aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kvm
diff options
context:
space:
mode:
authorIngo Molnar <mingo@elte.hu>2009-01-06 03:53:05 -0500
committerIngo Molnar <mingo@elte.hu>2009-01-06 03:53:05 -0500
commit3d7a96f5a485b7d06c2379f343d7312af89ec9e2 (patch)
tree5f097f68eb0f9fd3fa4a10f38672e300e9127b10 /arch/x86/kvm
parent723cbe0775514853c22dc45005af59c360916af1 (diff)
parent238c6d54830c624f34ac9cf123ac04aebfca5013 (diff)
Merge branch 'linus' into tracing/kmemtrace2
Diffstat (limited to 'arch/x86/kvm')
-rw-r--r--arch/x86/kvm/Makefile4
-rw-r--r--arch/x86/kvm/i8254.c19
-rw-r--r--arch/x86/kvm/i8259.c52
-rw-r--r--arch/x86/kvm/irq.h6
-rw-r--r--arch/x86/kvm/kvm_svm.h2
-rw-r--r--arch/x86/kvm/lapic.c58
-rw-r--r--arch/x86/kvm/mmu.c444
-rw-r--r--arch/x86/kvm/paging_tmpl.h44
-rw-r--r--arch/x86/kvm/svm.c48
-rw-r--r--arch/x86/kvm/svm.h328
-rw-r--r--arch/x86/kvm/vmx.c350
-rw-r--r--arch/x86/kvm/vmx.h359
-rw-r--r--arch/x86/kvm/x86.c120
-rw-r--r--arch/x86/kvm/x86_emulate.c297
14 files changed, 1059 insertions, 1072 deletions
diff --git a/arch/x86/kvm/Makefile b/arch/x86/kvm/Makefile
index c02343594b4d..d3ec292f00f2 100644
--- a/arch/x86/kvm/Makefile
+++ b/arch/x86/kvm/Makefile
@@ -7,8 +7,8 @@ common-objs = $(addprefix ../../../virt/kvm/, kvm_main.o ioapic.o \
7ifeq ($(CONFIG_KVM_TRACE),y) 7ifeq ($(CONFIG_KVM_TRACE),y)
8common-objs += $(addprefix ../../../virt/kvm/, kvm_trace.o) 8common-objs += $(addprefix ../../../virt/kvm/, kvm_trace.o)
9endif 9endif
10ifeq ($(CONFIG_DMAR),y) 10ifeq ($(CONFIG_IOMMU_API),y)
11common-objs += $(addprefix ../../../virt/kvm/, vtd.o) 11common-objs += $(addprefix ../../../virt/kvm/, iommu.o)
12endif 12endif
13 13
14EXTRA_CFLAGS += -Ivirt/kvm -Iarch/x86/kvm 14EXTRA_CFLAGS += -Ivirt/kvm -Iarch/x86/kvm
diff --git a/arch/x86/kvm/i8254.c b/arch/x86/kvm/i8254.c
index 59ebd37ad79e..e665d1c623ca 100644
--- a/arch/x86/kvm/i8254.c
+++ b/arch/x86/kvm/i8254.c
@@ -603,10 +603,29 @@ void kvm_free_pit(struct kvm *kvm)
603 603
604static void __inject_pit_timer_intr(struct kvm *kvm) 604static void __inject_pit_timer_intr(struct kvm *kvm)
605{ 605{
606 struct kvm_vcpu *vcpu;
607 int i;
608
606 mutex_lock(&kvm->lock); 609 mutex_lock(&kvm->lock);
607 kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 1); 610 kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 1);
608 kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 0); 611 kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 0);
609 mutex_unlock(&kvm->lock); 612 mutex_unlock(&kvm->lock);
613
614 /*
615 * Provides NMI watchdog support via Virtual Wire mode.
616 * The route is: PIT -> PIC -> LVT0 in NMI mode.
617 *
618 * Note: Our Virtual Wire implementation is simplified, only
619 * propagating PIT interrupts to all VCPUs when they have set
620 * LVT0 to NMI delivery. Other PIC interrupts are just sent to
621 * VCPU0, and only if its LVT0 is in EXTINT mode.
622 */
623 if (kvm->arch.vapics_in_nmi_mode > 0)
624 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
625 vcpu = kvm->vcpus[i];
626 if (vcpu)
627 kvm_apic_nmi_wd_deliver(vcpu);
628 }
610} 629}
611 630
612void kvm_inject_pit_timer_irqs(struct kvm_vcpu *vcpu) 631void kvm_inject_pit_timer_irqs(struct kvm_vcpu *vcpu)
diff --git a/arch/x86/kvm/i8259.c b/arch/x86/kvm/i8259.c
index 17e41e165f1a..179dcb0103fd 100644
--- a/arch/x86/kvm/i8259.c
+++ b/arch/x86/kvm/i8259.c
@@ -26,10 +26,40 @@
26 * Port from Qemu. 26 * Port from Qemu.
27 */ 27 */
28#include <linux/mm.h> 28#include <linux/mm.h>
29#include <linux/bitops.h>
29#include "irq.h" 30#include "irq.h"
30 31
31#include <linux/kvm_host.h> 32#include <linux/kvm_host.h>
32 33
34static void pic_lock(struct kvm_pic *s)
35{
36 spin_lock(&s->lock);
37}
38
39static void pic_unlock(struct kvm_pic *s)
40{
41 struct kvm *kvm = s->kvm;
42 unsigned acks = s->pending_acks;
43 bool wakeup = s->wakeup_needed;
44 struct kvm_vcpu *vcpu;
45
46 s->pending_acks = 0;
47 s->wakeup_needed = false;
48
49 spin_unlock(&s->lock);
50
51 while (acks) {
52 kvm_notify_acked_irq(kvm, __ffs(acks));
53 acks &= acks - 1;
54 }
55
56 if (wakeup) {
57 vcpu = s->kvm->vcpus[0];
58 if (vcpu)
59 kvm_vcpu_kick(vcpu);
60 }
61}
62
33static void pic_clear_isr(struct kvm_kpic_state *s, int irq) 63static void pic_clear_isr(struct kvm_kpic_state *s, int irq)
34{ 64{
35 s->isr &= ~(1 << irq); 65 s->isr &= ~(1 << irq);
@@ -136,17 +166,21 @@ static void pic_update_irq(struct kvm_pic *s)
136 166
137void kvm_pic_update_irq(struct kvm_pic *s) 167void kvm_pic_update_irq(struct kvm_pic *s)
138{ 168{
169 pic_lock(s);
139 pic_update_irq(s); 170 pic_update_irq(s);
171 pic_unlock(s);
140} 172}
141 173
142void kvm_pic_set_irq(void *opaque, int irq, int level) 174void kvm_pic_set_irq(void *opaque, int irq, int level)
143{ 175{
144 struct kvm_pic *s = opaque; 176 struct kvm_pic *s = opaque;
145 177
178 pic_lock(s);
146 if (irq >= 0 && irq < PIC_NUM_PINS) { 179 if (irq >= 0 && irq < PIC_NUM_PINS) {
147 pic_set_irq1(&s->pics[irq >> 3], irq & 7, level); 180 pic_set_irq1(&s->pics[irq >> 3], irq & 7, level);
148 pic_update_irq(s); 181 pic_update_irq(s);
149 } 182 }
183 pic_unlock(s);
150} 184}
151 185
152/* 186/*
@@ -172,6 +206,7 @@ int kvm_pic_read_irq(struct kvm *kvm)
172 int irq, irq2, intno; 206 int irq, irq2, intno;
173 struct kvm_pic *s = pic_irqchip(kvm); 207 struct kvm_pic *s = pic_irqchip(kvm);
174 208
209 pic_lock(s);
175 irq = pic_get_irq(&s->pics[0]); 210 irq = pic_get_irq(&s->pics[0]);
176 if (irq >= 0) { 211 if (irq >= 0) {
177 pic_intack(&s->pics[0], irq); 212 pic_intack(&s->pics[0], irq);
@@ -196,6 +231,7 @@ int kvm_pic_read_irq(struct kvm *kvm)
196 intno = s->pics[0].irq_base + irq; 231 intno = s->pics[0].irq_base + irq;
197 } 232 }
198 pic_update_irq(s); 233 pic_update_irq(s);
234 pic_unlock(s);
199 kvm_notify_acked_irq(kvm, irq); 235 kvm_notify_acked_irq(kvm, irq);
200 236
201 return intno; 237 return intno;
@@ -203,7 +239,7 @@ int kvm_pic_read_irq(struct kvm *kvm)
203 239
204void kvm_pic_reset(struct kvm_kpic_state *s) 240void kvm_pic_reset(struct kvm_kpic_state *s)
205{ 241{
206 int irq, irqbase; 242 int irq, irqbase, n;
207 struct kvm *kvm = s->pics_state->irq_request_opaque; 243 struct kvm *kvm = s->pics_state->irq_request_opaque;
208 struct kvm_vcpu *vcpu0 = kvm->vcpus[0]; 244 struct kvm_vcpu *vcpu0 = kvm->vcpus[0];
209 245
@@ -214,8 +250,10 @@ void kvm_pic_reset(struct kvm_kpic_state *s)
214 250
215 for (irq = 0; irq < PIC_NUM_PINS/2; irq++) { 251 for (irq = 0; irq < PIC_NUM_PINS/2; irq++) {
216 if (vcpu0 && kvm_apic_accept_pic_intr(vcpu0)) 252 if (vcpu0 && kvm_apic_accept_pic_intr(vcpu0))
217 if (s->irr & (1 << irq) || s->isr & (1 << irq)) 253 if (s->irr & (1 << irq) || s->isr & (1 << irq)) {
218 kvm_notify_acked_irq(kvm, irq+irqbase); 254 n = irq + irqbase;
255 s->pics_state->pending_acks |= 1 << n;
256 }
219 } 257 }
220 s->last_irr = 0; 258 s->last_irr = 0;
221 s->irr = 0; 259 s->irr = 0;
@@ -406,6 +444,7 @@ static void picdev_write(struct kvm_io_device *this,
406 printk(KERN_ERR "PIC: non byte write\n"); 444 printk(KERN_ERR "PIC: non byte write\n");
407 return; 445 return;
408 } 446 }
447 pic_lock(s);
409 switch (addr) { 448 switch (addr) {
410 case 0x20: 449 case 0x20:
411 case 0x21: 450 case 0x21:
@@ -418,6 +457,7 @@ static void picdev_write(struct kvm_io_device *this,
418 elcr_ioport_write(&s->pics[addr & 1], addr, data); 457 elcr_ioport_write(&s->pics[addr & 1], addr, data);
419 break; 458 break;
420 } 459 }
460 pic_unlock(s);
421} 461}
422 462
423static void picdev_read(struct kvm_io_device *this, 463static void picdev_read(struct kvm_io_device *this,
@@ -431,6 +471,7 @@ static void picdev_read(struct kvm_io_device *this,
431 printk(KERN_ERR "PIC: non byte read\n"); 471 printk(KERN_ERR "PIC: non byte read\n");
432 return; 472 return;
433 } 473 }
474 pic_lock(s);
434 switch (addr) { 475 switch (addr) {
435 case 0x20: 476 case 0x20:
436 case 0x21: 477 case 0x21:
@@ -444,6 +485,7 @@ static void picdev_read(struct kvm_io_device *this,
444 break; 485 break;
445 } 486 }
446 *(unsigned char *)val = data; 487 *(unsigned char *)val = data;
488 pic_unlock(s);
447} 489}
448 490
449/* 491/*
@@ -459,7 +501,7 @@ static void pic_irq_request(void *opaque, int level)
459 s->output = level; 501 s->output = level;
460 if (vcpu && level && (s->pics[0].isr_ack & (1 << irq))) { 502 if (vcpu && level && (s->pics[0].isr_ack & (1 << irq))) {
461 s->pics[0].isr_ack &= ~(1 << irq); 503 s->pics[0].isr_ack &= ~(1 << irq);
462 kvm_vcpu_kick(vcpu); 504 s->wakeup_needed = true;
463 } 505 }
464} 506}
465 507
@@ -469,6 +511,8 @@ struct kvm_pic *kvm_create_pic(struct kvm *kvm)
469 s = kzalloc(sizeof(struct kvm_pic), GFP_KERNEL); 511 s = kzalloc(sizeof(struct kvm_pic), GFP_KERNEL);
470 if (!s) 512 if (!s)
471 return NULL; 513 return NULL;
514 spin_lock_init(&s->lock);
515 s->kvm = kvm;
472 s->pics[0].elcr_mask = 0xf8; 516 s->pics[0].elcr_mask = 0xf8;
473 s->pics[1].elcr_mask = 0xde; 517 s->pics[1].elcr_mask = 0xde;
474 s->irq_request = pic_irq_request; 518 s->irq_request = pic_irq_request;
diff --git a/arch/x86/kvm/irq.h b/arch/x86/kvm/irq.h
index f17c8f5bbf31..2bf32a03ceec 100644
--- a/arch/x86/kvm/irq.h
+++ b/arch/x86/kvm/irq.h
@@ -25,6 +25,7 @@
25#include <linux/mm_types.h> 25#include <linux/mm_types.h>
26#include <linux/hrtimer.h> 26#include <linux/hrtimer.h>
27#include <linux/kvm_host.h> 27#include <linux/kvm_host.h>
28#include <linux/spinlock.h>
28 29
29#include "iodev.h" 30#include "iodev.h"
30#include "ioapic.h" 31#include "ioapic.h"
@@ -59,6 +60,10 @@ struct kvm_kpic_state {
59}; 60};
60 61
61struct kvm_pic { 62struct kvm_pic {
63 spinlock_t lock;
64 bool wakeup_needed;
65 unsigned pending_acks;
66 struct kvm *kvm;
62 struct kvm_kpic_state pics[2]; /* 0 is master pic, 1 is slave pic */ 67 struct kvm_kpic_state pics[2]; /* 0 is master pic, 1 is slave pic */
63 irq_request_func *irq_request; 68 irq_request_func *irq_request;
64 void *irq_request_opaque; 69 void *irq_request_opaque;
@@ -87,6 +92,7 @@ void kvm_pic_reset(struct kvm_kpic_state *s);
87void kvm_timer_intr_post(struct kvm_vcpu *vcpu, int vec); 92void kvm_timer_intr_post(struct kvm_vcpu *vcpu, int vec);
88void kvm_inject_pending_timer_irqs(struct kvm_vcpu *vcpu); 93void kvm_inject_pending_timer_irqs(struct kvm_vcpu *vcpu);
89void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu); 94void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu);
95void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu);
90void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu); 96void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu);
91void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu); 97void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu);
92void __kvm_migrate_timers(struct kvm_vcpu *vcpu); 98void __kvm_migrate_timers(struct kvm_vcpu *vcpu);
diff --git a/arch/x86/kvm/kvm_svm.h b/arch/x86/kvm/kvm_svm.h
index 65ef0fc2c036..8e5ee99551f6 100644
--- a/arch/x86/kvm/kvm_svm.h
+++ b/arch/x86/kvm/kvm_svm.h
@@ -7,7 +7,7 @@
7#include <linux/kvm_host.h> 7#include <linux/kvm_host.h>
8#include <asm/msr.h> 8#include <asm/msr.h>
9 9
10#include "svm.h" 10#include <asm/svm.h>
11 11
12static const u32 host_save_user_msrs[] = { 12static const u32 host_save_user_msrs[] = {
13#ifdef CONFIG_X86_64 13#ifdef CONFIG_X86_64
diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index 0fc3cab48943..afac68c0815c 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -130,6 +130,11 @@ static inline int apic_lvtt_period(struct kvm_lapic *apic)
130 return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC; 130 return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC;
131} 131}
132 132
133static inline int apic_lvt_nmi_mode(u32 lvt_val)
134{
135 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
136}
137
133static unsigned int apic_lvt_mask[APIC_LVT_NUM] = { 138static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
134 LVT_MASK | APIC_LVT_TIMER_PERIODIC, /* LVTT */ 139 LVT_MASK | APIC_LVT_TIMER_PERIODIC, /* LVTT */
135 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */ 140 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
@@ -354,6 +359,7 @@ static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
354 359
355 case APIC_DM_NMI: 360 case APIC_DM_NMI:
356 kvm_inject_nmi(vcpu); 361 kvm_inject_nmi(vcpu);
362 kvm_vcpu_kick(vcpu);
357 break; 363 break;
358 364
359 case APIC_DM_INIT: 365 case APIC_DM_INIT:
@@ -380,6 +386,14 @@ static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
380 } 386 }
381 break; 387 break;
382 388
389 case APIC_DM_EXTINT:
390 /*
391 * Should only be called by kvm_apic_local_deliver() with LVT0,
392 * before NMI watchdog was enabled. Already handled by
393 * kvm_apic_accept_pic_intr().
394 */
395 break;
396
383 default: 397 default:
384 printk(KERN_ERR "TODO: unsupported delivery mode %x\n", 398 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
385 delivery_mode); 399 delivery_mode);
@@ -663,6 +677,20 @@ static void start_apic_timer(struct kvm_lapic *apic)
663 apic->timer.period))); 677 apic->timer.period)));
664} 678}
665 679
680static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
681{
682 int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
683
684 if (apic_lvt_nmi_mode(lvt0_val)) {
685 if (!nmi_wd_enabled) {
686 apic_debug("Receive NMI setting on APIC_LVT0 "
687 "for cpu %d\n", apic->vcpu->vcpu_id);
688 apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
689 }
690 } else if (nmi_wd_enabled)
691 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
692}
693
666static void apic_mmio_write(struct kvm_io_device *this, 694static void apic_mmio_write(struct kvm_io_device *this,
667 gpa_t address, int len, const void *data) 695 gpa_t address, int len, const void *data)
668{ 696{
@@ -743,10 +771,11 @@ static void apic_mmio_write(struct kvm_io_device *this,
743 apic_set_reg(apic, APIC_ICR2, val & 0xff000000); 771 apic_set_reg(apic, APIC_ICR2, val & 0xff000000);
744 break; 772 break;
745 773
774 case APIC_LVT0:
775 apic_manage_nmi_watchdog(apic, val);
746 case APIC_LVTT: 776 case APIC_LVTT:
747 case APIC_LVTTHMR: 777 case APIC_LVTTHMR:
748 case APIC_LVTPC: 778 case APIC_LVTPC:
749 case APIC_LVT0:
750 case APIC_LVT1: 779 case APIC_LVT1:
751 case APIC_LVTERR: 780 case APIC_LVTERR:
752 /* TODO: Check vector */ 781 /* TODO: Check vector */
@@ -961,12 +990,26 @@ int apic_has_pending_timer(struct kvm_vcpu *vcpu)
961 return 0; 990 return 0;
962} 991}
963 992
964static int __inject_apic_timer_irq(struct kvm_lapic *apic) 993static int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
994{
995 u32 reg = apic_get_reg(apic, lvt_type);
996 int vector, mode, trig_mode;
997
998 if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
999 vector = reg & APIC_VECTOR_MASK;
1000 mode = reg & APIC_MODE_MASK;
1001 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1002 return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
1003 }
1004 return 0;
1005}
1006
1007void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
965{ 1008{
966 int vector; 1009 struct kvm_lapic *apic = vcpu->arch.apic;
967 1010
968 vector = apic_lvt_vector(apic, APIC_LVTT); 1011 if (apic)
969 return __apic_accept_irq(apic, APIC_DM_FIXED, vector, 1, 0); 1012 kvm_apic_local_deliver(apic, APIC_LVT0);
970} 1013}
971 1014
972static enum hrtimer_restart apic_timer_fn(struct hrtimer *data) 1015static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
@@ -1061,9 +1104,8 @@ void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1061{ 1104{
1062 struct kvm_lapic *apic = vcpu->arch.apic; 1105 struct kvm_lapic *apic = vcpu->arch.apic;
1063 1106
1064 if (apic && apic_lvt_enabled(apic, APIC_LVTT) && 1107 if (apic && atomic_read(&apic->timer.pending) > 0) {
1065 atomic_read(&apic->timer.pending) > 0) { 1108 if (kvm_apic_local_deliver(apic, APIC_LVTT))
1066 if (__inject_apic_timer_irq(apic))
1067 atomic_dec(&apic->timer.pending); 1109 atomic_dec(&apic->timer.pending);
1068 } 1110 }
1069} 1111}
diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 410ddbc1aa2e..83f11c7474a1 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -17,7 +17,6 @@
17 * 17 *
18 */ 18 */
19 19
20#include "vmx.h"
21#include "mmu.h" 20#include "mmu.h"
22 21
23#include <linux/kvm_host.h> 22#include <linux/kvm_host.h>
@@ -33,6 +32,7 @@
33#include <asm/page.h> 32#include <asm/page.h>
34#include <asm/cmpxchg.h> 33#include <asm/cmpxchg.h>
35#include <asm/io.h> 34#include <asm/io.h>
35#include <asm/vmx.h>
36 36
37/* 37/*
38 * When setting this variable to true it enables Two-Dimensional-Paging 38 * When setting this variable to true it enables Two-Dimensional-Paging
@@ -168,6 +168,7 @@ static u64 __read_mostly shadow_x_mask; /* mutual exclusive with nx_mask */
168static u64 __read_mostly shadow_user_mask; 168static u64 __read_mostly shadow_user_mask;
169static u64 __read_mostly shadow_accessed_mask; 169static u64 __read_mostly shadow_accessed_mask;
170static u64 __read_mostly shadow_dirty_mask; 170static u64 __read_mostly shadow_dirty_mask;
171static u64 __read_mostly shadow_mt_mask;
171 172
172void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte) 173void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte)
173{ 174{
@@ -183,13 +184,14 @@ void kvm_mmu_set_base_ptes(u64 base_pte)
183EXPORT_SYMBOL_GPL(kvm_mmu_set_base_ptes); 184EXPORT_SYMBOL_GPL(kvm_mmu_set_base_ptes);
184 185
185void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask, 186void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask,
186 u64 dirty_mask, u64 nx_mask, u64 x_mask) 187 u64 dirty_mask, u64 nx_mask, u64 x_mask, u64 mt_mask)
187{ 188{
188 shadow_user_mask = user_mask; 189 shadow_user_mask = user_mask;
189 shadow_accessed_mask = accessed_mask; 190 shadow_accessed_mask = accessed_mask;
190 shadow_dirty_mask = dirty_mask; 191 shadow_dirty_mask = dirty_mask;
191 shadow_nx_mask = nx_mask; 192 shadow_nx_mask = nx_mask;
192 shadow_x_mask = x_mask; 193 shadow_x_mask = x_mask;
194 shadow_mt_mask = mt_mask;
193} 195}
194EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes); 196EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes);
195 197
@@ -384,7 +386,9 @@ static void account_shadowed(struct kvm *kvm, gfn_t gfn)
384{ 386{
385 int *write_count; 387 int *write_count;
386 388
387 write_count = slot_largepage_idx(gfn, gfn_to_memslot(kvm, gfn)); 389 gfn = unalias_gfn(kvm, gfn);
390 write_count = slot_largepage_idx(gfn,
391 gfn_to_memslot_unaliased(kvm, gfn));
388 *write_count += 1; 392 *write_count += 1;
389} 393}
390 394
@@ -392,16 +396,20 @@ static void unaccount_shadowed(struct kvm *kvm, gfn_t gfn)
392{ 396{
393 int *write_count; 397 int *write_count;
394 398
395 write_count = slot_largepage_idx(gfn, gfn_to_memslot(kvm, gfn)); 399 gfn = unalias_gfn(kvm, gfn);
400 write_count = slot_largepage_idx(gfn,
401 gfn_to_memslot_unaliased(kvm, gfn));
396 *write_count -= 1; 402 *write_count -= 1;
397 WARN_ON(*write_count < 0); 403 WARN_ON(*write_count < 0);
398} 404}
399 405
400static int has_wrprotected_page(struct kvm *kvm, gfn_t gfn) 406static int has_wrprotected_page(struct kvm *kvm, gfn_t gfn)
401{ 407{
402 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); 408 struct kvm_memory_slot *slot;
403 int *largepage_idx; 409 int *largepage_idx;
404 410
411 gfn = unalias_gfn(kvm, gfn);
412 slot = gfn_to_memslot_unaliased(kvm, gfn);
405 if (slot) { 413 if (slot) {
406 largepage_idx = slot_largepage_idx(gfn, slot); 414 largepage_idx = slot_largepage_idx(gfn, slot);
407 return *largepage_idx; 415 return *largepage_idx;
@@ -613,7 +621,7 @@ static u64 *rmap_next(struct kvm *kvm, unsigned long *rmapp, u64 *spte)
613 return NULL; 621 return NULL;
614} 622}
615 623
616static void rmap_write_protect(struct kvm *kvm, u64 gfn) 624static int rmap_write_protect(struct kvm *kvm, u64 gfn)
617{ 625{
618 unsigned long *rmapp; 626 unsigned long *rmapp;
619 u64 *spte; 627 u64 *spte;
@@ -659,8 +667,7 @@ static void rmap_write_protect(struct kvm *kvm, u64 gfn)
659 spte = rmap_next(kvm, rmapp, spte); 667 spte = rmap_next(kvm, rmapp, spte);
660 } 668 }
661 669
662 if (write_protected) 670 return write_protected;
663 kvm_flush_remote_tlbs(kvm);
664} 671}
665 672
666static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp) 673static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp)
@@ -786,9 +793,11 @@ static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
786 sp->gfns = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache, PAGE_SIZE); 793 sp->gfns = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache, PAGE_SIZE);
787 set_page_private(virt_to_page(sp->spt), (unsigned long)sp); 794 set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
788 list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages); 795 list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages);
796 INIT_LIST_HEAD(&sp->oos_link);
789 ASSERT(is_empty_shadow_page(sp->spt)); 797 ASSERT(is_empty_shadow_page(sp->spt));
790 sp->slot_bitmap = 0; 798 bitmap_zero(sp->slot_bitmap, KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS);
791 sp->multimapped = 0; 799 sp->multimapped = 0;
800 sp->global = 1;
792 sp->parent_pte = parent_pte; 801 sp->parent_pte = parent_pte;
793 --vcpu->kvm->arch.n_free_mmu_pages; 802 --vcpu->kvm->arch.n_free_mmu_pages;
794 return sp; 803 return sp;
@@ -900,8 +909,9 @@ static void kvm_mmu_update_unsync_bitmap(u64 *spte)
900 struct kvm_mmu_page *sp = page_header(__pa(spte)); 909 struct kvm_mmu_page *sp = page_header(__pa(spte));
901 910
902 index = spte - sp->spt; 911 index = spte - sp->spt;
903 __set_bit(index, sp->unsync_child_bitmap); 912 if (!__test_and_set_bit(index, sp->unsync_child_bitmap))
904 sp->unsync_children = 1; 913 sp->unsync_children++;
914 WARN_ON(!sp->unsync_children);
905} 915}
906 916
907static void kvm_mmu_update_parents_unsync(struct kvm_mmu_page *sp) 917static void kvm_mmu_update_parents_unsync(struct kvm_mmu_page *sp)
@@ -928,7 +938,6 @@ static void kvm_mmu_update_parents_unsync(struct kvm_mmu_page *sp)
928 938
929static int unsync_walk_fn(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) 939static int unsync_walk_fn(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
930{ 940{
931 sp->unsync_children = 1;
932 kvm_mmu_update_parents_unsync(sp); 941 kvm_mmu_update_parents_unsync(sp);
933 return 1; 942 return 1;
934} 943}
@@ -959,38 +968,66 @@ static void nonpaging_invlpg(struct kvm_vcpu *vcpu, gva_t gva)
959{ 968{
960} 969}
961 970
971#define KVM_PAGE_ARRAY_NR 16
972
973struct kvm_mmu_pages {
974 struct mmu_page_and_offset {
975 struct kvm_mmu_page *sp;
976 unsigned int idx;
977 } page[KVM_PAGE_ARRAY_NR];
978 unsigned int nr;
979};
980
962#define for_each_unsync_children(bitmap, idx) \ 981#define for_each_unsync_children(bitmap, idx) \
963 for (idx = find_first_bit(bitmap, 512); \ 982 for (idx = find_first_bit(bitmap, 512); \
964 idx < 512; \ 983 idx < 512; \
965 idx = find_next_bit(bitmap, 512, idx+1)) 984 idx = find_next_bit(bitmap, 512, idx+1))
966 985
967static int mmu_unsync_walk(struct kvm_mmu_page *sp, 986int mmu_pages_add(struct kvm_mmu_pages *pvec, struct kvm_mmu_page *sp,
968 struct kvm_unsync_walk *walker) 987 int idx)
969{ 988{
970 int i, ret; 989 int i;
971 990
972 if (!sp->unsync_children) 991 if (sp->unsync)
973 return 0; 992 for (i=0; i < pvec->nr; i++)
993 if (pvec->page[i].sp == sp)
994 return 0;
995
996 pvec->page[pvec->nr].sp = sp;
997 pvec->page[pvec->nr].idx = idx;
998 pvec->nr++;
999 return (pvec->nr == KVM_PAGE_ARRAY_NR);
1000}
1001
1002static int __mmu_unsync_walk(struct kvm_mmu_page *sp,
1003 struct kvm_mmu_pages *pvec)
1004{
1005 int i, ret, nr_unsync_leaf = 0;
974 1006
975 for_each_unsync_children(sp->unsync_child_bitmap, i) { 1007 for_each_unsync_children(sp->unsync_child_bitmap, i) {
976 u64 ent = sp->spt[i]; 1008 u64 ent = sp->spt[i];
977 1009
978 if (is_shadow_present_pte(ent)) { 1010 if (is_shadow_present_pte(ent) && !is_large_pte(ent)) {
979 struct kvm_mmu_page *child; 1011 struct kvm_mmu_page *child;
980 child = page_header(ent & PT64_BASE_ADDR_MASK); 1012 child = page_header(ent & PT64_BASE_ADDR_MASK);
981 1013
982 if (child->unsync_children) { 1014 if (child->unsync_children) {
983 ret = mmu_unsync_walk(child, walker); 1015 if (mmu_pages_add(pvec, child, i))
984 if (ret) 1016 return -ENOSPC;
1017
1018 ret = __mmu_unsync_walk(child, pvec);
1019 if (!ret)
1020 __clear_bit(i, sp->unsync_child_bitmap);
1021 else if (ret > 0)
1022 nr_unsync_leaf += ret;
1023 else
985 return ret; 1024 return ret;
986 __clear_bit(i, sp->unsync_child_bitmap);
987 } 1025 }
988 1026
989 if (child->unsync) { 1027 if (child->unsync) {
990 ret = walker->entry(child, walker); 1028 nr_unsync_leaf++;
991 __clear_bit(i, sp->unsync_child_bitmap); 1029 if (mmu_pages_add(pvec, child, i))
992 if (ret) 1030 return -ENOSPC;
993 return ret;
994 } 1031 }
995 } 1032 }
996 } 1033 }
@@ -998,7 +1035,17 @@ static int mmu_unsync_walk(struct kvm_mmu_page *sp,
998 if (find_first_bit(sp->unsync_child_bitmap, 512) == 512) 1035 if (find_first_bit(sp->unsync_child_bitmap, 512) == 512)
999 sp->unsync_children = 0; 1036 sp->unsync_children = 0;
1000 1037
1001 return 0; 1038 return nr_unsync_leaf;
1039}
1040
1041static int mmu_unsync_walk(struct kvm_mmu_page *sp,
1042 struct kvm_mmu_pages *pvec)
1043{
1044 if (!sp->unsync_children)
1045 return 0;
1046
1047 mmu_pages_add(pvec, sp, 0);
1048 return __mmu_unsync_walk(sp, pvec);
1002} 1049}
1003 1050
1004static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, gfn_t gfn) 1051static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, gfn_t gfn)
@@ -1021,10 +1068,18 @@ static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, gfn_t gfn)
1021 return NULL; 1068 return NULL;
1022} 1069}
1023 1070
1071static void kvm_unlink_unsync_global(struct kvm *kvm, struct kvm_mmu_page *sp)
1072{
1073 list_del(&sp->oos_link);
1074 --kvm->stat.mmu_unsync_global;
1075}
1076
1024static void kvm_unlink_unsync_page(struct kvm *kvm, struct kvm_mmu_page *sp) 1077static void kvm_unlink_unsync_page(struct kvm *kvm, struct kvm_mmu_page *sp)
1025{ 1078{
1026 WARN_ON(!sp->unsync); 1079 WARN_ON(!sp->unsync);
1027 sp->unsync = 0; 1080 sp->unsync = 0;
1081 if (sp->global)
1082 kvm_unlink_unsync_global(kvm, sp);
1028 --kvm->stat.mmu_unsync; 1083 --kvm->stat.mmu_unsync;
1029} 1084}
1030 1085
@@ -1037,7 +1092,8 @@ static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
1037 return 1; 1092 return 1;
1038 } 1093 }
1039 1094
1040 rmap_write_protect(vcpu->kvm, sp->gfn); 1095 if (rmap_write_protect(vcpu->kvm, sp->gfn))
1096 kvm_flush_remote_tlbs(vcpu->kvm);
1041 kvm_unlink_unsync_page(vcpu->kvm, sp); 1097 kvm_unlink_unsync_page(vcpu->kvm, sp);
1042 if (vcpu->arch.mmu.sync_page(vcpu, sp)) { 1098 if (vcpu->arch.mmu.sync_page(vcpu, sp)) {
1043 kvm_mmu_zap_page(vcpu->kvm, sp); 1099 kvm_mmu_zap_page(vcpu->kvm, sp);
@@ -1048,30 +1104,89 @@ static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
1048 return 0; 1104 return 0;
1049} 1105}
1050 1106
1051struct sync_walker { 1107struct mmu_page_path {
1052 struct kvm_vcpu *vcpu; 1108 struct kvm_mmu_page *parent[PT64_ROOT_LEVEL-1];
1053 struct kvm_unsync_walk walker; 1109 unsigned int idx[PT64_ROOT_LEVEL-1];
1054}; 1110};
1055 1111
1056static int mmu_sync_fn(struct kvm_mmu_page *sp, struct kvm_unsync_walk *walk) 1112#define for_each_sp(pvec, sp, parents, i) \
1113 for (i = mmu_pages_next(&pvec, &parents, -1), \
1114 sp = pvec.page[i].sp; \
1115 i < pvec.nr && ({ sp = pvec.page[i].sp; 1;}); \
1116 i = mmu_pages_next(&pvec, &parents, i))
1117
1118int mmu_pages_next(struct kvm_mmu_pages *pvec, struct mmu_page_path *parents,
1119 int i)
1057{ 1120{
1058 struct sync_walker *sync_walk = container_of(walk, struct sync_walker, 1121 int n;
1059 walker);
1060 struct kvm_vcpu *vcpu = sync_walk->vcpu;
1061 1122
1062 kvm_sync_page(vcpu, sp); 1123 for (n = i+1; n < pvec->nr; n++) {
1063 return (need_resched() || spin_needbreak(&vcpu->kvm->mmu_lock)); 1124 struct kvm_mmu_page *sp = pvec->page[n].sp;
1125
1126 if (sp->role.level == PT_PAGE_TABLE_LEVEL) {
1127 parents->idx[0] = pvec->page[n].idx;
1128 return n;
1129 }
1130
1131 parents->parent[sp->role.level-2] = sp;
1132 parents->idx[sp->role.level-1] = pvec->page[n].idx;
1133 }
1134
1135 return n;
1064} 1136}
1065 1137
1066static void mmu_sync_children(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) 1138void mmu_pages_clear_parents(struct mmu_page_path *parents)
1067{ 1139{
1068 struct sync_walker walker = { 1140 struct kvm_mmu_page *sp;
1069 .walker = { .entry = mmu_sync_fn, }, 1141 unsigned int level = 0;
1070 .vcpu = vcpu, 1142
1071 }; 1143 do {
1144 unsigned int idx = parents->idx[level];
1145
1146 sp = parents->parent[level];
1147 if (!sp)
1148 return;
1149
1150 --sp->unsync_children;
1151 WARN_ON((int)sp->unsync_children < 0);
1152 __clear_bit(idx, sp->unsync_child_bitmap);
1153 level++;
1154 } while (level < PT64_ROOT_LEVEL-1 && !sp->unsync_children);
1155}
1156
1157static void kvm_mmu_pages_init(struct kvm_mmu_page *parent,
1158 struct mmu_page_path *parents,
1159 struct kvm_mmu_pages *pvec)
1160{
1161 parents->parent[parent->role.level-1] = NULL;
1162 pvec->nr = 0;
1163}
1164
1165static void mmu_sync_children(struct kvm_vcpu *vcpu,
1166 struct kvm_mmu_page *parent)
1167{
1168 int i;
1169 struct kvm_mmu_page *sp;
1170 struct mmu_page_path parents;
1171 struct kvm_mmu_pages pages;
1172
1173 kvm_mmu_pages_init(parent, &parents, &pages);
1174 while (mmu_unsync_walk(parent, &pages)) {
1175 int protected = 0;
1072 1176
1073 while (mmu_unsync_walk(sp, &walker.walker)) 1177 for_each_sp(pages, sp, parents, i)
1178 protected |= rmap_write_protect(vcpu->kvm, sp->gfn);
1179
1180 if (protected)
1181 kvm_flush_remote_tlbs(vcpu->kvm);
1182
1183 for_each_sp(pages, sp, parents, i) {
1184 kvm_sync_page(vcpu, sp);
1185 mmu_pages_clear_parents(&parents);
1186 }
1074 cond_resched_lock(&vcpu->kvm->mmu_lock); 1187 cond_resched_lock(&vcpu->kvm->mmu_lock);
1188 kvm_mmu_pages_init(parent, &parents, &pages);
1189 }
1075} 1190}
1076 1191
1077static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu, 1192static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
@@ -1129,7 +1244,8 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
1129 sp->role = role; 1244 sp->role = role;
1130 hlist_add_head(&sp->hash_link, bucket); 1245 hlist_add_head(&sp->hash_link, bucket);
1131 if (!metaphysical) { 1246 if (!metaphysical) {
1132 rmap_write_protect(vcpu->kvm, gfn); 1247 if (rmap_write_protect(vcpu->kvm, gfn))
1248 kvm_flush_remote_tlbs(vcpu->kvm);
1133 account_shadowed(vcpu->kvm, gfn); 1249 account_shadowed(vcpu->kvm, gfn);
1134 } 1250 }
1135 if (shadow_trap_nonpresent_pte != shadow_notrap_nonpresent_pte) 1251 if (shadow_trap_nonpresent_pte != shadow_notrap_nonpresent_pte)
@@ -1153,6 +1269,8 @@ static int walk_shadow(struct kvm_shadow_walk *walker,
1153 if (level == PT32E_ROOT_LEVEL) { 1269 if (level == PT32E_ROOT_LEVEL) {
1154 shadow_addr = vcpu->arch.mmu.pae_root[(addr >> 30) & 3]; 1270 shadow_addr = vcpu->arch.mmu.pae_root[(addr >> 30) & 3];
1155 shadow_addr &= PT64_BASE_ADDR_MASK; 1271 shadow_addr &= PT64_BASE_ADDR_MASK;
1272 if (!shadow_addr)
1273 return 1;
1156 --level; 1274 --level;
1157 } 1275 }
1158 1276
@@ -1237,33 +1355,29 @@ static void kvm_mmu_unlink_parents(struct kvm *kvm, struct kvm_mmu_page *sp)
1237 } 1355 }
1238} 1356}
1239 1357
1240struct zap_walker { 1358static int mmu_zap_unsync_children(struct kvm *kvm,
1241 struct kvm_unsync_walk walker; 1359 struct kvm_mmu_page *parent)
1242 struct kvm *kvm;
1243 int zapped;
1244};
1245
1246static int mmu_zap_fn(struct kvm_mmu_page *sp, struct kvm_unsync_walk *walk)
1247{ 1360{
1248 struct zap_walker *zap_walk = container_of(walk, struct zap_walker, 1361 int i, zapped = 0;
1249 walker); 1362 struct mmu_page_path parents;
1250 kvm_mmu_zap_page(zap_walk->kvm, sp); 1363 struct kvm_mmu_pages pages;
1251 zap_walk->zapped = 1;
1252 return 0;
1253}
1254 1364
1255static int mmu_zap_unsync_children(struct kvm *kvm, struct kvm_mmu_page *sp) 1365 if (parent->role.level == PT_PAGE_TABLE_LEVEL)
1256{
1257 struct zap_walker walker = {
1258 .walker = { .entry = mmu_zap_fn, },
1259 .kvm = kvm,
1260 .zapped = 0,
1261 };
1262
1263 if (sp->role.level == PT_PAGE_TABLE_LEVEL)
1264 return 0; 1366 return 0;
1265 mmu_unsync_walk(sp, &walker.walker); 1367
1266 return walker.zapped; 1368 kvm_mmu_pages_init(parent, &parents, &pages);
1369 while (mmu_unsync_walk(parent, &pages)) {
1370 struct kvm_mmu_page *sp;
1371
1372 for_each_sp(pages, sp, parents, i) {
1373 kvm_mmu_zap_page(kvm, sp);
1374 mmu_pages_clear_parents(&parents);
1375 }
1376 zapped += pages.nr;
1377 kvm_mmu_pages_init(parent, &parents, &pages);
1378 }
1379
1380 return zapped;
1267} 1381}
1268 1382
1269static int kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp) 1383static int kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp)
@@ -1362,7 +1476,7 @@ static void page_header_update_slot(struct kvm *kvm, void *pte, gfn_t gfn)
1362 int slot = memslot_id(kvm, gfn_to_memslot(kvm, gfn)); 1476 int slot = memslot_id(kvm, gfn_to_memslot(kvm, gfn));
1363 struct kvm_mmu_page *sp = page_header(__pa(pte)); 1477 struct kvm_mmu_page *sp = page_header(__pa(pte));
1364 1478
1365 __set_bit(slot, &sp->slot_bitmap); 1479 __set_bit(slot, sp->slot_bitmap);
1366} 1480}
1367 1481
1368static void mmu_convert_notrap(struct kvm_mmu_page *sp) 1482static void mmu_convert_notrap(struct kvm_mmu_page *sp)
@@ -1393,6 +1507,110 @@ struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva)
1393 return page; 1507 return page;
1394} 1508}
1395 1509
1510/*
1511 * The function is based on mtrr_type_lookup() in
1512 * arch/x86/kernel/cpu/mtrr/generic.c
1513 */
1514static int get_mtrr_type(struct mtrr_state_type *mtrr_state,
1515 u64 start, u64 end)
1516{
1517 int i;
1518 u64 base, mask;
1519 u8 prev_match, curr_match;
1520 int num_var_ranges = KVM_NR_VAR_MTRR;
1521
1522 if (!mtrr_state->enabled)
1523 return 0xFF;
1524
1525 /* Make end inclusive end, instead of exclusive */
1526 end--;
1527
1528 /* Look in fixed ranges. Just return the type as per start */
1529 if (mtrr_state->have_fixed && (start < 0x100000)) {
1530 int idx;
1531
1532 if (start < 0x80000) {
1533 idx = 0;
1534 idx += (start >> 16);
1535 return mtrr_state->fixed_ranges[idx];
1536 } else if (start < 0xC0000) {
1537 idx = 1 * 8;
1538 idx += ((start - 0x80000) >> 14);
1539 return mtrr_state->fixed_ranges[idx];
1540 } else if (start < 0x1000000) {
1541 idx = 3 * 8;
1542 idx += ((start - 0xC0000) >> 12);
1543 return mtrr_state->fixed_ranges[idx];
1544 }
1545 }
1546
1547 /*
1548 * Look in variable ranges
1549 * Look of multiple ranges matching this address and pick type
1550 * as per MTRR precedence
1551 */
1552 if (!(mtrr_state->enabled & 2))
1553 return mtrr_state->def_type;
1554
1555 prev_match = 0xFF;
1556 for (i = 0; i < num_var_ranges; ++i) {
1557 unsigned short start_state, end_state;
1558
1559 if (!(mtrr_state->var_ranges[i].mask_lo & (1 << 11)))
1560 continue;
1561
1562 base = (((u64)mtrr_state->var_ranges[i].base_hi) << 32) +
1563 (mtrr_state->var_ranges[i].base_lo & PAGE_MASK);
1564 mask = (((u64)mtrr_state->var_ranges[i].mask_hi) << 32) +
1565 (mtrr_state->var_ranges[i].mask_lo & PAGE_MASK);
1566
1567 start_state = ((start & mask) == (base & mask));
1568 end_state = ((end & mask) == (base & mask));
1569 if (start_state != end_state)
1570 return 0xFE;
1571
1572 if ((start & mask) != (base & mask))
1573 continue;
1574
1575 curr_match = mtrr_state->var_ranges[i].base_lo & 0xff;
1576 if (prev_match == 0xFF) {
1577 prev_match = curr_match;
1578 continue;
1579 }
1580
1581 if (prev_match == MTRR_TYPE_UNCACHABLE ||
1582 curr_match == MTRR_TYPE_UNCACHABLE)
1583 return MTRR_TYPE_UNCACHABLE;
1584
1585 if ((prev_match == MTRR_TYPE_WRBACK &&
1586 curr_match == MTRR_TYPE_WRTHROUGH) ||
1587 (prev_match == MTRR_TYPE_WRTHROUGH &&
1588 curr_match == MTRR_TYPE_WRBACK)) {
1589 prev_match = MTRR_TYPE_WRTHROUGH;
1590 curr_match = MTRR_TYPE_WRTHROUGH;
1591 }
1592
1593 if (prev_match != curr_match)
1594 return MTRR_TYPE_UNCACHABLE;
1595 }
1596
1597 if (prev_match != 0xFF)
1598 return prev_match;
1599
1600 return mtrr_state->def_type;
1601}
1602
1603static u8 get_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn)
1604{
1605 u8 mtrr;
1606
1607 mtrr = get_mtrr_type(&vcpu->arch.mtrr_state, gfn << PAGE_SHIFT,
1608 (gfn << PAGE_SHIFT) + PAGE_SIZE);
1609 if (mtrr == 0xfe || mtrr == 0xff)
1610 mtrr = MTRR_TYPE_WRBACK;
1611 return mtrr;
1612}
1613
1396static int kvm_unsync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) 1614static int kvm_unsync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
1397{ 1615{
1398 unsigned index; 1616 unsigned index;
@@ -1409,9 +1627,15 @@ static int kvm_unsync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
1409 if (s->role.word != sp->role.word) 1627 if (s->role.word != sp->role.word)
1410 return 1; 1628 return 1;
1411 } 1629 }
1412 kvm_mmu_mark_parents_unsync(vcpu, sp);
1413 ++vcpu->kvm->stat.mmu_unsync; 1630 ++vcpu->kvm->stat.mmu_unsync;
1414 sp->unsync = 1; 1631 sp->unsync = 1;
1632
1633 if (sp->global) {
1634 list_add(&sp->oos_link, &vcpu->kvm->arch.oos_global_pages);
1635 ++vcpu->kvm->stat.mmu_unsync_global;
1636 } else
1637 kvm_mmu_mark_parents_unsync(vcpu, sp);
1638
1415 mmu_convert_notrap(sp); 1639 mmu_convert_notrap(sp);
1416 return 0; 1640 return 0;
1417} 1641}
@@ -1437,11 +1661,24 @@ static int mmu_need_write_protect(struct kvm_vcpu *vcpu, gfn_t gfn,
1437static int set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte, 1661static int set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
1438 unsigned pte_access, int user_fault, 1662 unsigned pte_access, int user_fault,
1439 int write_fault, int dirty, int largepage, 1663 int write_fault, int dirty, int largepage,
1440 gfn_t gfn, pfn_t pfn, bool speculative, 1664 int global, gfn_t gfn, pfn_t pfn, bool speculative,
1441 bool can_unsync) 1665 bool can_unsync)
1442{ 1666{
1443 u64 spte; 1667 u64 spte;
1444 int ret = 0; 1668 int ret = 0;
1669 u64 mt_mask = shadow_mt_mask;
1670 struct kvm_mmu_page *sp = page_header(__pa(shadow_pte));
1671
1672 if (!(vcpu->arch.cr4 & X86_CR4_PGE))
1673 global = 0;
1674 if (!global && sp->global) {
1675 sp->global = 0;
1676 if (sp->unsync) {
1677 kvm_unlink_unsync_global(vcpu->kvm, sp);
1678 kvm_mmu_mark_parents_unsync(vcpu, sp);
1679 }
1680 }
1681
1445 /* 1682 /*
1446 * We don't set the accessed bit, since we sometimes want to see 1683 * We don't set the accessed bit, since we sometimes want to see
1447 * whether the guest actually used the pte (in order to detect 1684 * whether the guest actually used the pte (in order to detect
@@ -1460,6 +1697,11 @@ static int set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
1460 spte |= shadow_user_mask; 1697 spte |= shadow_user_mask;
1461 if (largepage) 1698 if (largepage)
1462 spte |= PT_PAGE_SIZE_MASK; 1699 spte |= PT_PAGE_SIZE_MASK;
1700 if (mt_mask) {
1701 mt_mask = get_memory_type(vcpu, gfn) <<
1702 kvm_x86_ops->get_mt_mask_shift();
1703 spte |= mt_mask;
1704 }
1463 1705
1464 spte |= (u64)pfn << PAGE_SHIFT; 1706 spte |= (u64)pfn << PAGE_SHIFT;
1465 1707
@@ -1474,6 +1716,15 @@ static int set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
1474 1716
1475 spte |= PT_WRITABLE_MASK; 1717 spte |= PT_WRITABLE_MASK;
1476 1718
1719 /*
1720 * Optimization: for pte sync, if spte was writable the hash
1721 * lookup is unnecessary (and expensive). Write protection
1722 * is responsibility of mmu_get_page / kvm_sync_page.
1723 * Same reasoning can be applied to dirty page accounting.
1724 */
1725 if (!can_unsync && is_writeble_pte(*shadow_pte))
1726 goto set_pte;
1727
1477 if (mmu_need_write_protect(vcpu, gfn, can_unsync)) { 1728 if (mmu_need_write_protect(vcpu, gfn, can_unsync)) {
1478 pgprintk("%s: found shadow page for %lx, marking ro\n", 1729 pgprintk("%s: found shadow page for %lx, marking ro\n",
1479 __func__, gfn); 1730 __func__, gfn);
@@ -1495,8 +1746,8 @@ set_pte:
1495static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte, 1746static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
1496 unsigned pt_access, unsigned pte_access, 1747 unsigned pt_access, unsigned pte_access,
1497 int user_fault, int write_fault, int dirty, 1748 int user_fault, int write_fault, int dirty,
1498 int *ptwrite, int largepage, gfn_t gfn, 1749 int *ptwrite, int largepage, int global,
1499 pfn_t pfn, bool speculative) 1750 gfn_t gfn, pfn_t pfn, bool speculative)
1500{ 1751{
1501 int was_rmapped = 0; 1752 int was_rmapped = 0;
1502 int was_writeble = is_writeble_pte(*shadow_pte); 1753 int was_writeble = is_writeble_pte(*shadow_pte);
@@ -1529,7 +1780,7 @@ static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
1529 } 1780 }
1530 } 1781 }
1531 if (set_spte(vcpu, shadow_pte, pte_access, user_fault, write_fault, 1782 if (set_spte(vcpu, shadow_pte, pte_access, user_fault, write_fault,
1532 dirty, largepage, gfn, pfn, speculative, true)) { 1783 dirty, largepage, global, gfn, pfn, speculative, true)) {
1533 if (write_fault) 1784 if (write_fault)
1534 *ptwrite = 1; 1785 *ptwrite = 1;
1535 kvm_x86_ops->tlb_flush(vcpu); 1786 kvm_x86_ops->tlb_flush(vcpu);
@@ -1586,7 +1837,7 @@ static int direct_map_entry(struct kvm_shadow_walk *_walk,
1586 || (walk->largepage && level == PT_DIRECTORY_LEVEL)) { 1837 || (walk->largepage && level == PT_DIRECTORY_LEVEL)) {
1587 mmu_set_spte(vcpu, sptep, ACC_ALL, ACC_ALL, 1838 mmu_set_spte(vcpu, sptep, ACC_ALL, ACC_ALL,
1588 0, walk->write, 1, &walk->pt_write, 1839 0, walk->write, 1, &walk->pt_write,
1589 walk->largepage, gfn, walk->pfn, false); 1840 walk->largepage, 0, gfn, walk->pfn, false);
1590 ++vcpu->stat.pf_fixed; 1841 ++vcpu->stat.pf_fixed;
1591 return 1; 1842 return 1;
1592 } 1843 }
@@ -1773,6 +2024,15 @@ static void mmu_sync_roots(struct kvm_vcpu *vcpu)
1773 } 2024 }
1774} 2025}
1775 2026
2027static void mmu_sync_global(struct kvm_vcpu *vcpu)
2028{
2029 struct kvm *kvm = vcpu->kvm;
2030 struct kvm_mmu_page *sp, *n;
2031
2032 list_for_each_entry_safe(sp, n, &kvm->arch.oos_global_pages, oos_link)
2033 kvm_sync_page(vcpu, sp);
2034}
2035
1776void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu) 2036void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu)
1777{ 2037{
1778 spin_lock(&vcpu->kvm->mmu_lock); 2038 spin_lock(&vcpu->kvm->mmu_lock);
@@ -1780,6 +2040,13 @@ void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu)
1780 spin_unlock(&vcpu->kvm->mmu_lock); 2040 spin_unlock(&vcpu->kvm->mmu_lock);
1781} 2041}
1782 2042
2043void kvm_mmu_sync_global(struct kvm_vcpu *vcpu)
2044{
2045 spin_lock(&vcpu->kvm->mmu_lock);
2046 mmu_sync_global(vcpu);
2047 spin_unlock(&vcpu->kvm->mmu_lock);
2048}
2049
1783static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr) 2050static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr)
1784{ 2051{
1785 return vaddr; 2052 return vaddr;
@@ -2178,7 +2445,8 @@ static void kvm_mmu_access_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2178} 2445}
2179 2446
2180void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa, 2447void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
2181 const u8 *new, int bytes) 2448 const u8 *new, int bytes,
2449 bool guest_initiated)
2182{ 2450{
2183 gfn_t gfn = gpa >> PAGE_SHIFT; 2451 gfn_t gfn = gpa >> PAGE_SHIFT;
2184 struct kvm_mmu_page *sp; 2452 struct kvm_mmu_page *sp;
@@ -2204,15 +2472,17 @@ void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
2204 kvm_mmu_free_some_pages(vcpu); 2472 kvm_mmu_free_some_pages(vcpu);
2205 ++vcpu->kvm->stat.mmu_pte_write; 2473 ++vcpu->kvm->stat.mmu_pte_write;
2206 kvm_mmu_audit(vcpu, "pre pte write"); 2474 kvm_mmu_audit(vcpu, "pre pte write");
2207 if (gfn == vcpu->arch.last_pt_write_gfn 2475 if (guest_initiated) {
2208 && !last_updated_pte_accessed(vcpu)) { 2476 if (gfn == vcpu->arch.last_pt_write_gfn
2209 ++vcpu->arch.last_pt_write_count; 2477 && !last_updated_pte_accessed(vcpu)) {
2210 if (vcpu->arch.last_pt_write_count >= 3) 2478 ++vcpu->arch.last_pt_write_count;
2211 flooded = 1; 2479 if (vcpu->arch.last_pt_write_count >= 3)
2212 } else { 2480 flooded = 1;
2213 vcpu->arch.last_pt_write_gfn = gfn; 2481 } else {
2214 vcpu->arch.last_pt_write_count = 1; 2482 vcpu->arch.last_pt_write_gfn = gfn;
2215 vcpu->arch.last_pte_updated = NULL; 2483 vcpu->arch.last_pt_write_count = 1;
2484 vcpu->arch.last_pte_updated = NULL;
2485 }
2216 } 2486 }
2217 index = kvm_page_table_hashfn(gfn); 2487 index = kvm_page_table_hashfn(gfn);
2218 bucket = &vcpu->kvm->arch.mmu_page_hash[index]; 2488 bucket = &vcpu->kvm->arch.mmu_page_hash[index];
@@ -2352,9 +2622,7 @@ EXPORT_SYMBOL_GPL(kvm_mmu_page_fault);
2352 2622
2353void kvm_mmu_invlpg(struct kvm_vcpu *vcpu, gva_t gva) 2623void kvm_mmu_invlpg(struct kvm_vcpu *vcpu, gva_t gva)
2354{ 2624{
2355 spin_lock(&vcpu->kvm->mmu_lock);
2356 vcpu->arch.mmu.invlpg(vcpu, gva); 2625 vcpu->arch.mmu.invlpg(vcpu, gva);
2357 spin_unlock(&vcpu->kvm->mmu_lock);
2358 kvm_mmu_flush_tlb(vcpu); 2626 kvm_mmu_flush_tlb(vcpu);
2359 ++vcpu->stat.invlpg; 2627 ++vcpu->stat.invlpg;
2360} 2628}
@@ -2451,7 +2719,7 @@ void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot)
2451 int i; 2719 int i;
2452 u64 *pt; 2720 u64 *pt;
2453 2721
2454 if (!test_bit(slot, &sp->slot_bitmap)) 2722 if (!test_bit(slot, sp->slot_bitmap))
2455 continue; 2723 continue;
2456 2724
2457 pt = sp->spt; 2725 pt = sp->spt;
@@ -2860,8 +3128,8 @@ static void audit_write_protection(struct kvm_vcpu *vcpu)
2860 if (sp->role.metaphysical) 3128 if (sp->role.metaphysical)
2861 continue; 3129 continue;
2862 3130
2863 slot = gfn_to_memslot(vcpu->kvm, sp->gfn);
2864 gfn = unalias_gfn(vcpu->kvm, sp->gfn); 3131 gfn = unalias_gfn(vcpu->kvm, sp->gfn);
3132 slot = gfn_to_memslot_unaliased(vcpu->kvm, sp->gfn);
2865 rmapp = &slot->rmap[gfn - slot->base_gfn]; 3133 rmapp = &slot->rmap[gfn - slot->base_gfn];
2866 if (*rmapp) 3134 if (*rmapp)
2867 printk(KERN_ERR "%s: (%s) shadow page has writable" 3135 printk(KERN_ERR "%s: (%s) shadow page has writable"
diff --git a/arch/x86/kvm/paging_tmpl.h b/arch/x86/kvm/paging_tmpl.h
index 84eee43bbe74..9fd78b6e17ad 100644
--- a/arch/x86/kvm/paging_tmpl.h
+++ b/arch/x86/kvm/paging_tmpl.h
@@ -82,6 +82,7 @@ struct shadow_walker {
82 int *ptwrite; 82 int *ptwrite;
83 pfn_t pfn; 83 pfn_t pfn;
84 u64 *sptep; 84 u64 *sptep;
85 gpa_t pte_gpa;
85}; 86};
86 87
87static gfn_t gpte_to_gfn(pt_element_t gpte) 88static gfn_t gpte_to_gfn(pt_element_t gpte)
@@ -222,7 +223,7 @@ walk:
222 if (ret) 223 if (ret)
223 goto walk; 224 goto walk;
224 pte |= PT_DIRTY_MASK; 225 pte |= PT_DIRTY_MASK;
225 kvm_mmu_pte_write(vcpu, pte_gpa, (u8 *)&pte, sizeof(pte)); 226 kvm_mmu_pte_write(vcpu, pte_gpa, (u8 *)&pte, sizeof(pte), 0);
226 walker->ptes[walker->level - 1] = pte; 227 walker->ptes[walker->level - 1] = pte;
227 } 228 }
228 229
@@ -274,7 +275,8 @@ static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page,
274 return; 275 return;
275 kvm_get_pfn(pfn); 276 kvm_get_pfn(pfn);
276 mmu_set_spte(vcpu, spte, page->role.access, pte_access, 0, 0, 277 mmu_set_spte(vcpu, spte, page->role.access, pte_access, 0, 0,
277 gpte & PT_DIRTY_MASK, NULL, largepage, gpte_to_gfn(gpte), 278 gpte & PT_DIRTY_MASK, NULL, largepage,
279 gpte & PT_GLOBAL_MASK, gpte_to_gfn(gpte),
278 pfn, true); 280 pfn, true);
279} 281}
280 282
@@ -301,8 +303,9 @@ static int FNAME(shadow_walk_entry)(struct kvm_shadow_walk *_sw,
301 mmu_set_spte(vcpu, sptep, access, gw->pte_access & access, 303 mmu_set_spte(vcpu, sptep, access, gw->pte_access & access,
302 sw->user_fault, sw->write_fault, 304 sw->user_fault, sw->write_fault,
303 gw->ptes[gw->level-1] & PT_DIRTY_MASK, 305 gw->ptes[gw->level-1] & PT_DIRTY_MASK,
304 sw->ptwrite, sw->largepage, gw->gfn, sw->pfn, 306 sw->ptwrite, sw->largepage,
305 false); 307 gw->ptes[gw->level-1] & PT_GLOBAL_MASK,
308 gw->gfn, sw->pfn, false);
306 sw->sptep = sptep; 309 sw->sptep = sptep;
307 return 1; 310 return 1;
308 } 311 }
@@ -466,10 +469,22 @@ static int FNAME(shadow_invlpg_entry)(struct kvm_shadow_walk *_sw,
466 struct kvm_vcpu *vcpu, u64 addr, 469 struct kvm_vcpu *vcpu, u64 addr,
467 u64 *sptep, int level) 470 u64 *sptep, int level)
468{ 471{
472 struct shadow_walker *sw =
473 container_of(_sw, struct shadow_walker, walker);
469 474
470 if (level == PT_PAGE_TABLE_LEVEL) { 475 /* FIXME: properly handle invlpg on large guest pages */
471 if (is_shadow_present_pte(*sptep)) 476 if (level == PT_PAGE_TABLE_LEVEL ||
477 ((level == PT_DIRECTORY_LEVEL) && is_large_pte(*sptep))) {
478 struct kvm_mmu_page *sp = page_header(__pa(sptep));
479
480 sw->pte_gpa = (sp->gfn << PAGE_SHIFT);
481 sw->pte_gpa += (sptep - sp->spt) * sizeof(pt_element_t);
482
483 if (is_shadow_present_pte(*sptep)) {
472 rmap_remove(vcpu->kvm, sptep); 484 rmap_remove(vcpu->kvm, sptep);
485 if (is_large_pte(*sptep))
486 --vcpu->kvm->stat.lpages;
487 }
473 set_shadow_pte(sptep, shadow_trap_nonpresent_pte); 488 set_shadow_pte(sptep, shadow_trap_nonpresent_pte);
474 return 1; 489 return 1;
475 } 490 }
@@ -480,11 +495,26 @@ static int FNAME(shadow_invlpg_entry)(struct kvm_shadow_walk *_sw,
480 495
481static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva) 496static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
482{ 497{
498 pt_element_t gpte;
483 struct shadow_walker walker = { 499 struct shadow_walker walker = {
484 .walker = { .entry = FNAME(shadow_invlpg_entry), }, 500 .walker = { .entry = FNAME(shadow_invlpg_entry), },
501 .pte_gpa = -1,
485 }; 502 };
486 503
504 spin_lock(&vcpu->kvm->mmu_lock);
487 walk_shadow(&walker.walker, vcpu, gva); 505 walk_shadow(&walker.walker, vcpu, gva);
506 spin_unlock(&vcpu->kvm->mmu_lock);
507 if (walker.pte_gpa == -1)
508 return;
509 if (kvm_read_guest_atomic(vcpu->kvm, walker.pte_gpa, &gpte,
510 sizeof(pt_element_t)))
511 return;
512 if (is_present_pte(gpte) && (gpte & PT_ACCESSED_MASK)) {
513 if (mmu_topup_memory_caches(vcpu))
514 return;
515 kvm_mmu_pte_write(vcpu, walker.pte_gpa, (const u8 *)&gpte,
516 sizeof(pt_element_t), 0);
517 }
488} 518}
489 519
490static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr) 520static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr)
@@ -580,7 +610,7 @@ static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
580 nr_present++; 610 nr_present++;
581 pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte); 611 pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte);
582 set_spte(vcpu, &sp->spt[i], pte_access, 0, 0, 612 set_spte(vcpu, &sp->spt[i], pte_access, 0, 0,
583 is_dirty_pte(gpte), 0, gfn, 613 is_dirty_pte(gpte), 0, gpte & PT_GLOBAL_MASK, gfn,
584 spte_to_pfn(sp->spt[i]), true, false); 614 spte_to_pfn(sp->spt[i]), true, false);
585 } 615 }
586 616
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index 9c4ce657d963..1452851ae258 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -28,6 +28,8 @@
28 28
29#include <asm/desc.h> 29#include <asm/desc.h>
30 30
31#include <asm/virtext.h>
32
31#define __ex(x) __kvm_handle_fault_on_reboot(x) 33#define __ex(x) __kvm_handle_fault_on_reboot(x)
32 34
33MODULE_AUTHOR("Qumranet"); 35MODULE_AUTHOR("Qumranet");
@@ -245,34 +247,19 @@ static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
245 247
246static int has_svm(void) 248static int has_svm(void)
247{ 249{
248 uint32_t eax, ebx, ecx, edx; 250 const char *msg;
249
250 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
251 printk(KERN_INFO "has_svm: not amd\n");
252 return 0;
253 }
254 251
255 cpuid(0x80000000, &eax, &ebx, &ecx, &edx); 252 if (!cpu_has_svm(&msg)) {
256 if (eax < SVM_CPUID_FUNC) { 253 printk(KERN_INFO "has_svn: %s\n", msg);
257 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
258 return 0; 254 return 0;
259 } 255 }
260 256
261 cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
262 if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
263 printk(KERN_DEBUG "has_svm: svm not available\n");
264 return 0;
265 }
266 return 1; 257 return 1;
267} 258}
268 259
269static void svm_hardware_disable(void *garbage) 260static void svm_hardware_disable(void *garbage)
270{ 261{
271 uint64_t efer; 262 cpu_svm_disable();
272
273 wrmsrl(MSR_VM_HSAVE_PA, 0);
274 rdmsrl(MSR_EFER, efer);
275 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
276} 263}
277 264
278static void svm_hardware_enable(void *garbage) 265static void svm_hardware_enable(void *garbage)
@@ -772,6 +759,22 @@ static void svm_get_segment(struct kvm_vcpu *vcpu,
772 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1; 759 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
773 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1; 760 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
774 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1; 761 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
762
763 /*
764 * SVM always stores 0 for the 'G' bit in the CS selector in
765 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
766 * Intel's VMENTRY has a check on the 'G' bit.
767 */
768 if (seg == VCPU_SREG_CS)
769 var->g = s->limit > 0xfffff;
770
771 /*
772 * Work around a bug where the busy flag in the tr selector
773 * isn't exposed
774 */
775 if (seg == VCPU_SREG_TR)
776 var->type |= 0x2;
777
775 var->unusable = !var->present; 778 var->unusable = !var->present;
776} 779}
777 780
@@ -1099,6 +1102,7 @@ static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1099 rep = (io_info & SVM_IOIO_REP_MASK) != 0; 1102 rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1100 down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0; 1103 down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1101 1104
1105 skip_emulated_instruction(&svm->vcpu);
1102 return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port); 1106 return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1103} 1107}
1104 1108
@@ -1912,6 +1916,11 @@ static int get_npt_level(void)
1912#endif 1916#endif
1913} 1917}
1914 1918
1919static int svm_get_mt_mask_shift(void)
1920{
1921 return 0;
1922}
1923
1915static struct kvm_x86_ops svm_x86_ops = { 1924static struct kvm_x86_ops svm_x86_ops = {
1916 .cpu_has_kvm_support = has_svm, 1925 .cpu_has_kvm_support = has_svm,
1917 .disabled_by_bios = is_disabled, 1926 .disabled_by_bios = is_disabled,
@@ -1967,6 +1976,7 @@ static struct kvm_x86_ops svm_x86_ops = {
1967 1976
1968 .set_tss_addr = svm_set_tss_addr, 1977 .set_tss_addr = svm_set_tss_addr,
1969 .get_tdp_level = get_npt_level, 1978 .get_tdp_level = get_npt_level,
1979 .get_mt_mask_shift = svm_get_mt_mask_shift,
1970}; 1980};
1971 1981
1972static int __init svm_init(void) 1982static int __init svm_init(void)
diff --git a/arch/x86/kvm/svm.h b/arch/x86/kvm/svm.h
deleted file mode 100644
index 1b8afa78e869..000000000000
--- a/arch/x86/kvm/svm.h
+++ /dev/null
@@ -1,328 +0,0 @@
1#ifndef __SVM_H
2#define __SVM_H
3
4enum {
5 INTERCEPT_INTR,
6 INTERCEPT_NMI,
7 INTERCEPT_SMI,
8 INTERCEPT_INIT,
9 INTERCEPT_VINTR,
10 INTERCEPT_SELECTIVE_CR0,
11 INTERCEPT_STORE_IDTR,
12 INTERCEPT_STORE_GDTR,
13 INTERCEPT_STORE_LDTR,
14 INTERCEPT_STORE_TR,
15 INTERCEPT_LOAD_IDTR,
16 INTERCEPT_LOAD_GDTR,
17 INTERCEPT_LOAD_LDTR,
18 INTERCEPT_LOAD_TR,
19 INTERCEPT_RDTSC,
20 INTERCEPT_RDPMC,
21 INTERCEPT_PUSHF,
22 INTERCEPT_POPF,
23 INTERCEPT_CPUID,
24 INTERCEPT_RSM,
25 INTERCEPT_IRET,
26 INTERCEPT_INTn,
27 INTERCEPT_INVD,
28 INTERCEPT_PAUSE,
29 INTERCEPT_HLT,
30 INTERCEPT_INVLPG,
31 INTERCEPT_INVLPGA,
32 INTERCEPT_IOIO_PROT,
33 INTERCEPT_MSR_PROT,
34 INTERCEPT_TASK_SWITCH,
35 INTERCEPT_FERR_FREEZE,
36 INTERCEPT_SHUTDOWN,
37 INTERCEPT_VMRUN,
38 INTERCEPT_VMMCALL,
39 INTERCEPT_VMLOAD,
40 INTERCEPT_VMSAVE,
41 INTERCEPT_STGI,
42 INTERCEPT_CLGI,
43 INTERCEPT_SKINIT,
44 INTERCEPT_RDTSCP,
45 INTERCEPT_ICEBP,
46 INTERCEPT_WBINVD,
47 INTERCEPT_MONITOR,
48 INTERCEPT_MWAIT,
49 INTERCEPT_MWAIT_COND,
50};
51
52
53struct __attribute__ ((__packed__)) vmcb_control_area {
54 u16 intercept_cr_read;
55 u16 intercept_cr_write;
56 u16 intercept_dr_read;
57 u16 intercept_dr_write;
58 u32 intercept_exceptions;
59 u64 intercept;
60 u8 reserved_1[44];
61 u64 iopm_base_pa;
62 u64 msrpm_base_pa;
63 u64 tsc_offset;
64 u32 asid;
65 u8 tlb_ctl;
66 u8 reserved_2[3];
67 u32 int_ctl;
68 u32 int_vector;
69 u32 int_state;
70 u8 reserved_3[4];
71 u32 exit_code;
72 u32 exit_code_hi;
73 u64 exit_info_1;
74 u64 exit_info_2;
75 u32 exit_int_info;
76 u32 exit_int_info_err;
77 u64 nested_ctl;
78 u8 reserved_4[16];
79 u32 event_inj;
80 u32 event_inj_err;
81 u64 nested_cr3;
82 u64 lbr_ctl;
83 u8 reserved_5[832];
84};
85
86
87#define TLB_CONTROL_DO_NOTHING 0
88#define TLB_CONTROL_FLUSH_ALL_ASID 1
89
90#define V_TPR_MASK 0x0f
91
92#define V_IRQ_SHIFT 8
93#define V_IRQ_MASK (1 << V_IRQ_SHIFT)
94
95#define V_INTR_PRIO_SHIFT 16
96#define V_INTR_PRIO_MASK (0x0f << V_INTR_PRIO_SHIFT)
97
98#define V_IGN_TPR_SHIFT 20
99#define V_IGN_TPR_MASK (1 << V_IGN_TPR_SHIFT)
100
101#define V_INTR_MASKING_SHIFT 24
102#define V_INTR_MASKING_MASK (1 << V_INTR_MASKING_SHIFT)
103
104#define SVM_INTERRUPT_SHADOW_MASK 1
105
106#define SVM_IOIO_STR_SHIFT 2
107#define SVM_IOIO_REP_SHIFT 3
108#define SVM_IOIO_SIZE_SHIFT 4
109#define SVM_IOIO_ASIZE_SHIFT 7
110
111#define SVM_IOIO_TYPE_MASK 1
112#define SVM_IOIO_STR_MASK (1 << SVM_IOIO_STR_SHIFT)
113#define SVM_IOIO_REP_MASK (1 << SVM_IOIO_REP_SHIFT)
114#define SVM_IOIO_SIZE_MASK (7 << SVM_IOIO_SIZE_SHIFT)
115#define SVM_IOIO_ASIZE_MASK (7 << SVM_IOIO_ASIZE_SHIFT)
116
117struct __attribute__ ((__packed__)) vmcb_seg {
118 u16 selector;
119 u16 attrib;
120 u32 limit;
121 u64 base;
122};
123
124struct __attribute__ ((__packed__)) vmcb_save_area {
125 struct vmcb_seg es;
126 struct vmcb_seg cs;
127 struct vmcb_seg ss;
128 struct vmcb_seg ds;
129 struct vmcb_seg fs;
130 struct vmcb_seg gs;
131 struct vmcb_seg gdtr;
132 struct vmcb_seg ldtr;
133 struct vmcb_seg idtr;
134 struct vmcb_seg tr;
135 u8 reserved_1[43];
136 u8 cpl;
137 u8 reserved_2[4];
138 u64 efer;
139 u8 reserved_3[112];
140 u64 cr4;
141 u64 cr3;
142 u64 cr0;
143 u64 dr7;
144 u64 dr6;
145 u64 rflags;
146 u64 rip;
147 u8 reserved_4[88];
148 u64 rsp;
149 u8 reserved_5[24];
150 u64 rax;
151 u64 star;
152 u64 lstar;
153 u64 cstar;
154 u64 sfmask;
155 u64 kernel_gs_base;
156 u64 sysenter_cs;
157 u64 sysenter_esp;
158 u64 sysenter_eip;
159 u64 cr2;
160 u8 reserved_6[32];
161 u64 g_pat;
162 u64 dbgctl;
163 u64 br_from;
164 u64 br_to;
165 u64 last_excp_from;
166 u64 last_excp_to;
167};
168
169struct __attribute__ ((__packed__)) vmcb {
170 struct vmcb_control_area control;
171 struct vmcb_save_area save;
172};
173
174#define SVM_CPUID_FEATURE_SHIFT 2
175#define SVM_CPUID_FUNC 0x8000000a
176
177#define MSR_EFER_SVME_MASK (1ULL << 12)
178#define MSR_VM_CR 0xc0010114
179#define MSR_VM_HSAVE_PA 0xc0010117ULL
180
181#define SVM_VM_CR_SVM_DISABLE 4
182
183#define SVM_SELECTOR_S_SHIFT 4
184#define SVM_SELECTOR_DPL_SHIFT 5
185#define SVM_SELECTOR_P_SHIFT 7
186#define SVM_SELECTOR_AVL_SHIFT 8
187#define SVM_SELECTOR_L_SHIFT 9
188#define SVM_SELECTOR_DB_SHIFT 10
189#define SVM_SELECTOR_G_SHIFT 11
190
191#define SVM_SELECTOR_TYPE_MASK (0xf)
192#define SVM_SELECTOR_S_MASK (1 << SVM_SELECTOR_S_SHIFT)
193#define SVM_SELECTOR_DPL_MASK (3 << SVM_SELECTOR_DPL_SHIFT)
194#define SVM_SELECTOR_P_MASK (1 << SVM_SELECTOR_P_SHIFT)
195#define SVM_SELECTOR_AVL_MASK (1 << SVM_SELECTOR_AVL_SHIFT)
196#define SVM_SELECTOR_L_MASK (1 << SVM_SELECTOR_L_SHIFT)
197#define SVM_SELECTOR_DB_MASK (1 << SVM_SELECTOR_DB_SHIFT)
198#define SVM_SELECTOR_G_MASK (1 << SVM_SELECTOR_G_SHIFT)
199
200#define SVM_SELECTOR_WRITE_MASK (1 << 1)
201#define SVM_SELECTOR_READ_MASK SVM_SELECTOR_WRITE_MASK
202#define SVM_SELECTOR_CODE_MASK (1 << 3)
203
204#define INTERCEPT_CR0_MASK 1
205#define INTERCEPT_CR3_MASK (1 << 3)
206#define INTERCEPT_CR4_MASK (1 << 4)
207#define INTERCEPT_CR8_MASK (1 << 8)
208
209#define INTERCEPT_DR0_MASK 1
210#define INTERCEPT_DR1_MASK (1 << 1)
211#define INTERCEPT_DR2_MASK (1 << 2)
212#define INTERCEPT_DR3_MASK (1 << 3)
213#define INTERCEPT_DR4_MASK (1 << 4)
214#define INTERCEPT_DR5_MASK (1 << 5)
215#define INTERCEPT_DR6_MASK (1 << 6)
216#define INTERCEPT_DR7_MASK (1 << 7)
217
218#define SVM_EVTINJ_VEC_MASK 0xff
219
220#define SVM_EVTINJ_TYPE_SHIFT 8
221#define SVM_EVTINJ_TYPE_MASK (7 << SVM_EVTINJ_TYPE_SHIFT)
222
223#define SVM_EVTINJ_TYPE_INTR (0 << SVM_EVTINJ_TYPE_SHIFT)
224#define SVM_EVTINJ_TYPE_NMI (2 << SVM_EVTINJ_TYPE_SHIFT)
225#define SVM_EVTINJ_TYPE_EXEPT (3 << SVM_EVTINJ_TYPE_SHIFT)
226#define SVM_EVTINJ_TYPE_SOFT (4 << SVM_EVTINJ_TYPE_SHIFT)
227
228#define SVM_EVTINJ_VALID (1 << 31)
229#define SVM_EVTINJ_VALID_ERR (1 << 11)
230
231#define SVM_EXITINTINFO_VEC_MASK SVM_EVTINJ_VEC_MASK
232
233#define SVM_EXITINTINFO_TYPE_INTR SVM_EVTINJ_TYPE_INTR
234#define SVM_EXITINTINFO_TYPE_NMI SVM_EVTINJ_TYPE_NMI
235#define SVM_EXITINTINFO_TYPE_EXEPT SVM_EVTINJ_TYPE_EXEPT
236#define SVM_EXITINTINFO_TYPE_SOFT SVM_EVTINJ_TYPE_SOFT
237
238#define SVM_EXITINTINFO_VALID SVM_EVTINJ_VALID
239#define SVM_EXITINTINFO_VALID_ERR SVM_EVTINJ_VALID_ERR
240
241#define SVM_EXITINFOSHIFT_TS_REASON_IRET 36
242#define SVM_EXITINFOSHIFT_TS_REASON_JMP 38
243
244#define SVM_EXIT_READ_CR0 0x000
245#define SVM_EXIT_READ_CR3 0x003
246#define SVM_EXIT_READ_CR4 0x004
247#define SVM_EXIT_READ_CR8 0x008
248#define SVM_EXIT_WRITE_CR0 0x010
249#define SVM_EXIT_WRITE_CR3 0x013
250#define SVM_EXIT_WRITE_CR4 0x014
251#define SVM_EXIT_WRITE_CR8 0x018
252#define SVM_EXIT_READ_DR0 0x020
253#define SVM_EXIT_READ_DR1 0x021
254#define SVM_EXIT_READ_DR2 0x022
255#define SVM_EXIT_READ_DR3 0x023
256#define SVM_EXIT_READ_DR4 0x024
257#define SVM_EXIT_READ_DR5 0x025
258#define SVM_EXIT_READ_DR6 0x026
259#define SVM_EXIT_READ_DR7 0x027
260#define SVM_EXIT_WRITE_DR0 0x030
261#define SVM_EXIT_WRITE_DR1 0x031
262#define SVM_EXIT_WRITE_DR2 0x032
263#define SVM_EXIT_WRITE_DR3 0x033
264#define SVM_EXIT_WRITE_DR4 0x034
265#define SVM_EXIT_WRITE_DR5 0x035
266#define SVM_EXIT_WRITE_DR6 0x036
267#define SVM_EXIT_WRITE_DR7 0x037
268#define SVM_EXIT_EXCP_BASE 0x040
269#define SVM_EXIT_INTR 0x060
270#define SVM_EXIT_NMI 0x061
271#define SVM_EXIT_SMI 0x062
272#define SVM_EXIT_INIT 0x063
273#define SVM_EXIT_VINTR 0x064
274#define SVM_EXIT_CR0_SEL_WRITE 0x065
275#define SVM_EXIT_IDTR_READ 0x066
276#define SVM_EXIT_GDTR_READ 0x067
277#define SVM_EXIT_LDTR_READ 0x068
278#define SVM_EXIT_TR_READ 0x069
279#define SVM_EXIT_IDTR_WRITE 0x06a
280#define SVM_EXIT_GDTR_WRITE 0x06b
281#define SVM_EXIT_LDTR_WRITE 0x06c
282#define SVM_EXIT_TR_WRITE 0x06d
283#define SVM_EXIT_RDTSC 0x06e
284#define SVM_EXIT_RDPMC 0x06f
285#define SVM_EXIT_PUSHF 0x070
286#define SVM_EXIT_POPF 0x071
287#define SVM_EXIT_CPUID 0x072
288#define SVM_EXIT_RSM 0x073
289#define SVM_EXIT_IRET 0x074
290#define SVM_EXIT_SWINT 0x075
291#define SVM_EXIT_INVD 0x076
292#define SVM_EXIT_PAUSE 0x077
293#define SVM_EXIT_HLT 0x078
294#define SVM_EXIT_INVLPG 0x079
295#define SVM_EXIT_INVLPGA 0x07a
296#define SVM_EXIT_IOIO 0x07b
297#define SVM_EXIT_MSR 0x07c
298#define SVM_EXIT_TASK_SWITCH 0x07d
299#define SVM_EXIT_FERR_FREEZE 0x07e
300#define SVM_EXIT_SHUTDOWN 0x07f
301#define SVM_EXIT_VMRUN 0x080
302#define SVM_EXIT_VMMCALL 0x081
303#define SVM_EXIT_VMLOAD 0x082
304#define SVM_EXIT_VMSAVE 0x083
305#define SVM_EXIT_STGI 0x084
306#define SVM_EXIT_CLGI 0x085
307#define SVM_EXIT_SKINIT 0x086
308#define SVM_EXIT_RDTSCP 0x087
309#define SVM_EXIT_ICEBP 0x088
310#define SVM_EXIT_WBINVD 0x089
311#define SVM_EXIT_MONITOR 0x08a
312#define SVM_EXIT_MWAIT 0x08b
313#define SVM_EXIT_MWAIT_COND 0x08c
314#define SVM_EXIT_NPF 0x400
315
316#define SVM_EXIT_ERR -1
317
318#define SVM_CR0_SELECTIVE_MASK (1 << 3 | 1) /* TS and MP */
319
320#define SVM_VMLOAD ".byte 0x0f, 0x01, 0xda"
321#define SVM_VMRUN ".byte 0x0f, 0x01, 0xd8"
322#define SVM_VMSAVE ".byte 0x0f, 0x01, 0xdb"
323#define SVM_CLGI ".byte 0x0f, 0x01, 0xdd"
324#define SVM_STGI ".byte 0x0f, 0x01, 0xdc"
325#define SVM_INVLPGA ".byte 0x0f, 0x01, 0xdf"
326
327#endif
328
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index a4018b01e1f9..6259d7467648 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -16,7 +16,6 @@
16 */ 16 */
17 17
18#include "irq.h" 18#include "irq.h"
19#include "vmx.h"
20#include "mmu.h" 19#include "mmu.h"
21 20
22#include <linux/kvm_host.h> 21#include <linux/kvm_host.h>
@@ -31,6 +30,8 @@
31 30
32#include <asm/io.h> 31#include <asm/io.h>
33#include <asm/desc.h> 32#include <asm/desc.h>
33#include <asm/vmx.h>
34#include <asm/virtext.h>
34 35
35#define __ex(x) __kvm_handle_fault_on_reboot(x) 36#define __ex(x) __kvm_handle_fault_on_reboot(x)
36 37
@@ -90,6 +91,11 @@ struct vcpu_vmx {
90 } rmode; 91 } rmode;
91 int vpid; 92 int vpid;
92 bool emulation_required; 93 bool emulation_required;
94
95 /* Support for vnmi-less CPUs */
96 int soft_vnmi_blocked;
97 ktime_t entry_time;
98 s64 vnmi_blocked_time;
93}; 99};
94 100
95static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu) 101static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
@@ -122,7 +128,7 @@ static struct vmcs_config {
122 u32 vmentry_ctrl; 128 u32 vmentry_ctrl;
123} vmcs_config; 129} vmcs_config;
124 130
125struct vmx_capability { 131static struct vmx_capability {
126 u32 ept; 132 u32 ept;
127 u32 vpid; 133 u32 vpid;
128} vmx_capability; 134} vmx_capability;
@@ -957,6 +963,13 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
957 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", msr_index, data); 963 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", msr_index, data);
958 964
959 break; 965 break;
966 case MSR_IA32_CR_PAT:
967 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
968 vmcs_write64(GUEST_IA32_PAT, data);
969 vcpu->arch.pat = data;
970 break;
971 }
972 /* Otherwise falls through to kvm_set_msr_common */
960 default: 973 default:
961 vmx_load_host_state(vmx); 974 vmx_load_host_state(vmx);
962 msr = find_msr_entry(vmx, msr_index); 975 msr = find_msr_entry(vmx, msr_index);
@@ -1032,8 +1045,7 @@ static int vmx_get_irq(struct kvm_vcpu *vcpu)
1032 1045
1033static __init int cpu_has_kvm_support(void) 1046static __init int cpu_has_kvm_support(void)
1034{ 1047{
1035 unsigned long ecx = cpuid_ecx(1); 1048 return cpu_has_vmx();
1036 return test_bit(5, &ecx); /* CPUID.1:ECX.VMX[bit 5] -> VT */
1037} 1049}
1038 1050
1039static __init int vmx_disabled_by_bios(void) 1051static __init int vmx_disabled_by_bios(void)
@@ -1079,13 +1091,22 @@ static void vmclear_local_vcpus(void)
1079 __vcpu_clear(vmx); 1091 __vcpu_clear(vmx);
1080} 1092}
1081 1093
1082static void hardware_disable(void *garbage) 1094
1095/* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1096 * tricks.
1097 */
1098static void kvm_cpu_vmxoff(void)
1083{ 1099{
1084 vmclear_local_vcpus();
1085 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc"); 1100 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1086 write_cr4(read_cr4() & ~X86_CR4_VMXE); 1101 write_cr4(read_cr4() & ~X86_CR4_VMXE);
1087} 1102}
1088 1103
1104static void hardware_disable(void *garbage)
1105{
1106 vmclear_local_vcpus();
1107 kvm_cpu_vmxoff();
1108}
1109
1089static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt, 1110static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1090 u32 msr, u32 *result) 1111 u32 msr, u32 *result)
1091{ 1112{
@@ -1176,12 +1197,13 @@ static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1176#ifdef CONFIG_X86_64 1197#ifdef CONFIG_X86_64
1177 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE; 1198 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1178#endif 1199#endif
1179 opt = 0; 1200 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1180 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS, 1201 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1181 &_vmexit_control) < 0) 1202 &_vmexit_control) < 0)
1182 return -EIO; 1203 return -EIO;
1183 1204
1184 min = opt = 0; 1205 min = 0;
1206 opt = VM_ENTRY_LOAD_IA32_PAT;
1185 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS, 1207 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1186 &_vmentry_control) < 0) 1208 &_vmentry_control) < 0)
1187 return -EIO; 1209 return -EIO;
@@ -2087,8 +2109,9 @@ static void vmx_disable_intercept_for_msr(struct page *msr_bitmap, u32 msr)
2087 */ 2109 */
2088static int vmx_vcpu_setup(struct vcpu_vmx *vmx) 2110static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2089{ 2111{
2090 u32 host_sysenter_cs; 2112 u32 host_sysenter_cs, msr_low, msr_high;
2091 u32 junk; 2113 u32 junk;
2114 u64 host_pat;
2092 unsigned long a; 2115 unsigned long a;
2093 struct descriptor_table dt; 2116 struct descriptor_table dt;
2094 int i; 2117 int i;
@@ -2176,6 +2199,20 @@ static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2176 rdmsrl(MSR_IA32_SYSENTER_EIP, a); 2199 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2177 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */ 2200 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2178 2201
2202 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2203 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2204 host_pat = msr_low | ((u64) msr_high << 32);
2205 vmcs_write64(HOST_IA32_PAT, host_pat);
2206 }
2207 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2208 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2209 host_pat = msr_low | ((u64) msr_high << 32);
2210 /* Write the default value follow host pat */
2211 vmcs_write64(GUEST_IA32_PAT, host_pat);
2212 /* Keep arch.pat sync with GUEST_IA32_PAT */
2213 vmx->vcpu.arch.pat = host_pat;
2214 }
2215
2179 for (i = 0; i < NR_VMX_MSR; ++i) { 2216 for (i = 0; i < NR_VMX_MSR; ++i) {
2180 u32 index = vmx_msr_index[i]; 2217 u32 index = vmx_msr_index[i];
2181 u32 data_low, data_high; 2218 u32 data_low, data_high;
@@ -2230,6 +2267,8 @@ static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2230 2267
2231 vmx->vcpu.arch.rmode.active = 0; 2268 vmx->vcpu.arch.rmode.active = 0;
2232 2269
2270 vmx->soft_vnmi_blocked = 0;
2271
2233 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val(); 2272 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2234 kvm_set_cr8(&vmx->vcpu, 0); 2273 kvm_set_cr8(&vmx->vcpu, 0);
2235 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE; 2274 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
@@ -2335,6 +2374,29 @@ out:
2335 return ret; 2374 return ret;
2336} 2375}
2337 2376
2377static void enable_irq_window(struct kvm_vcpu *vcpu)
2378{
2379 u32 cpu_based_vm_exec_control;
2380
2381 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2382 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2383 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2384}
2385
2386static void enable_nmi_window(struct kvm_vcpu *vcpu)
2387{
2388 u32 cpu_based_vm_exec_control;
2389
2390 if (!cpu_has_virtual_nmis()) {
2391 enable_irq_window(vcpu);
2392 return;
2393 }
2394
2395 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2396 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2397 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2398}
2399
2338static void vmx_inject_irq(struct kvm_vcpu *vcpu, int irq) 2400static void vmx_inject_irq(struct kvm_vcpu *vcpu, int irq)
2339{ 2401{
2340 struct vcpu_vmx *vmx = to_vmx(vcpu); 2402 struct vcpu_vmx *vmx = to_vmx(vcpu);
@@ -2358,10 +2420,54 @@ static void vmx_inject_irq(struct kvm_vcpu *vcpu, int irq)
2358 2420
2359static void vmx_inject_nmi(struct kvm_vcpu *vcpu) 2421static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2360{ 2422{
2423 struct vcpu_vmx *vmx = to_vmx(vcpu);
2424
2425 if (!cpu_has_virtual_nmis()) {
2426 /*
2427 * Tracking the NMI-blocked state in software is built upon
2428 * finding the next open IRQ window. This, in turn, depends on
2429 * well-behaving guests: They have to keep IRQs disabled at
2430 * least as long as the NMI handler runs. Otherwise we may
2431 * cause NMI nesting, maybe breaking the guest. But as this is
2432 * highly unlikely, we can live with the residual risk.
2433 */
2434 vmx->soft_vnmi_blocked = 1;
2435 vmx->vnmi_blocked_time = 0;
2436 }
2437
2438 ++vcpu->stat.nmi_injections;
2439 if (vcpu->arch.rmode.active) {
2440 vmx->rmode.irq.pending = true;
2441 vmx->rmode.irq.vector = NMI_VECTOR;
2442 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2443 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2444 NMI_VECTOR | INTR_TYPE_SOFT_INTR |
2445 INTR_INFO_VALID_MASK);
2446 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2447 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2448 return;
2449 }
2361 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 2450 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2362 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR); 2451 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2363} 2452}
2364 2453
2454static void vmx_update_window_states(struct kvm_vcpu *vcpu)
2455{
2456 u32 guest_intr = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2457
2458 vcpu->arch.nmi_window_open =
2459 !(guest_intr & (GUEST_INTR_STATE_STI |
2460 GUEST_INTR_STATE_MOV_SS |
2461 GUEST_INTR_STATE_NMI));
2462 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2463 vcpu->arch.nmi_window_open = 0;
2464
2465 vcpu->arch.interrupt_window_open =
2466 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2467 !(guest_intr & (GUEST_INTR_STATE_STI |
2468 GUEST_INTR_STATE_MOV_SS)));
2469}
2470
2365static void kvm_do_inject_irq(struct kvm_vcpu *vcpu) 2471static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
2366{ 2472{
2367 int word_index = __ffs(vcpu->arch.irq_summary); 2473 int word_index = __ffs(vcpu->arch.irq_summary);
@@ -2374,40 +2480,49 @@ static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
2374 kvm_queue_interrupt(vcpu, irq); 2480 kvm_queue_interrupt(vcpu, irq);
2375} 2481}
2376 2482
2377
2378static void do_interrupt_requests(struct kvm_vcpu *vcpu, 2483static void do_interrupt_requests(struct kvm_vcpu *vcpu,
2379 struct kvm_run *kvm_run) 2484 struct kvm_run *kvm_run)
2380{ 2485{
2381 u32 cpu_based_vm_exec_control; 2486 vmx_update_window_states(vcpu);
2382
2383 vcpu->arch.interrupt_window_open =
2384 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2385 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
2386 2487
2387 if (vcpu->arch.interrupt_window_open && 2488 if (vcpu->arch.nmi_pending && !vcpu->arch.nmi_injected) {
2388 vcpu->arch.irq_summary && !vcpu->arch.interrupt.pending) 2489 if (vcpu->arch.interrupt.pending) {
2389 kvm_do_inject_irq(vcpu); 2490 enable_nmi_window(vcpu);
2491 } else if (vcpu->arch.nmi_window_open) {
2492 vcpu->arch.nmi_pending = false;
2493 vcpu->arch.nmi_injected = true;
2494 } else {
2495 enable_nmi_window(vcpu);
2496 return;
2497 }
2498 }
2499 if (vcpu->arch.nmi_injected) {
2500 vmx_inject_nmi(vcpu);
2501 if (vcpu->arch.nmi_pending)
2502 enable_nmi_window(vcpu);
2503 else if (vcpu->arch.irq_summary
2504 || kvm_run->request_interrupt_window)
2505 enable_irq_window(vcpu);
2506 return;
2507 }
2390 2508
2391 if (vcpu->arch.interrupt_window_open && vcpu->arch.interrupt.pending) 2509 if (vcpu->arch.interrupt_window_open) {
2392 vmx_inject_irq(vcpu, vcpu->arch.interrupt.nr); 2510 if (vcpu->arch.irq_summary && !vcpu->arch.interrupt.pending)
2511 kvm_do_inject_irq(vcpu);
2393 2512
2394 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); 2513 if (vcpu->arch.interrupt.pending)
2514 vmx_inject_irq(vcpu, vcpu->arch.interrupt.nr);
2515 }
2395 if (!vcpu->arch.interrupt_window_open && 2516 if (!vcpu->arch.interrupt_window_open &&
2396 (vcpu->arch.irq_summary || kvm_run->request_interrupt_window)) 2517 (vcpu->arch.irq_summary || kvm_run->request_interrupt_window))
2397 /* 2518 enable_irq_window(vcpu);
2398 * Interrupts blocked. Wait for unblock.
2399 */
2400 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2401 else
2402 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2403 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2404} 2519}
2405 2520
2406static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr) 2521static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2407{ 2522{
2408 int ret; 2523 int ret;
2409 struct kvm_userspace_memory_region tss_mem = { 2524 struct kvm_userspace_memory_region tss_mem = {
2410 .slot = 8, 2525 .slot = TSS_PRIVATE_MEMSLOT,
2411 .guest_phys_addr = addr, 2526 .guest_phys_addr = addr,
2412 .memory_size = PAGE_SIZE * 3, 2527 .memory_size = PAGE_SIZE * 3,
2413 .flags = 0, 2528 .flags = 0,
@@ -2492,7 +2607,7 @@ static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2492 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary); 2607 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
2493 } 2608 }
2494 2609
2495 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) /* nmi */ 2610 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
2496 return 1; /* already handled by vmx_vcpu_run() */ 2611 return 1; /* already handled by vmx_vcpu_run() */
2497 2612
2498 if (is_no_device(intr_info)) { 2613 if (is_no_device(intr_info)) {
@@ -2581,6 +2696,7 @@ static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2581 rep = (exit_qualification & 32) != 0; 2696 rep = (exit_qualification & 32) != 0;
2582 port = exit_qualification >> 16; 2697 port = exit_qualification >> 16;
2583 2698
2699 skip_emulated_instruction(vcpu);
2584 return kvm_emulate_pio(vcpu, kvm_run, in, size, port); 2700 return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
2585} 2701}
2586 2702
@@ -2767,6 +2883,7 @@ static int handle_interrupt_window(struct kvm_vcpu *vcpu,
2767 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control); 2883 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2768 2884
2769 KVMTRACE_0D(PEND_INTR, vcpu, handler); 2885 KVMTRACE_0D(PEND_INTR, vcpu, handler);
2886 ++vcpu->stat.irq_window_exits;
2770 2887
2771 /* 2888 /*
2772 * If the user space waits to inject interrupts, exit as soon as 2889 * If the user space waits to inject interrupts, exit as soon as
@@ -2775,7 +2892,6 @@ static int handle_interrupt_window(struct kvm_vcpu *vcpu,
2775 if (kvm_run->request_interrupt_window && 2892 if (kvm_run->request_interrupt_window &&
2776 !vcpu->arch.irq_summary) { 2893 !vcpu->arch.irq_summary) {
2777 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN; 2894 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2778 ++vcpu->stat.irq_window_exits;
2779 return 0; 2895 return 0;
2780 } 2896 }
2781 return 1; 2897 return 1;
@@ -2832,6 +2948,7 @@ static int handle_apic_access(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2832 2948
2833static int handle_task_switch(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) 2949static int handle_task_switch(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2834{ 2950{
2951 struct vcpu_vmx *vmx = to_vmx(vcpu);
2835 unsigned long exit_qualification; 2952 unsigned long exit_qualification;
2836 u16 tss_selector; 2953 u16 tss_selector;
2837 int reason; 2954 int reason;
@@ -2839,6 +2956,15 @@ static int handle_task_switch(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2839 exit_qualification = vmcs_readl(EXIT_QUALIFICATION); 2956 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2840 2957
2841 reason = (u32)exit_qualification >> 30; 2958 reason = (u32)exit_qualification >> 30;
2959 if (reason == TASK_SWITCH_GATE && vmx->vcpu.arch.nmi_injected &&
2960 (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
2961 (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK)
2962 == INTR_TYPE_NMI_INTR) {
2963 vcpu->arch.nmi_injected = false;
2964 if (cpu_has_virtual_nmis())
2965 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
2966 GUEST_INTR_STATE_NMI);
2967 }
2842 tss_selector = exit_qualification; 2968 tss_selector = exit_qualification;
2843 2969
2844 return kvm_task_switch(vcpu, tss_selector, reason); 2970 return kvm_task_switch(vcpu, tss_selector, reason);
@@ -2927,16 +3053,12 @@ static void handle_invalid_guest_state(struct kvm_vcpu *vcpu,
2927 while (!guest_state_valid(vcpu)) { 3053 while (!guest_state_valid(vcpu)) {
2928 err = emulate_instruction(vcpu, kvm_run, 0, 0, 0); 3054 err = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
2929 3055
2930 switch (err) { 3056 if (err == EMULATE_DO_MMIO)
2931 case EMULATE_DONE: 3057 break;
2932 break; 3058
2933 case EMULATE_DO_MMIO: 3059 if (err != EMULATE_DONE) {
2934 kvm_report_emulation_failure(vcpu, "mmio"); 3060 kvm_report_emulation_failure(vcpu, "emulation failure");
2935 /* TODO: Handle MMIO */ 3061 return;
2936 return;
2937 default:
2938 kvm_report_emulation_failure(vcpu, "emulation failure");
2939 return;
2940 } 3062 }
2941 3063
2942 if (signal_pending(current)) 3064 if (signal_pending(current))
@@ -2948,8 +3070,10 @@ static void handle_invalid_guest_state(struct kvm_vcpu *vcpu,
2948 local_irq_disable(); 3070 local_irq_disable();
2949 preempt_disable(); 3071 preempt_disable();
2950 3072
2951 /* Guest state should be valid now, no more emulation should be needed */ 3073 /* Guest state should be valid now except if we need to
2952 vmx->emulation_required = 0; 3074 * emulate an MMIO */
3075 if (guest_state_valid(vcpu))
3076 vmx->emulation_required = 0;
2953} 3077}
2954 3078
2955/* 3079/*
@@ -2996,6 +3120,11 @@ static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2996 KVMTRACE_3D(VMEXIT, vcpu, exit_reason, (u32)kvm_rip_read(vcpu), 3120 KVMTRACE_3D(VMEXIT, vcpu, exit_reason, (u32)kvm_rip_read(vcpu),
2997 (u32)((u64)kvm_rip_read(vcpu) >> 32), entryexit); 3121 (u32)((u64)kvm_rip_read(vcpu) >> 32), entryexit);
2998 3122
3123 /* If we need to emulate an MMIO from handle_invalid_guest_state
3124 * we just return 0 */
3125 if (vmx->emulation_required && emulate_invalid_guest_state)
3126 return 0;
3127
2999 /* Access CR3 don't cause VMExit in paging mode, so we need 3128 /* Access CR3 don't cause VMExit in paging mode, so we need
3000 * to sync with guest real CR3. */ 3129 * to sync with guest real CR3. */
3001 if (vm_need_ept() && is_paging(vcpu)) { 3130 if (vm_need_ept() && is_paging(vcpu)) {
@@ -3012,9 +3141,32 @@ static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3012 3141
3013 if ((vectoring_info & VECTORING_INFO_VALID_MASK) && 3142 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3014 (exit_reason != EXIT_REASON_EXCEPTION_NMI && 3143 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3015 exit_reason != EXIT_REASON_EPT_VIOLATION)) 3144 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3016 printk(KERN_WARNING "%s: unexpected, valid vectoring info and " 3145 exit_reason != EXIT_REASON_TASK_SWITCH))
3017 "exit reason is 0x%x\n", __func__, exit_reason); 3146 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3147 "(0x%x) and exit reason is 0x%x\n",
3148 __func__, vectoring_info, exit_reason);
3149
3150 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3151 if (vcpu->arch.interrupt_window_open) {
3152 vmx->soft_vnmi_blocked = 0;
3153 vcpu->arch.nmi_window_open = 1;
3154 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3155 vcpu->arch.nmi_pending) {
3156 /*
3157 * This CPU don't support us in finding the end of an
3158 * NMI-blocked window if the guest runs with IRQs
3159 * disabled. So we pull the trigger after 1 s of
3160 * futile waiting, but inform the user about this.
3161 */
3162 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3163 "state on VCPU %d after 1 s timeout\n",
3164 __func__, vcpu->vcpu_id);
3165 vmx->soft_vnmi_blocked = 0;
3166 vmx->vcpu.arch.nmi_window_open = 1;
3167 }
3168 }
3169
3018 if (exit_reason < kvm_vmx_max_exit_handlers 3170 if (exit_reason < kvm_vmx_max_exit_handlers
3019 && kvm_vmx_exit_handlers[exit_reason]) 3171 && kvm_vmx_exit_handlers[exit_reason])
3020 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run); 3172 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
@@ -3042,51 +3194,6 @@ static void update_tpr_threshold(struct kvm_vcpu *vcpu)
3042 vmcs_write32(TPR_THRESHOLD, (max_irr > tpr) ? tpr >> 4 : max_irr >> 4); 3194 vmcs_write32(TPR_THRESHOLD, (max_irr > tpr) ? tpr >> 4 : max_irr >> 4);
3043} 3195}
3044 3196
3045static void enable_irq_window(struct kvm_vcpu *vcpu)
3046{
3047 u32 cpu_based_vm_exec_control;
3048
3049 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3050 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
3051 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3052}
3053
3054static void enable_nmi_window(struct kvm_vcpu *vcpu)
3055{
3056 u32 cpu_based_vm_exec_control;
3057
3058 if (!cpu_has_virtual_nmis())
3059 return;
3060
3061 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3062 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
3063 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3064}
3065
3066static int vmx_nmi_enabled(struct kvm_vcpu *vcpu)
3067{
3068 u32 guest_intr = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
3069 return !(guest_intr & (GUEST_INTR_STATE_NMI |
3070 GUEST_INTR_STATE_MOV_SS |
3071 GUEST_INTR_STATE_STI));
3072}
3073
3074static int vmx_irq_enabled(struct kvm_vcpu *vcpu)
3075{
3076 u32 guest_intr = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
3077 return (!(guest_intr & (GUEST_INTR_STATE_MOV_SS |
3078 GUEST_INTR_STATE_STI)) &&
3079 (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF));
3080}
3081
3082static void enable_intr_window(struct kvm_vcpu *vcpu)
3083{
3084 if (vcpu->arch.nmi_pending)
3085 enable_nmi_window(vcpu);
3086 else if (kvm_cpu_has_interrupt(vcpu))
3087 enable_irq_window(vcpu);
3088}
3089
3090static void vmx_complete_interrupts(struct vcpu_vmx *vmx) 3197static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3091{ 3198{
3092 u32 exit_intr_info; 3199 u32 exit_intr_info;
@@ -3109,7 +3216,9 @@ static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3109 if (unblock_nmi && vector != DF_VECTOR) 3216 if (unblock_nmi && vector != DF_VECTOR)
3110 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, 3217 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3111 GUEST_INTR_STATE_NMI); 3218 GUEST_INTR_STATE_NMI);
3112 } 3219 } else if (unlikely(vmx->soft_vnmi_blocked))
3220 vmx->vnmi_blocked_time +=
3221 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3113 3222
3114 idt_vectoring_info = vmx->idt_vectoring_info; 3223 idt_vectoring_info = vmx->idt_vectoring_info;
3115 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK; 3224 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
@@ -3147,26 +3256,29 @@ static void vmx_intr_assist(struct kvm_vcpu *vcpu)
3147{ 3256{
3148 update_tpr_threshold(vcpu); 3257 update_tpr_threshold(vcpu);
3149 3258
3150 if (cpu_has_virtual_nmis()) { 3259 vmx_update_window_states(vcpu);
3151 if (vcpu->arch.nmi_pending && !vcpu->arch.nmi_injected) { 3260
3152 if (vcpu->arch.interrupt.pending) { 3261 if (vcpu->arch.nmi_pending && !vcpu->arch.nmi_injected) {
3153 enable_nmi_window(vcpu); 3262 if (vcpu->arch.interrupt.pending) {
3154 } else if (vmx_nmi_enabled(vcpu)) { 3263 enable_nmi_window(vcpu);
3155 vcpu->arch.nmi_pending = false; 3264 } else if (vcpu->arch.nmi_window_open) {
3156 vcpu->arch.nmi_injected = true; 3265 vcpu->arch.nmi_pending = false;
3157 } else { 3266 vcpu->arch.nmi_injected = true;
3158 enable_intr_window(vcpu); 3267 } else {
3159 return; 3268 enable_nmi_window(vcpu);
3160 }
3161 }
3162 if (vcpu->arch.nmi_injected) {
3163 vmx_inject_nmi(vcpu);
3164 enable_intr_window(vcpu);
3165 return; 3269 return;
3166 } 3270 }
3167 } 3271 }
3272 if (vcpu->arch.nmi_injected) {
3273 vmx_inject_nmi(vcpu);
3274 if (vcpu->arch.nmi_pending)
3275 enable_nmi_window(vcpu);
3276 else if (kvm_cpu_has_interrupt(vcpu))
3277 enable_irq_window(vcpu);
3278 return;
3279 }
3168 if (!vcpu->arch.interrupt.pending && kvm_cpu_has_interrupt(vcpu)) { 3280 if (!vcpu->arch.interrupt.pending && kvm_cpu_has_interrupt(vcpu)) {
3169 if (vmx_irq_enabled(vcpu)) 3281 if (vcpu->arch.interrupt_window_open)
3170 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu)); 3282 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu));
3171 else 3283 else
3172 enable_irq_window(vcpu); 3284 enable_irq_window(vcpu);
@@ -3174,6 +3286,8 @@ static void vmx_intr_assist(struct kvm_vcpu *vcpu)
3174 if (vcpu->arch.interrupt.pending) { 3286 if (vcpu->arch.interrupt.pending) {
3175 vmx_inject_irq(vcpu, vcpu->arch.interrupt.nr); 3287 vmx_inject_irq(vcpu, vcpu->arch.interrupt.nr);
3176 kvm_timer_intr_post(vcpu, vcpu->arch.interrupt.nr); 3288 kvm_timer_intr_post(vcpu, vcpu->arch.interrupt.nr);
3289 if (kvm_cpu_has_interrupt(vcpu))
3290 enable_irq_window(vcpu);
3177 } 3291 }
3178} 3292}
3179 3293
@@ -3213,6 +3327,10 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3213 struct vcpu_vmx *vmx = to_vmx(vcpu); 3327 struct vcpu_vmx *vmx = to_vmx(vcpu);
3214 u32 intr_info; 3328 u32 intr_info;
3215 3329
3330 /* Record the guest's net vcpu time for enforced NMI injections. */
3331 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3332 vmx->entry_time = ktime_get();
3333
3216 /* Handle invalid guest state instead of entering VMX */ 3334 /* Handle invalid guest state instead of entering VMX */
3217 if (vmx->emulation_required && emulate_invalid_guest_state) { 3335 if (vmx->emulation_required && emulate_invalid_guest_state) {
3218 handle_invalid_guest_state(vcpu, kvm_run); 3336 handle_invalid_guest_state(vcpu, kvm_run);
@@ -3327,9 +3445,7 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3327 if (vmx->rmode.irq.pending) 3445 if (vmx->rmode.irq.pending)
3328 fixup_rmode_irq(vmx); 3446 fixup_rmode_irq(vmx);
3329 3447
3330 vcpu->arch.interrupt_window_open = 3448 vmx_update_window_states(vcpu);
3331 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
3332 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS)) == 0;
3333 3449
3334 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS)); 3450 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
3335 vmx->launched = 1; 3451 vmx->launched = 1;
@@ -3337,7 +3453,7 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3337 intr_info = vmcs_read32(VM_EXIT_INTR_INFO); 3453 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3338 3454
3339 /* We need to handle NMIs before interrupts are enabled */ 3455 /* We need to handle NMIs before interrupts are enabled */
3340 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200 && 3456 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3341 (intr_info & INTR_INFO_VALID_MASK)) { 3457 (intr_info & INTR_INFO_VALID_MASK)) {
3342 KVMTRACE_0D(NMI, vcpu, handler); 3458 KVMTRACE_0D(NMI, vcpu, handler);
3343 asm("int $2"); 3459 asm("int $2");
@@ -3455,6 +3571,11 @@ static int get_ept_level(void)
3455 return VMX_EPT_DEFAULT_GAW + 1; 3571 return VMX_EPT_DEFAULT_GAW + 1;
3456} 3572}
3457 3573
3574static int vmx_get_mt_mask_shift(void)
3575{
3576 return VMX_EPT_MT_EPTE_SHIFT;
3577}
3578
3458static struct kvm_x86_ops vmx_x86_ops = { 3579static struct kvm_x86_ops vmx_x86_ops = {
3459 .cpu_has_kvm_support = cpu_has_kvm_support, 3580 .cpu_has_kvm_support = cpu_has_kvm_support,
3460 .disabled_by_bios = vmx_disabled_by_bios, 3581 .disabled_by_bios = vmx_disabled_by_bios,
@@ -3510,6 +3631,7 @@ static struct kvm_x86_ops vmx_x86_ops = {
3510 3631
3511 .set_tss_addr = vmx_set_tss_addr, 3632 .set_tss_addr = vmx_set_tss_addr,
3512 .get_tdp_level = get_ept_level, 3633 .get_tdp_level = get_ept_level,
3634 .get_mt_mask_shift = vmx_get_mt_mask_shift,
3513}; 3635};
3514 3636
3515static int __init vmx_init(void) 3637static int __init vmx_init(void)
@@ -3566,10 +3688,10 @@ static int __init vmx_init(void)
3566 bypass_guest_pf = 0; 3688 bypass_guest_pf = 0;
3567 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK | 3689 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
3568 VMX_EPT_WRITABLE_MASK | 3690 VMX_EPT_WRITABLE_MASK |
3569 VMX_EPT_DEFAULT_MT << VMX_EPT_MT_EPTE_SHIFT |
3570 VMX_EPT_IGMT_BIT); 3691 VMX_EPT_IGMT_BIT);
3571 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull, 3692 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
3572 VMX_EPT_EXECUTABLE_MASK); 3693 VMX_EPT_EXECUTABLE_MASK,
3694 VMX_EPT_DEFAULT_MT << VMX_EPT_MT_EPTE_SHIFT);
3573 kvm_enable_tdp(); 3695 kvm_enable_tdp();
3574 } else 3696 } else
3575 kvm_disable_tdp(); 3697 kvm_disable_tdp();
diff --git a/arch/x86/kvm/vmx.h b/arch/x86/kvm/vmx.h
deleted file mode 100644
index ec5edc339da6..000000000000
--- a/arch/x86/kvm/vmx.h
+++ /dev/null
@@ -1,359 +0,0 @@
1#ifndef VMX_H
2#define VMX_H
3
4/*
5 * vmx.h: VMX Architecture related definitions
6 * Copyright (c) 2004, Intel Corporation.
7 *
8 * This program is free software; you can redistribute it and/or modify it
9 * under the terms and conditions of the GNU General Public License,
10 * version 2, as published by the Free Software Foundation.
11 *
12 * This program is distributed in the hope it will be useful, but WITHOUT
13 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * more details.
16 *
17 * You should have received a copy of the GNU General Public License along with
18 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
19 * Place - Suite 330, Boston, MA 02111-1307 USA.
20 *
21 * A few random additions are:
22 * Copyright (C) 2006 Qumranet
23 * Avi Kivity <avi@qumranet.com>
24 * Yaniv Kamay <yaniv@qumranet.com>
25 *
26 */
27
28/*
29 * Definitions of Primary Processor-Based VM-Execution Controls.
30 */
31#define CPU_BASED_VIRTUAL_INTR_PENDING 0x00000004
32#define CPU_BASED_USE_TSC_OFFSETING 0x00000008
33#define CPU_BASED_HLT_EXITING 0x00000080
34#define CPU_BASED_INVLPG_EXITING 0x00000200
35#define CPU_BASED_MWAIT_EXITING 0x00000400
36#define CPU_BASED_RDPMC_EXITING 0x00000800
37#define CPU_BASED_RDTSC_EXITING 0x00001000
38#define CPU_BASED_CR3_LOAD_EXITING 0x00008000
39#define CPU_BASED_CR3_STORE_EXITING 0x00010000
40#define CPU_BASED_CR8_LOAD_EXITING 0x00080000
41#define CPU_BASED_CR8_STORE_EXITING 0x00100000
42#define CPU_BASED_TPR_SHADOW 0x00200000
43#define CPU_BASED_VIRTUAL_NMI_PENDING 0x00400000
44#define CPU_BASED_MOV_DR_EXITING 0x00800000
45#define CPU_BASED_UNCOND_IO_EXITING 0x01000000
46#define CPU_BASED_USE_IO_BITMAPS 0x02000000
47#define CPU_BASED_USE_MSR_BITMAPS 0x10000000
48#define CPU_BASED_MONITOR_EXITING 0x20000000
49#define CPU_BASED_PAUSE_EXITING 0x40000000
50#define CPU_BASED_ACTIVATE_SECONDARY_CONTROLS 0x80000000
51/*
52 * Definitions of Secondary Processor-Based VM-Execution Controls.
53 */
54#define SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES 0x00000001
55#define SECONDARY_EXEC_ENABLE_EPT 0x00000002
56#define SECONDARY_EXEC_ENABLE_VPID 0x00000020
57#define SECONDARY_EXEC_WBINVD_EXITING 0x00000040
58
59
60#define PIN_BASED_EXT_INTR_MASK 0x00000001
61#define PIN_BASED_NMI_EXITING 0x00000008
62#define PIN_BASED_VIRTUAL_NMIS 0x00000020
63
64#define VM_EXIT_HOST_ADDR_SPACE_SIZE 0x00000200
65#define VM_EXIT_ACK_INTR_ON_EXIT 0x00008000
66
67#define VM_ENTRY_IA32E_MODE 0x00000200
68#define VM_ENTRY_SMM 0x00000400
69#define VM_ENTRY_DEACT_DUAL_MONITOR 0x00000800
70
71/* VMCS Encodings */
72enum vmcs_field {
73 VIRTUAL_PROCESSOR_ID = 0x00000000,
74 GUEST_ES_SELECTOR = 0x00000800,
75 GUEST_CS_SELECTOR = 0x00000802,
76 GUEST_SS_SELECTOR = 0x00000804,
77 GUEST_DS_SELECTOR = 0x00000806,
78 GUEST_FS_SELECTOR = 0x00000808,
79 GUEST_GS_SELECTOR = 0x0000080a,
80 GUEST_LDTR_SELECTOR = 0x0000080c,
81 GUEST_TR_SELECTOR = 0x0000080e,
82 HOST_ES_SELECTOR = 0x00000c00,
83 HOST_CS_SELECTOR = 0x00000c02,
84 HOST_SS_SELECTOR = 0x00000c04,
85 HOST_DS_SELECTOR = 0x00000c06,
86 HOST_FS_SELECTOR = 0x00000c08,
87 HOST_GS_SELECTOR = 0x00000c0a,
88 HOST_TR_SELECTOR = 0x00000c0c,
89 IO_BITMAP_A = 0x00002000,
90 IO_BITMAP_A_HIGH = 0x00002001,
91 IO_BITMAP_B = 0x00002002,
92 IO_BITMAP_B_HIGH = 0x00002003,
93 MSR_BITMAP = 0x00002004,
94 MSR_BITMAP_HIGH = 0x00002005,
95 VM_EXIT_MSR_STORE_ADDR = 0x00002006,
96 VM_EXIT_MSR_STORE_ADDR_HIGH = 0x00002007,
97 VM_EXIT_MSR_LOAD_ADDR = 0x00002008,
98 VM_EXIT_MSR_LOAD_ADDR_HIGH = 0x00002009,
99 VM_ENTRY_MSR_LOAD_ADDR = 0x0000200a,
100 VM_ENTRY_MSR_LOAD_ADDR_HIGH = 0x0000200b,
101 TSC_OFFSET = 0x00002010,
102 TSC_OFFSET_HIGH = 0x00002011,
103 VIRTUAL_APIC_PAGE_ADDR = 0x00002012,
104 VIRTUAL_APIC_PAGE_ADDR_HIGH = 0x00002013,
105 APIC_ACCESS_ADDR = 0x00002014,
106 APIC_ACCESS_ADDR_HIGH = 0x00002015,
107 EPT_POINTER = 0x0000201a,
108 EPT_POINTER_HIGH = 0x0000201b,
109 GUEST_PHYSICAL_ADDRESS = 0x00002400,
110 GUEST_PHYSICAL_ADDRESS_HIGH = 0x00002401,
111 VMCS_LINK_POINTER = 0x00002800,
112 VMCS_LINK_POINTER_HIGH = 0x00002801,
113 GUEST_IA32_DEBUGCTL = 0x00002802,
114 GUEST_IA32_DEBUGCTL_HIGH = 0x00002803,
115 GUEST_PDPTR0 = 0x0000280a,
116 GUEST_PDPTR0_HIGH = 0x0000280b,
117 GUEST_PDPTR1 = 0x0000280c,
118 GUEST_PDPTR1_HIGH = 0x0000280d,
119 GUEST_PDPTR2 = 0x0000280e,
120 GUEST_PDPTR2_HIGH = 0x0000280f,
121 GUEST_PDPTR3 = 0x00002810,
122 GUEST_PDPTR3_HIGH = 0x00002811,
123 PIN_BASED_VM_EXEC_CONTROL = 0x00004000,
124 CPU_BASED_VM_EXEC_CONTROL = 0x00004002,
125 EXCEPTION_BITMAP = 0x00004004,
126 PAGE_FAULT_ERROR_CODE_MASK = 0x00004006,
127 PAGE_FAULT_ERROR_CODE_MATCH = 0x00004008,
128 CR3_TARGET_COUNT = 0x0000400a,
129 VM_EXIT_CONTROLS = 0x0000400c,
130 VM_EXIT_MSR_STORE_COUNT = 0x0000400e,
131 VM_EXIT_MSR_LOAD_COUNT = 0x00004010,
132 VM_ENTRY_CONTROLS = 0x00004012,
133 VM_ENTRY_MSR_LOAD_COUNT = 0x00004014,
134 VM_ENTRY_INTR_INFO_FIELD = 0x00004016,
135 VM_ENTRY_EXCEPTION_ERROR_CODE = 0x00004018,
136 VM_ENTRY_INSTRUCTION_LEN = 0x0000401a,
137 TPR_THRESHOLD = 0x0000401c,
138 SECONDARY_VM_EXEC_CONTROL = 0x0000401e,
139 VM_INSTRUCTION_ERROR = 0x00004400,
140 VM_EXIT_REASON = 0x00004402,
141 VM_EXIT_INTR_INFO = 0x00004404,
142 VM_EXIT_INTR_ERROR_CODE = 0x00004406,
143 IDT_VECTORING_INFO_FIELD = 0x00004408,
144 IDT_VECTORING_ERROR_CODE = 0x0000440a,
145 VM_EXIT_INSTRUCTION_LEN = 0x0000440c,
146 VMX_INSTRUCTION_INFO = 0x0000440e,
147 GUEST_ES_LIMIT = 0x00004800,
148 GUEST_CS_LIMIT = 0x00004802,
149 GUEST_SS_LIMIT = 0x00004804,
150 GUEST_DS_LIMIT = 0x00004806,
151 GUEST_FS_LIMIT = 0x00004808,
152 GUEST_GS_LIMIT = 0x0000480a,
153 GUEST_LDTR_LIMIT = 0x0000480c,
154 GUEST_TR_LIMIT = 0x0000480e,
155 GUEST_GDTR_LIMIT = 0x00004810,
156 GUEST_IDTR_LIMIT = 0x00004812,
157 GUEST_ES_AR_BYTES = 0x00004814,
158 GUEST_CS_AR_BYTES = 0x00004816,
159 GUEST_SS_AR_BYTES = 0x00004818,
160 GUEST_DS_AR_BYTES = 0x0000481a,
161 GUEST_FS_AR_BYTES = 0x0000481c,
162 GUEST_GS_AR_BYTES = 0x0000481e,
163 GUEST_LDTR_AR_BYTES = 0x00004820,
164 GUEST_TR_AR_BYTES = 0x00004822,
165 GUEST_INTERRUPTIBILITY_INFO = 0x00004824,
166 GUEST_ACTIVITY_STATE = 0X00004826,
167 GUEST_SYSENTER_CS = 0x0000482A,
168 HOST_IA32_SYSENTER_CS = 0x00004c00,
169 CR0_GUEST_HOST_MASK = 0x00006000,
170 CR4_GUEST_HOST_MASK = 0x00006002,
171 CR0_READ_SHADOW = 0x00006004,
172 CR4_READ_SHADOW = 0x00006006,
173 CR3_TARGET_VALUE0 = 0x00006008,
174 CR3_TARGET_VALUE1 = 0x0000600a,
175 CR3_TARGET_VALUE2 = 0x0000600c,
176 CR3_TARGET_VALUE3 = 0x0000600e,
177 EXIT_QUALIFICATION = 0x00006400,
178 GUEST_LINEAR_ADDRESS = 0x0000640a,
179 GUEST_CR0 = 0x00006800,
180 GUEST_CR3 = 0x00006802,
181 GUEST_CR4 = 0x00006804,
182 GUEST_ES_BASE = 0x00006806,
183 GUEST_CS_BASE = 0x00006808,
184 GUEST_SS_BASE = 0x0000680a,
185 GUEST_DS_BASE = 0x0000680c,
186 GUEST_FS_BASE = 0x0000680e,
187 GUEST_GS_BASE = 0x00006810,
188 GUEST_LDTR_BASE = 0x00006812,
189 GUEST_TR_BASE = 0x00006814,
190 GUEST_GDTR_BASE = 0x00006816,
191 GUEST_IDTR_BASE = 0x00006818,
192 GUEST_DR7 = 0x0000681a,
193 GUEST_RSP = 0x0000681c,
194 GUEST_RIP = 0x0000681e,
195 GUEST_RFLAGS = 0x00006820,
196 GUEST_PENDING_DBG_EXCEPTIONS = 0x00006822,
197 GUEST_SYSENTER_ESP = 0x00006824,
198 GUEST_SYSENTER_EIP = 0x00006826,
199 HOST_CR0 = 0x00006c00,
200 HOST_CR3 = 0x00006c02,
201 HOST_CR4 = 0x00006c04,
202 HOST_FS_BASE = 0x00006c06,
203 HOST_GS_BASE = 0x00006c08,
204 HOST_TR_BASE = 0x00006c0a,
205 HOST_GDTR_BASE = 0x00006c0c,
206 HOST_IDTR_BASE = 0x00006c0e,
207 HOST_IA32_SYSENTER_ESP = 0x00006c10,
208 HOST_IA32_SYSENTER_EIP = 0x00006c12,
209 HOST_RSP = 0x00006c14,
210 HOST_RIP = 0x00006c16,
211};
212
213#define VMX_EXIT_REASONS_FAILED_VMENTRY 0x80000000
214
215#define EXIT_REASON_EXCEPTION_NMI 0
216#define EXIT_REASON_EXTERNAL_INTERRUPT 1
217#define EXIT_REASON_TRIPLE_FAULT 2
218
219#define EXIT_REASON_PENDING_INTERRUPT 7
220#define EXIT_REASON_NMI_WINDOW 8
221#define EXIT_REASON_TASK_SWITCH 9
222#define EXIT_REASON_CPUID 10
223#define EXIT_REASON_HLT 12
224#define EXIT_REASON_INVLPG 14
225#define EXIT_REASON_RDPMC 15
226#define EXIT_REASON_RDTSC 16
227#define EXIT_REASON_VMCALL 18
228#define EXIT_REASON_VMCLEAR 19
229#define EXIT_REASON_VMLAUNCH 20
230#define EXIT_REASON_VMPTRLD 21
231#define EXIT_REASON_VMPTRST 22
232#define EXIT_REASON_VMREAD 23
233#define EXIT_REASON_VMRESUME 24
234#define EXIT_REASON_VMWRITE 25
235#define EXIT_REASON_VMOFF 26
236#define EXIT_REASON_VMON 27
237#define EXIT_REASON_CR_ACCESS 28
238#define EXIT_REASON_DR_ACCESS 29
239#define EXIT_REASON_IO_INSTRUCTION 30
240#define EXIT_REASON_MSR_READ 31
241#define EXIT_REASON_MSR_WRITE 32
242#define EXIT_REASON_MWAIT_INSTRUCTION 36
243#define EXIT_REASON_TPR_BELOW_THRESHOLD 43
244#define EXIT_REASON_APIC_ACCESS 44
245#define EXIT_REASON_EPT_VIOLATION 48
246#define EXIT_REASON_EPT_MISCONFIG 49
247#define EXIT_REASON_WBINVD 54
248
249/*
250 * Interruption-information format
251 */
252#define INTR_INFO_VECTOR_MASK 0xff /* 7:0 */
253#define INTR_INFO_INTR_TYPE_MASK 0x700 /* 10:8 */
254#define INTR_INFO_DELIVER_CODE_MASK 0x800 /* 11 */
255#define INTR_INFO_UNBLOCK_NMI 0x1000 /* 12 */
256#define INTR_INFO_VALID_MASK 0x80000000 /* 31 */
257#define INTR_INFO_RESVD_BITS_MASK 0x7ffff000
258
259#define VECTORING_INFO_VECTOR_MASK INTR_INFO_VECTOR_MASK
260#define VECTORING_INFO_TYPE_MASK INTR_INFO_INTR_TYPE_MASK
261#define VECTORING_INFO_DELIVER_CODE_MASK INTR_INFO_DELIVER_CODE_MASK
262#define VECTORING_INFO_VALID_MASK INTR_INFO_VALID_MASK
263
264#define INTR_TYPE_EXT_INTR (0 << 8) /* external interrupt */
265#define INTR_TYPE_NMI_INTR (2 << 8) /* NMI */
266#define INTR_TYPE_EXCEPTION (3 << 8) /* processor exception */
267#define INTR_TYPE_SOFT_INTR (4 << 8) /* software interrupt */
268
269/* GUEST_INTERRUPTIBILITY_INFO flags. */
270#define GUEST_INTR_STATE_STI 0x00000001
271#define GUEST_INTR_STATE_MOV_SS 0x00000002
272#define GUEST_INTR_STATE_SMI 0x00000004
273#define GUEST_INTR_STATE_NMI 0x00000008
274
275/*
276 * Exit Qualifications for MOV for Control Register Access
277 */
278#define CONTROL_REG_ACCESS_NUM 0x7 /* 2:0, number of control reg.*/
279#define CONTROL_REG_ACCESS_TYPE 0x30 /* 5:4, access type */
280#define CONTROL_REG_ACCESS_REG 0xf00 /* 10:8, general purpose reg. */
281#define LMSW_SOURCE_DATA_SHIFT 16
282#define LMSW_SOURCE_DATA (0xFFFF << LMSW_SOURCE_DATA_SHIFT) /* 16:31 lmsw source */
283#define REG_EAX (0 << 8)
284#define REG_ECX (1 << 8)
285#define REG_EDX (2 << 8)
286#define REG_EBX (3 << 8)
287#define REG_ESP (4 << 8)
288#define REG_EBP (5 << 8)
289#define REG_ESI (6 << 8)
290#define REG_EDI (7 << 8)
291#define REG_R8 (8 << 8)
292#define REG_R9 (9 << 8)
293#define REG_R10 (10 << 8)
294#define REG_R11 (11 << 8)
295#define REG_R12 (12 << 8)
296#define REG_R13 (13 << 8)
297#define REG_R14 (14 << 8)
298#define REG_R15 (15 << 8)
299
300/*
301 * Exit Qualifications for MOV for Debug Register Access
302 */
303#define DEBUG_REG_ACCESS_NUM 0x7 /* 2:0, number of debug reg. */
304#define DEBUG_REG_ACCESS_TYPE 0x10 /* 4, direction of access */
305#define TYPE_MOV_TO_DR (0 << 4)
306#define TYPE_MOV_FROM_DR (1 << 4)
307#define DEBUG_REG_ACCESS_REG 0xf00 /* 11:8, general purpose reg. */
308
309
310/* segment AR */
311#define SEGMENT_AR_L_MASK (1 << 13)
312
313#define AR_TYPE_ACCESSES_MASK 1
314#define AR_TYPE_READABLE_MASK (1 << 1)
315#define AR_TYPE_WRITEABLE_MASK (1 << 2)
316#define AR_TYPE_CODE_MASK (1 << 3)
317#define AR_TYPE_MASK 0x0f
318#define AR_TYPE_BUSY_64_TSS 11
319#define AR_TYPE_BUSY_32_TSS 11
320#define AR_TYPE_BUSY_16_TSS 3
321#define AR_TYPE_LDT 2
322
323#define AR_UNUSABLE_MASK (1 << 16)
324#define AR_S_MASK (1 << 4)
325#define AR_P_MASK (1 << 7)
326#define AR_L_MASK (1 << 13)
327#define AR_DB_MASK (1 << 14)
328#define AR_G_MASK (1 << 15)
329#define AR_DPL_SHIFT 5
330#define AR_DPL(ar) (((ar) >> AR_DPL_SHIFT) & 3)
331
332#define AR_RESERVD_MASK 0xfffe0f00
333
334#define APIC_ACCESS_PAGE_PRIVATE_MEMSLOT 9
335#define IDENTITY_PAGETABLE_PRIVATE_MEMSLOT 10
336
337#define VMX_NR_VPIDS (1 << 16)
338#define VMX_VPID_EXTENT_SINGLE_CONTEXT 1
339#define VMX_VPID_EXTENT_ALL_CONTEXT 2
340
341#define VMX_EPT_EXTENT_INDIVIDUAL_ADDR 0
342#define VMX_EPT_EXTENT_CONTEXT 1
343#define VMX_EPT_EXTENT_GLOBAL 2
344#define VMX_EPT_EXTENT_INDIVIDUAL_BIT (1ull << 24)
345#define VMX_EPT_EXTENT_CONTEXT_BIT (1ull << 25)
346#define VMX_EPT_EXTENT_GLOBAL_BIT (1ull << 26)
347#define VMX_EPT_DEFAULT_GAW 3
348#define VMX_EPT_MAX_GAW 0x4
349#define VMX_EPT_MT_EPTE_SHIFT 3
350#define VMX_EPT_GAW_EPTP_SHIFT 3
351#define VMX_EPT_DEFAULT_MT 0x6ull
352#define VMX_EPT_READABLE_MASK 0x1ull
353#define VMX_EPT_WRITABLE_MASK 0x2ull
354#define VMX_EPT_EXECUTABLE_MASK 0x4ull
355#define VMX_EPT_IGMT_BIT (1ull << 6)
356
357#define VMX_EPT_IDENTITY_PAGETABLE_ADDR 0xfffbc000ul
358
359#endif
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index f1f8ff2f1fa2..cc17546a2406 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -34,11 +34,13 @@
34#include <linux/module.h> 34#include <linux/module.h>
35#include <linux/mman.h> 35#include <linux/mman.h>
36#include <linux/highmem.h> 36#include <linux/highmem.h>
37#include <linux/iommu.h>
37#include <linux/intel-iommu.h> 38#include <linux/intel-iommu.h>
38 39
39#include <asm/uaccess.h> 40#include <asm/uaccess.h>
40#include <asm/msr.h> 41#include <asm/msr.h>
41#include <asm/desc.h> 42#include <asm/desc.h>
43#include <asm/mtrr.h>
42 44
43#define MAX_IO_MSRS 256 45#define MAX_IO_MSRS 256
44#define CR0_RESERVED_BITS \ 46#define CR0_RESERVED_BITS \
@@ -86,6 +88,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
86 { "halt_wakeup", VCPU_STAT(halt_wakeup) }, 88 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
87 { "hypercalls", VCPU_STAT(hypercalls) }, 89 { "hypercalls", VCPU_STAT(hypercalls) },
88 { "request_irq", VCPU_STAT(request_irq_exits) }, 90 { "request_irq", VCPU_STAT(request_irq_exits) },
91 { "request_nmi", VCPU_STAT(request_nmi_exits) },
89 { "irq_exits", VCPU_STAT(irq_exits) }, 92 { "irq_exits", VCPU_STAT(irq_exits) },
90 { "host_state_reload", VCPU_STAT(host_state_reload) }, 93 { "host_state_reload", VCPU_STAT(host_state_reload) },
91 { "efer_reload", VCPU_STAT(efer_reload) }, 94 { "efer_reload", VCPU_STAT(efer_reload) },
@@ -93,6 +96,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
93 { "insn_emulation", VCPU_STAT(insn_emulation) }, 96 { "insn_emulation", VCPU_STAT(insn_emulation) },
94 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) }, 97 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
95 { "irq_injections", VCPU_STAT(irq_injections) }, 98 { "irq_injections", VCPU_STAT(irq_injections) },
99 { "nmi_injections", VCPU_STAT(nmi_injections) },
96 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) }, 100 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
97 { "mmu_pte_write", VM_STAT(mmu_pte_write) }, 101 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
98 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) }, 102 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
@@ -101,6 +105,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
101 { "mmu_recycled", VM_STAT(mmu_recycled) }, 105 { "mmu_recycled", VM_STAT(mmu_recycled) },
102 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) }, 106 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
103 { "mmu_unsync", VM_STAT(mmu_unsync) }, 107 { "mmu_unsync", VM_STAT(mmu_unsync) },
108 { "mmu_unsync_global", VM_STAT(mmu_unsync_global) },
104 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) }, 109 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
105 { "largepages", VM_STAT(lpages) }, 110 { "largepages", VM_STAT(lpages) },
106 { NULL } 111 { NULL }
@@ -312,6 +317,7 @@ void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
312 kvm_x86_ops->set_cr0(vcpu, cr0); 317 kvm_x86_ops->set_cr0(vcpu, cr0);
313 vcpu->arch.cr0 = cr0; 318 vcpu->arch.cr0 = cr0;
314 319
320 kvm_mmu_sync_global(vcpu);
315 kvm_mmu_reset_context(vcpu); 321 kvm_mmu_reset_context(vcpu);
316 return; 322 return;
317} 323}
@@ -355,6 +361,7 @@ void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
355 } 361 }
356 kvm_x86_ops->set_cr4(vcpu, cr4); 362 kvm_x86_ops->set_cr4(vcpu, cr4);
357 vcpu->arch.cr4 = cr4; 363 vcpu->arch.cr4 = cr4;
364 kvm_mmu_sync_global(vcpu);
358 kvm_mmu_reset_context(vcpu); 365 kvm_mmu_reset_context(vcpu);
359} 366}
360EXPORT_SYMBOL_GPL(kvm_set_cr4); 367EXPORT_SYMBOL_GPL(kvm_set_cr4);
@@ -449,7 +456,7 @@ static u32 msrs_to_save[] = {
449 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR, 456 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
450#endif 457#endif
451 MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK, 458 MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
452 MSR_IA32_PERF_STATUS, 459 MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT
453}; 460};
454 461
455static unsigned num_msrs_to_save; 462static unsigned num_msrs_to_save;
@@ -648,10 +655,38 @@ static bool msr_mtrr_valid(unsigned msr)
648 655
649static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data) 656static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
650{ 657{
658 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
659
651 if (!msr_mtrr_valid(msr)) 660 if (!msr_mtrr_valid(msr))
652 return 1; 661 return 1;
653 662
654 vcpu->arch.mtrr[msr - 0x200] = data; 663 if (msr == MSR_MTRRdefType) {
664 vcpu->arch.mtrr_state.def_type = data;
665 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
666 } else if (msr == MSR_MTRRfix64K_00000)
667 p[0] = data;
668 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
669 p[1 + msr - MSR_MTRRfix16K_80000] = data;
670 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
671 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
672 else if (msr == MSR_IA32_CR_PAT)
673 vcpu->arch.pat = data;
674 else { /* Variable MTRRs */
675 int idx, is_mtrr_mask;
676 u64 *pt;
677
678 idx = (msr - 0x200) / 2;
679 is_mtrr_mask = msr - 0x200 - 2 * idx;
680 if (!is_mtrr_mask)
681 pt =
682 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
683 else
684 pt =
685 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
686 *pt = data;
687 }
688
689 kvm_mmu_reset_context(vcpu);
655 return 0; 690 return 0;
656} 691}
657 692
@@ -747,10 +782,37 @@ int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
747 782
748static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata) 783static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
749{ 784{
785 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
786
750 if (!msr_mtrr_valid(msr)) 787 if (!msr_mtrr_valid(msr))
751 return 1; 788 return 1;
752 789
753 *pdata = vcpu->arch.mtrr[msr - 0x200]; 790 if (msr == MSR_MTRRdefType)
791 *pdata = vcpu->arch.mtrr_state.def_type +
792 (vcpu->arch.mtrr_state.enabled << 10);
793 else if (msr == MSR_MTRRfix64K_00000)
794 *pdata = p[0];
795 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
796 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
797 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
798 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
799 else if (msr == MSR_IA32_CR_PAT)
800 *pdata = vcpu->arch.pat;
801 else { /* Variable MTRRs */
802 int idx, is_mtrr_mask;
803 u64 *pt;
804
805 idx = (msr - 0x200) / 2;
806 is_mtrr_mask = msr - 0x200 - 2 * idx;
807 if (!is_mtrr_mask)
808 pt =
809 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
810 else
811 pt =
812 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
813 *pdata = *pt;
814 }
815
754 return 0; 816 return 0;
755} 817}
756 818
@@ -903,7 +965,6 @@ int kvm_dev_ioctl_check_extension(long ext)
903 case KVM_CAP_IRQCHIP: 965 case KVM_CAP_IRQCHIP:
904 case KVM_CAP_HLT: 966 case KVM_CAP_HLT:
905 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL: 967 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
906 case KVM_CAP_USER_MEMORY:
907 case KVM_CAP_SET_TSS_ADDR: 968 case KVM_CAP_SET_TSS_ADDR:
908 case KVM_CAP_EXT_CPUID: 969 case KVM_CAP_EXT_CPUID:
909 case KVM_CAP_CLOCKSOURCE: 970 case KVM_CAP_CLOCKSOURCE:
@@ -929,7 +990,7 @@ int kvm_dev_ioctl_check_extension(long ext)
929 r = !tdp_enabled; 990 r = !tdp_enabled;
930 break; 991 break;
931 case KVM_CAP_IOMMU: 992 case KVM_CAP_IOMMU:
932 r = intel_iommu_found(); 993 r = iommu_found();
933 break; 994 break;
934 default: 995 default:
935 r = 0; 996 r = 0;
@@ -1188,6 +1249,7 @@ static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1188 int t, times = entry->eax & 0xff; 1249 int t, times = entry->eax & 0xff;
1189 1250
1190 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; 1251 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1252 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1191 for (t = 1; t < times && *nent < maxnent; ++t) { 1253 for (t = 1; t < times && *nent < maxnent; ++t) {
1192 do_cpuid_1_ent(&entry[t], function, 0); 1254 do_cpuid_1_ent(&entry[t], function, 0);
1193 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; 1255 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
@@ -1218,7 +1280,7 @@ static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1218 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 1280 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1219 /* read more entries until level_type is zero */ 1281 /* read more entries until level_type is zero */
1220 for (i = 1; *nent < maxnent; ++i) { 1282 for (i = 1; *nent < maxnent; ++i) {
1221 level_type = entry[i - 1].ecx & 0xff; 1283 level_type = entry[i - 1].ecx & 0xff00;
1222 if (!level_type) 1284 if (!level_type)
1223 break; 1285 break;
1224 do_cpuid_1_ent(&entry[i], function, i); 1286 do_cpuid_1_ent(&entry[i], function, i);
@@ -1318,6 +1380,15 @@ static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1318 return 0; 1380 return 0;
1319} 1381}
1320 1382
1383static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1384{
1385 vcpu_load(vcpu);
1386 kvm_inject_nmi(vcpu);
1387 vcpu_put(vcpu);
1388
1389 return 0;
1390}
1391
1321static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu, 1392static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1322 struct kvm_tpr_access_ctl *tac) 1393 struct kvm_tpr_access_ctl *tac)
1323{ 1394{
@@ -1377,6 +1448,13 @@ long kvm_arch_vcpu_ioctl(struct file *filp,
1377 r = 0; 1448 r = 0;
1378 break; 1449 break;
1379 } 1450 }
1451 case KVM_NMI: {
1452 r = kvm_vcpu_ioctl_nmi(vcpu);
1453 if (r)
1454 goto out;
1455 r = 0;
1456 break;
1457 }
1380 case KVM_SET_CPUID: { 1458 case KVM_SET_CPUID: {
1381 struct kvm_cpuid __user *cpuid_arg = argp; 1459 struct kvm_cpuid __user *cpuid_arg = argp;
1382 struct kvm_cpuid cpuid; 1460 struct kvm_cpuid cpuid;
@@ -1968,7 +2046,7 @@ int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1968 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes); 2046 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1969 if (ret < 0) 2047 if (ret < 0)
1970 return 0; 2048 return 0;
1971 kvm_mmu_pte_write(vcpu, gpa, val, bytes); 2049 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
1972 return 1; 2050 return 1;
1973} 2051}
1974 2052
@@ -2404,8 +2482,6 @@ int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2404 val = kvm_register_read(vcpu, VCPU_REGS_RAX); 2482 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2405 memcpy(vcpu->arch.pio_data, &val, 4); 2483 memcpy(vcpu->arch.pio_data, &val, 4);
2406 2484
2407 kvm_x86_ops->skip_emulated_instruction(vcpu);
2408
2409 pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in); 2485 pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2410 if (pio_dev) { 2486 if (pio_dev) {
2411 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data); 2487 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
@@ -2541,7 +2617,7 @@ int kvm_arch_init(void *opaque)
2541 kvm_mmu_set_nonpresent_ptes(0ull, 0ull); 2617 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2542 kvm_mmu_set_base_ptes(PT_PRESENT_MASK); 2618 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2543 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK, 2619 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2544 PT_DIRTY_MASK, PT64_NX_MASK, 0); 2620 PT_DIRTY_MASK, PT64_NX_MASK, 0, 0);
2545 return 0; 2621 return 0;
2546 2622
2547out: 2623out:
@@ -2729,7 +2805,7 @@ static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2729 2805
2730 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT; 2806 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2731 /* when no next entry is found, the current entry[i] is reselected */ 2807 /* when no next entry is found, the current entry[i] is reselected */
2732 for (j = i + 1; j == i; j = (j + 1) % nent) { 2808 for (j = i + 1; ; j = (j + 1) % nent) {
2733 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j]; 2809 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2734 if (ej->function == e->function) { 2810 if (ej->function == e->function) {
2735 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; 2811 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
@@ -2973,7 +3049,7 @@ static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2973 pr_debug("vcpu %d received sipi with vector # %x\n", 3049 pr_debug("vcpu %d received sipi with vector # %x\n",
2974 vcpu->vcpu_id, vcpu->arch.sipi_vector); 3050 vcpu->vcpu_id, vcpu->arch.sipi_vector);
2975 kvm_lapic_reset(vcpu); 3051 kvm_lapic_reset(vcpu);
2976 r = kvm_x86_ops->vcpu_reset(vcpu); 3052 r = kvm_arch_vcpu_reset(vcpu);
2977 if (r) 3053 if (r)
2978 return r; 3054 return r;
2979 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; 3055 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
@@ -3275,9 +3351,9 @@ static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3275 kvm_desct->padding = 0; 3351 kvm_desct->padding = 0;
3276} 3352}
3277 3353
3278static void get_segment_descritptor_dtable(struct kvm_vcpu *vcpu, 3354static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3279 u16 selector, 3355 u16 selector,
3280 struct descriptor_table *dtable) 3356 struct descriptor_table *dtable)
3281{ 3357{
3282 if (selector & 1 << 2) { 3358 if (selector & 1 << 2) {
3283 struct kvm_segment kvm_seg; 3359 struct kvm_segment kvm_seg;
@@ -3302,7 +3378,7 @@ static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3302 struct descriptor_table dtable; 3378 struct descriptor_table dtable;
3303 u16 index = selector >> 3; 3379 u16 index = selector >> 3;
3304 3380
3305 get_segment_descritptor_dtable(vcpu, selector, &dtable); 3381 get_segment_descriptor_dtable(vcpu, selector, &dtable);
3306 3382
3307 if (dtable.limit < index * 8 + 7) { 3383 if (dtable.limit < index * 8 + 7) {
3308 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc); 3384 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
@@ -3321,7 +3397,7 @@ static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3321 struct descriptor_table dtable; 3397 struct descriptor_table dtable;
3322 u16 index = selector >> 3; 3398 u16 index = selector >> 3;
3323 3399
3324 get_segment_descritptor_dtable(vcpu, selector, &dtable); 3400 get_segment_descriptor_dtable(vcpu, selector, &dtable);
3325 3401
3326 if (dtable.limit < index * 8 + 7) 3402 if (dtable.limit < index * 8 + 7)
3327 return 1; 3403 return 1;
@@ -3900,6 +3976,7 @@ int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3900 /* We do fxsave: this must be aligned. */ 3976 /* We do fxsave: this must be aligned. */
3901 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF); 3977 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3902 3978
3979 vcpu->arch.mtrr_state.have_fixed = 1;
3903 vcpu_load(vcpu); 3980 vcpu_load(vcpu);
3904 r = kvm_arch_vcpu_reset(vcpu); 3981 r = kvm_arch_vcpu_reset(vcpu);
3905 if (r == 0) 3982 if (r == 0)
@@ -3925,6 +4002,9 @@ void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3925 4002
3926int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu) 4003int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3927{ 4004{
4005 vcpu->arch.nmi_pending = false;
4006 vcpu->arch.nmi_injected = false;
4007
3928 return kvm_x86_ops->vcpu_reset(vcpu); 4008 return kvm_x86_ops->vcpu_reset(vcpu);
3929} 4009}
3930 4010
@@ -4012,6 +4092,7 @@ struct kvm *kvm_arch_create_vm(void)
4012 return ERR_PTR(-ENOMEM); 4092 return ERR_PTR(-ENOMEM);
4013 4093
4014 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); 4094 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4095 INIT_LIST_HEAD(&kvm->arch.oos_global_pages);
4015 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head); 4096 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4016 4097
4017 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */ 4098 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
@@ -4048,8 +4129,8 @@ static void kvm_free_vcpus(struct kvm *kvm)
4048 4129
4049void kvm_arch_destroy_vm(struct kvm *kvm) 4130void kvm_arch_destroy_vm(struct kvm *kvm)
4050{ 4131{
4051 kvm_iommu_unmap_guest(kvm);
4052 kvm_free_all_assigned_devices(kvm); 4132 kvm_free_all_assigned_devices(kvm);
4133 kvm_iommu_unmap_guest(kvm);
4053 kvm_free_pit(kvm); 4134 kvm_free_pit(kvm);
4054 kfree(kvm->arch.vpic); 4135 kfree(kvm->arch.vpic);
4055 kfree(kvm->arch.vioapic); 4136 kfree(kvm->arch.vioapic);
@@ -4127,7 +4208,8 @@ void kvm_arch_flush_shadow(struct kvm *kvm)
4127int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu) 4208int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4128{ 4209{
4129 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE 4210 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4130 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED; 4211 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4212 || vcpu->arch.nmi_pending;
4131} 4213}
4132 4214
4133static void vcpu_kick_intr(void *info) 4215static void vcpu_kick_intr(void *info)
diff --git a/arch/x86/kvm/x86_emulate.c b/arch/x86/kvm/x86_emulate.c
index ea051173b0da..d174db7a3370 100644
--- a/arch/x86/kvm/x86_emulate.c
+++ b/arch/x86/kvm/x86_emulate.c
@@ -58,6 +58,7 @@
58#define SrcMem32 (4<<4) /* Memory operand (32-bit). */ 58#define SrcMem32 (4<<4) /* Memory operand (32-bit). */
59#define SrcImm (5<<4) /* Immediate operand. */ 59#define SrcImm (5<<4) /* Immediate operand. */
60#define SrcImmByte (6<<4) /* 8-bit sign-extended immediate operand. */ 60#define SrcImmByte (6<<4) /* 8-bit sign-extended immediate operand. */
61#define SrcOne (7<<4) /* Implied '1' */
61#define SrcMask (7<<4) 62#define SrcMask (7<<4)
62/* Generic ModRM decode. */ 63/* Generic ModRM decode. */
63#define ModRM (1<<7) 64#define ModRM (1<<7)
@@ -70,17 +71,23 @@
70#define Group (1<<14) /* Bits 3:5 of modrm byte extend opcode */ 71#define Group (1<<14) /* Bits 3:5 of modrm byte extend opcode */
71#define GroupDual (1<<15) /* Alternate decoding of mod == 3 */ 72#define GroupDual (1<<15) /* Alternate decoding of mod == 3 */
72#define GroupMask 0xff /* Group number stored in bits 0:7 */ 73#define GroupMask 0xff /* Group number stored in bits 0:7 */
74/* Source 2 operand type */
75#define Src2None (0<<29)
76#define Src2CL (1<<29)
77#define Src2ImmByte (2<<29)
78#define Src2One (3<<29)
79#define Src2Mask (7<<29)
73 80
74enum { 81enum {
75 Group1_80, Group1_81, Group1_82, Group1_83, 82 Group1_80, Group1_81, Group1_82, Group1_83,
76 Group1A, Group3_Byte, Group3, Group4, Group5, Group7, 83 Group1A, Group3_Byte, Group3, Group4, Group5, Group7,
77}; 84};
78 85
79static u16 opcode_table[256] = { 86static u32 opcode_table[256] = {
80 /* 0x00 - 0x07 */ 87 /* 0x00 - 0x07 */
81 ByteOp | DstMem | SrcReg | ModRM, DstMem | SrcReg | ModRM, 88 ByteOp | DstMem | SrcReg | ModRM, DstMem | SrcReg | ModRM,
82 ByteOp | DstReg | SrcMem | ModRM, DstReg | SrcMem | ModRM, 89 ByteOp | DstReg | SrcMem | ModRM, DstReg | SrcMem | ModRM,
83 0, 0, 0, 0, 90 ByteOp | DstAcc | SrcImm, DstAcc | SrcImm, 0, 0,
84 /* 0x08 - 0x0F */ 91 /* 0x08 - 0x0F */
85 ByteOp | DstMem | SrcReg | ModRM, DstMem | SrcReg | ModRM, 92 ByteOp | DstMem | SrcReg | ModRM, DstMem | SrcReg | ModRM,
86 ByteOp | DstReg | SrcMem | ModRM, DstReg | SrcMem | ModRM, 93 ByteOp | DstReg | SrcMem | ModRM, DstReg | SrcMem | ModRM,
@@ -195,7 +202,7 @@ static u16 opcode_table[256] = {
195 ImplicitOps, ImplicitOps, Group | Group4, Group | Group5, 202 ImplicitOps, ImplicitOps, Group | Group4, Group | Group5,
196}; 203};
197 204
198static u16 twobyte_table[256] = { 205static u32 twobyte_table[256] = {
199 /* 0x00 - 0x0F */ 206 /* 0x00 - 0x0F */
200 0, Group | GroupDual | Group7, 0, 0, 0, 0, ImplicitOps, 0, 207 0, Group | GroupDual | Group7, 0, 0, 0, 0, ImplicitOps, 0,
201 ImplicitOps, ImplicitOps, 0, 0, 0, ImplicitOps | ModRM, 0, 0, 208 ImplicitOps, ImplicitOps, 0, 0, 0, ImplicitOps | ModRM, 0, 0,
@@ -230,9 +237,14 @@ static u16 twobyte_table[256] = {
230 /* 0x90 - 0x9F */ 237 /* 0x90 - 0x9F */
231 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 238 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
232 /* 0xA0 - 0xA7 */ 239 /* 0xA0 - 0xA7 */
233 0, 0, 0, DstMem | SrcReg | ModRM | BitOp, 0, 0, 0, 0, 240 0, 0, 0, DstMem | SrcReg | ModRM | BitOp,
241 DstMem | SrcReg | Src2ImmByte | ModRM,
242 DstMem | SrcReg | Src2CL | ModRM, 0, 0,
234 /* 0xA8 - 0xAF */ 243 /* 0xA8 - 0xAF */
235 0, 0, 0, DstMem | SrcReg | ModRM | BitOp, 0, 0, ModRM, 0, 244 0, 0, 0, DstMem | SrcReg | ModRM | BitOp,
245 DstMem | SrcReg | Src2ImmByte | ModRM,
246 DstMem | SrcReg | Src2CL | ModRM,
247 ModRM, 0,
236 /* 0xB0 - 0xB7 */ 248 /* 0xB0 - 0xB7 */
237 ByteOp | DstMem | SrcReg | ModRM, DstMem | SrcReg | ModRM, 0, 249 ByteOp | DstMem | SrcReg | ModRM, DstMem | SrcReg | ModRM, 0,
238 DstMem | SrcReg | ModRM | BitOp, 250 DstMem | SrcReg | ModRM | BitOp,
@@ -253,7 +265,7 @@ static u16 twobyte_table[256] = {
253 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 265 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
254}; 266};
255 267
256static u16 group_table[] = { 268static u32 group_table[] = {
257 [Group1_80*8] = 269 [Group1_80*8] =
258 ByteOp | DstMem | SrcImm | ModRM, ByteOp | DstMem | SrcImm | ModRM, 270 ByteOp | DstMem | SrcImm | ModRM, ByteOp | DstMem | SrcImm | ModRM,
259 ByteOp | DstMem | SrcImm | ModRM, ByteOp | DstMem | SrcImm | ModRM, 271 ByteOp | DstMem | SrcImm | ModRM, ByteOp | DstMem | SrcImm | ModRM,
@@ -297,9 +309,9 @@ static u16 group_table[] = {
297 SrcMem16 | ModRM | Mov, SrcMem | ModRM | ByteOp, 309 SrcMem16 | ModRM | Mov, SrcMem | ModRM | ByteOp,
298}; 310};
299 311
300static u16 group2_table[] = { 312static u32 group2_table[] = {
301 [Group7*8] = 313 [Group7*8] =
302 SrcNone | ModRM, 0, 0, 0, 314 SrcNone | ModRM, 0, 0, SrcNone | ModRM,
303 SrcNone | ModRM | DstMem | Mov, 0, 315 SrcNone | ModRM | DstMem | Mov, 0,
304 SrcMem16 | ModRM | Mov, 0, 316 SrcMem16 | ModRM | Mov, 0,
305}; 317};
@@ -359,49 +371,48 @@ static u16 group2_table[] = {
359 "andl %"_msk",%"_LO32 _tmp"; " \ 371 "andl %"_msk",%"_LO32 _tmp"; " \
360 "orl %"_LO32 _tmp",%"_sav"; " 372 "orl %"_LO32 _tmp",%"_sav"; "
361 373
374#ifdef CONFIG_X86_64
375#define ON64(x) x
376#else
377#define ON64(x)
378#endif
379
380#define ____emulate_2op(_op, _src, _dst, _eflags, _x, _y, _suffix) \
381 do { \
382 __asm__ __volatile__ ( \
383 _PRE_EFLAGS("0", "4", "2") \
384 _op _suffix " %"_x"3,%1; " \
385 _POST_EFLAGS("0", "4", "2") \
386 : "=m" (_eflags), "=m" ((_dst).val), \
387 "=&r" (_tmp) \
388 : _y ((_src).val), "i" (EFLAGS_MASK)); \
389 } while (0)
390
391
362/* Raw emulation: instruction has two explicit operands. */ 392/* Raw emulation: instruction has two explicit operands. */
363#define __emulate_2op_nobyte(_op,_src,_dst,_eflags,_wx,_wy,_lx,_ly,_qx,_qy) \ 393#define __emulate_2op_nobyte(_op,_src,_dst,_eflags,_wx,_wy,_lx,_ly,_qx,_qy) \
364 do { \ 394 do { \
365 unsigned long _tmp; \ 395 unsigned long _tmp; \
366 \ 396 \
367 switch ((_dst).bytes) { \ 397 switch ((_dst).bytes) { \
368 case 2: \ 398 case 2: \
369 __asm__ __volatile__ ( \ 399 ____emulate_2op(_op,_src,_dst,_eflags,_wx,_wy,"w"); \
370 _PRE_EFLAGS("0", "4", "2") \ 400 break; \
371 _op"w %"_wx"3,%1; " \ 401 case 4: \
372 _POST_EFLAGS("0", "4", "2") \ 402 ____emulate_2op(_op,_src,_dst,_eflags,_lx,_ly,"l"); \
373 : "=m" (_eflags), "=m" ((_dst).val), \ 403 break; \
374 "=&r" (_tmp) \ 404 case 8: \
375 : _wy ((_src).val), "i" (EFLAGS_MASK)); \ 405 ON64(____emulate_2op(_op,_src,_dst,_eflags,_qx,_qy,"q")); \
376 break; \ 406 break; \
377 case 4: \ 407 } \
378 __asm__ __volatile__ ( \
379 _PRE_EFLAGS("0", "4", "2") \
380 _op"l %"_lx"3,%1; " \
381 _POST_EFLAGS("0", "4", "2") \
382 : "=m" (_eflags), "=m" ((_dst).val), \
383 "=&r" (_tmp) \
384 : _ly ((_src).val), "i" (EFLAGS_MASK)); \
385 break; \
386 case 8: \
387 __emulate_2op_8byte(_op, _src, _dst, \
388 _eflags, _qx, _qy); \
389 break; \
390 } \
391 } while (0) 408 } while (0)
392 409
393#define __emulate_2op(_op,_src,_dst,_eflags,_bx,_by,_wx,_wy,_lx,_ly,_qx,_qy) \ 410#define __emulate_2op(_op,_src,_dst,_eflags,_bx,_by,_wx,_wy,_lx,_ly,_qx,_qy) \
394 do { \ 411 do { \
395 unsigned long __tmp; \ 412 unsigned long _tmp; \
396 switch ((_dst).bytes) { \ 413 switch ((_dst).bytes) { \
397 case 1: \ 414 case 1: \
398 __asm__ __volatile__ ( \ 415 ____emulate_2op(_op,_src,_dst,_eflags,_bx,_by,"b"); \
399 _PRE_EFLAGS("0", "4", "2") \
400 _op"b %"_bx"3,%1; " \
401 _POST_EFLAGS("0", "4", "2") \
402 : "=m" (_eflags), "=m" ((_dst).val), \
403 "=&r" (__tmp) \
404 : _by ((_src).val), "i" (EFLAGS_MASK)); \
405 break; \ 416 break; \
406 default: \ 417 default: \
407 __emulate_2op_nobyte(_op, _src, _dst, _eflags, \ 418 __emulate_2op_nobyte(_op, _src, _dst, _eflags, \
@@ -425,71 +436,68 @@ static u16 group2_table[] = {
425 __emulate_2op_nobyte(_op, _src, _dst, _eflags, \ 436 __emulate_2op_nobyte(_op, _src, _dst, _eflags, \
426 "w", "r", _LO32, "r", "", "r") 437 "w", "r", _LO32, "r", "", "r")
427 438
428/* Instruction has only one explicit operand (no source operand). */ 439/* Instruction has three operands and one operand is stored in ECX register */
429#define emulate_1op(_op, _dst, _eflags) \ 440#define __emulate_2op_cl(_op, _cl, _src, _dst, _eflags, _suffix, _type) \
430 do { \ 441 do { \
431 unsigned long _tmp; \ 442 unsigned long _tmp; \
432 \ 443 _type _clv = (_cl).val; \
433 switch ((_dst).bytes) { \ 444 _type _srcv = (_src).val; \
434 case 1: \ 445 _type _dstv = (_dst).val; \
435 __asm__ __volatile__ ( \ 446 \
436 _PRE_EFLAGS("0", "3", "2") \ 447 __asm__ __volatile__ ( \
437 _op"b %1; " \ 448 _PRE_EFLAGS("0", "5", "2") \
438 _POST_EFLAGS("0", "3", "2") \ 449 _op _suffix " %4,%1 \n" \
439 : "=m" (_eflags), "=m" ((_dst).val), \ 450 _POST_EFLAGS("0", "5", "2") \
440 "=&r" (_tmp) \ 451 : "=m" (_eflags), "+r" (_dstv), "=&r" (_tmp) \
441 : "i" (EFLAGS_MASK)); \ 452 : "c" (_clv) , "r" (_srcv), "i" (EFLAGS_MASK) \
442 break; \ 453 ); \
443 case 2: \ 454 \
444 __asm__ __volatile__ ( \ 455 (_cl).val = (unsigned long) _clv; \
445 _PRE_EFLAGS("0", "3", "2") \ 456 (_src).val = (unsigned long) _srcv; \
446 _op"w %1; " \ 457 (_dst).val = (unsigned long) _dstv; \
447 _POST_EFLAGS("0", "3", "2") \
448 : "=m" (_eflags), "=m" ((_dst).val), \
449 "=&r" (_tmp) \
450 : "i" (EFLAGS_MASK)); \
451 break; \
452 case 4: \
453 __asm__ __volatile__ ( \
454 _PRE_EFLAGS("0", "3", "2") \
455 _op"l %1; " \
456 _POST_EFLAGS("0", "3", "2") \
457 : "=m" (_eflags), "=m" ((_dst).val), \
458 "=&r" (_tmp) \
459 : "i" (EFLAGS_MASK)); \
460 break; \
461 case 8: \
462 __emulate_1op_8byte(_op, _dst, _eflags); \
463 break; \
464 } \
465 } while (0) 458 } while (0)
466 459
467/* Emulate an instruction with quadword operands (x86/64 only). */ 460#define emulate_2op_cl(_op, _cl, _src, _dst, _eflags) \
468#if defined(CONFIG_X86_64) 461 do { \
469#define __emulate_2op_8byte(_op, _src, _dst, _eflags, _qx, _qy) \ 462 switch ((_dst).bytes) { \
470 do { \ 463 case 2: \
471 __asm__ __volatile__ ( \ 464 __emulate_2op_cl(_op, _cl, _src, _dst, _eflags, \
472 _PRE_EFLAGS("0", "4", "2") \ 465 "w", unsigned short); \
473 _op"q %"_qx"3,%1; " \ 466 break; \
474 _POST_EFLAGS("0", "4", "2") \ 467 case 4: \
475 : "=m" (_eflags), "=m" ((_dst).val), "=&r" (_tmp) \ 468 __emulate_2op_cl(_op, _cl, _src, _dst, _eflags, \
476 : _qy ((_src).val), "i" (EFLAGS_MASK)); \ 469 "l", unsigned int); \
470 break; \
471 case 8: \
472 ON64(__emulate_2op_cl(_op, _cl, _src, _dst, _eflags, \
473 "q", unsigned long)); \
474 break; \
475 } \
477 } while (0) 476 } while (0)
478 477
479#define __emulate_1op_8byte(_op, _dst, _eflags) \ 478#define __emulate_1op(_op, _dst, _eflags, _suffix) \
480 do { \ 479 do { \
481 __asm__ __volatile__ ( \ 480 unsigned long _tmp; \
482 _PRE_EFLAGS("0", "3", "2") \ 481 \
483 _op"q %1; " \ 482 __asm__ __volatile__ ( \
484 _POST_EFLAGS("0", "3", "2") \ 483 _PRE_EFLAGS("0", "3", "2") \
485 : "=m" (_eflags), "=m" ((_dst).val), "=&r" (_tmp) \ 484 _op _suffix " %1; " \
486 : "i" (EFLAGS_MASK)); \ 485 _POST_EFLAGS("0", "3", "2") \
486 : "=m" (_eflags), "+m" ((_dst).val), \
487 "=&r" (_tmp) \
488 : "i" (EFLAGS_MASK)); \
487 } while (0) 489 } while (0)
488 490
489#elif defined(__i386__) 491/* Instruction has only one explicit operand (no source operand). */
490#define __emulate_2op_8byte(_op, _src, _dst, _eflags, _qx, _qy) 492#define emulate_1op(_op, _dst, _eflags) \
491#define __emulate_1op_8byte(_op, _dst, _eflags) 493 do { \
492#endif /* __i386__ */ 494 switch ((_dst).bytes) { \
495 case 1: __emulate_1op(_op, _dst, _eflags, "b"); break; \
496 case 2: __emulate_1op(_op, _dst, _eflags, "w"); break; \
497 case 4: __emulate_1op(_op, _dst, _eflags, "l"); break; \
498 case 8: ON64(__emulate_1op(_op, _dst, _eflags, "q")); break; \
499 } \
500 } while (0)
493 501
494/* Fetch next part of the instruction being emulated. */ 502/* Fetch next part of the instruction being emulated. */
495#define insn_fetch(_type, _size, _eip) \ 503#define insn_fetch(_type, _size, _eip) \
@@ -1041,6 +1049,33 @@ done_prefixes:
1041 c->src.bytes = 1; 1049 c->src.bytes = 1;
1042 c->src.val = insn_fetch(s8, 1, c->eip); 1050 c->src.val = insn_fetch(s8, 1, c->eip);
1043 break; 1051 break;
1052 case SrcOne:
1053 c->src.bytes = 1;
1054 c->src.val = 1;
1055 break;
1056 }
1057
1058 /*
1059 * Decode and fetch the second source operand: register, memory
1060 * or immediate.
1061 */
1062 switch (c->d & Src2Mask) {
1063 case Src2None:
1064 break;
1065 case Src2CL:
1066 c->src2.bytes = 1;
1067 c->src2.val = c->regs[VCPU_REGS_RCX] & 0x8;
1068 break;
1069 case Src2ImmByte:
1070 c->src2.type = OP_IMM;
1071 c->src2.ptr = (unsigned long *)c->eip;
1072 c->src2.bytes = 1;
1073 c->src2.val = insn_fetch(u8, 1, c->eip);
1074 break;
1075 case Src2One:
1076 c->src2.bytes = 1;
1077 c->src2.val = 1;
1078 break;
1044 } 1079 }
1045 1080
1046 /* Decode and fetch the destination operand: register or memory. */ 1081 /* Decode and fetch the destination operand: register or memory. */
@@ -1100,20 +1135,33 @@ static inline void emulate_push(struct x86_emulate_ctxt *ctxt)
1100 c->regs[VCPU_REGS_RSP]); 1135 c->regs[VCPU_REGS_RSP]);
1101} 1136}
1102 1137
1103static inline int emulate_grp1a(struct x86_emulate_ctxt *ctxt, 1138static int emulate_pop(struct x86_emulate_ctxt *ctxt,
1104 struct x86_emulate_ops *ops) 1139 struct x86_emulate_ops *ops)
1105{ 1140{
1106 struct decode_cache *c = &ctxt->decode; 1141 struct decode_cache *c = &ctxt->decode;
1107 int rc; 1142 int rc;
1108 1143
1109 rc = ops->read_std(register_address(c, ss_base(ctxt), 1144 rc = ops->read_emulated(register_address(c, ss_base(ctxt),
1110 c->regs[VCPU_REGS_RSP]), 1145 c->regs[VCPU_REGS_RSP]),
1111 &c->dst.val, c->dst.bytes, ctxt->vcpu); 1146 &c->src.val, c->src.bytes, ctxt->vcpu);
1112 if (rc != 0) 1147 if (rc != 0)
1113 return rc; 1148 return rc;
1114 1149
1115 register_address_increment(c, &c->regs[VCPU_REGS_RSP], c->dst.bytes); 1150 register_address_increment(c, &c->regs[VCPU_REGS_RSP], c->src.bytes);
1151 return rc;
1152}
1153
1154static inline int emulate_grp1a(struct x86_emulate_ctxt *ctxt,
1155 struct x86_emulate_ops *ops)
1156{
1157 struct decode_cache *c = &ctxt->decode;
1158 int rc;
1116 1159
1160 c->src.bytes = c->dst.bytes;
1161 rc = emulate_pop(ctxt, ops);
1162 if (rc != 0)
1163 return rc;
1164 c->dst.val = c->src.val;
1117 return 0; 1165 return 0;
1118} 1166}
1119 1167
@@ -1415,24 +1463,15 @@ special_insn:
1415 emulate_1op("dec", c->dst, ctxt->eflags); 1463 emulate_1op("dec", c->dst, ctxt->eflags);
1416 break; 1464 break;
1417 case 0x50 ... 0x57: /* push reg */ 1465 case 0x50 ... 0x57: /* push reg */
1418 c->dst.type = OP_MEM; 1466 emulate_push(ctxt);
1419 c->dst.bytes = c->op_bytes;
1420 c->dst.val = c->src.val;
1421 register_address_increment(c, &c->regs[VCPU_REGS_RSP],
1422 -c->op_bytes);
1423 c->dst.ptr = (void *) register_address(
1424 c, ss_base(ctxt), c->regs[VCPU_REGS_RSP]);
1425 break; 1467 break;
1426 case 0x58 ... 0x5f: /* pop reg */ 1468 case 0x58 ... 0x5f: /* pop reg */
1427 pop_instruction: 1469 pop_instruction:
1428 if ((rc = ops->read_std(register_address(c, ss_base(ctxt), 1470 c->src.bytes = c->op_bytes;
1429 c->regs[VCPU_REGS_RSP]), c->dst.ptr, 1471 rc = emulate_pop(ctxt, ops);
1430 c->op_bytes, ctxt->vcpu)) != 0) 1472 if (rc != 0)
1431 goto done; 1473 goto done;
1432 1474 c->dst.val = c->src.val;
1433 register_address_increment(c, &c->regs[VCPU_REGS_RSP],
1434 c->op_bytes);
1435 c->dst.type = OP_NONE; /* Disable writeback. */
1436 break; 1475 break;
1437 case 0x63: /* movsxd */ 1476 case 0x63: /* movsxd */
1438 if (ctxt->mode != X86EMUL_MODE_PROT64) 1477 if (ctxt->mode != X86EMUL_MODE_PROT64)
@@ -1591,7 +1630,9 @@ special_insn:
1591 emulate_push(ctxt); 1630 emulate_push(ctxt);
1592 break; 1631 break;
1593 case 0x9d: /* popf */ 1632 case 0x9d: /* popf */
1633 c->dst.type = OP_REG;
1594 c->dst.ptr = (unsigned long *) &ctxt->eflags; 1634 c->dst.ptr = (unsigned long *) &ctxt->eflags;
1635 c->dst.bytes = c->op_bytes;
1595 goto pop_instruction; 1636 goto pop_instruction;
1596 case 0xa0 ... 0xa1: /* mov */ 1637 case 0xa0 ... 0xa1: /* mov */
1597 c->dst.ptr = (unsigned long *)&c->regs[VCPU_REGS_RAX]; 1638 c->dst.ptr = (unsigned long *)&c->regs[VCPU_REGS_RAX];
@@ -1689,7 +1730,9 @@ special_insn:
1689 emulate_grp2(ctxt); 1730 emulate_grp2(ctxt);
1690 break; 1731 break;
1691 case 0xc3: /* ret */ 1732 case 0xc3: /* ret */
1733 c->dst.type = OP_REG;
1692 c->dst.ptr = &c->eip; 1734 c->dst.ptr = &c->eip;
1735 c->dst.bytes = c->op_bytes;
1693 goto pop_instruction; 1736 goto pop_instruction;
1694 case 0xc6 ... 0xc7: /* mov (sole member of Grp11) */ 1737 case 0xc6 ... 0xc7: /* mov (sole member of Grp11) */
1695 mov: 1738 mov:
@@ -1778,7 +1821,7 @@ special_insn:
1778 c->eip = saved_eip; 1821 c->eip = saved_eip;
1779 goto cannot_emulate; 1822 goto cannot_emulate;
1780 } 1823 }
1781 return 0; 1824 break;
1782 case 0xf4: /* hlt */ 1825 case 0xf4: /* hlt */
1783 ctxt->vcpu->arch.halt_request = 1; 1826 ctxt->vcpu->arch.halt_request = 1;
1784 break; 1827 break;
@@ -1999,12 +2042,20 @@ twobyte_insn:
1999 c->src.val &= (c->dst.bytes << 3) - 1; 2042 c->src.val &= (c->dst.bytes << 3) - 1;
2000 emulate_2op_SrcV_nobyte("bt", c->src, c->dst, ctxt->eflags); 2043 emulate_2op_SrcV_nobyte("bt", c->src, c->dst, ctxt->eflags);
2001 break; 2044 break;
2045 case 0xa4: /* shld imm8, r, r/m */
2046 case 0xa5: /* shld cl, r, r/m */
2047 emulate_2op_cl("shld", c->src2, c->src, c->dst, ctxt->eflags);
2048 break;
2002 case 0xab: 2049 case 0xab:
2003 bts: /* bts */ 2050 bts: /* bts */
2004 /* only subword offset */ 2051 /* only subword offset */
2005 c->src.val &= (c->dst.bytes << 3) - 1; 2052 c->src.val &= (c->dst.bytes << 3) - 1;
2006 emulate_2op_SrcV_nobyte("bts", c->src, c->dst, ctxt->eflags); 2053 emulate_2op_SrcV_nobyte("bts", c->src, c->dst, ctxt->eflags);
2007 break; 2054 break;
2055 case 0xac: /* shrd imm8, r, r/m */
2056 case 0xad: /* shrd cl, r, r/m */
2057 emulate_2op_cl("shrd", c->src2, c->src, c->dst, ctxt->eflags);
2058 break;
2008 case 0xae: /* clflush */ 2059 case 0xae: /* clflush */
2009 break; 2060 break;
2010 case 0xb0 ... 0xb1: /* cmpxchg */ 2061 case 0xb0 ... 0xb1: /* cmpxchg */
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-15 07:38:30 -0500