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 static void recalculate_apic_map(struct kvm *kvm)
137 struct kvm_apic_map *new, *old = NULL;
138 struct kvm_vcpu *vcpu;
141 new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
143 mutex_lock(&kvm->arch.apic_map_lock);
149 /* flat mode is default */
152 new->lid_mask = 0xff;
153 new->broadcast = APIC_BROADCAST;
155 kvm_for_each_vcpu(i, vcpu, kvm) {
156 struct kvm_lapic *apic = vcpu->arch.apic;
158 if (!kvm_apic_present(vcpu))
161 if (apic_x2apic_mode(apic)) {
164 new->cid_mask = new->lid_mask = 0xffff;
165 new->broadcast = X2APIC_BROADCAST;
166 } else if (kvm_apic_get_reg(apic, APIC_LDR)) {
167 if (kvm_apic_get_reg(apic, APIC_DFR) ==
175 new->lid_mask = 0xff;
180 * All APICs have to be configured in the same mode by an OS.
181 * We take advatage of this while building logical id loockup
182 * table. After reset APICs are in software disabled mode, so if
183 * we find apic with different setting we assume this is the mode
184 * OS wants all apics to be in; build lookup table accordingly.
186 if (kvm_apic_sw_enabled(apic))
190 kvm_for_each_vcpu(i, vcpu, kvm) {
191 struct kvm_lapic *apic = vcpu->arch.apic;
195 aid = kvm_apic_id(apic);
196 ldr = kvm_apic_get_reg(apic, APIC_LDR);
197 cid = apic_cluster_id(new, ldr);
198 lid = apic_logical_id(new, ldr);
200 if (aid < ARRAY_SIZE(new->phys_map))
201 new->phys_map[aid] = apic;
202 if (lid && cid < ARRAY_SIZE(new->logical_map))
203 new->logical_map[cid][ffs(lid) - 1] = apic;
206 old = rcu_dereference_protected(kvm->arch.apic_map,
207 lockdep_is_held(&kvm->arch.apic_map_lock));
208 rcu_assign_pointer(kvm->arch.apic_map, new);
209 mutex_unlock(&kvm->arch.apic_map_lock);
214 kvm_vcpu_request_scan_ioapic(kvm);
217 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
219 bool enabled = val & APIC_SPIV_APIC_ENABLED;
221 apic_set_reg(apic, APIC_SPIV, val);
223 if (enabled != apic->sw_enabled) {
224 apic->sw_enabled = enabled;
226 static_key_slow_dec_deferred(&apic_sw_disabled);
227 recalculate_apic_map(apic->vcpu->kvm);
229 static_key_slow_inc(&apic_sw_disabled.key);
233 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
235 apic_set_reg(apic, APIC_ID, id << 24);
236 recalculate_apic_map(apic->vcpu->kvm);
239 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
241 apic_set_reg(apic, APIC_LDR, id);
242 recalculate_apic_map(apic->vcpu->kvm);
245 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
247 return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
250 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
252 return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
255 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
257 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
260 static inline int apic_lvtt_period(struct kvm_lapic *apic)
262 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
265 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
267 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
270 static inline int apic_lvt_nmi_mode(u32 lvt_val)
272 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
275 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
277 struct kvm_lapic *apic = vcpu->arch.apic;
278 struct kvm_cpuid_entry2 *feat;
279 u32 v = APIC_VERSION;
281 if (!kvm_vcpu_has_lapic(vcpu))
284 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
285 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
286 v |= APIC_LVR_DIRECTED_EOI;
287 apic_set_reg(apic, APIC_LVR, v);
290 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
291 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
292 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
293 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
294 LINT_MASK, LINT_MASK, /* LVT0-1 */
295 LVT_MASK /* LVTERR */
298 static int find_highest_vector(void *bitmap)
303 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
304 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
305 reg = bitmap + REG_POS(vec);
307 return fls(*reg) - 1 + vec;
313 static u8 count_vectors(void *bitmap)
319 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
320 reg = bitmap + REG_POS(vec);
321 count += hweight32(*reg);
327 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
330 struct kvm_lapic *apic = vcpu->arch.apic;
332 for (i = 0; i <= 7; i++) {
333 pir_val = xchg(&pir[i], 0);
335 *((u32 *)(apic->regs + APIC_IRR + i * 0x10)) |= pir_val;
338 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
340 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
342 apic_set_vector(vec, apic->regs + APIC_IRR);
344 * irr_pending must be true if any interrupt is pending; set it after
345 * APIC_IRR to avoid race with apic_clear_irr
347 apic->irr_pending = true;
350 static inline int apic_search_irr(struct kvm_lapic *apic)
352 return find_highest_vector(apic->regs + APIC_IRR);
355 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
360 * Note that irr_pending is just a hint. It will be always
361 * true with virtual interrupt delivery enabled.
363 if (!apic->irr_pending)
366 kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
367 result = apic_search_irr(apic);
368 ASSERT(result == -1 || result >= 16);
373 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
375 struct kvm_vcpu *vcpu;
379 if (unlikely(kvm_apic_vid_enabled(vcpu->kvm))) {
380 /* try to update RVI */
381 apic_clear_vector(vec, apic->regs + APIC_IRR);
382 kvm_make_request(KVM_REQ_EVENT, vcpu);
384 apic->irr_pending = false;
385 apic_clear_vector(vec, apic->regs + APIC_IRR);
386 if (apic_search_irr(apic) != -1)
387 apic->irr_pending = true;
391 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
393 struct kvm_vcpu *vcpu;
395 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
401 * With APIC virtualization enabled, all caching is disabled
402 * because the processor can modify ISR under the hood. Instead
405 if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
406 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
409 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
411 * ISR (in service register) bit is set when injecting an interrupt.
412 * The highest vector is injected. Thus the latest bit set matches
413 * the highest bit in ISR.
415 apic->highest_isr_cache = vec;
419 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
424 * Note that isr_count is always 1, and highest_isr_cache
425 * is always -1, with APIC virtualization enabled.
427 if (!apic->isr_count)
429 if (likely(apic->highest_isr_cache != -1))
430 return apic->highest_isr_cache;
432 result = find_highest_vector(apic->regs + APIC_ISR);
433 ASSERT(result == -1 || result >= 16);
438 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
440 struct kvm_vcpu *vcpu;
441 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
447 * We do get here for APIC virtualization enabled if the guest
448 * uses the Hyper-V APIC enlightenment. In this case we may need
449 * to trigger a new interrupt delivery by writing the SVI field;
450 * on the other hand isr_count and highest_isr_cache are unused
451 * and must be left alone.
453 if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
454 kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
455 apic_find_highest_isr(apic));
458 BUG_ON(apic->isr_count < 0);
459 apic->highest_isr_cache = -1;
463 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
467 /* This may race with setting of irr in __apic_accept_irq() and
468 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
469 * will cause vmexit immediately and the value will be recalculated
470 * on the next vmentry.
472 if (!kvm_vcpu_has_lapic(vcpu))
474 highest_irr = apic_find_highest_irr(vcpu->arch.apic);
479 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
480 int vector, int level, int trig_mode,
481 unsigned long *dest_map);
483 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
484 unsigned long *dest_map)
486 struct kvm_lapic *apic = vcpu->arch.apic;
488 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
489 irq->level, irq->trig_mode, dest_map);
492 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
495 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
499 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
502 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
506 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
508 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
511 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
514 if (pv_eoi_get_user(vcpu, &val) < 0)
515 apic_debug("Can't read EOI MSR value: 0x%llx\n",
516 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
520 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
522 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
523 apic_debug("Can't set EOI MSR value: 0x%llx\n",
524 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
527 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
530 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
532 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
533 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
534 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
537 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
540 void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
542 struct kvm_lapic *apic = vcpu->arch.apic;
545 for (i = 0; i < 8; i++)
546 apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
549 static void apic_update_ppr(struct kvm_lapic *apic)
551 u32 tpr, isrv, ppr, old_ppr;
554 old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
555 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
556 isr = apic_find_highest_isr(apic);
557 isrv = (isr != -1) ? isr : 0;
559 if ((tpr & 0xf0) >= (isrv & 0xf0))
564 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
565 apic, ppr, isr, isrv);
567 if (old_ppr != ppr) {
568 apic_set_reg(apic, APIC_PROCPRI, ppr);
570 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
574 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
576 apic_set_reg(apic, APIC_TASKPRI, tpr);
577 apic_update_ppr(apic);
580 static int kvm_apic_broadcast(struct kvm_lapic *apic, u32 dest)
582 return dest == (apic_x2apic_mode(apic) ?
583 X2APIC_BROADCAST : APIC_BROADCAST);
586 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 dest)
588 return kvm_apic_id(apic) == dest || kvm_apic_broadcast(apic, dest);
591 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
596 if (kvm_apic_broadcast(apic, mda))
599 if (apic_x2apic_mode(apic)) {
600 logical_id = kvm_apic_get_reg(apic, APIC_LDR);
601 return logical_id & mda;
604 logical_id = GET_APIC_LOGICAL_ID(kvm_apic_get_reg(apic, APIC_LDR));
606 switch (kvm_apic_get_reg(apic, APIC_DFR)) {
608 if (logical_id & mda)
611 case APIC_DFR_CLUSTER:
612 if (((logical_id >> 4) == (mda >> 0x4))
613 && (logical_id & mda & 0xf))
617 apic_debug("Bad DFR vcpu %d: %08x\n",
618 apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
625 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
626 int short_hand, unsigned int dest, int dest_mode)
629 struct kvm_lapic *target = vcpu->arch.apic;
631 apic_debug("target %p, source %p, dest 0x%x, "
632 "dest_mode 0x%x, short_hand 0x%x\n",
633 target, source, dest, dest_mode, short_hand);
636 switch (short_hand) {
637 case APIC_DEST_NOSHORT:
640 result = kvm_apic_match_physical_addr(target, dest);
643 result = kvm_apic_match_logical_addr(target, dest);
646 result = (target == source);
648 case APIC_DEST_ALLINC:
651 case APIC_DEST_ALLBUT:
652 result = (target != source);
655 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
663 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
664 struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
666 struct kvm_apic_map *map;
667 unsigned long bitmap = 1;
668 struct kvm_lapic **dst;
674 if (irq->shorthand == APIC_DEST_SELF) {
675 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
683 map = rcu_dereference(kvm->arch.apic_map);
688 if (irq->dest_id == map->broadcast)
693 if (irq->dest_mode == 0) { /* physical mode */
694 if (irq->dest_id >= ARRAY_SIZE(map->phys_map))
697 dst = &map->phys_map[irq->dest_id];
699 u32 mda = irq->dest_id << (32 - map->ldr_bits);
700 u16 cid = apic_cluster_id(map, mda);
702 if (cid >= ARRAY_SIZE(map->logical_map))
705 dst = map->logical_map[cid];
707 bitmap = apic_logical_id(map, mda);
709 if (irq->delivery_mode == APIC_DM_LOWEST) {
711 for_each_set_bit(i, &bitmap, 16) {
716 else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
720 bitmap = (l >= 0) ? 1 << l : 0;
724 for_each_set_bit(i, &bitmap, 16) {
729 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
737 * Add a pending IRQ into lapic.
738 * Return 1 if successfully added and 0 if discarded.
740 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
741 int vector, int level, int trig_mode,
742 unsigned long *dest_map)
745 struct kvm_vcpu *vcpu = apic->vcpu;
747 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
749 switch (delivery_mode) {
751 vcpu->arch.apic_arb_prio++;
753 /* FIXME add logic for vcpu on reset */
754 if (unlikely(!apic_enabled(apic)))
760 __set_bit(vcpu->vcpu_id, dest_map);
762 if (kvm_x86_ops->deliver_posted_interrupt)
763 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
765 apic_set_irr(vector, apic);
767 kvm_make_request(KVM_REQ_EVENT, vcpu);
774 vcpu->arch.pv.pv_unhalted = 1;
775 kvm_make_request(KVM_REQ_EVENT, vcpu);
780 apic_debug("Ignoring guest SMI\n");
785 kvm_inject_nmi(vcpu);
790 if (!trig_mode || level) {
792 /* assumes that there are only KVM_APIC_INIT/SIPI */
793 apic->pending_events = (1UL << KVM_APIC_INIT);
794 /* make sure pending_events is visible before sending
797 kvm_make_request(KVM_REQ_EVENT, vcpu);
800 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
805 case APIC_DM_STARTUP:
806 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
807 vcpu->vcpu_id, vector);
809 apic->sipi_vector = vector;
810 /* make sure sipi_vector is visible for the receiver */
812 set_bit(KVM_APIC_SIPI, &apic->pending_events);
813 kvm_make_request(KVM_REQ_EVENT, vcpu);
819 * Should only be called by kvm_apic_local_deliver() with LVT0,
820 * before NMI watchdog was enabled. Already handled by
821 * kvm_apic_accept_pic_intr().
826 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
833 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
835 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
838 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
840 if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
841 kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
843 if (apic_test_vector(vector, apic->regs + APIC_TMR))
844 trigger_mode = IOAPIC_LEVEL_TRIG;
846 trigger_mode = IOAPIC_EDGE_TRIG;
847 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
851 static int apic_set_eoi(struct kvm_lapic *apic)
853 int vector = apic_find_highest_isr(apic);
855 trace_kvm_eoi(apic, vector);
858 * Not every write EOI will has corresponding ISR,
859 * one example is when Kernel check timer on setup_IO_APIC
864 apic_clear_isr(vector, apic);
865 apic_update_ppr(apic);
867 kvm_ioapic_send_eoi(apic, vector);
868 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
873 * this interface assumes a trap-like exit, which has already finished
874 * desired side effect including vISR and vPPR update.
876 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
878 struct kvm_lapic *apic = vcpu->arch.apic;
880 trace_kvm_eoi(apic, vector);
882 kvm_ioapic_send_eoi(apic, vector);
883 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
885 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
887 static void apic_send_ipi(struct kvm_lapic *apic)
889 u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
890 u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
891 struct kvm_lapic_irq irq;
893 irq.vector = icr_low & APIC_VECTOR_MASK;
894 irq.delivery_mode = icr_low & APIC_MODE_MASK;
895 irq.dest_mode = icr_low & APIC_DEST_MASK;
896 irq.level = icr_low & APIC_INT_ASSERT;
897 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
898 irq.shorthand = icr_low & APIC_SHORT_MASK;
899 if (apic_x2apic_mode(apic))
900 irq.dest_id = icr_high;
902 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
904 trace_kvm_apic_ipi(icr_low, irq.dest_id);
906 apic_debug("icr_high 0x%x, icr_low 0x%x, "
907 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
908 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
909 icr_high, icr_low, irq.shorthand, irq.dest_id,
910 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
913 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
916 static u32 apic_get_tmcct(struct kvm_lapic *apic)
922 ASSERT(apic != NULL);
924 /* if initial count is 0, current count should also be 0 */
925 if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
926 apic->lapic_timer.period == 0)
929 remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
930 if (ktime_to_ns(remaining) < 0)
931 remaining = ktime_set(0, 0);
933 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
934 tmcct = div64_u64(ns,
935 (APIC_BUS_CYCLE_NS * apic->divide_count));
940 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
942 struct kvm_vcpu *vcpu = apic->vcpu;
943 struct kvm_run *run = vcpu->run;
945 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
946 run->tpr_access.rip = kvm_rip_read(vcpu);
947 run->tpr_access.is_write = write;
950 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
952 if (apic->vcpu->arch.tpr_access_reporting)
953 __report_tpr_access(apic, write);
956 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
960 if (offset >= LAPIC_MMIO_LENGTH)
965 if (apic_x2apic_mode(apic))
966 val = kvm_apic_id(apic);
968 val = kvm_apic_id(apic) << 24;
971 apic_debug("Access APIC ARBPRI register which is for P6\n");
974 case APIC_TMCCT: /* Timer CCR */
975 if (apic_lvtt_tscdeadline(apic))
978 val = apic_get_tmcct(apic);
981 apic_update_ppr(apic);
982 val = kvm_apic_get_reg(apic, offset);
985 report_tpr_access(apic, false);
988 val = kvm_apic_get_reg(apic, offset);
995 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
997 return container_of(dev, struct kvm_lapic, dev);
1000 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1003 unsigned char alignment = offset & 0xf;
1005 /* this bitmask has a bit cleared for each reserved register */
1006 static const u64 rmask = 0x43ff01ffffffe70cULL;
1008 if ((alignment + len) > 4) {
1009 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1014 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1015 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1020 result = __apic_read(apic, offset & ~0xf);
1022 trace_kvm_apic_read(offset, result);
1028 memcpy(data, (char *)&result + alignment, len);
1031 printk(KERN_ERR "Local APIC read with len = %x, "
1032 "should be 1,2, or 4 instead\n", len);
1038 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1040 return kvm_apic_hw_enabled(apic) &&
1041 addr >= apic->base_address &&
1042 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1045 static int apic_mmio_read(struct kvm_io_device *this,
1046 gpa_t address, int len, void *data)
1048 struct kvm_lapic *apic = to_lapic(this);
1049 u32 offset = address - apic->base_address;
1051 if (!apic_mmio_in_range(apic, address))
1054 apic_reg_read(apic, offset, len, data);
1059 static void update_divide_count(struct kvm_lapic *apic)
1061 u32 tmp1, tmp2, tdcr;
1063 tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1065 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1066 apic->divide_count = 0x1 << (tmp2 & 0x7);
1068 apic_debug("timer divide count is 0x%x\n",
1069 apic->divide_count);
1072 static void apic_timer_expired(struct kvm_lapic *apic)
1074 struct kvm_vcpu *vcpu = apic->vcpu;
1075 wait_queue_head_t *q = &vcpu->wq;
1078 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1081 if (atomic_read(&apic->lapic_timer.pending))
1084 atomic_inc(&apic->lapic_timer.pending);
1085 /* FIXME: this code should not know anything about vcpus */
1086 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1088 if (waitqueue_active(q))
1089 wake_up_interruptible(q);
1092 static void start_apic_timer(struct kvm_lapic *apic)
1095 atomic_set(&apic->lapic_timer.pending, 0);
1097 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1098 /* lapic timer in oneshot or periodic mode */
1099 now = apic->lapic_timer.timer.base->get_time();
1100 apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1101 * APIC_BUS_CYCLE_NS * apic->divide_count;
1103 if (!apic->lapic_timer.period)
1106 * Do not allow the guest to program periodic timers with small
1107 * interval, since the hrtimers are not throttled by the host
1110 if (apic_lvtt_period(apic)) {
1111 s64 min_period = min_timer_period_us * 1000LL;
1113 if (apic->lapic_timer.period < min_period) {
1114 pr_info_ratelimited(
1115 "kvm: vcpu %i: requested %lld ns "
1116 "lapic timer period limited to %lld ns\n",
1117 apic->vcpu->vcpu_id,
1118 apic->lapic_timer.period, min_period);
1119 apic->lapic_timer.period = min_period;
1123 hrtimer_start(&apic->lapic_timer.timer,
1124 ktime_add_ns(now, apic->lapic_timer.period),
1127 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1129 "timer initial count 0x%x, period %lldns, "
1130 "expire @ 0x%016" PRIx64 ".\n", __func__,
1131 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1132 kvm_apic_get_reg(apic, APIC_TMICT),
1133 apic->lapic_timer.period,
1134 ktime_to_ns(ktime_add_ns(now,
1135 apic->lapic_timer.period)));
1136 } else if (apic_lvtt_tscdeadline(apic)) {
1137 /* lapic timer in tsc deadline mode */
1138 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1140 struct kvm_vcpu *vcpu = apic->vcpu;
1141 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1142 unsigned long flags;
1144 if (unlikely(!tscdeadline || !this_tsc_khz))
1147 local_irq_save(flags);
1149 now = apic->lapic_timer.timer.base->get_time();
1150 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1151 if (likely(tscdeadline > guest_tsc)) {
1152 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1153 do_div(ns, this_tsc_khz);
1154 hrtimer_start(&apic->lapic_timer.timer,
1155 ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
1157 apic_timer_expired(apic);
1159 local_irq_restore(flags);
1163 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1165 int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1167 if (apic_lvt_nmi_mode(lvt0_val)) {
1168 if (!nmi_wd_enabled) {
1169 apic_debug("Receive NMI setting on APIC_LVT0 "
1170 "for cpu %d\n", apic->vcpu->vcpu_id);
1171 apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1173 } else if (nmi_wd_enabled)
1174 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1177 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1181 trace_kvm_apic_write(reg, val);
1184 case APIC_ID: /* Local APIC ID */
1185 if (!apic_x2apic_mode(apic))
1186 kvm_apic_set_id(apic, val >> 24);
1192 report_tpr_access(apic, true);
1193 apic_set_tpr(apic, val & 0xff);
1201 if (!apic_x2apic_mode(apic))
1202 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1208 if (!apic_x2apic_mode(apic)) {
1209 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1210 recalculate_apic_map(apic->vcpu->kvm);
1217 if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1218 mask |= APIC_SPIV_DIRECTED_EOI;
1219 apic_set_spiv(apic, val & mask);
1220 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1224 for (i = 0; i < APIC_LVT_NUM; i++) {
1225 lvt_val = kvm_apic_get_reg(apic,
1226 APIC_LVTT + 0x10 * i);
1227 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1228 lvt_val | APIC_LVT_MASKED);
1230 atomic_set(&apic->lapic_timer.pending, 0);
1236 /* No delay here, so we always clear the pending bit */
1237 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1238 apic_send_ipi(apic);
1242 if (!apic_x2apic_mode(apic))
1244 apic_set_reg(apic, APIC_ICR2, val);
1248 apic_manage_nmi_watchdog(apic, val);
1253 /* TODO: Check vector */
1254 if (!kvm_apic_sw_enabled(apic))
1255 val |= APIC_LVT_MASKED;
1257 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1258 apic_set_reg(apic, reg, val);
1263 u32 timer_mode = val & apic->lapic_timer.timer_mode_mask;
1265 if (apic->lapic_timer.timer_mode != timer_mode) {
1266 apic->lapic_timer.timer_mode = timer_mode;
1267 hrtimer_cancel(&apic->lapic_timer.timer);
1270 if (!kvm_apic_sw_enabled(apic))
1271 val |= APIC_LVT_MASKED;
1272 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1273 apic_set_reg(apic, APIC_LVTT, val);
1278 if (apic_lvtt_tscdeadline(apic))
1281 hrtimer_cancel(&apic->lapic_timer.timer);
1282 apic_set_reg(apic, APIC_TMICT, val);
1283 start_apic_timer(apic);
1288 apic_debug("KVM_WRITE:TDCR %x\n", val);
1289 apic_set_reg(apic, APIC_TDCR, val);
1290 update_divide_count(apic);
1294 if (apic_x2apic_mode(apic) && val != 0) {
1295 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1301 if (apic_x2apic_mode(apic)) {
1302 apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1311 apic_debug("Local APIC Write to read-only register %x\n", reg);
1315 static int apic_mmio_write(struct kvm_io_device *this,
1316 gpa_t address, int len, const void *data)
1318 struct kvm_lapic *apic = to_lapic(this);
1319 unsigned int offset = address - apic->base_address;
1322 if (!apic_mmio_in_range(apic, address))
1326 * APIC register must be aligned on 128-bits boundary.
1327 * 32/64/128 bits registers must be accessed thru 32 bits.
1330 if (len != 4 || (offset & 0xf)) {
1331 /* Don't shout loud, $infamous_os would cause only noise. */
1332 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1338 /* too common printing */
1339 if (offset != APIC_EOI)
1340 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1341 "0x%x\n", __func__, offset, len, val);
1343 apic_reg_write(apic, offset & 0xff0, val);
1348 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1350 if (kvm_vcpu_has_lapic(vcpu))
1351 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1353 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1355 /* emulate APIC access in a trap manner */
1356 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1360 /* hw has done the conditional check and inst decode */
1363 apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1365 /* TODO: optimize to just emulate side effect w/o one more write */
1366 apic_reg_write(vcpu->arch.apic, offset, val);
1368 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1370 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1372 struct kvm_lapic *apic = vcpu->arch.apic;
1374 if (!vcpu->arch.apic)
1377 hrtimer_cancel(&apic->lapic_timer.timer);
1379 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1380 static_key_slow_dec_deferred(&apic_hw_disabled);
1382 if (!apic->sw_enabled)
1383 static_key_slow_dec_deferred(&apic_sw_disabled);
1386 free_page((unsigned long)apic->regs);
1392 *----------------------------------------------------------------------
1394 *----------------------------------------------------------------------
1397 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1399 struct kvm_lapic *apic = vcpu->arch.apic;
1401 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1402 apic_lvtt_period(apic))
1405 return apic->lapic_timer.tscdeadline;
1408 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1410 struct kvm_lapic *apic = vcpu->arch.apic;
1412 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1413 apic_lvtt_period(apic))
1416 hrtimer_cancel(&apic->lapic_timer.timer);
1417 apic->lapic_timer.tscdeadline = data;
1418 start_apic_timer(apic);
1421 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1423 struct kvm_lapic *apic = vcpu->arch.apic;
1425 if (!kvm_vcpu_has_lapic(vcpu))
1428 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1429 | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1432 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1436 if (!kvm_vcpu_has_lapic(vcpu))
1439 tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1441 return (tpr & 0xf0) >> 4;
1444 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1446 u64 old_value = vcpu->arch.apic_base;
1447 struct kvm_lapic *apic = vcpu->arch.apic;
1450 value |= MSR_IA32_APICBASE_BSP;
1451 vcpu->arch.apic_base = value;
1455 if (!kvm_vcpu_is_bsp(apic->vcpu))
1456 value &= ~MSR_IA32_APICBASE_BSP;
1457 vcpu->arch.apic_base = value;
1459 /* update jump label if enable bit changes */
1460 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1461 if (value & MSR_IA32_APICBASE_ENABLE)
1462 static_key_slow_dec_deferred(&apic_hw_disabled);
1464 static_key_slow_inc(&apic_hw_disabled.key);
1465 recalculate_apic_map(vcpu->kvm);
1468 if ((old_value ^ value) & X2APIC_ENABLE) {
1469 if (value & X2APIC_ENABLE) {
1470 u32 id = kvm_apic_id(apic);
1471 u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1472 kvm_apic_set_ldr(apic, ldr);
1473 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1475 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1478 apic->base_address = apic->vcpu->arch.apic_base &
1479 MSR_IA32_APICBASE_BASE;
1481 if ((value & MSR_IA32_APICBASE_ENABLE) &&
1482 apic->base_address != APIC_DEFAULT_PHYS_BASE)
1483 pr_warn_once("APIC base relocation is unsupported by KVM");
1485 /* with FSB delivery interrupt, we can restart APIC functionality */
1486 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1487 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1491 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1493 struct kvm_lapic *apic;
1496 apic_debug("%s\n", __func__);
1499 apic = vcpu->arch.apic;
1500 ASSERT(apic != NULL);
1502 /* Stop the timer in case it's a reset to an active apic */
1503 hrtimer_cancel(&apic->lapic_timer.timer);
1505 kvm_apic_set_id(apic, vcpu->vcpu_id);
1506 kvm_apic_set_version(apic->vcpu);
1508 for (i = 0; i < APIC_LVT_NUM; i++)
1509 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1510 apic->lapic_timer.timer_mode = 0;
1511 apic_set_reg(apic, APIC_LVT0,
1512 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1514 apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1515 apic_set_spiv(apic, 0xff);
1516 apic_set_reg(apic, APIC_TASKPRI, 0);
1517 kvm_apic_set_ldr(apic, 0);
1518 apic_set_reg(apic, APIC_ESR, 0);
1519 apic_set_reg(apic, APIC_ICR, 0);
1520 apic_set_reg(apic, APIC_ICR2, 0);
1521 apic_set_reg(apic, APIC_TDCR, 0);
1522 apic_set_reg(apic, APIC_TMICT, 0);
1523 for (i = 0; i < 8; i++) {
1524 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1525 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1526 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1528 apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1529 apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm);
1530 apic->highest_isr_cache = -1;
1531 update_divide_count(apic);
1532 atomic_set(&apic->lapic_timer.pending, 0);
1533 if (kvm_vcpu_is_bsp(vcpu))
1534 kvm_lapic_set_base(vcpu,
1535 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1536 vcpu->arch.pv_eoi.msr_val = 0;
1537 apic_update_ppr(apic);
1539 vcpu->arch.apic_arb_prio = 0;
1540 vcpu->arch.apic_attention = 0;
1542 apic_debug("%s: vcpu=%p, id=%d, base_msr="
1543 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1544 vcpu, kvm_apic_id(apic),
1545 vcpu->arch.apic_base, apic->base_address);
1549 *----------------------------------------------------------------------
1551 *----------------------------------------------------------------------
1554 static bool lapic_is_periodic(struct kvm_lapic *apic)
1556 return apic_lvtt_period(apic);
1559 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1561 struct kvm_lapic *apic = vcpu->arch.apic;
1563 if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1564 apic_lvt_enabled(apic, APIC_LVTT))
1565 return atomic_read(&apic->lapic_timer.pending);
1570 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1572 u32 reg = kvm_apic_get_reg(apic, lvt_type);
1573 int vector, mode, trig_mode;
1575 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1576 vector = reg & APIC_VECTOR_MASK;
1577 mode = reg & APIC_MODE_MASK;
1578 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1579 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1585 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1587 struct kvm_lapic *apic = vcpu->arch.apic;
1590 kvm_apic_local_deliver(apic, APIC_LVT0);
1593 static const struct kvm_io_device_ops apic_mmio_ops = {
1594 .read = apic_mmio_read,
1595 .write = apic_mmio_write,
1598 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1600 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1601 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1603 apic_timer_expired(apic);
1605 if (lapic_is_periodic(apic)) {
1606 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1607 return HRTIMER_RESTART;
1609 return HRTIMER_NORESTART;
1612 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1614 struct kvm_lapic *apic;
1616 ASSERT(vcpu != NULL);
1617 apic_debug("apic_init %d\n", vcpu->vcpu_id);
1619 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1623 vcpu->arch.apic = apic;
1625 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1627 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1629 goto nomem_free_apic;
1633 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1635 apic->lapic_timer.timer.function = apic_timer_fn;
1638 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1639 * thinking that APIC satet has changed.
1641 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1642 kvm_lapic_set_base(vcpu,
1643 APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1645 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1646 kvm_lapic_reset(vcpu);
1647 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1656 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1658 struct kvm_lapic *apic = vcpu->arch.apic;
1661 if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1664 apic_update_ppr(apic);
1665 highest_irr = apic_find_highest_irr(apic);
1666 if ((highest_irr == -1) ||
1667 ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1672 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1674 u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1677 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1679 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1680 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1685 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1687 struct kvm_lapic *apic = vcpu->arch.apic;
1689 if (!kvm_vcpu_has_lapic(vcpu))
1692 if (atomic_read(&apic->lapic_timer.pending) > 0) {
1693 kvm_apic_local_deliver(apic, APIC_LVTT);
1694 if (apic_lvtt_tscdeadline(apic))
1695 apic->lapic_timer.tscdeadline = 0;
1696 atomic_set(&apic->lapic_timer.pending, 0);
1700 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1702 int vector = kvm_apic_has_interrupt(vcpu);
1703 struct kvm_lapic *apic = vcpu->arch.apic;
1709 * We get here even with APIC virtualization enabled, if doing
1710 * nested virtualization and L1 runs with the "acknowledge interrupt
1711 * on exit" mode. Then we cannot inject the interrupt via RVI,
1712 * because the process would deliver it through the IDT.
1715 apic_set_isr(vector, apic);
1716 apic_update_ppr(apic);
1717 apic_clear_irr(vector, apic);
1721 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1722 struct kvm_lapic_state *s)
1724 struct kvm_lapic *apic = vcpu->arch.apic;
1726 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1727 /* set SPIV separately to get count of SW disabled APICs right */
1728 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1729 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1730 /* call kvm_apic_set_id() to put apic into apic_map */
1731 kvm_apic_set_id(apic, kvm_apic_id(apic));
1732 kvm_apic_set_version(vcpu);
1734 apic_update_ppr(apic);
1735 hrtimer_cancel(&apic->lapic_timer.timer);
1736 update_divide_count(apic);
1737 start_apic_timer(apic);
1738 apic->irr_pending = true;
1739 apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm) ?
1740 1 : count_vectors(apic->regs + APIC_ISR);
1741 apic->highest_isr_cache = -1;
1742 if (kvm_x86_ops->hwapic_irr_update)
1743 kvm_x86_ops->hwapic_irr_update(vcpu,
1744 apic_find_highest_irr(apic));
1745 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, apic_find_highest_isr(apic));
1746 kvm_make_request(KVM_REQ_EVENT, vcpu);
1747 kvm_rtc_eoi_tracking_restore_one(vcpu);
1750 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1752 struct hrtimer *timer;
1754 if (!kvm_vcpu_has_lapic(vcpu))
1757 timer = &vcpu->arch.apic->lapic_timer.timer;
1758 if (hrtimer_cancel(timer))
1759 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1763 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1765 * Detect whether guest triggered PV EOI since the
1766 * last entry. If yes, set EOI on guests's behalf.
1767 * Clear PV EOI in guest memory in any case.
1769 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1770 struct kvm_lapic *apic)
1775 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1776 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1778 * KVM_APIC_PV_EOI_PENDING is unset:
1779 * -> host disabled PV EOI.
1780 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1781 * -> host enabled PV EOI, guest did not execute EOI yet.
1782 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1783 * -> host enabled PV EOI, guest executed EOI.
1785 BUG_ON(!pv_eoi_enabled(vcpu));
1786 pending = pv_eoi_get_pending(vcpu);
1788 * Clear pending bit in any case: it will be set again on vmentry.
1789 * While this might not be ideal from performance point of view,
1790 * this makes sure pv eoi is only enabled when we know it's safe.
1792 pv_eoi_clr_pending(vcpu);
1795 vector = apic_set_eoi(apic);
1796 trace_kvm_pv_eoi(apic, vector);
1799 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1803 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1804 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1806 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1809 kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1812 apic_set_tpr(vcpu->arch.apic, data & 0xff);
1816 * apic_sync_pv_eoi_to_guest - called before vmentry
1818 * Detect whether it's safe to enable PV EOI and
1821 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1822 struct kvm_lapic *apic)
1824 if (!pv_eoi_enabled(vcpu) ||
1825 /* IRR set or many bits in ISR: could be nested. */
1826 apic->irr_pending ||
1827 /* Cache not set: could be safe but we don't bother. */
1828 apic->highest_isr_cache == -1 ||
1829 /* Need EOI to update ioapic. */
1830 kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1832 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1833 * so we need not do anything here.
1838 pv_eoi_set_pending(apic->vcpu);
1841 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1844 int max_irr, max_isr;
1845 struct kvm_lapic *apic = vcpu->arch.apic;
1847 apic_sync_pv_eoi_to_guest(vcpu, apic);
1849 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1852 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1853 max_irr = apic_find_highest_irr(apic);
1856 max_isr = apic_find_highest_isr(apic);
1859 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1861 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1865 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1868 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1869 &vcpu->arch.apic->vapic_cache,
1870 vapic_addr, sizeof(u32)))
1872 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1874 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1877 vcpu->arch.apic->vapic_addr = vapic_addr;
1881 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1883 struct kvm_lapic *apic = vcpu->arch.apic;
1884 u32 reg = (msr - APIC_BASE_MSR) << 4;
1886 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1889 if (reg == APIC_ICR2)
1892 /* if this is ICR write vector before command */
1893 if (reg == APIC_ICR)
1894 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1895 return apic_reg_write(apic, reg, (u32)data);
1898 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1900 struct kvm_lapic *apic = vcpu->arch.apic;
1901 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1903 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1906 if (reg == APIC_DFR || reg == APIC_ICR2) {
1907 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
1912 if (apic_reg_read(apic, reg, 4, &low))
1914 if (reg == APIC_ICR)
1915 apic_reg_read(apic, APIC_ICR2, 4, &high);
1917 *data = (((u64)high) << 32) | low;
1922 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1924 struct kvm_lapic *apic = vcpu->arch.apic;
1926 if (!kvm_vcpu_has_lapic(vcpu))
1929 /* if this is ICR write vector before command */
1930 if (reg == APIC_ICR)
1931 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1932 return apic_reg_write(apic, reg, (u32)data);
1935 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1937 struct kvm_lapic *apic = vcpu->arch.apic;
1940 if (!kvm_vcpu_has_lapic(vcpu))
1943 if (apic_reg_read(apic, reg, 4, &low))
1945 if (reg == APIC_ICR)
1946 apic_reg_read(apic, APIC_ICR2, 4, &high);
1948 *data = (((u64)high) << 32) | low;
1953 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
1955 u64 addr = data & ~KVM_MSR_ENABLED;
1956 if (!IS_ALIGNED(addr, 4))
1959 vcpu->arch.pv_eoi.msr_val = data;
1960 if (!pv_eoi_enabled(vcpu))
1962 return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
1966 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
1968 struct kvm_lapic *apic = vcpu->arch.apic;
1972 if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
1975 pe = xchg(&apic->pending_events, 0);
1977 if (test_bit(KVM_APIC_INIT, &pe)) {
1978 kvm_lapic_reset(vcpu);
1979 kvm_vcpu_reset(vcpu);
1980 if (kvm_vcpu_is_bsp(apic->vcpu))
1981 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1983 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
1985 if (test_bit(KVM_APIC_SIPI, &pe) &&
1986 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
1987 /* evaluate pending_events before reading the vector */
1989 sipi_vector = apic->sipi_vector;
1990 apic_debug("vcpu %d received sipi with vector # %x\n",
1991 vcpu->vcpu_id, sipi_vector);
1992 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
1993 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1997 void kvm_lapic_init(void)
1999 /* do not patch jump label more than once per second */
2000 jump_label_rate_limit(&apic_hw_disabled, HZ);
2001 jump_label_rate_limit(&apic_sw_disabled, HZ);