 8; i++) {
1503 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1504 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1505 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1507 apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1508 apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm);
1509 apic->highest_isr_cache = -1;
1510 update_divide_count(apic);
1511 atomic_set(&apic->lapic_timer.pending, 0);
1512 if (kvm_vcpu_is_bsp(vcpu))
1513 kvm_lapic_set_base(vcpu,
1514 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1515 vcpu->arch.pv_eoi.msr_val = 0;
1516 apic_update_ppr(apic);
1518 vcpu->arch.apic_arb_prio = 0;
1519 vcpu->arch.apic_attention = 0;
1521 apic_debug("%s: vcpu=%p, id=%d, base_msr="
1522 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1523 vcpu, kvm_apic_id(apic),
1524 vcpu->arch.apic_base, apic->base_address);
1528 *----------------------------------------------------------------------
1530 *----------------------------------------------------------------------
1533 static bool lapic_is_periodic(struct kvm_lapic *apic)
1535 return apic_lvtt_period(apic);
1538 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1540 struct kvm_lapic *apic = vcpu->arch.apic;
1542 if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1543 apic_lvt_enabled(apic, APIC_LVTT))
1544 return atomic_read(&apic->lapic_timer.pending);
1549 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1551 u32 reg = kvm_apic_get_reg(apic, lvt_type);
1552 int vector, mode, trig_mode;
1554 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1555 vector = reg & APIC_VECTOR_MASK;
1556 mode = reg & APIC_MODE_MASK;
1557 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1558 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1564 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1566 struct kvm_lapic *apic = vcpu->arch.apic;
1569 kvm_apic_local_deliver(apic, APIC_LVT0);
1572 static const struct kvm_io_device_ops apic_mmio_ops = {
1573 .read = apic_mmio_read,
1574 .write = apic_mmio_write,
1577 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1579 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1580 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1582 apic_timer_expired(apic);
1584 if (lapic_is_periodic(apic)) {
1585 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1586 return HRTIMER_RESTART;
1588 return HRTIMER_NORESTART;
1591 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1593 struct kvm_lapic *apic;
1595 ASSERT(vcpu != NULL);
1596 apic_debug("apic_init %d\n", vcpu->vcpu_id);
1598 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1602 vcpu->arch.apic = apic;
1604 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1606 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1608 goto nomem_free_apic;
1612 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1614 apic->lapic_timer.timer.function = apic_timer_fn;
1617 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1618 * thinking that APIC satet has changed.
1620 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1621 kvm_lapic_set_base(vcpu,
1622 APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1624 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1625 kvm_lapic_reset(vcpu);
1626 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1635 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1637 struct kvm_lapic *apic = vcpu->arch.apic;
1640 if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1643 apic_update_ppr(apic);
1644 highest_irr = apic_find_highest_irr(apic);
1645 if ((highest_irr == -1) ||
1646 ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1651 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1653 u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1656 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1658 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1659 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1664 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1666 struct kvm_lapic *apic = vcpu->arch.apic;
1668 if (!kvm_vcpu_has_lapic(vcpu))
1671 if (atomic_read(&apic->lapic_timer.pending) > 0) {
1672 kvm_apic_local_deliver(apic, APIC_LVTT);
1673 if (apic_lvtt_tscdeadline(apic))
1674 apic->lapic_timer.tscdeadline = 0;
1675 atomic_set(&apic->lapic_timer.pending, 0);
1679 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1681 int vector = kvm_apic_has_interrupt(vcpu);
1682 struct kvm_lapic *apic = vcpu->arch.apic;
1688 * We get here even with APIC virtualization enabled, if doing
1689 * nested virtualization and L1 runs with the "acknowledge interrupt
1690 * on exit" mode. Then we cannot inject the interrupt via RVI,
1691 * because the process would deliver it through the IDT.
1694 apic_set_isr(vector, apic);
1695 apic_update_ppr(apic);
1696 apic_clear_irr(vector, apic);
1700 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1701 struct kvm_lapic_state *s)
1703 struct kvm_lapic *apic = vcpu->arch.apic;
1705 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1706 /* set SPIV separately to get count of SW disabled APICs right */
1707 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1708 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1709 /* call kvm_apic_set_id() to put apic into apic_map */
1710 kvm_apic_set_id(apic, kvm_apic_id(apic));
1711 kvm_apic_set_version(vcpu);
1713 apic_update_ppr(apic);
1714 hrtimer_cancel(&apic->lapic_timer.timer);
1715 update_divide_count(apic);
1716 start_apic_timer(apic);
1717 apic->irr_pending = true;
1718 apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm) ?
1719 1 : count_vectors(apic->regs + APIC_ISR);
1720 apic->highest_isr_cache = -1;
1721 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, apic_find_highest_isr(apic));
1722 kvm_make_request(KVM_REQ_EVENT, vcpu);
1723 kvm_rtc_eoi_tracking_restore_one(vcpu);
1726 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1728 struct hrtimer *timer;
1730 if (!kvm_vcpu_has_lapic(vcpu))
1733 timer = &vcpu->arch.apic->lapic_timer.timer;
1734 if (hrtimer_cancel(timer))
1735 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1739 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1741 * Detect whether guest triggered PV EOI since the
1742 * last entry. If yes, set EOI on guests's behalf.
1743 * Clear PV EOI in guest memory in any case.
1745 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1746 struct kvm_lapic *apic)
1751 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1752 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1754 * KVM_APIC_PV_EOI_PENDING is unset:
1755 * -> host disabled PV EOI.
1756 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1757 * -> host enabled PV EOI, guest did not execute EOI yet.
1758 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1759 * -> host enabled PV EOI, guest executed EOI.
1761 BUG_ON(!pv_eoi_enabled(vcpu));
1762 pending = pv_eoi_get_pending(vcpu);
1764 * Clear pending bit in any case: it will be set again on vmentry.
1765 * While this might not be ideal from performance point of view,
1766 * this makes sure pv eoi is only enabled when we know it's safe.
1768 pv_eoi_clr_pending(vcpu);
1771 vector = apic_set_eoi(apic);
1772 trace_kvm_pv_eoi(apic, vector);
1775 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1779 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1780 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1782 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1785 kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1788 apic_set_tpr(vcpu->arch.apic, data & 0xff);
1792 * apic_sync_pv_eoi_to_guest - called before vmentry
1794 * Detect whether it's safe to enable PV EOI and
1797 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1798 struct kvm_lapic *apic)
1800 if (!pv_eoi_enabled(vcpu) ||
1801 /* IRR set or many bits in ISR: could be nested. */
1802 apic->irr_pending ||
1803 /* Cache not set: could be safe but we don't bother. */
1804 apic->highest_isr_cache == -1 ||
1805 /* Need EOI to update ioapic. */
1806 kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1808 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1809 * so we need not do anything here.
1814 pv_eoi_set_pending(apic->vcpu);
1817 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1820 int max_irr, max_isr;
1821 struct kvm_lapic *apic = vcpu->arch.apic;
1823 apic_sync_pv_eoi_to_guest(vcpu, apic);
1825 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1828 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1829 max_irr = apic_find_highest_irr(apic);
1832 max_isr = apic_find_highest_isr(apic);
1835 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1837 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1841 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1844 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1845 &vcpu->arch.apic->vapic_cache,
1846 vapic_addr, sizeof(u32)))
1848 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1850 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1853 vcpu->arch.apic->vapic_addr = vapic_addr;
1857 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1859 struct kvm_lapic *apic = vcpu->arch.apic;
1860 u32 reg = (msr - APIC_BASE_MSR) << 4;
1862 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1865 /* if this is ICR write vector before command */
1867 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1868 return apic_reg_write(apic, reg, (u32)data);
1871 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1873 struct kvm_lapic *apic = vcpu->arch.apic;
1874 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1876 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1879 if (apic_reg_read(apic, reg, 4, &low))
1882 apic_reg_read(apic, APIC_ICR2, 4, &high);
1884 *data = (((u64)high) << 32) | low;
1889 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1891 struct kvm_lapic *apic = vcpu->arch.apic;
1893 if (!kvm_vcpu_has_lapic(vcpu))
1896 /* if this is ICR write vector before command */
1897 if (reg == APIC_ICR)
1898 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1899 return apic_reg_write(apic, reg, (u32)data);
1902 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1904 struct kvm_lapic *apic = vcpu->arch.apic;
1907 if (!kvm_vcpu_has_lapic(vcpu))
1910 if (apic_reg_read(apic, reg, 4, &low))
1912 if (reg == APIC_ICR)
1913 apic_reg_read(apic, APIC_ICR2, 4, &high);
1915 *data = (((u64)high) << 32) | low;
1920 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
1922 u64 addr = data & ~KVM_MSR_ENABLED;
1923 if (!IS_ALIGNED(addr, 4))
1926 vcpu->arch.pv_eoi.msr_val = data;
1927 if (!pv_eoi_enabled(vcpu))
1929 return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
1933 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
1935 struct kvm_lapic *apic = vcpu->arch.apic;
1936 unsigned int sipi_vector;
1939 if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
1942 pe = xchg(&apic->pending_events, 0);
1944 if (test_bit(KVM_APIC_INIT, &pe)) {
1945 kvm_lapic_reset(vcpu);
1946 kvm_vcpu_reset(vcpu);
1947 if (kvm_vcpu_is_bsp(apic->vcpu))
1948 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1950 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
1952 if (test_bit(KVM_APIC_SIPI, &pe) &&
1953 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
1954 /* evaluate pending_events before reading the vector */
1956 sipi_vector = apic->sipi_vector;
1957 apic_debug("vcpu %d received sipi with vector # %x\n",
1958 vcpu->vcpu_id, sipi_vector);
1959 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
1960 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1964 void kvm_lapic_init(void)
1966 /* do not patch jump label more than once per second */
1967 jump_label_rate_limit(&apic_hw_disabled, HZ);
1968 jump_label_rate_limit(&apic_sw_disabled, HZ);