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;
160 if (!kvm_apic_present(vcpu))
163 if (apic_x2apic_mode(apic)) {
166 new->cid_mask = (1 << KVM_X2APIC_CID_BITS) - 1;
167 new->lid_mask = 0xffff;
168 new->broadcast = X2APIC_BROADCAST;
169 } else if (kvm_apic_get_reg(apic, APIC_LDR)) {
170 if (kvm_apic_get_reg(apic, APIC_DFR) ==
178 new->lid_mask = 0xff;
183 * All APICs have to be configured in the same mode by an OS.
184 * We take advatage of this while building logical id loockup
185 * table. After reset APICs are in software disabled mode, so if
186 * we find apic with different setting we assume this is the mode
187 * OS wants all apics to be in; build lookup table accordingly.
189 if (kvm_apic_sw_enabled(apic))
193 kvm_for_each_vcpu(i, vcpu, kvm) {
194 struct kvm_lapic *apic = vcpu->arch.apic;
198 new->phys_map[kvm_apic_id(apic)] = apic;
200 ldr = kvm_apic_get_reg(apic, APIC_LDR);
201 cid = apic_cluster_id(new, ldr);
202 lid = apic_logical_id(new, ldr);
205 new->logical_map[cid][ffs(lid) - 1] = apic;
208 old = rcu_dereference_protected(kvm->arch.apic_map,
209 lockdep_is_held(&kvm->arch.apic_map_lock));
210 rcu_assign_pointer(kvm->arch.apic_map, new);
211 mutex_unlock(&kvm->arch.apic_map_lock);
216 kvm_vcpu_request_scan_ioapic(kvm);
219 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
221 bool enabled = val & APIC_SPIV_APIC_ENABLED;
223 apic_set_reg(apic, APIC_SPIV, val);
225 if (enabled != apic->sw_enabled) {
226 apic->sw_enabled = enabled;
228 static_key_slow_dec_deferred(&apic_sw_disabled);
229 recalculate_apic_map(apic->vcpu->kvm);
231 static_key_slow_inc(&apic_sw_disabled.key);
235 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
237 apic_set_reg(apic, APIC_ID, id << 24);
238 recalculate_apic_map(apic->vcpu->kvm);
241 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
243 apic_set_reg(apic, APIC_LDR, id);
244 recalculate_apic_map(apic->vcpu->kvm);
247 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
249 return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
252 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
254 return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
257 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
259 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
262 static inline int apic_lvtt_period(struct kvm_lapic *apic)
264 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
267 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
269 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
272 static inline int apic_lvt_nmi_mode(u32 lvt_val)
274 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
277 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
279 struct kvm_lapic *apic = vcpu->arch.apic;
280 struct kvm_cpuid_entry2 *feat;
281 u32 v = APIC_VERSION;
283 if (!kvm_vcpu_has_lapic(vcpu))
286 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
287 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
288 v |= APIC_LVR_DIRECTED_EOI;
289 apic_set_reg(apic, APIC_LVR, v);
292 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
293 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
294 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
295 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
296 LINT_MASK, LINT_MASK, /* LVT0-1 */
297 LVT_MASK /* LVTERR */
300 static int find_highest_vector(void *bitmap)
305 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
306 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
307 reg = bitmap + REG_POS(vec);
309 return fls(*reg) - 1 + vec;
315 static u8 count_vectors(void *bitmap)
321 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
322 reg = bitmap + REG_POS(vec);
323 count += hweight32(*reg);
329 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
332 struct kvm_lapic *apic = vcpu->arch.apic;
334 for (i = 0; i <= 7; i++) {
335 pir_val = xchg(&pir[i], 0);
337 *((u32 *)(apic->regs + APIC_IRR + i * 0x10)) |= pir_val;
340 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
342 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
344 apic->irr_pending = true;
345 apic_set_vector(vec, apic->regs + APIC_IRR);
348 static inline int apic_search_irr(struct kvm_lapic *apic)
350 return find_highest_vector(apic->regs + APIC_IRR);
353 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
358 * Note that irr_pending is just a hint. It will be always
359 * true with virtual interrupt delivery enabled.
361 if (!apic->irr_pending)
364 kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
365 result = apic_search_irr(apic);
366 ASSERT(result == -1 || result >= 16);
371 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
373 struct kvm_vcpu *vcpu;
377 apic_clear_vector(vec, apic->regs + APIC_IRR);
378 if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
379 /* try to update RVI */
380 kvm_make_request(KVM_REQ_EVENT, vcpu);
382 vec = apic_search_irr(apic);
383 apic->irr_pending = (vec != -1);
387 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
389 struct kvm_vcpu *vcpu;
391 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
397 * With APIC virtualization enabled, all caching is disabled
398 * because the processor can modify ISR under the hood. Instead
401 if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
402 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
405 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
407 * ISR (in service register) bit is set when injecting an interrupt.
408 * The highest vector is injected. Thus the latest bit set matches
409 * the highest bit in ISR.
411 apic->highest_isr_cache = vec;
415 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
420 * Note that isr_count is always 1, and highest_isr_cache
421 * is always -1, with APIC virtualization enabled.
423 if (!apic->isr_count)
425 if (likely(apic->highest_isr_cache != -1))
426 return apic->highest_isr_cache;
428 result = find_highest_vector(apic->regs + APIC_ISR);
429 ASSERT(result == -1 || result >= 16);
434 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
436 struct kvm_vcpu *vcpu;
437 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
443 * We do get here for APIC virtualization enabled if the guest
444 * uses the Hyper-V APIC enlightenment. In this case we may need
445 * to trigger a new interrupt delivery by writing the SVI field;
446 * on the other hand isr_count and highest_isr_cache are unused
447 * and must be left alone.
449 if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
450 kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
451 apic_find_highest_isr(apic));
454 BUG_ON(apic->isr_count < 0);
455 apic->highest_isr_cache = -1;
459 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
463 /* This may race with setting of irr in __apic_accept_irq() and
464 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
465 * will cause vmexit immediately and the value will be recalculated
466 * on the next vmentry.
468 if (!kvm_vcpu_has_lapic(vcpu))
470 highest_irr = apic_find_highest_irr(vcpu->arch.apic);
475 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
476 int vector, int level, int trig_mode,
477 unsigned long *dest_map);
479 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
480 unsigned long *dest_map)
482 struct kvm_lapic *apic = vcpu->arch.apic;
484 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
485 irq->level, irq->trig_mode, dest_map);
488 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
491 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
495 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
498 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
502 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
504 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
507 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
510 if (pv_eoi_get_user(vcpu, &val) < 0)
511 apic_debug("Can't read EOI MSR value: 0x%llx\n",
512 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
516 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
518 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
519 apic_debug("Can't set EOI MSR value: 0x%llx\n",
520 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
523 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
526 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
528 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
529 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
530 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
533 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
536 void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
538 struct kvm_lapic *apic = vcpu->arch.apic;
541 for (i = 0; i < 8; i++)
542 apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
545 static void apic_update_ppr(struct kvm_lapic *apic)
547 u32 tpr, isrv, ppr, old_ppr;
550 old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
551 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
552 isr = apic_find_highest_isr(apic);
553 isrv = (isr != -1) ? isr : 0;
555 if ((tpr & 0xf0) >= (isrv & 0xf0))
560 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
561 apic, ppr, isr, isrv);
563 if (old_ppr != ppr) {
564 apic_set_reg(apic, APIC_PROCPRI, ppr);
566 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
570 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
572 apic_set_reg(apic, APIC_TASKPRI, tpr);
573 apic_update_ppr(apic);
576 static int kvm_apic_broadcast(struct kvm_lapic *apic, u32 dest)
578 return dest == (apic_x2apic_mode(apic) ?
579 X2APIC_BROADCAST : APIC_BROADCAST);
582 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 dest)
584 return kvm_apic_id(apic) == dest || kvm_apic_broadcast(apic, dest);
587 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
592 if (kvm_apic_broadcast(apic, mda))
595 if (apic_x2apic_mode(apic)) {
596 logical_id = kvm_apic_get_reg(apic, APIC_LDR);
597 return logical_id & mda;
600 logical_id = GET_APIC_LOGICAL_ID(kvm_apic_get_reg(apic, APIC_LDR));
602 switch (kvm_apic_get_reg(apic, APIC_DFR)) {
604 if (logical_id & mda)
607 case APIC_DFR_CLUSTER:
608 if (((logical_id >> 4) == (mda >> 0x4))
609 && (logical_id & mda & 0xf))
613 apic_debug("Bad DFR vcpu %d: %08x\n",
614 apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
621 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
622 int short_hand, unsigned int dest, int dest_mode)
625 struct kvm_lapic *target = vcpu->arch.apic;
627 apic_debug("target %p, source %p, dest 0x%x, "
628 "dest_mode 0x%x, short_hand 0x%x\n",
629 target, source, dest, dest_mode, short_hand);
632 switch (short_hand) {
633 case APIC_DEST_NOSHORT:
636 result = kvm_apic_match_physical_addr(target, dest);
639 result = kvm_apic_match_logical_addr(target, dest);
642 result = (target == source);
644 case APIC_DEST_ALLINC:
647 case APIC_DEST_ALLBUT:
648 result = (target != source);
651 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
659 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
660 struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
662 struct kvm_apic_map *map;
663 unsigned long bitmap = 1;
664 struct kvm_lapic **dst;
670 if (irq->shorthand == APIC_DEST_SELF) {
671 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
679 map = rcu_dereference(kvm->arch.apic_map);
684 if (irq->dest_id == map->broadcast)
687 if (irq->dest_mode == 0) { /* physical mode */
688 if (irq->delivery_mode == APIC_DM_LOWEST)
690 dst = &map->phys_map[irq->dest_id & 0xff];
692 u32 mda = irq->dest_id << (32 - map->ldr_bits);
694 dst = map->logical_map[apic_cluster_id(map, mda)];
696 bitmap = apic_logical_id(map, mda);
698 if (irq->delivery_mode == APIC_DM_LOWEST) {
700 for_each_set_bit(i, &bitmap, 16) {
705 else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
709 bitmap = (l >= 0) ? 1 << l : 0;
713 for_each_set_bit(i, &bitmap, 16) {
718 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
728 * Add a pending IRQ into lapic.
729 * Return 1 if successfully added and 0 if discarded.
731 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
732 int vector, int level, int trig_mode,
733 unsigned long *dest_map)
736 struct kvm_vcpu *vcpu = apic->vcpu;
738 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
740 switch (delivery_mode) {
742 vcpu->arch.apic_arb_prio++;
744 /* FIXME add logic for vcpu on reset */
745 if (unlikely(!apic_enabled(apic)))
751 __set_bit(vcpu->vcpu_id, dest_map);
753 if (kvm_x86_ops->deliver_posted_interrupt)
754 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
756 apic_set_irr(vector, apic);
758 kvm_make_request(KVM_REQ_EVENT, vcpu);
765 vcpu->arch.pv.pv_unhalted = 1;
766 kvm_make_request(KVM_REQ_EVENT, vcpu);
771 apic_debug("Ignoring guest SMI\n");
776 kvm_inject_nmi(vcpu);
781 if (!trig_mode || level) {
783 /* assumes that there are only KVM_APIC_INIT/SIPI */
784 apic->pending_events = (1UL << KVM_APIC_INIT);
785 /* make sure pending_events is visible before sending
788 kvm_make_request(KVM_REQ_EVENT, vcpu);
791 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
796 case APIC_DM_STARTUP:
797 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
798 vcpu->vcpu_id, vector);
800 apic->sipi_vector = vector;
801 /* make sure sipi_vector is visible for the receiver */
803 set_bit(KVM_APIC_SIPI, &apic->pending_events);
804 kvm_make_request(KVM_REQ_EVENT, vcpu);
810 * Should only be called by kvm_apic_local_deliver() with LVT0,
811 * before NMI watchdog was enabled. Already handled by
812 * kvm_apic_accept_pic_intr().
817 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
824 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
826 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
829 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
831 if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
832 kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
834 if (apic_test_vector(vector, apic->regs + APIC_TMR))
835 trigger_mode = IOAPIC_LEVEL_TRIG;
837 trigger_mode = IOAPIC_EDGE_TRIG;
838 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
842 static int apic_set_eoi(struct kvm_lapic *apic)
844 int vector = apic_find_highest_isr(apic);
846 trace_kvm_eoi(apic, vector);
849 * Not every write EOI will has corresponding ISR,
850 * one example is when Kernel check timer on setup_IO_APIC
855 apic_clear_isr(vector, apic);
856 apic_update_ppr(apic);
858 kvm_ioapic_send_eoi(apic, vector);
859 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
864 * this interface assumes a trap-like exit, which has already finished
865 * desired side effect including vISR and vPPR update.
867 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
869 struct kvm_lapic *apic = vcpu->arch.apic;
871 trace_kvm_eoi(apic, vector);
873 kvm_ioapic_send_eoi(apic, vector);
874 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
876 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
878 static void apic_send_ipi(struct kvm_lapic *apic)
880 u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
881 u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
882 struct kvm_lapic_irq irq;
884 irq.vector = icr_low & APIC_VECTOR_MASK;
885 irq.delivery_mode = icr_low & APIC_MODE_MASK;
886 irq.dest_mode = icr_low & APIC_DEST_MASK;
887 irq.level = icr_low & APIC_INT_ASSERT;
888 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
889 irq.shorthand = icr_low & APIC_SHORT_MASK;
890 if (apic_x2apic_mode(apic))
891 irq.dest_id = icr_high;
893 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
895 trace_kvm_apic_ipi(icr_low, irq.dest_id);
897 apic_debug("icr_high 0x%x, icr_low 0x%x, "
898 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
899 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
900 icr_high, icr_low, irq.shorthand, irq.dest_id,
901 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
904 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
907 static u32 apic_get_tmcct(struct kvm_lapic *apic)
913 ASSERT(apic != NULL);
915 /* if initial count is 0, current count should also be 0 */
916 if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
917 apic->lapic_timer.period == 0)
920 remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
921 if (ktime_to_ns(remaining) < 0)
922 remaining = ktime_set(0, 0);
924 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
925 tmcct = div64_u64(ns,
926 (APIC_BUS_CYCLE_NS * apic->divide_count));
931 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
933 struct kvm_vcpu *vcpu = apic->vcpu;
934 struct kvm_run *run = vcpu->run;
936 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
937 run->tpr_access.rip = kvm_rip_read(vcpu);
938 run->tpr_access.is_write = write;
941 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
943 if (apic->vcpu->arch.tpr_access_reporting)
944 __report_tpr_access(apic, write);
947 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
951 if (offset >= LAPIC_MMIO_LENGTH)
956 if (apic_x2apic_mode(apic))
957 val = kvm_apic_id(apic);
959 val = kvm_apic_id(apic) << 24;
962 apic_debug("Access APIC ARBPRI register which is for P6\n");
965 case APIC_TMCCT: /* Timer CCR */
966 if (apic_lvtt_tscdeadline(apic))
969 val = apic_get_tmcct(apic);
972 apic_update_ppr(apic);
973 val = kvm_apic_get_reg(apic, offset);
976 report_tpr_access(apic, false);
979 val = kvm_apic_get_reg(apic, offset);
986 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
988 return container_of(dev, struct kvm_lapic, dev);
991 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
994 unsigned char alignment = offset & 0xf;
996 /* this bitmask has a bit cleared for each reserved register */
997 static const u64 rmask = 0x43ff01ffffffe70cULL;
999 if ((alignment + len) > 4) {
1000 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1005 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1006 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1011 result = __apic_read(apic, offset & ~0xf);
1013 trace_kvm_apic_read(offset, result);
1019 memcpy(data, (char *)&result + alignment, len);
1022 printk(KERN_ERR "Local APIC read with len = %x, "
1023 "should be 1,2, or 4 instead\n", len);
1029 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1031 return kvm_apic_hw_enabled(apic) &&
1032 addr >= apic->base_address &&
1033 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1036 static int apic_mmio_read(struct kvm_io_device *this,
1037 gpa_t address, int len, void *data)
1039 struct kvm_lapic *apic = to_lapic(this);
1040 u32 offset = address - apic->base_address;
1042 if (!apic_mmio_in_range(apic, address))
1045 apic_reg_read(apic, offset, len, data);
1050 static void update_divide_count(struct kvm_lapic *apic)
1052 u32 tmp1, tmp2, tdcr;
1054 tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1056 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1057 apic->divide_count = 0x1 << (tmp2 & 0x7);
1059 apic_debug("timer divide count is 0x%x\n",
1060 apic->divide_count);
1063 static void apic_timer_expired(struct kvm_lapic *apic)
1065 struct kvm_vcpu *vcpu = apic->vcpu;
1066 wait_queue_head_t *q = &vcpu->wq;
1069 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1072 if (atomic_read(&apic->lapic_timer.pending))
1075 atomic_inc(&apic->lapic_timer.pending);
1076 /* FIXME: this code should not know anything about vcpus */
1077 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1079 if (waitqueue_active(q))
1080 wake_up_interruptible(q);
1083 static void start_apic_timer(struct kvm_lapic *apic)
1086 atomic_set(&apic->lapic_timer.pending, 0);
1088 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1089 /* lapic timer in oneshot or periodic mode */
1090 now = apic->lapic_timer.timer.base->get_time();
1091 apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1092 * APIC_BUS_CYCLE_NS * apic->divide_count;
1094 if (!apic->lapic_timer.period)
1097 * Do not allow the guest to program periodic timers with small
1098 * interval, since the hrtimers are not throttled by the host
1101 if (apic_lvtt_period(apic)) {
1102 s64 min_period = min_timer_period_us * 1000LL;
1104 if (apic->lapic_timer.period < min_period) {
1105 pr_info_ratelimited(
1106 "kvm: vcpu %i: requested %lld ns "
1107 "lapic timer period limited to %lld ns\n",
1108 apic->vcpu->vcpu_id,
1109 apic->lapic_timer.period, min_period);
1110 apic->lapic_timer.period = min_period;
1114 hrtimer_start(&apic->lapic_timer.timer,
1115 ktime_add_ns(now, apic->lapic_timer.period),
1118 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1120 "timer initial count 0x%x, period %lldns, "
1121 "expire @ 0x%016" PRIx64 ".\n", __func__,
1122 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1123 kvm_apic_get_reg(apic, APIC_TMICT),
1124 apic->lapic_timer.period,
1125 ktime_to_ns(ktime_add_ns(now,
1126 apic->lapic_timer.period)));
1127 } else if (apic_lvtt_tscdeadline(apic)) {
1128 /* lapic timer in tsc deadline mode */
1129 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1131 struct kvm_vcpu *vcpu = apic->vcpu;
1132 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1133 unsigned long flags;
1135 if (unlikely(!tscdeadline || !this_tsc_khz))
1138 local_irq_save(flags);
1140 now = apic->lapic_timer.timer.base->get_time();
1141 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1142 if (likely(tscdeadline > guest_tsc)) {
1143 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1144 do_div(ns, this_tsc_khz);
1145 hrtimer_start(&apic->lapic_timer.timer,
1146 ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
1148 apic_timer_expired(apic);
1150 local_irq_restore(flags);
1154 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1156 int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1158 if (apic_lvt_nmi_mode(lvt0_val)) {
1159 if (!nmi_wd_enabled) {
1160 apic_debug("Receive NMI setting on APIC_LVT0 "
1161 "for cpu %d\n", apic->vcpu->vcpu_id);
1162 apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1164 } else if (nmi_wd_enabled)
1165 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1168 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1172 trace_kvm_apic_write(reg, val);
1175 case APIC_ID: /* Local APIC ID */
1176 if (!apic_x2apic_mode(apic))
1177 kvm_apic_set_id(apic, val >> 24);
1183 report_tpr_access(apic, true);
1184 apic_set_tpr(apic, val & 0xff);
1192 if (!apic_x2apic_mode(apic))
1193 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1199 if (!apic_x2apic_mode(apic)) {
1200 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1201 recalculate_apic_map(apic->vcpu->kvm);
1208 if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1209 mask |= APIC_SPIV_DIRECTED_EOI;
1210 apic_set_spiv(apic, val & mask);
1211 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1215 for (i = 0; i < APIC_LVT_NUM; i++) {
1216 lvt_val = kvm_apic_get_reg(apic,
1217 APIC_LVTT + 0x10 * i);
1218 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1219 lvt_val | APIC_LVT_MASKED);
1221 atomic_set(&apic->lapic_timer.pending, 0);
1227 /* No delay here, so we always clear the pending bit */
1228 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1229 apic_send_ipi(apic);
1233 if (!apic_x2apic_mode(apic))
1235 apic_set_reg(apic, APIC_ICR2, val);
1239 apic_manage_nmi_watchdog(apic, val);
1244 /* TODO: Check vector */
1245 if (!kvm_apic_sw_enabled(apic))
1246 val |= APIC_LVT_MASKED;
1248 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1249 apic_set_reg(apic, reg, val);
1254 u32 timer_mode = val & apic->lapic_timer.timer_mode_mask;
1256 if (apic->lapic_timer.timer_mode != timer_mode) {
1257 apic->lapic_timer.timer_mode = timer_mode;
1258 hrtimer_cancel(&apic->lapic_timer.timer);
1261 if (!kvm_apic_sw_enabled(apic))
1262 val |= APIC_LVT_MASKED;
1263 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1264 apic_set_reg(apic, APIC_LVTT, val);
1269 if (apic_lvtt_tscdeadline(apic))
1272 hrtimer_cancel(&apic->lapic_timer.timer);
1273 apic_set_reg(apic, APIC_TMICT, val);
1274 start_apic_timer(apic);
1279 apic_debug("KVM_WRITE:TDCR %x\n", val);
1280 apic_set_reg(apic, APIC_TDCR, val);
1281 update_divide_count(apic);
1285 if (apic_x2apic_mode(apic) && val != 0) {
1286 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1292 if (apic_x2apic_mode(apic)) {
1293 apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1302 apic_debug("Local APIC Write to read-only register %x\n", reg);
1306 static int apic_mmio_write(struct kvm_io_device *this,
1307 gpa_t address, int len, const void *data)
1309 struct kvm_lapic *apic = to_lapic(this);
1310 unsigned int offset = address - apic->base_address;
1313 if (!apic_mmio_in_range(apic, address))
1317 * APIC register must be aligned on 128-bits boundary.
1318 * 32/64/128 bits registers must be accessed thru 32 bits.
1321 if (len != 4 || (offset & 0xf)) {
1322 /* Don't shout loud, $infamous_os would cause only noise. */
1323 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1329 /* too common printing */
1330 if (offset != APIC_EOI)
1331 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1332 "0x%x\n", __func__, offset, len, val);
1334 apic_reg_write(apic, offset & 0xff0, val);
1339 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1341 if (kvm_vcpu_has_lapic(vcpu))
1342 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1344 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1346 /* emulate APIC access in a trap manner */
1347 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1351 /* hw has done the conditional check and inst decode */
1354 apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1356 /* TODO: optimize to just emulate side effect w/o one more write */
1357 apic_reg_write(vcpu->arch.apic, offset, val);
1359 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1361 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1363 struct kvm_lapic *apic = vcpu->arch.apic;
1365 if (!vcpu->arch.apic)
1368 hrtimer_cancel(&apic->lapic_timer.timer);
1370 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1371 static_key_slow_dec_deferred(&apic_hw_disabled);
1373 if (!apic->sw_enabled)
1374 static_key_slow_dec_deferred(&apic_sw_disabled);
1377 free_page((unsigned long)apic->regs);
1383 *----------------------------------------------------------------------
1385 *----------------------------------------------------------------------
1388 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1390 struct kvm_lapic *apic = vcpu->arch.apic;
1392 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1393 apic_lvtt_period(apic))
1396 return apic->lapic_timer.tscdeadline;
1399 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1401 struct kvm_lapic *apic = vcpu->arch.apic;
1403 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1404 apic_lvtt_period(apic))
1407 hrtimer_cancel(&apic->lapic_timer.timer);
1408 apic->lapic_timer.tscdeadline = data;
1409 start_apic_timer(apic);
1412 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1414 struct kvm_lapic *apic = vcpu->arch.apic;
1416 if (!kvm_vcpu_has_lapic(vcpu))
1419 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1420 | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1423 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1427 if (!kvm_vcpu_has_lapic(vcpu))
1430 tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1432 return (tpr & 0xf0) >> 4;
1435 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1437 u64 old_value = vcpu->arch.apic_base;
1438 struct kvm_lapic *apic = vcpu->arch.apic;
1441 value |= MSR_IA32_APICBASE_BSP;
1442 vcpu->arch.apic_base = value;
1446 if (!kvm_vcpu_is_bsp(apic->vcpu))
1447 value &= ~MSR_IA32_APICBASE_BSP;
1448 vcpu->arch.apic_base = value;
1450 /* update jump label if enable bit changes */
1451 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1452 if (value & MSR_IA32_APICBASE_ENABLE)
1453 static_key_slow_dec_deferred(&apic_hw_disabled);
1455 static_key_slow_inc(&apic_hw_disabled.key);
1456 recalculate_apic_map(vcpu->kvm);
1459 if ((old_value ^ value) & X2APIC_ENABLE) {
1460 if (value & X2APIC_ENABLE) {
1461 u32 id = kvm_apic_id(apic);
1462 u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1463 kvm_apic_set_ldr(apic, ldr);
1464 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1466 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1469 apic->base_address = apic->vcpu->arch.apic_base &
1470 MSR_IA32_APICBASE_BASE;
1472 if ((value & MSR_IA32_APICBASE_ENABLE) &&
1473 apic->base_address != APIC_DEFAULT_PHYS_BASE)
1474 pr_warn_once("APIC base relocation is unsupported by KVM");
1476 /* with FSB delivery interrupt, we can restart APIC functionality */
1477 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1478 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1482 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1484 struct kvm_lapic *apic;
1487 apic_debug("%s\n", __func__);
1490 apic = vcpu->arch.apic;
1491 ASSERT(apic != NULL);
1493 /* Stop the timer in case it's a reset to an active apic */
1494 hrtimer_cancel(&apic->lapic_timer.timer);
1496 kvm_apic_set_id(apic, vcpu->vcpu_id);
1497 kvm_apic_set_version(apic->vcpu);
1499 for (i = 0; i < APIC_LVT_NUM; i++)
1500 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1501 apic->lapic_timer.timer_mode = 0;
1502 apic_set_reg(apic, APIC_LVT0,
1503 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1505 apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1506 apic_set_spiv(apic, 0xff);
1507 apic_set_reg(apic, APIC_TASKPRI, 0);
1508 kvm_apic_set_ldr(apic, 0);
1509 apic_set_reg(apic, APIC_ESR, 0);
1510 apic_set_reg(apic, APIC_ICR, 0);
1511 apic_set_reg(apic, APIC_ICR2, 0);
1512 apic_set_reg(apic, APIC_TDCR, 0);
1513 apic_set_reg(apic, APIC_TMICT, 0);
1514 for (i = 0; i < 8; i++) {
1515 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1516 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1517 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1519 apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1520 apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm);
1521 apic->highest_isr_cache = -1;
1522 update_divide_count(apic);
1523 atomic_set(&apic->lapic_timer.pending, 0);
1524 if (kvm_vcpu_is_bsp(vcpu))
1525 kvm_lapic_set_base(vcpu,
1526 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1527 vcpu->arch.pv_eoi.msr_val = 0;
1528 apic_update_ppr(apic);
1530 vcpu->arch.apic_arb_prio = 0;
1531 vcpu->arch.apic_attention = 0;
1533 apic_debug("%s: vcpu=%p, id=%d, base_msr="
1534 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1535 vcpu, kvm_apic_id(apic),
1536 vcpu->arch.apic_base, apic->base_address);
1540 *----------------------------------------------------------------------
1542 *----------------------------------------------------------------------
1545 static bool lapic_is_periodic(struct kvm_lapic *apic)
1547 return apic_lvtt_period(apic);
1550 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1552 struct kvm_lapic *apic = vcpu->arch.apic;
1554 if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1555 apic_lvt_enabled(apic, APIC_LVTT))
1556 return atomic_read(&apic->lapic_timer.pending);
1561 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1563 u32 reg = kvm_apic_get_reg(apic, lvt_type);
1564 int vector, mode, trig_mode;
1566 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1567 vector = reg & APIC_VECTOR_MASK;
1568 mode = reg & APIC_MODE_MASK;
1569 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1570 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1576 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1578 struct kvm_lapic *apic = vcpu->arch.apic;
1581 kvm_apic_local_deliver(apic, APIC_LVT0);
1584 static const struct kvm_io_device_ops apic_mmio_ops = {
1585 .read = apic_mmio_read,
1586 .write = apic_mmio_write,
1589 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1591 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1592 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1594 apic_timer_expired(apic);
1596 if (lapic_is_periodic(apic)) {
1597 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1598 return HRTIMER_RESTART;
1600 return HRTIMER_NORESTART;
1603 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1605 struct kvm_lapic *apic;
1607 ASSERT(vcpu != NULL);
1608 apic_debug("apic_init %d\n", vcpu->vcpu_id);
1610 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1614 vcpu->arch.apic = apic;
1616 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1618 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1620 goto nomem_free_apic;
1624 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1626 apic->lapic_timer.timer.function = apic_timer_fn;
1629 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1630 * thinking that APIC satet has changed.
1632 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1633 kvm_lapic_set_base(vcpu,
1634 APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1636 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1637 kvm_lapic_reset(vcpu);
1638 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1647 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1649 struct kvm_lapic *apic = vcpu->arch.apic;
1652 if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1655 apic_update_ppr(apic);
1656 highest_irr = apic_find_highest_irr(apic);
1657 if ((highest_irr == -1) ||
1658 ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1663 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1665 u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1668 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1670 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1671 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1676 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1678 struct kvm_lapic *apic = vcpu->arch.apic;
1680 if (!kvm_vcpu_has_lapic(vcpu))
1683 if (atomic_read(&apic->lapic_timer.pending) > 0) {
1684 kvm_apic_local_deliver(apic, APIC_LVTT);
1685 if (apic_lvtt_tscdeadline(apic))
1686 apic->lapic_timer.tscdeadline = 0;
1687 atomic_set(&apic->lapic_timer.pending, 0);
1691 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1693 int vector = kvm_apic_has_interrupt(vcpu);
1694 struct kvm_lapic *apic = vcpu->arch.apic;
1700 * We get here even with APIC virtualization enabled, if doing
1701 * nested virtualization and L1 runs with the "acknowledge interrupt
1702 * on exit" mode. Then we cannot inject the interrupt via RVI,
1703 * because the process would deliver it through the IDT.
1706 apic_set_isr(vector, apic);
1707 apic_update_ppr(apic);
1708 apic_clear_irr(vector, apic);
1712 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1713 struct kvm_lapic_state *s)
1715 struct kvm_lapic *apic = vcpu->arch.apic;
1717 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1718 /* set SPIV separately to get count of SW disabled APICs right */
1719 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1720 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1721 /* call kvm_apic_set_id() to put apic into apic_map */
1722 kvm_apic_set_id(apic, kvm_apic_id(apic));
1723 kvm_apic_set_version(vcpu);
1725 apic_update_ppr(apic);
1726 hrtimer_cancel(&apic->lapic_timer.timer);
1727 update_divide_count(apic);
1728 start_apic_timer(apic);
1729 apic->irr_pending = true;
1730 apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm) ?
1731 1 : count_vectors(apic->regs + APIC_ISR);
1732 apic->highest_isr_cache = -1;
1733 if (kvm_x86_ops->hwapic_irr_update)
1734 kvm_x86_ops->hwapic_irr_update(vcpu,
1735 apic_find_highest_irr(apic));
1736 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, apic_find_highest_isr(apic));
1737 kvm_make_request(KVM_REQ_EVENT, vcpu);
1738 kvm_rtc_eoi_tracking_restore_one(vcpu);
1741 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1743 struct hrtimer *timer;
1745 if (!kvm_vcpu_has_lapic(vcpu))
1748 timer = &vcpu->arch.apic->lapic_timer.timer;
1749 if (hrtimer_cancel(timer))
1750 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1754 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1756 * Detect whether guest triggered PV EOI since the
1757 * last entry. If yes, set EOI on guests's behalf.
1758 * Clear PV EOI in guest memory in any case.
1760 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1761 struct kvm_lapic *apic)
1766 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1767 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1769 * KVM_APIC_PV_EOI_PENDING is unset:
1770 * -> host disabled PV EOI.
1771 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1772 * -> host enabled PV EOI, guest did not execute EOI yet.
1773 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1774 * -> host enabled PV EOI, guest executed EOI.
1776 BUG_ON(!pv_eoi_enabled(vcpu));
1777 pending = pv_eoi_get_pending(vcpu);
1779 * Clear pending bit in any case: it will be set again on vmentry.
1780 * While this might not be ideal from performance point of view,
1781 * this makes sure pv eoi is only enabled when we know it's safe.
1783 pv_eoi_clr_pending(vcpu);
1786 vector = apic_set_eoi(apic);
1787 trace_kvm_pv_eoi(apic, vector);
1790 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1794 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1795 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1797 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1800 kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1803 apic_set_tpr(vcpu->arch.apic, data & 0xff);
1807 * apic_sync_pv_eoi_to_guest - called before vmentry
1809 * Detect whether it's safe to enable PV EOI and
1812 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1813 struct kvm_lapic *apic)
1815 if (!pv_eoi_enabled(vcpu) ||
1816 /* IRR set or many bits in ISR: could be nested. */
1817 apic->irr_pending ||
1818 /* Cache not set: could be safe but we don't bother. */
1819 apic->highest_isr_cache == -1 ||
1820 /* Need EOI to update ioapic. */
1821 kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1823 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1824 * so we need not do anything here.
1829 pv_eoi_set_pending(apic->vcpu);
1832 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1835 int max_irr, max_isr;
1836 struct kvm_lapic *apic = vcpu->arch.apic;
1838 apic_sync_pv_eoi_to_guest(vcpu, apic);
1840 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1843 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1844 max_irr = apic_find_highest_irr(apic);
1847 max_isr = apic_find_highest_isr(apic);
1850 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1852 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1856 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1859 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1860 &vcpu->arch.apic->vapic_cache,
1861 vapic_addr, sizeof(u32)))
1863 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1865 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1868 vcpu->arch.apic->vapic_addr = vapic_addr;
1872 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1874 struct kvm_lapic *apic = vcpu->arch.apic;
1875 u32 reg = (msr - APIC_BASE_MSR) << 4;
1877 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1880 /* if this is ICR write vector before command */
1882 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1883 return apic_reg_write(apic, reg, (u32)data);
1886 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1888 struct kvm_lapic *apic = vcpu->arch.apic;
1889 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1891 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1894 if (apic_reg_read(apic, reg, 4, &low))
1897 apic_reg_read(apic, APIC_ICR2, 4, &high);
1899 *data = (((u64)high) << 32) | low;
1904 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1906 struct kvm_lapic *apic = vcpu->arch.apic;
1908 if (!kvm_vcpu_has_lapic(vcpu))
1911 /* if this is ICR write vector before command */
1912 if (reg == APIC_ICR)
1913 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1914 return apic_reg_write(apic, reg, (u32)data);
1917 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1919 struct kvm_lapic *apic = vcpu->arch.apic;
1922 if (!kvm_vcpu_has_lapic(vcpu))
1925 if (apic_reg_read(apic, reg, 4, &low))
1927 if (reg == APIC_ICR)
1928 apic_reg_read(apic, APIC_ICR2, 4, &high);
1930 *data = (((u64)high) << 32) | low;
1935 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
1937 u64 addr = data & ~KVM_MSR_ENABLED;
1938 if (!IS_ALIGNED(addr, 4))
1941 vcpu->arch.pv_eoi.msr_val = data;
1942 if (!pv_eoi_enabled(vcpu))
1944 return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
1948 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
1950 struct kvm_lapic *apic = vcpu->arch.apic;
1951 unsigned int sipi_vector;
1954 if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
1957 pe = xchg(&apic->pending_events, 0);
1959 if (test_bit(KVM_APIC_INIT, &pe)) {
1960 kvm_lapic_reset(vcpu);
1961 kvm_vcpu_reset(vcpu);
1962 if (kvm_vcpu_is_bsp(apic->vcpu))
1963 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1965 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
1967 if (test_bit(KVM_APIC_SIPI, &pe) &&
1968 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
1969 /* evaluate pending_events before reading the vector */
1971 sipi_vector = apic->sipi_vector;
1972 apic_debug("vcpu %d received sipi with vector # %x\n",
1973 vcpu->vcpu_id, sipi_vector);
1974 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
1975 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1979 void kvm_lapic_init(void)
1981 /* do not patch jump label more than once per second */
1982 jump_label_rate_limit(&apic_hw_disabled, HZ);
1983 jump_label_rate_limit(&apic_sw_disabled, HZ);