3 * Local APIC virtualization
5 * Copyright (C) 2006 Qumranet, Inc.
6 * Copyright (C) 2007 Novell
7 * Copyright (C) 2007 Intel
10 * Dor Laor <dor.laor@qumranet.com>
11 * Gregory Haskins <ghaskins@novell.com>
12 * Yaozu (Eddie) Dong <eddie.dong@intel.com>
14 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
16 * This work is licensed under the terms of the GNU GPL, version 2. See
17 * the COPYING file in the top-level directory.
20 #include <linux/kvm_host.h>
21 #include <linux/kvm.h>
23 #include <linux/highmem.h>
24 #include <linux/smp.h>
25 #include <linux/hrtimer.h>
27 #include <linux/module.h>
28 #include <linux/math64.h>
29 #include <asm/processor.h>
32 #include <asm/current.h>
33 #include <asm/apicdef.h>
34 #include <asm/atomic.h>
35 #include "kvm_cache_regs.h"
41 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
43 #define mod_64(x, y) ((x) % (y))
51 #define APIC_BUS_CYCLE_NS 1
53 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
54 #define apic_debug(fmt, arg...)
56 #define APIC_LVT_NUM 6
57 /* 14 is the version for Xeon and Pentium 8.4.8*/
58 #define APIC_VERSION (0x14UL | ((APIC_LVT_NUM - 1) << 16))
59 #define LAPIC_MMIO_LENGTH (1 << 12)
60 /* followed define is not in apicdef.h */
61 #define APIC_SHORT_MASK 0xc0000
62 #define APIC_DEST_NOSHORT 0x0
63 #define APIC_DEST_MASK 0x800
64 #define MAX_APIC_VECTOR 256
66 #define VEC_POS(v) ((v) & (32 - 1))
67 #define REG_POS(v) (((v) >> 5) << 4)
69 static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
71 return *((u32 *) (apic->regs + reg_off));
74 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
76 *((u32 *) (apic->regs + reg_off)) = val;
79 static inline int apic_test_and_set_vector(int vec, void *bitmap)
81 return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
84 static inline int apic_test_and_clear_vector(int vec, void *bitmap)
86 return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
89 static inline void apic_set_vector(int vec, void *bitmap)
91 set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
94 static inline void apic_clear_vector(int vec, void *bitmap)
96 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
99 static inline int apic_hw_enabled(struct kvm_lapic *apic)
101 return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
104 static inline int apic_sw_enabled(struct kvm_lapic *apic)
106 return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
109 static inline int apic_enabled(struct kvm_lapic *apic)
111 return apic_sw_enabled(apic) && apic_hw_enabled(apic);
115 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
118 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
119 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
121 static inline int kvm_apic_id(struct kvm_lapic *apic)
123 return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
126 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
128 return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
131 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
133 return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
136 static inline int apic_lvtt_period(struct kvm_lapic *apic)
138 return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC;
141 static inline int apic_lvt_nmi_mode(u32 lvt_val)
143 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
146 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
148 struct kvm_lapic *apic = vcpu->arch.apic;
149 struct kvm_cpuid_entry2 *feat;
150 u32 v = APIC_VERSION;
152 if (!irqchip_in_kernel(vcpu->kvm))
155 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
156 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
157 v |= APIC_LVR_DIRECTED_EOI;
158 apic_set_reg(apic, APIC_LVR, v);
161 static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
162 LVT_MASK | APIC_LVT_TIMER_PERIODIC, /* LVTT */
163 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
164 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
165 LINT_MASK, LINT_MASK, /* LVT0-1 */
166 LVT_MASK /* LVTERR */
169 static int find_highest_vector(void *bitmap)
172 int word_offset = MAX_APIC_VECTOR >> 5;
174 while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
177 if (likely(!word_offset && !word[0]))
180 return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
183 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
185 apic->irr_pending = true;
186 return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
189 static inline int apic_search_irr(struct kvm_lapic *apic)
191 return find_highest_vector(apic->regs + APIC_IRR);
194 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
198 if (!apic->irr_pending)
201 result = apic_search_irr(apic);
202 ASSERT(result == -1 || result >= 16);
207 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
209 apic->irr_pending = false;
210 apic_clear_vector(vec, apic->regs + APIC_IRR);
211 if (apic_search_irr(apic) != -1)
212 apic->irr_pending = true;
215 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
217 struct kvm_lapic *apic = vcpu->arch.apic;
220 /* This may race with setting of irr in __apic_accept_irq() and
221 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
222 * will cause vmexit immediately and the value will be recalculated
223 * on the next vmentry.
227 highest_irr = apic_find_highest_irr(apic);
232 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
233 int vector, int level, int trig_mode);
235 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
237 struct kvm_lapic *apic = vcpu->arch.apic;
239 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
240 irq->level, irq->trig_mode);
243 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
247 result = find_highest_vector(apic->regs + APIC_ISR);
248 ASSERT(result == -1 || result >= 16);
253 static void apic_update_ppr(struct kvm_lapic *apic)
258 tpr = apic_get_reg(apic, APIC_TASKPRI);
259 isr = apic_find_highest_isr(apic);
260 isrv = (isr != -1) ? isr : 0;
262 if ((tpr & 0xf0) >= (isrv & 0xf0))
267 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
268 apic, ppr, isr, isrv);
270 apic_set_reg(apic, APIC_PROCPRI, ppr);
273 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
275 apic_set_reg(apic, APIC_TASKPRI, tpr);
276 apic_update_ppr(apic);
279 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
281 return dest == 0xff || kvm_apic_id(apic) == dest;
284 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
289 logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
291 switch (apic_get_reg(apic, APIC_DFR)) {
293 if (logical_id & mda)
296 case APIC_DFR_CLUSTER:
297 if (((logical_id >> 4) == (mda >> 0x4))
298 && (logical_id & mda & 0xf))
302 printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n",
303 apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
310 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
311 int short_hand, int dest, int dest_mode)
314 struct kvm_lapic *target = vcpu->arch.apic;
316 apic_debug("target %p, source %p, dest 0x%x, "
317 "dest_mode 0x%x, short_hand 0x%x\n",
318 target, source, dest, dest_mode, short_hand);
321 switch (short_hand) {
322 case APIC_DEST_NOSHORT:
325 result = kvm_apic_match_physical_addr(target, dest);
328 result = kvm_apic_match_logical_addr(target, dest);
331 result = (target == source);
333 case APIC_DEST_ALLINC:
336 case APIC_DEST_ALLBUT:
337 result = (target != source);
340 printk(KERN_WARNING "Bad dest shorthand value %x\n",
349 * Add a pending IRQ into lapic.
350 * Return 1 if successfully added and 0 if discarded.
352 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
353 int vector, int level, int trig_mode)
356 struct kvm_vcpu *vcpu = apic->vcpu;
358 switch (delivery_mode) {
360 vcpu->arch.apic_arb_prio++;
362 /* FIXME add logic for vcpu on reset */
363 if (unlikely(!apic_enabled(apic)))
366 result = !apic_test_and_set_irr(vector, apic);
369 apic_debug("level trig mode repeatedly for "
370 "vector %d", vector);
375 apic_debug("level trig mode for vector %d", vector);
376 apic_set_vector(vector, apic->regs + APIC_TMR);
378 apic_clear_vector(vector, apic->regs + APIC_TMR);
383 printk(KERN_DEBUG "Ignoring delivery mode 3\n");
387 printk(KERN_DEBUG "Ignoring guest SMI\n");
392 kvm_inject_nmi(vcpu);
399 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
401 "INIT on a runnable vcpu %d\n",
403 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
406 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
411 case APIC_DM_STARTUP:
412 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
413 vcpu->vcpu_id, vector);
414 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
416 vcpu->arch.sipi_vector = vector;
417 vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
424 * Should only be called by kvm_apic_local_deliver() with LVT0,
425 * before NMI watchdog was enabled. Already handled by
426 * kvm_apic_accept_pic_intr().
431 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
438 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
440 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
443 static void apic_set_eoi(struct kvm_lapic *apic)
445 int vector = apic_find_highest_isr(apic);
448 * Not every write EOI will has corresponding ISR,
449 * one example is when Kernel check timer on setup_IO_APIC
454 apic_clear_vector(vector, apic->regs + APIC_ISR);
455 apic_update_ppr(apic);
457 if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
458 trigger_mode = IOAPIC_LEVEL_TRIG;
460 trigger_mode = IOAPIC_EDGE_TRIG;
461 if (!(apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI)) {
462 mutex_lock(&apic->vcpu->kvm->irq_lock);
463 kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
464 mutex_unlock(&apic->vcpu->kvm->irq_lock);
468 static void apic_send_ipi(struct kvm_lapic *apic)
470 u32 icr_low = apic_get_reg(apic, APIC_ICR);
471 u32 icr_high = apic_get_reg(apic, APIC_ICR2);
472 struct kvm_lapic_irq irq;
474 irq.vector = icr_low & APIC_VECTOR_MASK;
475 irq.delivery_mode = icr_low & APIC_MODE_MASK;
476 irq.dest_mode = icr_low & APIC_DEST_MASK;
477 irq.level = icr_low & APIC_INT_ASSERT;
478 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
479 irq.shorthand = icr_low & APIC_SHORT_MASK;
480 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
482 apic_debug("icr_high 0x%x, icr_low 0x%x, "
483 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
484 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
485 icr_high, icr_low, irq.shorthand, irq.dest_id,
486 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
489 mutex_lock(&apic->vcpu->kvm->irq_lock);
490 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
491 mutex_unlock(&apic->vcpu->kvm->irq_lock);
494 static u32 apic_get_tmcct(struct kvm_lapic *apic)
500 ASSERT(apic != NULL);
502 /* if initial count is 0, current count should also be 0 */
503 if (apic_get_reg(apic, APIC_TMICT) == 0)
506 remaining = hrtimer_expires_remaining(&apic->lapic_timer.timer);
507 if (ktime_to_ns(remaining) < 0)
508 remaining = ktime_set(0, 0);
510 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
511 tmcct = div64_u64(ns,
512 (APIC_BUS_CYCLE_NS * apic->divide_count));
517 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
519 struct kvm_vcpu *vcpu = apic->vcpu;
520 struct kvm_run *run = vcpu->run;
522 set_bit(KVM_REQ_REPORT_TPR_ACCESS, &vcpu->requests);
523 run->tpr_access.rip = kvm_rip_read(vcpu);
524 run->tpr_access.is_write = write;
527 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
529 if (apic->vcpu->arch.tpr_access_reporting)
530 __report_tpr_access(apic, write);
533 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
537 if (offset >= LAPIC_MMIO_LENGTH)
542 printk(KERN_WARNING "Access APIC ARBPRI register "
543 "which is for P6\n");
546 case APIC_TMCCT: /* Timer CCR */
547 val = apic_get_tmcct(apic);
551 report_tpr_access(apic, false);
554 apic_update_ppr(apic);
555 val = apic_get_reg(apic, offset);
562 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
564 return container_of(dev, struct kvm_lapic, dev);
567 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
569 return apic_hw_enabled(apic) &&
570 addr >= apic->base_address &&
571 addr < apic->base_address + LAPIC_MMIO_LENGTH;
574 static int apic_mmio_read(struct kvm_io_device *this,
575 gpa_t address, int len, void *data)
577 struct kvm_lapic *apic = to_lapic(this);
578 unsigned int offset = address - apic->base_address;
579 unsigned char alignment = offset & 0xf;
581 if (!apic_mmio_in_range(apic, address))
584 if ((alignment + len) > 4) {
585 printk(KERN_ERR "KVM_APIC_READ: alignment error %lx %d",
586 (unsigned long)address, len);
589 result = __apic_read(apic, offset & ~0xf);
591 trace_kvm_apic_read(offset, result);
597 memcpy(data, (char *)&result + alignment, len);
600 printk(KERN_ERR "Local APIC read with len = %x, "
601 "should be 1,2, or 4 instead\n", len);
607 static void update_divide_count(struct kvm_lapic *apic)
609 u32 tmp1, tmp2, tdcr;
611 tdcr = apic_get_reg(apic, APIC_TDCR);
613 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
614 apic->divide_count = 0x1 << (tmp2 & 0x7);
616 apic_debug("timer divide count is 0x%x\n",
620 static void start_apic_timer(struct kvm_lapic *apic)
622 ktime_t now = apic->lapic_timer.timer.base->get_time();
624 apic->lapic_timer.period = apic_get_reg(apic, APIC_TMICT) *
625 APIC_BUS_CYCLE_NS * apic->divide_count;
626 atomic_set(&apic->lapic_timer.pending, 0);
628 if (!apic->lapic_timer.period)
631 hrtimer_start(&apic->lapic_timer.timer,
632 ktime_add_ns(now, apic->lapic_timer.period),
635 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
637 "timer initial count 0x%x, period %lldns, "
638 "expire @ 0x%016" PRIx64 ".\n", __func__,
639 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
640 apic_get_reg(apic, APIC_TMICT),
641 apic->lapic_timer.period,
642 ktime_to_ns(ktime_add_ns(now,
643 apic->lapic_timer.period)));
646 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
648 int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
650 if (apic_lvt_nmi_mode(lvt0_val)) {
651 if (!nmi_wd_enabled) {
652 apic_debug("Receive NMI setting on APIC_LVT0 "
653 "for cpu %d\n", apic->vcpu->vcpu_id);
654 apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
656 } else if (nmi_wd_enabled)
657 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
660 static int apic_mmio_write(struct kvm_io_device *this,
661 gpa_t address, int len, const void *data)
663 struct kvm_lapic *apic = to_lapic(this);
664 unsigned int offset = address - apic->base_address;
665 unsigned char alignment = offset & 0xf;
667 if (!apic_mmio_in_range(apic, address))
671 * APIC register must be aligned on 128-bits boundary.
672 * 32/64/128 bits registers must be accessed thru 32 bits.
675 if (len != 4 || alignment) {
676 /* Don't shout loud, $infamous_os would cause only noise. */
677 apic_debug("apic write: bad size=%d %lx\n",
684 /* too common printing */
685 if (offset != APIC_EOI)
686 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
687 "0x%x\n", __func__, offset, len, val);
691 trace_kvm_apic_write(offset, val);
694 case APIC_ID: /* Local APIC ID */
695 apic_set_reg(apic, APIC_ID, val);
699 report_tpr_access(apic, true);
700 apic_set_tpr(apic, val & 0xff);
708 apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
712 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
717 if (apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
718 mask |= APIC_SPIV_DIRECTED_EOI;
719 apic_set_reg(apic, APIC_SPIV, val & mask);
720 if (!(val & APIC_SPIV_APIC_ENABLED)) {
724 for (i = 0; i < APIC_LVT_NUM; i++) {
725 lvt_val = apic_get_reg(apic,
726 APIC_LVTT + 0x10 * i);
727 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
728 lvt_val | APIC_LVT_MASKED);
730 atomic_set(&apic->lapic_timer.pending, 0);
736 /* No delay here, so we always clear the pending bit */
737 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
742 apic_set_reg(apic, APIC_ICR2, val & 0xff000000);
746 apic_manage_nmi_watchdog(apic, val);
752 /* TODO: Check vector */
753 if (!apic_sw_enabled(apic))
754 val |= APIC_LVT_MASKED;
756 val &= apic_lvt_mask[(offset - APIC_LVTT) >> 4];
757 apic_set_reg(apic, offset, val);
762 hrtimer_cancel(&apic->lapic_timer.timer);
763 apic_set_reg(apic, APIC_TMICT, val);
764 start_apic_timer(apic);
769 printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
770 apic_set_reg(apic, APIC_TDCR, val);
771 update_divide_count(apic);
775 apic_debug("Local APIC Write to read-only register %x\n",
782 void kvm_free_lapic(struct kvm_vcpu *vcpu)
784 if (!vcpu->arch.apic)
787 hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
789 if (vcpu->arch.apic->regs_page)
790 __free_page(vcpu->arch.apic->regs_page);
792 kfree(vcpu->arch.apic);
796 *----------------------------------------------------------------------
798 *----------------------------------------------------------------------
801 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
803 struct kvm_lapic *apic = vcpu->arch.apic;
807 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
808 | (apic_get_reg(apic, APIC_TASKPRI) & 4));
811 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
813 struct kvm_lapic *apic = vcpu->arch.apic;
818 tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
820 return (tpr & 0xf0) >> 4;
823 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
825 struct kvm_lapic *apic = vcpu->arch.apic;
828 value |= MSR_IA32_APICBASE_BSP;
829 vcpu->arch.apic_base = value;
833 if (!kvm_vcpu_is_bsp(apic->vcpu))
834 value &= ~MSR_IA32_APICBASE_BSP;
836 vcpu->arch.apic_base = value;
837 apic->base_address = apic->vcpu->arch.apic_base &
838 MSR_IA32_APICBASE_BASE;
840 /* with FSB delivery interrupt, we can restart APIC functionality */
841 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
842 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
846 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
848 struct kvm_lapic *apic;
851 apic_debug("%s\n", __func__);
854 apic = vcpu->arch.apic;
855 ASSERT(apic != NULL);
857 /* Stop the timer in case it's a reset to an active apic */
858 hrtimer_cancel(&apic->lapic_timer.timer);
860 apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
861 kvm_apic_set_version(apic->vcpu);
863 for (i = 0; i < APIC_LVT_NUM; i++)
864 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
865 apic_set_reg(apic, APIC_LVT0,
866 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
868 apic_set_reg(apic, APIC_DFR, 0xffffffffU);
869 apic_set_reg(apic, APIC_SPIV, 0xff);
870 apic_set_reg(apic, APIC_TASKPRI, 0);
871 apic_set_reg(apic, APIC_LDR, 0);
872 apic_set_reg(apic, APIC_ESR, 0);
873 apic_set_reg(apic, APIC_ICR, 0);
874 apic_set_reg(apic, APIC_ICR2, 0);
875 apic_set_reg(apic, APIC_TDCR, 0);
876 apic_set_reg(apic, APIC_TMICT, 0);
877 for (i = 0; i < 8; i++) {
878 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
879 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
880 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
882 apic->irr_pending = false;
883 update_divide_count(apic);
884 atomic_set(&apic->lapic_timer.pending, 0);
885 if (kvm_vcpu_is_bsp(vcpu))
886 vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
887 apic_update_ppr(apic);
889 vcpu->arch.apic_arb_prio = 0;
891 apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
892 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
893 vcpu, kvm_apic_id(apic),
894 vcpu->arch.apic_base, apic->base_address);
897 bool kvm_apic_present(struct kvm_vcpu *vcpu)
899 return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
902 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
904 return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
908 *----------------------------------------------------------------------
910 *----------------------------------------------------------------------
913 static bool lapic_is_periodic(struct kvm_timer *ktimer)
915 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
917 return apic_lvtt_period(apic);
920 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
922 struct kvm_lapic *lapic = vcpu->arch.apic;
924 if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
925 return atomic_read(&lapic->lapic_timer.pending);
930 static int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
932 u32 reg = apic_get_reg(apic, lvt_type);
933 int vector, mode, trig_mode;
935 if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
936 vector = reg & APIC_VECTOR_MASK;
937 mode = reg & APIC_MODE_MASK;
938 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
939 return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
944 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
946 struct kvm_lapic *apic = vcpu->arch.apic;
949 kvm_apic_local_deliver(apic, APIC_LVT0);
952 static struct kvm_timer_ops lapic_timer_ops = {
953 .is_periodic = lapic_is_periodic,
956 static const struct kvm_io_device_ops apic_mmio_ops = {
957 .read = apic_mmio_read,
958 .write = apic_mmio_write,
961 int kvm_create_lapic(struct kvm_vcpu *vcpu)
963 struct kvm_lapic *apic;
965 ASSERT(vcpu != NULL);
966 apic_debug("apic_init %d\n", vcpu->vcpu_id);
968 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
972 vcpu->arch.apic = apic;
974 apic->regs_page = alloc_page(GFP_KERNEL);
975 if (apic->regs_page == NULL) {
976 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
978 goto nomem_free_apic;
980 apic->regs = page_address(apic->regs_page);
981 memset(apic->regs, 0, PAGE_SIZE);
984 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
986 apic->lapic_timer.timer.function = kvm_timer_fn;
987 apic->lapic_timer.t_ops = &lapic_timer_ops;
988 apic->lapic_timer.kvm = vcpu->kvm;
989 apic->lapic_timer.vcpu = vcpu;
991 apic->base_address = APIC_DEFAULT_PHYS_BASE;
992 vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
994 kvm_lapic_reset(vcpu);
995 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1004 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1006 struct kvm_lapic *apic = vcpu->arch.apic;
1009 if (!apic || !apic_enabled(apic))
1012 apic_update_ppr(apic);
1013 highest_irr = apic_find_highest_irr(apic);
1014 if ((highest_irr == -1) ||
1015 ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
1020 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1022 u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1025 if (kvm_vcpu_is_bsp(vcpu)) {
1026 if (!apic_hw_enabled(vcpu->arch.apic))
1028 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1029 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1035 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1037 struct kvm_lapic *apic = vcpu->arch.apic;
1039 if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1040 if (kvm_apic_local_deliver(apic, APIC_LVTT))
1041 atomic_dec(&apic->lapic_timer.pending);
1045 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1047 int vector = kvm_apic_has_interrupt(vcpu);
1048 struct kvm_lapic *apic = vcpu->arch.apic;
1053 apic_set_vector(vector, apic->regs + APIC_ISR);
1054 apic_update_ppr(apic);
1055 apic_clear_irr(vector, apic);
1059 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1061 struct kvm_lapic *apic = vcpu->arch.apic;
1063 apic->base_address = vcpu->arch.apic_base &
1064 MSR_IA32_APICBASE_BASE;
1065 kvm_apic_set_version(vcpu);
1067 apic_update_ppr(apic);
1068 hrtimer_cancel(&apic->lapic_timer.timer);
1069 update_divide_count(apic);
1070 start_apic_timer(apic);
1073 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1075 struct kvm_lapic *apic = vcpu->arch.apic;
1076 struct hrtimer *timer;
1081 timer = &apic->lapic_timer.timer;
1082 if (hrtimer_cancel(timer))
1083 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1086 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1091 if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1094 vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1095 data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1096 kunmap_atomic(vapic, KM_USER0);
1098 apic_set_tpr(vcpu->arch.apic, data & 0xff);
1101 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1104 int max_irr, max_isr;
1105 struct kvm_lapic *apic;
1108 if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1111 apic = vcpu->arch.apic;
1112 tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1113 max_irr = apic_find_highest_irr(apic);
1116 max_isr = apic_find_highest_isr(apic);
1119 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1121 vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1122 *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1123 kunmap_atomic(vapic, KM_USER0);
1126 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1128 if (!irqchip_in_kernel(vcpu->kvm))
1131 vcpu->arch.apic->vapic_addr = vapic_addr;