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 VEC_POS(v) ((v) & (32 - 1))
72 #define REG_POS(v) (((v) >> 5) << 4)
74 static unsigned int min_timer_period_us = 500;
75 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
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_and_set_vector(int vec, void *bitmap)
84 return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
87 static inline int apic_test_and_clear_vector(int vec, void *bitmap)
89 return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
92 static inline int apic_test_vector(int vec, void *bitmap)
94 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
97 static inline void apic_set_vector(int vec, void *bitmap)
99 set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
102 static inline void apic_clear_vector(int vec, void *bitmap)
104 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
107 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
109 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
112 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
114 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
117 struct static_key_deferred apic_hw_disabled __read_mostly;
118 struct static_key_deferred apic_sw_disabled __read_mostly;
120 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
122 if ((kvm_apic_get_reg(apic, APIC_SPIV) ^ val) & APIC_SPIV_APIC_ENABLED) {
123 if (val & APIC_SPIV_APIC_ENABLED)
124 static_key_slow_dec_deferred(&apic_sw_disabled);
126 static_key_slow_inc(&apic_sw_disabled.key);
128 apic_set_reg(apic, APIC_SPIV, val);
131 static inline int apic_enabled(struct kvm_lapic *apic)
133 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
137 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
140 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
141 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
143 static inline int kvm_apic_id(struct kvm_lapic *apic)
145 return (kvm_apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
148 static void recalculate_apic_map(struct kvm *kvm)
150 struct kvm_apic_map *new, *old = NULL;
151 struct kvm_vcpu *vcpu;
154 new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
156 mutex_lock(&kvm->arch.apic_map_lock);
162 /* flat mode is default */
165 new->lid_mask = 0xff;
167 kvm_for_each_vcpu(i, vcpu, kvm) {
168 struct kvm_lapic *apic = vcpu->arch.apic;
172 if (!kvm_apic_present(vcpu))
176 * All APICs have to be configured in the same mode by an OS.
177 * We take advatage of this while building logical id loockup
178 * table. After reset APICs are in xapic/flat mode, so if we
179 * find apic with different setting we assume this is the mode
180 * OS wants all apics to be in; build lookup table accordingly.
182 if (apic_x2apic_mode(apic)) {
185 new->cid_mask = new->lid_mask = 0xffff;
186 } else if (kvm_apic_sw_enabled(apic) &&
187 !new->cid_mask /* flat mode */ &&
188 kvm_apic_get_reg(apic, APIC_DFR) == APIC_DFR_CLUSTER) {
194 new->phys_map[kvm_apic_id(apic)] = apic;
196 ldr = kvm_apic_get_reg(apic, APIC_LDR);
197 cid = apic_cluster_id(new, ldr);
198 lid = apic_logical_id(new, ldr);
201 new->logical_map[cid][ffs(lid) - 1] = apic;
204 old = rcu_dereference_protected(kvm->arch.apic_map,
205 lockdep_is_held(&kvm->arch.apic_map_lock));
206 rcu_assign_pointer(kvm->arch.apic_map, new);
207 mutex_unlock(&kvm->arch.apic_map_lock);
212 kvm_ioapic_make_eoibitmap_request(kvm);
215 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
217 apic_set_reg(apic, APIC_ID, id << 24);
218 recalculate_apic_map(apic->vcpu->kvm);
221 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
223 apic_set_reg(apic, APIC_LDR, id);
224 recalculate_apic_map(apic->vcpu->kvm);
227 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
229 return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
232 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
234 return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
237 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
239 return ((kvm_apic_get_reg(apic, APIC_LVTT) &
240 apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_ONESHOT);
243 static inline int apic_lvtt_period(struct kvm_lapic *apic)
245 return ((kvm_apic_get_reg(apic, APIC_LVTT) &
246 apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_PERIODIC);
249 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
251 return ((kvm_apic_get_reg(apic, APIC_LVTT) &
252 apic->lapic_timer.timer_mode_mask) ==
253 APIC_LVT_TIMER_TSCDEADLINE);
256 static inline int apic_lvt_nmi_mode(u32 lvt_val)
258 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
261 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
263 struct kvm_lapic *apic = vcpu->arch.apic;
264 struct kvm_cpuid_entry2 *feat;
265 u32 v = APIC_VERSION;
267 if (!kvm_vcpu_has_lapic(vcpu))
270 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
271 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
272 v |= APIC_LVR_DIRECTED_EOI;
273 apic_set_reg(apic, APIC_LVR, v);
276 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
277 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
278 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
279 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
280 LINT_MASK, LINT_MASK, /* LVT0-1 */
281 LVT_MASK /* LVTERR */
284 static int find_highest_vector(void *bitmap)
289 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
290 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
291 reg = bitmap + REG_POS(vec);
293 return fls(*reg) - 1 + vec;
299 static u8 count_vectors(void *bitmap)
305 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
306 reg = bitmap + REG_POS(vec);
307 count += hweight32(*reg);
313 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
315 apic->irr_pending = true;
316 return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
319 static inline int apic_search_irr(struct kvm_lapic *apic)
321 return find_highest_vector(apic->regs + APIC_IRR);
324 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
329 * Note that irr_pending is just a hint. It will be always
330 * true with virtual interrupt delivery enabled.
332 if (!apic->irr_pending)
335 result = apic_search_irr(apic);
336 ASSERT(result == -1 || result >= 16);
341 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
343 apic->irr_pending = false;
344 apic_clear_vector(vec, apic->regs + APIC_IRR);
345 if (apic_search_irr(apic) != -1)
346 apic->irr_pending = true;
349 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
351 if (!__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
353 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
355 * ISR (in service register) bit is set when injecting an interrupt.
356 * The highest vector is injected. Thus the latest bit set matches
357 * the highest bit in ISR.
359 apic->highest_isr_cache = vec;
362 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
364 if (__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
366 BUG_ON(apic->isr_count < 0);
367 apic->highest_isr_cache = -1;
370 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
374 /* This may race with setting of irr in __apic_accept_irq() and
375 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
376 * will cause vmexit immediately and the value will be recalculated
377 * on the next vmentry.
379 if (!kvm_vcpu_has_lapic(vcpu))
381 highest_irr = apic_find_highest_irr(vcpu->arch.apic);
386 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
387 int vector, int level, int trig_mode);
389 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
391 struct kvm_lapic *apic = vcpu->arch.apic;
393 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
394 irq->level, irq->trig_mode);
397 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
400 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
404 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
407 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
411 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
413 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
416 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
419 if (pv_eoi_get_user(vcpu, &val) < 0)
420 apic_debug("Can't read EOI MSR value: 0x%llx\n",
421 (unsigned long long)vcpi->arch.pv_eoi.msr_val);
425 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
427 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
428 apic_debug("Can't set EOI MSR value: 0x%llx\n",
429 (unsigned long long)vcpi->arch.pv_eoi.msr_val);
432 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
435 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
437 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
438 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
439 (unsigned long long)vcpi->arch.pv_eoi.msr_val);
442 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
445 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
449 /* Note that isr_count is always 1 with vid enabled */
450 if (!apic->isr_count)
452 if (likely(apic->highest_isr_cache != -1))
453 return apic->highest_isr_cache;
455 result = find_highest_vector(apic->regs + APIC_ISR);
456 ASSERT(result == -1 || result >= 16);
461 static void apic_update_ppr(struct kvm_lapic *apic)
463 u32 tpr, isrv, ppr, old_ppr;
466 old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
467 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
468 isr = apic_find_highest_isr(apic);
469 isrv = (isr != -1) ? isr : 0;
471 if ((tpr & 0xf0) >= (isrv & 0xf0))
476 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
477 apic, ppr, isr, isrv);
479 if (old_ppr != ppr) {
480 apic_set_reg(apic, APIC_PROCPRI, ppr);
482 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
486 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
488 apic_set_reg(apic, APIC_TASKPRI, tpr);
489 apic_update_ppr(apic);
492 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
494 return dest == 0xff || kvm_apic_id(apic) == dest;
497 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
502 if (apic_x2apic_mode(apic)) {
503 logical_id = kvm_apic_get_reg(apic, APIC_LDR);
504 return logical_id & mda;
507 logical_id = GET_APIC_LOGICAL_ID(kvm_apic_get_reg(apic, APIC_LDR));
509 switch (kvm_apic_get_reg(apic, APIC_DFR)) {
511 if (logical_id & mda)
514 case APIC_DFR_CLUSTER:
515 if (((logical_id >> 4) == (mda >> 0x4))
516 && (logical_id & mda & 0xf))
520 apic_debug("Bad DFR vcpu %d: %08x\n",
521 apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
528 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
529 int short_hand, int dest, int dest_mode)
532 struct kvm_lapic *target = vcpu->arch.apic;
534 apic_debug("target %p, source %p, dest 0x%x, "
535 "dest_mode 0x%x, short_hand 0x%x\n",
536 target, source, dest, dest_mode, short_hand);
539 switch (short_hand) {
540 case APIC_DEST_NOSHORT:
543 result = kvm_apic_match_physical_addr(target, dest);
546 result = kvm_apic_match_logical_addr(target, dest);
549 result = (target == source);
551 case APIC_DEST_ALLINC:
554 case APIC_DEST_ALLBUT:
555 result = (target != source);
558 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
566 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
567 struct kvm_lapic_irq *irq, int *r)
569 struct kvm_apic_map *map;
570 unsigned long bitmap = 1;
571 struct kvm_lapic **dst;
577 if (irq->shorthand == APIC_DEST_SELF) {
578 *r = kvm_apic_set_irq(src->vcpu, irq);
586 map = rcu_dereference(kvm->arch.apic_map);
591 if (irq->dest_mode == 0) { /* physical mode */
592 if (irq->delivery_mode == APIC_DM_LOWEST ||
593 irq->dest_id == 0xff)
595 dst = &map->phys_map[irq->dest_id & 0xff];
597 u32 mda = irq->dest_id << (32 - map->ldr_bits);
599 dst = map->logical_map[apic_cluster_id(map, mda)];
601 bitmap = apic_logical_id(map, mda);
603 if (irq->delivery_mode == APIC_DM_LOWEST) {
605 for_each_set_bit(i, &bitmap, 16) {
610 else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
614 bitmap = (l >= 0) ? 1 << l : 0;
618 for_each_set_bit(i, &bitmap, 16) {
623 *r += kvm_apic_set_irq(dst[i]->vcpu, irq);
633 * Add a pending IRQ into lapic.
634 * Return 1 if successfully added and 0 if discarded.
636 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
637 int vector, int level, int trig_mode)
640 struct kvm_vcpu *vcpu = apic->vcpu;
642 switch (delivery_mode) {
644 vcpu->arch.apic_arb_prio++;
646 /* FIXME add logic for vcpu on reset */
647 if (unlikely(!apic_enabled(apic)))
651 apic_debug("level trig mode for vector %d", vector);
652 apic_set_vector(vector, apic->regs + APIC_TMR);
654 apic_clear_vector(vector, apic->regs + APIC_TMR);
656 result = !apic_test_and_set_irr(vector, apic);
657 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
658 trig_mode, vector, !result);
661 apic_debug("level trig mode repeatedly for "
662 "vector %d", vector);
666 kvm_make_request(KVM_REQ_EVENT, vcpu);
671 apic_debug("Ignoring delivery mode 3\n");
675 apic_debug("Ignoring guest SMI\n");
680 kvm_inject_nmi(vcpu);
685 if (!trig_mode || level) {
687 /* assumes that there are only KVM_APIC_INIT/SIPI */
688 apic->pending_events = (1UL << KVM_APIC_INIT);
689 /* make sure pending_events is visible before sending
692 kvm_make_request(KVM_REQ_EVENT, vcpu);
695 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
700 case APIC_DM_STARTUP:
701 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
702 vcpu->vcpu_id, vector);
704 apic->sipi_vector = vector;
705 /* make sure sipi_vector is visible for the receiver */
707 set_bit(KVM_APIC_SIPI, &apic->pending_events);
708 kvm_make_request(KVM_REQ_EVENT, vcpu);
714 * Should only be called by kvm_apic_local_deliver() with LVT0,
715 * before NMI watchdog was enabled. Already handled by
716 * kvm_apic_accept_pic_intr().
721 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
728 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
730 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
733 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
735 if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
736 kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
738 if (apic_test_vector(vector, apic->regs + APIC_TMR))
739 trigger_mode = IOAPIC_LEVEL_TRIG;
741 trigger_mode = IOAPIC_EDGE_TRIG;
742 kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
746 static int apic_set_eoi(struct kvm_lapic *apic)
748 int vector = apic_find_highest_isr(apic);
750 trace_kvm_eoi(apic, vector);
753 * Not every write EOI will has corresponding ISR,
754 * one example is when Kernel check timer on setup_IO_APIC
759 apic_clear_isr(vector, apic);
760 apic_update_ppr(apic);
762 kvm_ioapic_send_eoi(apic, vector);
763 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
768 * this interface assumes a trap-like exit, which has already finished
769 * desired side effect including vISR and vPPR update.
771 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
773 struct kvm_lapic *apic = vcpu->arch.apic;
775 trace_kvm_eoi(apic, vector);
777 kvm_ioapic_send_eoi(apic, vector);
778 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
780 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
782 static void apic_send_ipi(struct kvm_lapic *apic)
784 u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
785 u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
786 struct kvm_lapic_irq irq;
788 irq.vector = icr_low & APIC_VECTOR_MASK;
789 irq.delivery_mode = icr_low & APIC_MODE_MASK;
790 irq.dest_mode = icr_low & APIC_DEST_MASK;
791 irq.level = icr_low & APIC_INT_ASSERT;
792 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
793 irq.shorthand = icr_low & APIC_SHORT_MASK;
794 if (apic_x2apic_mode(apic))
795 irq.dest_id = icr_high;
797 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
799 trace_kvm_apic_ipi(icr_low, irq.dest_id);
801 apic_debug("icr_high 0x%x, icr_low 0x%x, "
802 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
803 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
804 icr_high, icr_low, irq.shorthand, irq.dest_id,
805 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
808 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
811 static u32 apic_get_tmcct(struct kvm_lapic *apic)
817 ASSERT(apic != NULL);
819 /* if initial count is 0, current count should also be 0 */
820 if (kvm_apic_get_reg(apic, APIC_TMICT) == 0)
823 remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
824 if (ktime_to_ns(remaining) < 0)
825 remaining = ktime_set(0, 0);
827 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
828 tmcct = div64_u64(ns,
829 (APIC_BUS_CYCLE_NS * apic->divide_count));
834 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
836 struct kvm_vcpu *vcpu = apic->vcpu;
837 struct kvm_run *run = vcpu->run;
839 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
840 run->tpr_access.rip = kvm_rip_read(vcpu);
841 run->tpr_access.is_write = write;
844 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
846 if (apic->vcpu->arch.tpr_access_reporting)
847 __report_tpr_access(apic, write);
850 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
854 if (offset >= LAPIC_MMIO_LENGTH)
859 if (apic_x2apic_mode(apic))
860 val = kvm_apic_id(apic);
862 val = kvm_apic_id(apic) << 24;
865 apic_debug("Access APIC ARBPRI register which is for P6\n");
868 case APIC_TMCCT: /* Timer CCR */
869 if (apic_lvtt_tscdeadline(apic))
872 val = apic_get_tmcct(apic);
875 apic_update_ppr(apic);
876 val = kvm_apic_get_reg(apic, offset);
879 report_tpr_access(apic, false);
882 val = kvm_apic_get_reg(apic, offset);
889 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
891 return container_of(dev, struct kvm_lapic, dev);
894 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
897 unsigned char alignment = offset & 0xf;
899 /* this bitmask has a bit cleared for each reserved register */
900 static const u64 rmask = 0x43ff01ffffffe70cULL;
902 if ((alignment + len) > 4) {
903 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
908 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
909 apic_debug("KVM_APIC_READ: read reserved register %x\n",
914 result = __apic_read(apic, offset & ~0xf);
916 trace_kvm_apic_read(offset, result);
922 memcpy(data, (char *)&result + alignment, len);
925 printk(KERN_ERR "Local APIC read with len = %x, "
926 "should be 1,2, or 4 instead\n", len);
932 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
934 return kvm_apic_hw_enabled(apic) &&
935 addr >= apic->base_address &&
936 addr < apic->base_address + LAPIC_MMIO_LENGTH;
939 static int apic_mmio_read(struct kvm_io_device *this,
940 gpa_t address, int len, void *data)
942 struct kvm_lapic *apic = to_lapic(this);
943 u32 offset = address - apic->base_address;
945 if (!apic_mmio_in_range(apic, address))
948 apic_reg_read(apic, offset, len, data);
953 static void update_divide_count(struct kvm_lapic *apic)
955 u32 tmp1, tmp2, tdcr;
957 tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
959 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
960 apic->divide_count = 0x1 << (tmp2 & 0x7);
962 apic_debug("timer divide count is 0x%x\n",
966 static void start_apic_timer(struct kvm_lapic *apic)
969 atomic_set(&apic->lapic_timer.pending, 0);
971 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
972 /* lapic timer in oneshot or periodic mode */
973 now = apic->lapic_timer.timer.base->get_time();
974 apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
975 * APIC_BUS_CYCLE_NS * apic->divide_count;
977 if (!apic->lapic_timer.period)
980 * Do not allow the guest to program periodic timers with small
981 * interval, since the hrtimers are not throttled by the host
984 if (apic_lvtt_period(apic)) {
985 s64 min_period = min_timer_period_us * 1000LL;
987 if (apic->lapic_timer.period < min_period) {
989 "kvm: vcpu %i: requested %lld ns "
990 "lapic timer period limited to %lld ns\n",
992 apic->lapic_timer.period, min_period);
993 apic->lapic_timer.period = min_period;
997 hrtimer_start(&apic->lapic_timer.timer,
998 ktime_add_ns(now, apic->lapic_timer.period),
1001 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1003 "timer initial count 0x%x, period %lldns, "
1004 "expire @ 0x%016" PRIx64 ".\n", __func__,
1005 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1006 kvm_apic_get_reg(apic, APIC_TMICT),
1007 apic->lapic_timer.period,
1008 ktime_to_ns(ktime_add_ns(now,
1009 apic->lapic_timer.period)));
1010 } else if (apic_lvtt_tscdeadline(apic)) {
1011 /* lapic timer in tsc deadline mode */
1012 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1014 struct kvm_vcpu *vcpu = apic->vcpu;
1015 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1016 unsigned long flags;
1018 if (unlikely(!tscdeadline || !this_tsc_khz))
1021 local_irq_save(flags);
1023 now = apic->lapic_timer.timer.base->get_time();
1024 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1025 if (likely(tscdeadline > guest_tsc)) {
1026 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1027 do_div(ns, this_tsc_khz);
1029 hrtimer_start(&apic->lapic_timer.timer,
1030 ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
1032 local_irq_restore(flags);
1036 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1038 int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1040 if (apic_lvt_nmi_mode(lvt0_val)) {
1041 if (!nmi_wd_enabled) {
1042 apic_debug("Receive NMI setting on APIC_LVT0 "
1043 "for cpu %d\n", apic->vcpu->vcpu_id);
1044 apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1046 } else if (nmi_wd_enabled)
1047 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1050 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1054 trace_kvm_apic_write(reg, val);
1057 case APIC_ID: /* Local APIC ID */
1058 if (!apic_x2apic_mode(apic))
1059 kvm_apic_set_id(apic, val >> 24);
1065 report_tpr_access(apic, true);
1066 apic_set_tpr(apic, val & 0xff);
1074 if (!apic_x2apic_mode(apic))
1075 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1081 if (!apic_x2apic_mode(apic)) {
1082 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1083 recalculate_apic_map(apic->vcpu->kvm);
1090 if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1091 mask |= APIC_SPIV_DIRECTED_EOI;
1092 apic_set_spiv(apic, val & mask);
1093 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1097 for (i = 0; i < APIC_LVT_NUM; i++) {
1098 lvt_val = kvm_apic_get_reg(apic,
1099 APIC_LVTT + 0x10 * i);
1100 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1101 lvt_val | APIC_LVT_MASKED);
1103 atomic_set(&apic->lapic_timer.pending, 0);
1109 /* No delay here, so we always clear the pending bit */
1110 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1111 apic_send_ipi(apic);
1115 if (!apic_x2apic_mode(apic))
1117 apic_set_reg(apic, APIC_ICR2, val);
1121 apic_manage_nmi_watchdog(apic, val);
1126 /* TODO: Check vector */
1127 if (!kvm_apic_sw_enabled(apic))
1128 val |= APIC_LVT_MASKED;
1130 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1131 apic_set_reg(apic, reg, val);
1136 if ((kvm_apic_get_reg(apic, APIC_LVTT) &
1137 apic->lapic_timer.timer_mode_mask) !=
1138 (val & apic->lapic_timer.timer_mode_mask))
1139 hrtimer_cancel(&apic->lapic_timer.timer);
1141 if (!kvm_apic_sw_enabled(apic))
1142 val |= APIC_LVT_MASKED;
1143 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1144 apic_set_reg(apic, APIC_LVTT, val);
1148 if (apic_lvtt_tscdeadline(apic))
1151 hrtimer_cancel(&apic->lapic_timer.timer);
1152 apic_set_reg(apic, APIC_TMICT, val);
1153 start_apic_timer(apic);
1158 apic_debug("KVM_WRITE:TDCR %x\n", val);
1159 apic_set_reg(apic, APIC_TDCR, val);
1160 update_divide_count(apic);
1164 if (apic_x2apic_mode(apic) && val != 0) {
1165 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1171 if (apic_x2apic_mode(apic)) {
1172 apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1181 apic_debug("Local APIC Write to read-only register %x\n", reg);
1185 static int apic_mmio_write(struct kvm_io_device *this,
1186 gpa_t address, int len, const void *data)
1188 struct kvm_lapic *apic = to_lapic(this);
1189 unsigned int offset = address - apic->base_address;
1192 if (!apic_mmio_in_range(apic, address))
1196 * APIC register must be aligned on 128-bits boundary.
1197 * 32/64/128 bits registers must be accessed thru 32 bits.
1200 if (len != 4 || (offset & 0xf)) {
1201 /* Don't shout loud, $infamous_os would cause only noise. */
1202 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1208 /* too common printing */
1209 if (offset != APIC_EOI)
1210 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1211 "0x%x\n", __func__, offset, len, val);
1213 apic_reg_write(apic, offset & 0xff0, val);
1218 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1220 if (kvm_vcpu_has_lapic(vcpu))
1221 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1223 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1225 /* emulate APIC access in a trap manner */
1226 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1230 /* hw has done the conditional check and inst decode */
1233 apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1235 /* TODO: optimize to just emulate side effect w/o one more write */
1236 apic_reg_write(vcpu->arch.apic, offset, val);
1238 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1240 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1242 struct kvm_lapic *apic = vcpu->arch.apic;
1244 if (!vcpu->arch.apic)
1247 hrtimer_cancel(&apic->lapic_timer.timer);
1249 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1250 static_key_slow_dec_deferred(&apic_hw_disabled);
1252 if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED))
1253 static_key_slow_dec_deferred(&apic_sw_disabled);
1256 free_page((unsigned long)apic->regs);
1262 *----------------------------------------------------------------------
1264 *----------------------------------------------------------------------
1267 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1269 struct kvm_lapic *apic = vcpu->arch.apic;
1271 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1272 apic_lvtt_period(apic))
1275 return apic->lapic_timer.tscdeadline;
1278 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1280 struct kvm_lapic *apic = vcpu->arch.apic;
1282 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1283 apic_lvtt_period(apic))
1286 hrtimer_cancel(&apic->lapic_timer.timer);
1287 apic->lapic_timer.tscdeadline = data;
1288 start_apic_timer(apic);
1291 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1293 struct kvm_lapic *apic = vcpu->arch.apic;
1295 if (!kvm_vcpu_has_lapic(vcpu))
1298 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1299 | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1302 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1306 if (!kvm_vcpu_has_lapic(vcpu))
1309 tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1311 return (tpr & 0xf0) >> 4;
1314 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1316 u64 old_value = vcpu->arch.apic_base;
1317 struct kvm_lapic *apic = vcpu->arch.apic;
1320 value |= MSR_IA32_APICBASE_BSP;
1321 vcpu->arch.apic_base = value;
1325 /* update jump label if enable bit changes */
1326 if ((vcpu->arch.apic_base ^ value) & MSR_IA32_APICBASE_ENABLE) {
1327 if (value & MSR_IA32_APICBASE_ENABLE)
1328 static_key_slow_dec_deferred(&apic_hw_disabled);
1330 static_key_slow_inc(&apic_hw_disabled.key);
1331 recalculate_apic_map(vcpu->kvm);
1334 if (!kvm_vcpu_is_bsp(apic->vcpu))
1335 value &= ~MSR_IA32_APICBASE_BSP;
1337 vcpu->arch.apic_base = value;
1338 if ((old_value ^ value) & X2APIC_ENABLE) {
1339 if (value & X2APIC_ENABLE) {
1340 u32 id = kvm_apic_id(apic);
1341 u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1342 kvm_apic_set_ldr(apic, ldr);
1343 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1345 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1348 apic->base_address = apic->vcpu->arch.apic_base &
1349 MSR_IA32_APICBASE_BASE;
1351 /* with FSB delivery interrupt, we can restart APIC functionality */
1352 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1353 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1357 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1359 struct kvm_lapic *apic;
1362 apic_debug("%s\n", __func__);
1365 apic = vcpu->arch.apic;
1366 ASSERT(apic != NULL);
1368 /* Stop the timer in case it's a reset to an active apic */
1369 hrtimer_cancel(&apic->lapic_timer.timer);
1371 kvm_apic_set_id(apic, vcpu->vcpu_id);
1372 kvm_apic_set_version(apic->vcpu);
1374 for (i = 0; i < APIC_LVT_NUM; i++)
1375 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1376 apic_set_reg(apic, APIC_LVT0,
1377 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1379 apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1380 apic_set_spiv(apic, 0xff);
1381 apic_set_reg(apic, APIC_TASKPRI, 0);
1382 kvm_apic_set_ldr(apic, 0);
1383 apic_set_reg(apic, APIC_ESR, 0);
1384 apic_set_reg(apic, APIC_ICR, 0);
1385 apic_set_reg(apic, APIC_ICR2, 0);
1386 apic_set_reg(apic, APIC_TDCR, 0);
1387 apic_set_reg(apic, APIC_TMICT, 0);
1388 for (i = 0; i < 8; i++) {
1389 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1390 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1391 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1393 apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1394 apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm);
1395 apic->highest_isr_cache = -1;
1396 update_divide_count(apic);
1397 atomic_set(&apic->lapic_timer.pending, 0);
1398 if (kvm_vcpu_is_bsp(vcpu))
1399 kvm_lapic_set_base(vcpu,
1400 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1401 vcpu->arch.pv_eoi.msr_val = 0;
1402 apic_update_ppr(apic);
1404 vcpu->arch.apic_arb_prio = 0;
1405 vcpu->arch.apic_attention = 0;
1407 apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
1408 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1409 vcpu, kvm_apic_id(apic),
1410 vcpu->arch.apic_base, apic->base_address);
1414 *----------------------------------------------------------------------
1416 *----------------------------------------------------------------------
1419 static bool lapic_is_periodic(struct kvm_lapic *apic)
1421 return apic_lvtt_period(apic);
1424 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1426 struct kvm_lapic *apic = vcpu->arch.apic;
1428 if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1429 apic_lvt_enabled(apic, APIC_LVTT))
1430 return atomic_read(&apic->lapic_timer.pending);
1435 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1437 u32 reg = kvm_apic_get_reg(apic, lvt_type);
1438 int vector, mode, trig_mode;
1440 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1441 vector = reg & APIC_VECTOR_MASK;
1442 mode = reg & APIC_MODE_MASK;
1443 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1444 return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
1449 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1451 struct kvm_lapic *apic = vcpu->arch.apic;
1454 kvm_apic_local_deliver(apic, APIC_LVT0);
1457 static const struct kvm_io_device_ops apic_mmio_ops = {
1458 .read = apic_mmio_read,
1459 .write = apic_mmio_write,
1462 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1464 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1465 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1466 struct kvm_vcpu *vcpu = apic->vcpu;
1467 wait_queue_head_t *q = &vcpu->wq;
1470 * There is a race window between reading and incrementing, but we do
1471 * not care about potentially losing timer events in the !reinject
1472 * case anyway. Note: KVM_REQ_PENDING_TIMER is implicitly checked
1473 * in vcpu_enter_guest.
1475 if (!atomic_read(&ktimer->pending)) {
1476 atomic_inc(&ktimer->pending);
1477 /* FIXME: this code should not know anything about vcpus */
1478 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1481 if (waitqueue_active(q))
1482 wake_up_interruptible(q);
1484 if (lapic_is_periodic(apic)) {
1485 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1486 return HRTIMER_RESTART;
1488 return HRTIMER_NORESTART;
1491 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1493 struct kvm_lapic *apic;
1495 ASSERT(vcpu != NULL);
1496 apic_debug("apic_init %d\n", vcpu->vcpu_id);
1498 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1502 vcpu->arch.apic = apic;
1504 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1506 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1508 goto nomem_free_apic;
1512 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1514 apic->lapic_timer.timer.function = apic_timer_fn;
1517 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1518 * thinking that APIC satet has changed.
1520 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1521 kvm_lapic_set_base(vcpu,
1522 APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1524 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1525 kvm_lapic_reset(vcpu);
1526 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1535 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1537 struct kvm_lapic *apic = vcpu->arch.apic;
1540 if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1543 apic_update_ppr(apic);
1544 highest_irr = apic_find_highest_irr(apic);
1545 if ((highest_irr == -1) ||
1546 ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1551 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1553 u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1556 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1558 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1559 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1564 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1566 struct kvm_lapic *apic = vcpu->arch.apic;
1568 if (!kvm_vcpu_has_lapic(vcpu))
1571 if (atomic_read(&apic->lapic_timer.pending) > 0) {
1572 if (kvm_apic_local_deliver(apic, APIC_LVTT))
1573 atomic_dec(&apic->lapic_timer.pending);
1577 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1579 int vector = kvm_apic_has_interrupt(vcpu);
1580 struct kvm_lapic *apic = vcpu->arch.apic;
1585 apic_set_isr(vector, apic);
1586 apic_update_ppr(apic);
1587 apic_clear_irr(vector, apic);
1591 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1592 struct kvm_lapic_state *s)
1594 struct kvm_lapic *apic = vcpu->arch.apic;
1596 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1597 /* set SPIV separately to get count of SW disabled APICs right */
1598 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1599 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1600 /* call kvm_apic_set_id() to put apic into apic_map */
1601 kvm_apic_set_id(apic, kvm_apic_id(apic));
1602 kvm_apic_set_version(vcpu);
1604 apic_update_ppr(apic);
1605 hrtimer_cancel(&apic->lapic_timer.timer);
1606 update_divide_count(apic);
1607 start_apic_timer(apic);
1608 apic->irr_pending = true;
1609 apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm) ?
1610 1 : count_vectors(apic->regs + APIC_ISR);
1611 apic->highest_isr_cache = -1;
1612 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, apic_find_highest_isr(apic));
1613 kvm_make_request(KVM_REQ_EVENT, vcpu);
1616 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1618 struct hrtimer *timer;
1620 if (!kvm_vcpu_has_lapic(vcpu))
1623 timer = &vcpu->arch.apic->lapic_timer.timer;
1624 if (hrtimer_cancel(timer))
1625 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1629 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1631 * Detect whether guest triggered PV EOI since the
1632 * last entry. If yes, set EOI on guests's behalf.
1633 * Clear PV EOI in guest memory in any case.
1635 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1636 struct kvm_lapic *apic)
1641 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1642 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1644 * KVM_APIC_PV_EOI_PENDING is unset:
1645 * -> host disabled PV EOI.
1646 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1647 * -> host enabled PV EOI, guest did not execute EOI yet.
1648 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1649 * -> host enabled PV EOI, guest executed EOI.
1651 BUG_ON(!pv_eoi_enabled(vcpu));
1652 pending = pv_eoi_get_pending(vcpu);
1654 * Clear pending bit in any case: it will be set again on vmentry.
1655 * While this might not be ideal from performance point of view,
1656 * this makes sure pv eoi is only enabled when we know it's safe.
1658 pv_eoi_clr_pending(vcpu);
1661 vector = apic_set_eoi(apic);
1662 trace_kvm_pv_eoi(apic, vector);
1665 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1670 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1671 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1673 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1676 vapic = kmap_atomic(vcpu->arch.apic->vapic_page);
1677 data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1678 kunmap_atomic(vapic);
1680 apic_set_tpr(vcpu->arch.apic, data & 0xff);
1684 * apic_sync_pv_eoi_to_guest - called before vmentry
1686 * Detect whether it's safe to enable PV EOI and
1689 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1690 struct kvm_lapic *apic)
1692 if (!pv_eoi_enabled(vcpu) ||
1693 /* IRR set or many bits in ISR: could be nested. */
1694 apic->irr_pending ||
1695 /* Cache not set: could be safe but we don't bother. */
1696 apic->highest_isr_cache == -1 ||
1697 /* Need EOI to update ioapic. */
1698 kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1700 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1701 * so we need not do anything here.
1706 pv_eoi_set_pending(apic->vcpu);
1709 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1712 int max_irr, max_isr;
1713 struct kvm_lapic *apic = vcpu->arch.apic;
1716 apic_sync_pv_eoi_to_guest(vcpu, apic);
1718 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1721 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1722 max_irr = apic_find_highest_irr(apic);
1725 max_isr = apic_find_highest_isr(apic);
1728 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1730 vapic = kmap_atomic(vcpu->arch.apic->vapic_page);
1731 *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1732 kunmap_atomic(vapic);
1735 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1737 vcpu->arch.apic->vapic_addr = vapic_addr;
1739 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1741 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1744 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1746 struct kvm_lapic *apic = vcpu->arch.apic;
1747 u32 reg = (msr - APIC_BASE_MSR) << 4;
1749 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1752 /* if this is ICR write vector before command */
1754 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1755 return apic_reg_write(apic, reg, (u32)data);
1758 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1760 struct kvm_lapic *apic = vcpu->arch.apic;
1761 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1763 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1766 if (apic_reg_read(apic, reg, 4, &low))
1769 apic_reg_read(apic, APIC_ICR2, 4, &high);
1771 *data = (((u64)high) << 32) | low;
1776 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1778 struct kvm_lapic *apic = vcpu->arch.apic;
1780 if (!kvm_vcpu_has_lapic(vcpu))
1783 /* if this is ICR write vector before command */
1784 if (reg == APIC_ICR)
1785 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1786 return apic_reg_write(apic, reg, (u32)data);
1789 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1791 struct kvm_lapic *apic = vcpu->arch.apic;
1794 if (!kvm_vcpu_has_lapic(vcpu))
1797 if (apic_reg_read(apic, reg, 4, &low))
1799 if (reg == APIC_ICR)
1800 apic_reg_read(apic, APIC_ICR2, 4, &high);
1802 *data = (((u64)high) << 32) | low;
1807 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
1809 u64 addr = data & ~KVM_MSR_ENABLED;
1810 if (!IS_ALIGNED(addr, 4))
1813 vcpu->arch.pv_eoi.msr_val = data;
1814 if (!pv_eoi_enabled(vcpu))
1816 return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
1820 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
1822 struct kvm_lapic *apic = vcpu->arch.apic;
1823 unsigned int sipi_vector;
1825 if (!kvm_vcpu_has_lapic(vcpu))
1828 if (test_and_clear_bit(KVM_APIC_INIT, &apic->pending_events)) {
1829 kvm_lapic_reset(vcpu);
1830 kvm_vcpu_reset(vcpu);
1831 if (kvm_vcpu_is_bsp(apic->vcpu))
1832 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1834 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
1836 if (test_and_clear_bit(KVM_APIC_SIPI, &apic->pending_events) &&
1837 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
1838 /* evaluate pending_events before reading the vector */
1840 sipi_vector = apic->sipi_vector;
1841 pr_debug("vcpu %d received sipi with vector # %x\n",
1842 vcpu->vcpu_id, sipi_vector);
1843 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
1844 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1848 void kvm_lapic_init(void)
1850 /* do not patch jump label more than once per second */
1851 jump_label_rate_limit(&apic_hw_disabled, HZ);
1852 jump_label_rate_limit(&apic_sw_disabled, HZ);