blob: 959cb59cfaeb29fc8c95d209737560a090ba1870 [file] [log] [blame]
Avi Kivity6aa8b732006-12-10 02:21:36 -08001/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 *
9 * Authors:
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
15 *
16 */
17
Eddie Dong85f455f2007-07-06 12:20:49 +030018#include "irq.h"
Zhang Xiantao1d737c82007-12-14 09:35:10 +080019#include "mmu.h"
Avi Kivitye4956062007-06-28 14:15:57 -040020
Avi Kivityedf88412007-12-16 11:02:48 +020021#include <linux/kvm_host.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080022#include <linux/module.h>
Ahmed S. Darwish9d8f5492007-02-19 14:37:46 +020023#include <linux/kernel.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080024#include <linux/mm.h>
25#include <linux/highmem.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040026#include <linux/sched.h>
Avi Kivityc7addb92007-09-16 18:58:32 +020027#include <linux/moduleparam.h>
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -030028#include "kvm_cache_regs.h"
Avi Kivity35920a32008-07-03 14:50:12 +030029#include "x86.h"
Avi Kivitye4956062007-06-28 14:15:57 -040030
Avi Kivity6aa8b732006-12-10 02:21:36 -080031#include <asm/io.h>
Anthony Liguori3b3be0d2006-12-13 00:33:43 -080032#include <asm/desc.h>
Eduardo Habkost13673a92008-11-17 19:03:13 -020033#include <asm/vmx.h>
Eduardo Habkost6210e372008-11-17 19:03:16 -020034#include <asm/virtext.h>
Andi Kleena0861c02009-06-08 17:37:09 +080035#include <asm/mce.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080036
Avi Kivity4ecac3f2008-05-13 13:23:38 +030037#define __ex(x) __kvm_handle_fault_on_reboot(x)
38
Avi Kivity6aa8b732006-12-10 02:21:36 -080039MODULE_AUTHOR("Qumranet");
40MODULE_LICENSE("GPL");
41
Avi Kivity4462d212009-03-23 17:53:37 +020042static int __read_mostly bypass_guest_pf = 1;
Avi Kivityc1f8bc02009-03-23 15:41:17 +020043module_param(bypass_guest_pf, bool, S_IRUGO);
Avi Kivityc7addb92007-09-16 18:58:32 +020044
Avi Kivity4462d212009-03-23 17:53:37 +020045static int __read_mostly enable_vpid = 1;
Avi Kivity736caef2009-03-23 17:39:48 +020046module_param_named(vpid, enable_vpid, bool, 0444);
Sheng Yang2384d2b2008-01-17 15:14:33 +080047
Avi Kivity4462d212009-03-23 17:53:37 +020048static int __read_mostly flexpriority_enabled = 1;
Avi Kivity736caef2009-03-23 17:39:48 +020049module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
Avi Kivity4c9fc8e2008-03-24 18:15:14 +020050
Avi Kivity4462d212009-03-23 17:53:37 +020051static int __read_mostly enable_ept = 1;
Avi Kivity736caef2009-03-23 17:39:48 +020052module_param_named(ept, enable_ept, bool, S_IRUGO);
Sheng Yangd56f5462008-04-25 10:13:16 +080053
Avi Kivity4462d212009-03-23 17:53:37 +020054static int __read_mostly emulate_invalid_guest_state = 0;
Avi Kivityc1f8bc02009-03-23 15:41:17 +020055module_param(emulate_invalid_guest_state, bool, S_IRUGO);
Mohammed Gamal04fa4d32008-08-17 16:39:48 +030056
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -040057struct vmcs {
58 u32 revision_id;
59 u32 abort;
60 char data[0];
61};
62
63struct vcpu_vmx {
Rusty Russellfb3f0f52007-07-27 17:16:56 +100064 struct kvm_vcpu vcpu;
Avi Kivity543e4242008-05-13 16:22:47 +030065 struct list_head local_vcpus_link;
Avi Kivity313dbd42008-07-17 18:04:30 +030066 unsigned long host_rsp;
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -040067 int launched;
Avi Kivity29bd8a72007-09-10 17:27:03 +030068 u8 fail;
Avi Kivity1155f762007-11-22 11:30:47 +020069 u32 idt_vectoring_info;
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -040070 struct kvm_msr_entry *guest_msrs;
71 struct kvm_msr_entry *host_msrs;
72 int nmsrs;
73 int save_nmsrs;
74 int msr_offset_efer;
75#ifdef CONFIG_X86_64
76 int msr_offset_kernel_gs_base;
77#endif
78 struct vmcs *vmcs;
79 struct {
80 int loaded;
81 u16 fs_sel, gs_sel, ldt_sel;
Laurent Vivier152d3f22007-08-23 16:33:11 +020082 int gs_ldt_reload_needed;
83 int fs_reload_needed;
Avi Kivity51c6cf62007-08-29 03:48:05 +030084 int guest_efer_loaded;
Mike Dayd77c26f2007-10-08 09:02:08 -040085 } host_state;
Avi Kivity9c8cba32007-11-22 11:42:59 +020086 struct {
87 struct {
88 bool pending;
89 u8 vector;
90 unsigned rip;
91 } irq;
92 } rmode;
Sheng Yang2384d2b2008-01-17 15:14:33 +080093 int vpid;
Mohammed Gamal04fa4d32008-08-17 16:39:48 +030094 bool emulation_required;
Avi Kivity8b3079a2009-01-05 12:10:54 +020095 enum emulation_result invalid_state_emulation_result;
Jan Kiszka3b86cd92008-09-26 09:30:57 +020096
97 /* Support for vnmi-less CPUs */
98 int soft_vnmi_blocked;
99 ktime_t entry_time;
100 s64 vnmi_blocked_time;
Andi Kleena0861c02009-06-08 17:37:09 +0800101 u32 exit_reason;
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400102};
103
104static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
105{
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000106 return container_of(vcpu, struct vcpu_vmx, vcpu);
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400107}
108
Sheng Yangb7ebfb02008-04-25 21:44:52 +0800109static int init_rmode(struct kvm *kvm);
Sheng Yang4e1096d2008-07-06 19:16:51 +0800110static u64 construct_eptp(unsigned long root_hpa);
Avi Kivity75880a02007-06-20 11:20:04 +0300111
Avi Kivity6aa8b732006-12-10 02:21:36 -0800112static DEFINE_PER_CPU(struct vmcs *, vmxarea);
113static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
Avi Kivity543e4242008-05-13 16:22:47 +0300114static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800115
Avi Kivity3e7c73e2009-02-24 21:46:19 +0200116static unsigned long *vmx_io_bitmap_a;
117static unsigned long *vmx_io_bitmap_b;
Avi Kivity58972972009-02-24 22:26:47 +0200118static unsigned long *vmx_msr_bitmap_legacy;
119static unsigned long *vmx_msr_bitmap_longmode;
He, Qingfdef3ad2007-04-30 09:45:24 +0300120
Sheng Yang2384d2b2008-01-17 15:14:33 +0800121static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
122static DEFINE_SPINLOCK(vmx_vpid_lock);
123
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +0300124static struct vmcs_config {
Avi Kivity6aa8b732006-12-10 02:21:36 -0800125 int size;
126 int order;
127 u32 revision_id;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +0300128 u32 pin_based_exec_ctrl;
129 u32 cpu_based_exec_ctrl;
Sheng Yangf78e0e22007-10-29 09:40:42 +0800130 u32 cpu_based_2nd_exec_ctrl;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +0300131 u32 vmexit_ctrl;
132 u32 vmentry_ctrl;
133} vmcs_config;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800134
Hannes Ederefff9e52008-11-28 17:02:06 +0100135static struct vmx_capability {
Sheng Yangd56f5462008-04-25 10:13:16 +0800136 u32 ept;
137 u32 vpid;
138} vmx_capability;
139
Avi Kivity6aa8b732006-12-10 02:21:36 -0800140#define VMX_SEGMENT_FIELD(seg) \
141 [VCPU_SREG_##seg] = { \
142 .selector = GUEST_##seg##_SELECTOR, \
143 .base = GUEST_##seg##_BASE, \
144 .limit = GUEST_##seg##_LIMIT, \
145 .ar_bytes = GUEST_##seg##_AR_BYTES, \
146 }
147
148static struct kvm_vmx_segment_field {
149 unsigned selector;
150 unsigned base;
151 unsigned limit;
152 unsigned ar_bytes;
153} kvm_vmx_segment_fields[] = {
154 VMX_SEGMENT_FIELD(CS),
155 VMX_SEGMENT_FIELD(DS),
156 VMX_SEGMENT_FIELD(ES),
157 VMX_SEGMENT_FIELD(FS),
158 VMX_SEGMENT_FIELD(GS),
159 VMX_SEGMENT_FIELD(SS),
160 VMX_SEGMENT_FIELD(TR),
161 VMX_SEGMENT_FIELD(LDTR),
162};
163
Avi Kivity6de4f3a2009-05-31 22:58:47 +0300164static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
165
Avi Kivity4d56c8a2007-04-19 14:28:44 +0300166/*
167 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
168 * away by decrementing the array size.
169 */
Avi Kivity6aa8b732006-12-10 02:21:36 -0800170static const u32 vmx_msr_index[] = {
Avi Kivity05b3e0c2006-12-13 00:33:45 -0800171#ifdef CONFIG_X86_64
Avi Kivity6aa8b732006-12-10 02:21:36 -0800172 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
173#endif
174 MSR_EFER, MSR_K6_STAR,
175};
Ahmed S. Darwish9d8f5492007-02-19 14:37:46 +0200176#define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800177
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400178static void load_msrs(struct kvm_msr_entry *e, int n)
179{
180 int i;
181
182 for (i = 0; i < n; ++i)
183 wrmsrl(e[i].index, e[i].data);
184}
185
186static void save_msrs(struct kvm_msr_entry *e, int n)
187{
188 int i;
189
190 for (i = 0; i < n; ++i)
191 rdmsrl(e[i].index, e[i].data);
192}
193
Avi Kivity6aa8b732006-12-10 02:21:36 -0800194static inline int is_page_fault(u32 intr_info)
195{
196 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
197 INTR_INFO_VALID_MASK)) ==
Jan Kiszka8ab2d2e2008-12-15 13:52:10 +0100198 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800199}
200
Anthony Liguori2ab455c2007-04-27 09:29:49 +0300201static inline int is_no_device(u32 intr_info)
202{
203 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
204 INTR_INFO_VALID_MASK)) ==
Jan Kiszka8ab2d2e2008-12-15 13:52:10 +0100205 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
Anthony Liguori2ab455c2007-04-27 09:29:49 +0300206}
207
Anthony Liguori7aa81cc2007-09-17 14:57:50 -0500208static inline int is_invalid_opcode(u32 intr_info)
209{
210 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
211 INTR_INFO_VALID_MASK)) ==
Jan Kiszka8ab2d2e2008-12-15 13:52:10 +0100212 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
Anthony Liguori7aa81cc2007-09-17 14:57:50 -0500213}
214
Avi Kivity6aa8b732006-12-10 02:21:36 -0800215static inline int is_external_interrupt(u32 intr_info)
216{
217 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
218 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
219}
220
Andi Kleena0861c02009-06-08 17:37:09 +0800221static inline int is_machine_check(u32 intr_info)
222{
223 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
224 INTR_INFO_VALID_MASK)) ==
225 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
226}
227
Sheng Yang25c5f222008-03-28 13:18:56 +0800228static inline int cpu_has_vmx_msr_bitmap(void)
229{
Sheng Yang04547152009-04-01 15:52:31 +0800230 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
Sheng Yang25c5f222008-03-28 13:18:56 +0800231}
232
Yang, Sheng6e5d8652007-09-12 18:03:11 +0800233static inline int cpu_has_vmx_tpr_shadow(void)
234{
Sheng Yang04547152009-04-01 15:52:31 +0800235 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
Yang, Sheng6e5d8652007-09-12 18:03:11 +0800236}
237
238static inline int vm_need_tpr_shadow(struct kvm *kvm)
239{
Sheng Yang04547152009-04-01 15:52:31 +0800240 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
Yang, Sheng6e5d8652007-09-12 18:03:11 +0800241}
242
Sheng Yangf78e0e22007-10-29 09:40:42 +0800243static inline int cpu_has_secondary_exec_ctrls(void)
244{
Sheng Yang04547152009-04-01 15:52:31 +0800245 return vmcs_config.cpu_based_exec_ctrl &
246 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
Sheng Yangf78e0e22007-10-29 09:40:42 +0800247}
248
Avi Kivity774ead32007-12-26 13:57:04 +0200249static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800250{
Sheng Yang04547152009-04-01 15:52:31 +0800251 return vmcs_config.cpu_based_2nd_exec_ctrl &
252 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
253}
254
255static inline bool cpu_has_vmx_flexpriority(void)
256{
257 return cpu_has_vmx_tpr_shadow() &&
258 cpu_has_vmx_virtualize_apic_accesses();
Sheng Yangf78e0e22007-10-29 09:40:42 +0800259}
260
Sheng Yangd56f5462008-04-25 10:13:16 +0800261static inline int cpu_has_vmx_invept_individual_addr(void)
262{
Sheng Yang04547152009-04-01 15:52:31 +0800263 return !!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT);
Sheng Yangd56f5462008-04-25 10:13:16 +0800264}
265
266static inline int cpu_has_vmx_invept_context(void)
267{
Sheng Yang04547152009-04-01 15:52:31 +0800268 return !!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT);
Sheng Yangd56f5462008-04-25 10:13:16 +0800269}
270
271static inline int cpu_has_vmx_invept_global(void)
272{
Sheng Yang04547152009-04-01 15:52:31 +0800273 return !!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT);
Sheng Yangd56f5462008-04-25 10:13:16 +0800274}
275
276static inline int cpu_has_vmx_ept(void)
277{
Sheng Yang04547152009-04-01 15:52:31 +0800278 return vmcs_config.cpu_based_2nd_exec_ctrl &
279 SECONDARY_EXEC_ENABLE_EPT;
Sheng Yangd56f5462008-04-25 10:13:16 +0800280}
281
Sheng Yangf78e0e22007-10-29 09:40:42 +0800282static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
283{
Sheng Yang04547152009-04-01 15:52:31 +0800284 return flexpriority_enabled &&
285 (cpu_has_vmx_virtualize_apic_accesses()) &&
286 (irqchip_in_kernel(kvm));
Sheng Yangf78e0e22007-10-29 09:40:42 +0800287}
288
Sheng Yang2384d2b2008-01-17 15:14:33 +0800289static inline int cpu_has_vmx_vpid(void)
290{
Sheng Yang04547152009-04-01 15:52:31 +0800291 return vmcs_config.cpu_based_2nd_exec_ctrl &
292 SECONDARY_EXEC_ENABLE_VPID;
Sheng Yang2384d2b2008-01-17 15:14:33 +0800293}
294
Sheng Yangf08864b2008-05-15 18:23:25 +0800295static inline int cpu_has_virtual_nmis(void)
296{
297 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
298}
299
Sheng Yang04547152009-04-01 15:52:31 +0800300static inline bool report_flexpriority(void)
301{
302 return flexpriority_enabled;
303}
304
Rusty Russell8b9cf982007-07-30 16:31:43 +1000305static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
Avi Kivity7725f0b2006-12-13 00:34:01 -0800306{
307 int i;
308
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400309 for (i = 0; i < vmx->nmsrs; ++i)
310 if (vmx->guest_msrs[i].index == msr)
Eddie Donga75beee2007-05-17 18:55:15 +0300311 return i;
312 return -1;
313}
314
Sheng Yang2384d2b2008-01-17 15:14:33 +0800315static inline void __invvpid(int ext, u16 vpid, gva_t gva)
316{
317 struct {
318 u64 vpid : 16;
319 u64 rsvd : 48;
320 u64 gva;
321 } operand = { vpid, 0, gva };
322
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300323 asm volatile (__ex(ASM_VMX_INVVPID)
Sheng Yang2384d2b2008-01-17 15:14:33 +0800324 /* CF==1 or ZF==1 --> rc = -1 */
325 "; ja 1f ; ud2 ; 1:"
326 : : "a"(&operand), "c"(ext) : "cc", "memory");
327}
328
Sheng Yang14394422008-04-28 12:24:45 +0800329static inline void __invept(int ext, u64 eptp, gpa_t gpa)
330{
331 struct {
332 u64 eptp, gpa;
333 } operand = {eptp, gpa};
334
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300335 asm volatile (__ex(ASM_VMX_INVEPT)
Sheng Yang14394422008-04-28 12:24:45 +0800336 /* CF==1 or ZF==1 --> rc = -1 */
337 "; ja 1f ; ud2 ; 1:\n"
338 : : "a" (&operand), "c" (ext) : "cc", "memory");
339}
340
Rusty Russell8b9cf982007-07-30 16:31:43 +1000341static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
Eddie Donga75beee2007-05-17 18:55:15 +0300342{
343 int i;
344
Rusty Russell8b9cf982007-07-30 16:31:43 +1000345 i = __find_msr_index(vmx, msr);
Eddie Donga75beee2007-05-17 18:55:15 +0300346 if (i >= 0)
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400347 return &vmx->guest_msrs[i];
Al Viro8b6d44c2007-02-09 16:38:40 +0000348 return NULL;
Avi Kivity7725f0b2006-12-13 00:34:01 -0800349}
350
Avi Kivity6aa8b732006-12-10 02:21:36 -0800351static void vmcs_clear(struct vmcs *vmcs)
352{
353 u64 phys_addr = __pa(vmcs);
354 u8 error;
355
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300356 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
Avi Kivity6aa8b732006-12-10 02:21:36 -0800357 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
358 : "cc", "memory");
359 if (error)
360 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
361 vmcs, phys_addr);
362}
363
364static void __vcpu_clear(void *arg)
365{
Rusty Russell8b9cf982007-07-30 16:31:43 +1000366 struct vcpu_vmx *vmx = arg;
Ingo Molnard3b2c332007-01-05 16:36:23 -0800367 int cpu = raw_smp_processor_id();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800368
Rusty Russell8b9cf982007-07-30 16:31:43 +1000369 if (vmx->vcpu.cpu == cpu)
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400370 vmcs_clear(vmx->vmcs);
371 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800372 per_cpu(current_vmcs, cpu) = NULL;
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800373 rdtscll(vmx->vcpu.arch.host_tsc);
Avi Kivity543e4242008-05-13 16:22:47 +0300374 list_del(&vmx->local_vcpus_link);
375 vmx->vcpu.cpu = -1;
376 vmx->launched = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800377}
378
Rusty Russell8b9cf982007-07-30 16:31:43 +1000379static void vcpu_clear(struct vcpu_vmx *vmx)
Avi Kivity8d0be2b2007-02-12 00:54:46 -0800380{
Avi Kivityeae5ecb2007-09-30 10:50:12 +0200381 if (vmx->vcpu.cpu == -1)
382 return;
Jens Axboe8691e5a2008-06-06 11:18:06 +0200383 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
Avi Kivity8d0be2b2007-02-12 00:54:46 -0800384}
385
Sheng Yang2384d2b2008-01-17 15:14:33 +0800386static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
387{
388 if (vmx->vpid == 0)
389 return;
390
391 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
392}
393
Sheng Yang14394422008-04-28 12:24:45 +0800394static inline void ept_sync_global(void)
395{
396 if (cpu_has_vmx_invept_global())
397 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
398}
399
400static inline void ept_sync_context(u64 eptp)
401{
Avi Kivity089d0342009-03-23 18:26:32 +0200402 if (enable_ept) {
Sheng Yang14394422008-04-28 12:24:45 +0800403 if (cpu_has_vmx_invept_context())
404 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
405 else
406 ept_sync_global();
407 }
408}
409
410static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
411{
Avi Kivity089d0342009-03-23 18:26:32 +0200412 if (enable_ept) {
Sheng Yang14394422008-04-28 12:24:45 +0800413 if (cpu_has_vmx_invept_individual_addr())
414 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
415 eptp, gpa);
416 else
417 ept_sync_context(eptp);
418 }
419}
420
Avi Kivity6aa8b732006-12-10 02:21:36 -0800421static unsigned long vmcs_readl(unsigned long field)
422{
423 unsigned long value;
424
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300425 asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800426 : "=a"(value) : "d"(field) : "cc");
427 return value;
428}
429
430static u16 vmcs_read16(unsigned long field)
431{
432 return vmcs_readl(field);
433}
434
435static u32 vmcs_read32(unsigned long field)
436{
437 return vmcs_readl(field);
438}
439
440static u64 vmcs_read64(unsigned long field)
441{
Avi Kivity05b3e0c2006-12-13 00:33:45 -0800442#ifdef CONFIG_X86_64
Avi Kivity6aa8b732006-12-10 02:21:36 -0800443 return vmcs_readl(field);
444#else
445 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
446#endif
447}
448
Avi Kivitye52de1b2007-01-05 16:36:56 -0800449static noinline void vmwrite_error(unsigned long field, unsigned long value)
450{
451 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
452 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
453 dump_stack();
454}
455
Avi Kivity6aa8b732006-12-10 02:21:36 -0800456static void vmcs_writel(unsigned long field, unsigned long value)
457{
458 u8 error;
459
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300460 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
Mike Dayd77c26f2007-10-08 09:02:08 -0400461 : "=q"(error) : "a"(value), "d"(field) : "cc");
Avi Kivitye52de1b2007-01-05 16:36:56 -0800462 if (unlikely(error))
463 vmwrite_error(field, value);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800464}
465
466static void vmcs_write16(unsigned long field, u16 value)
467{
468 vmcs_writel(field, value);
469}
470
471static void vmcs_write32(unsigned long field, u32 value)
472{
473 vmcs_writel(field, value);
474}
475
476static void vmcs_write64(unsigned long field, u64 value)
477{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800478 vmcs_writel(field, value);
Avi Kivity7682f2d2008-05-12 19:25:43 +0300479#ifndef CONFIG_X86_64
Avi Kivity6aa8b732006-12-10 02:21:36 -0800480 asm volatile ("");
481 vmcs_writel(field+1, value >> 32);
482#endif
483}
484
Anthony Liguori2ab455c2007-04-27 09:29:49 +0300485static void vmcs_clear_bits(unsigned long field, u32 mask)
486{
487 vmcs_writel(field, vmcs_readl(field) & ~mask);
488}
489
490static void vmcs_set_bits(unsigned long field, u32 mask)
491{
492 vmcs_writel(field, vmcs_readl(field) | mask);
493}
494
Avi Kivityabd3f2d2007-05-02 17:57:40 +0300495static void update_exception_bitmap(struct kvm_vcpu *vcpu)
496{
497 u32 eb;
498
Andi Kleena0861c02009-06-08 17:37:09 +0800499 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR);
Avi Kivityabd3f2d2007-05-02 17:57:40 +0300500 if (!vcpu->fpu_active)
501 eb |= 1u << NM_VECTOR;
Jan Kiszkad0bfb942008-12-15 13:52:10 +0100502 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
503 if (vcpu->guest_debug &
504 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
505 eb |= 1u << DB_VECTOR;
506 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
507 eb |= 1u << BP_VECTOR;
508 }
Nitin A Kamble56b237e2009-06-04 15:04:08 -0700509 if (vcpu->arch.rmode.vm86_active)
Avi Kivityabd3f2d2007-05-02 17:57:40 +0300510 eb = ~0;
Avi Kivity089d0342009-03-23 18:26:32 +0200511 if (enable_ept)
Sheng Yang14394422008-04-28 12:24:45 +0800512 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
Avi Kivityabd3f2d2007-05-02 17:57:40 +0300513 vmcs_write32(EXCEPTION_BITMAP, eb);
514}
515
Avi Kivity33ed6322007-05-02 16:54:03 +0300516static void reload_tss(void)
517{
Avi Kivity33ed6322007-05-02 16:54:03 +0300518 /*
519 * VT restores TR but not its size. Useless.
520 */
521 struct descriptor_table gdt;
Avi Kivitya5f61302008-02-20 17:57:21 +0200522 struct desc_struct *descs;
Avi Kivity33ed6322007-05-02 16:54:03 +0300523
Avi Kivityd6e88ae2008-07-10 16:53:33 +0300524 kvm_get_gdt(&gdt);
Avi Kivity33ed6322007-05-02 16:54:03 +0300525 descs = (void *)gdt.base;
526 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
527 load_TR_desc();
Avi Kivity33ed6322007-05-02 16:54:03 +0300528}
529
Rusty Russell8b9cf982007-07-30 16:31:43 +1000530static void load_transition_efer(struct vcpu_vmx *vmx)
Eddie Dong2cc51562007-05-21 07:28:09 +0300531{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400532 int efer_offset = vmx->msr_offset_efer;
Avi Kivity51c6cf62007-08-29 03:48:05 +0300533 u64 host_efer = vmx->host_msrs[efer_offset].data;
534 u64 guest_efer = vmx->guest_msrs[efer_offset].data;
535 u64 ignore_bits;
Eddie Dong2cc51562007-05-21 07:28:09 +0300536
Avi Kivity51c6cf62007-08-29 03:48:05 +0300537 if (efer_offset < 0)
538 return;
539 /*
540 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
541 * outside long mode
542 */
543 ignore_bits = EFER_NX | EFER_SCE;
544#ifdef CONFIG_X86_64
545 ignore_bits |= EFER_LMA | EFER_LME;
546 /* SCE is meaningful only in long mode on Intel */
547 if (guest_efer & EFER_LMA)
548 ignore_bits &= ~(u64)EFER_SCE;
549#endif
550 if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
551 return;
552
553 vmx->host_state.guest_efer_loaded = 1;
554 guest_efer &= ~ignore_bits;
555 guest_efer |= host_efer & ignore_bits;
556 wrmsrl(MSR_EFER, guest_efer);
Rusty Russell8b9cf982007-07-30 16:31:43 +1000557 vmx->vcpu.stat.efer_reload++;
Eddie Dong2cc51562007-05-21 07:28:09 +0300558}
559
Avi Kivity51c6cf62007-08-29 03:48:05 +0300560static void reload_host_efer(struct vcpu_vmx *vmx)
561{
562 if (vmx->host_state.guest_efer_loaded) {
563 vmx->host_state.guest_efer_loaded = 0;
564 load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1);
565 }
566}
567
Avi Kivity04d2cc72007-09-10 18:10:54 +0300568static void vmx_save_host_state(struct kvm_vcpu *vcpu)
Avi Kivity33ed6322007-05-02 16:54:03 +0300569{
Avi Kivity04d2cc72007-09-10 18:10:54 +0300570 struct vcpu_vmx *vmx = to_vmx(vcpu);
571
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400572 if (vmx->host_state.loaded)
Avi Kivity33ed6322007-05-02 16:54:03 +0300573 return;
574
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400575 vmx->host_state.loaded = 1;
Avi Kivity33ed6322007-05-02 16:54:03 +0300576 /*
577 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
578 * allow segment selectors with cpl > 0 or ti == 1.
579 */
Avi Kivityd6e88ae2008-07-10 16:53:33 +0300580 vmx->host_state.ldt_sel = kvm_read_ldt();
Laurent Vivier152d3f22007-08-23 16:33:11 +0200581 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
Avi Kivityd6e88ae2008-07-10 16:53:33 +0300582 vmx->host_state.fs_sel = kvm_read_fs();
Laurent Vivier152d3f22007-08-23 16:33:11 +0200583 if (!(vmx->host_state.fs_sel & 7)) {
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400584 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
Laurent Vivier152d3f22007-08-23 16:33:11 +0200585 vmx->host_state.fs_reload_needed = 0;
586 } else {
Avi Kivity33ed6322007-05-02 16:54:03 +0300587 vmcs_write16(HOST_FS_SELECTOR, 0);
Laurent Vivier152d3f22007-08-23 16:33:11 +0200588 vmx->host_state.fs_reload_needed = 1;
Avi Kivity33ed6322007-05-02 16:54:03 +0300589 }
Avi Kivityd6e88ae2008-07-10 16:53:33 +0300590 vmx->host_state.gs_sel = kvm_read_gs();
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400591 if (!(vmx->host_state.gs_sel & 7))
592 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
Avi Kivity33ed6322007-05-02 16:54:03 +0300593 else {
594 vmcs_write16(HOST_GS_SELECTOR, 0);
Laurent Vivier152d3f22007-08-23 16:33:11 +0200595 vmx->host_state.gs_ldt_reload_needed = 1;
Avi Kivity33ed6322007-05-02 16:54:03 +0300596 }
597
598#ifdef CONFIG_X86_64
599 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
600 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
601#else
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400602 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
603 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
Avi Kivity33ed6322007-05-02 16:54:03 +0300604#endif
Avi Kivity707c0872007-05-02 17:33:43 +0300605
606#ifdef CONFIG_X86_64
Mike Dayd77c26f2007-10-08 09:02:08 -0400607 if (is_long_mode(&vmx->vcpu))
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400608 save_msrs(vmx->host_msrs +
609 vmx->msr_offset_kernel_gs_base, 1);
Mike Dayd77c26f2007-10-08 09:02:08 -0400610
Avi Kivity707c0872007-05-02 17:33:43 +0300611#endif
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400612 load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
Avi Kivity51c6cf62007-08-29 03:48:05 +0300613 load_transition_efer(vmx);
Avi Kivity33ed6322007-05-02 16:54:03 +0300614}
615
Avi Kivitya9b21b62008-06-24 11:48:49 +0300616static void __vmx_load_host_state(struct vcpu_vmx *vmx)
Avi Kivity33ed6322007-05-02 16:54:03 +0300617{
Avi Kivity15ad7142007-07-11 18:17:21 +0300618 unsigned long flags;
Avi Kivity33ed6322007-05-02 16:54:03 +0300619
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400620 if (!vmx->host_state.loaded)
Avi Kivity33ed6322007-05-02 16:54:03 +0300621 return;
622
Avi Kivitye1beb1d2007-11-18 13:50:24 +0200623 ++vmx->vcpu.stat.host_state_reload;
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400624 vmx->host_state.loaded = 0;
Laurent Vivier152d3f22007-08-23 16:33:11 +0200625 if (vmx->host_state.fs_reload_needed)
Avi Kivityd6e88ae2008-07-10 16:53:33 +0300626 kvm_load_fs(vmx->host_state.fs_sel);
Laurent Vivier152d3f22007-08-23 16:33:11 +0200627 if (vmx->host_state.gs_ldt_reload_needed) {
Avi Kivityd6e88ae2008-07-10 16:53:33 +0300628 kvm_load_ldt(vmx->host_state.ldt_sel);
Avi Kivity33ed6322007-05-02 16:54:03 +0300629 /*
630 * If we have to reload gs, we must take care to
631 * preserve our gs base.
632 */
Avi Kivity15ad7142007-07-11 18:17:21 +0300633 local_irq_save(flags);
Avi Kivityd6e88ae2008-07-10 16:53:33 +0300634 kvm_load_gs(vmx->host_state.gs_sel);
Avi Kivity33ed6322007-05-02 16:54:03 +0300635#ifdef CONFIG_X86_64
636 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
637#endif
Avi Kivity15ad7142007-07-11 18:17:21 +0300638 local_irq_restore(flags);
Avi Kivity33ed6322007-05-02 16:54:03 +0300639 }
Laurent Vivier152d3f22007-08-23 16:33:11 +0200640 reload_tss();
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400641 save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
642 load_msrs(vmx->host_msrs, vmx->save_nmsrs);
Avi Kivity51c6cf62007-08-29 03:48:05 +0300643 reload_host_efer(vmx);
Avi Kivity33ed6322007-05-02 16:54:03 +0300644}
645
Avi Kivitya9b21b62008-06-24 11:48:49 +0300646static void vmx_load_host_state(struct vcpu_vmx *vmx)
647{
648 preempt_disable();
649 __vmx_load_host_state(vmx);
650 preempt_enable();
651}
652
Avi Kivity6aa8b732006-12-10 02:21:36 -0800653/*
654 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
655 * vcpu mutex is already taken.
656 */
Avi Kivity15ad7142007-07-11 18:17:21 +0300657static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800658{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400659 struct vcpu_vmx *vmx = to_vmx(vcpu);
660 u64 phys_addr = __pa(vmx->vmcs);
Avi Kivity019960a2008-03-04 10:44:51 +0200661 u64 tsc_this, delta, new_offset;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800662
Eddie Donga3d7f852007-09-03 16:15:12 +0300663 if (vcpu->cpu != cpu) {
Rusty Russell8b9cf982007-07-30 16:31:43 +1000664 vcpu_clear(vmx);
Marcelo Tosatti2f599712008-05-27 12:10:20 -0300665 kvm_migrate_timers(vcpu);
Sheng Yang2384d2b2008-01-17 15:14:33 +0800666 vpid_sync_vcpu_all(vmx);
Avi Kivity543e4242008-05-13 16:22:47 +0300667 local_irq_disable();
668 list_add(&vmx->local_vcpus_link,
669 &per_cpu(vcpus_on_cpu, cpu));
670 local_irq_enable();
Eddie Donga3d7f852007-09-03 16:15:12 +0300671 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800672
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400673 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
Avi Kivity6aa8b732006-12-10 02:21:36 -0800674 u8 error;
675
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400676 per_cpu(current_vmcs, cpu) = vmx->vmcs;
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300677 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
Avi Kivity6aa8b732006-12-10 02:21:36 -0800678 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
679 : "cc");
680 if (error)
681 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400682 vmx->vmcs, phys_addr);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800683 }
684
685 if (vcpu->cpu != cpu) {
686 struct descriptor_table dt;
687 unsigned long sysenter_esp;
688
689 vcpu->cpu = cpu;
690 /*
691 * Linux uses per-cpu TSS and GDT, so set these when switching
692 * processors.
693 */
Avi Kivityd6e88ae2008-07-10 16:53:33 +0300694 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
695 kvm_get_gdt(&dt);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800696 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
697
698 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
699 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
Avi Kivity77002702007-06-13 19:55:28 +0300700
701 /*
702 * Make sure the time stamp counter is monotonous.
703 */
704 rdtscll(tsc_this);
Avi Kivity019960a2008-03-04 10:44:51 +0200705 if (tsc_this < vcpu->arch.host_tsc) {
706 delta = vcpu->arch.host_tsc - tsc_this;
707 new_offset = vmcs_read64(TSC_OFFSET) + delta;
708 vmcs_write64(TSC_OFFSET, new_offset);
709 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800710 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800711}
712
713static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
714{
Avi Kivitya9b21b62008-06-24 11:48:49 +0300715 __vmx_load_host_state(to_vmx(vcpu));
Avi Kivity6aa8b732006-12-10 02:21:36 -0800716}
717
Avi Kivity5fd86fc2007-05-02 20:40:00 +0300718static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
719{
720 if (vcpu->fpu_active)
721 return;
722 vcpu->fpu_active = 1;
Rusty Russell707d92fa2007-07-17 23:19:08 +1000723 vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800724 if (vcpu->arch.cr0 & X86_CR0_TS)
Rusty Russell707d92fa2007-07-17 23:19:08 +1000725 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
Avi Kivity5fd86fc2007-05-02 20:40:00 +0300726 update_exception_bitmap(vcpu);
727}
728
729static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
730{
731 if (!vcpu->fpu_active)
732 return;
733 vcpu->fpu_active = 0;
Rusty Russell707d92fa2007-07-17 23:19:08 +1000734 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
Avi Kivity5fd86fc2007-05-02 20:40:00 +0300735 update_exception_bitmap(vcpu);
736}
737
Avi Kivity6aa8b732006-12-10 02:21:36 -0800738static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
739{
740 return vmcs_readl(GUEST_RFLAGS);
741}
742
743static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
744{
Nitin A Kamble56b237e2009-06-04 15:04:08 -0700745 if (vcpu->arch.rmode.vm86_active)
Glauber de Oliveira Costa053de042008-01-30 13:31:27 +0100746 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800747 vmcs_writel(GUEST_RFLAGS, rflags);
748}
749
Glauber Costa2809f5d2009-05-12 16:21:05 -0400750static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
751{
752 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
753 int ret = 0;
754
755 if (interruptibility & GUEST_INTR_STATE_STI)
756 ret |= X86_SHADOW_INT_STI;
757 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
758 ret |= X86_SHADOW_INT_MOV_SS;
759
760 return ret & mask;
761}
762
763static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
764{
765 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
766 u32 interruptibility = interruptibility_old;
767
768 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
769
770 if (mask & X86_SHADOW_INT_MOV_SS)
771 interruptibility |= GUEST_INTR_STATE_MOV_SS;
772 if (mask & X86_SHADOW_INT_STI)
773 interruptibility |= GUEST_INTR_STATE_STI;
774
775 if ((interruptibility != interruptibility_old))
776 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
777}
778
Avi Kivity6aa8b732006-12-10 02:21:36 -0800779static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
780{
781 unsigned long rip;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800782
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -0300783 rip = kvm_rip_read(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800784 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -0300785 kvm_rip_write(vcpu, rip);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800786
Glauber Costa2809f5d2009-05-12 16:21:05 -0400787 /* skipping an emulated instruction also counts */
788 vmx_set_interrupt_shadow(vcpu, 0);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800789}
790
Avi Kivity298101d2007-11-25 13:41:11 +0200791static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
792 bool has_error_code, u32 error_code)
793{
Jan Kiszka77ab6db2008-07-14 12:28:51 +0200794 struct vcpu_vmx *vmx = to_vmx(vcpu);
Jan Kiszka8ab2d2e2008-12-15 13:52:10 +0100795 u32 intr_info = nr | INTR_INFO_VALID_MASK;
Jan Kiszka77ab6db2008-07-14 12:28:51 +0200796
Jan Kiszka8ab2d2e2008-12-15 13:52:10 +0100797 if (has_error_code) {
Jan Kiszka77ab6db2008-07-14 12:28:51 +0200798 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
Jan Kiszka8ab2d2e2008-12-15 13:52:10 +0100799 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
800 }
Jan Kiszka77ab6db2008-07-14 12:28:51 +0200801
Nitin A Kamble56b237e2009-06-04 15:04:08 -0700802 if (vcpu->arch.rmode.vm86_active) {
Jan Kiszka77ab6db2008-07-14 12:28:51 +0200803 vmx->rmode.irq.pending = true;
804 vmx->rmode.irq.vector = nr;
805 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
Gleb Natapovae0bb3e2009-05-19 11:07:10 +0300806 if (kvm_exception_is_soft(nr))
807 vmx->rmode.irq.rip +=
808 vmx->vcpu.arch.event_exit_inst_len;
Jan Kiszka8ab2d2e2008-12-15 13:52:10 +0100809 intr_info |= INTR_TYPE_SOFT_INTR;
810 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
Jan Kiszka77ab6db2008-07-14 12:28:51 +0200811 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
812 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
813 return;
814 }
815
Gleb Natapov66fd3f72009-05-11 13:35:50 +0300816 if (kvm_exception_is_soft(nr)) {
817 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
818 vmx->vcpu.arch.event_exit_inst_len);
Jan Kiszka8ab2d2e2008-12-15 13:52:10 +0100819 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
820 } else
821 intr_info |= INTR_TYPE_HARD_EXCEPTION;
822
823 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
Avi Kivity298101d2007-11-25 13:41:11 +0200824}
825
Avi Kivity6aa8b732006-12-10 02:21:36 -0800826/*
Eddie Donga75beee2007-05-17 18:55:15 +0300827 * Swap MSR entry in host/guest MSR entry array.
828 */
Gabriel C54e11fa2007-08-01 16:23:10 +0200829#ifdef CONFIG_X86_64
Rusty Russell8b9cf982007-07-30 16:31:43 +1000830static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
Eddie Donga75beee2007-05-17 18:55:15 +0300831{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400832 struct kvm_msr_entry tmp;
833
834 tmp = vmx->guest_msrs[to];
835 vmx->guest_msrs[to] = vmx->guest_msrs[from];
836 vmx->guest_msrs[from] = tmp;
837 tmp = vmx->host_msrs[to];
838 vmx->host_msrs[to] = vmx->host_msrs[from];
839 vmx->host_msrs[from] = tmp;
Eddie Donga75beee2007-05-17 18:55:15 +0300840}
Gabriel C54e11fa2007-08-01 16:23:10 +0200841#endif
Eddie Donga75beee2007-05-17 18:55:15 +0300842
843/*
Avi Kivitye38aea32007-04-19 13:22:48 +0300844 * Set up the vmcs to automatically save and restore system
845 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
846 * mode, as fiddling with msrs is very expensive.
847 */
Rusty Russell8b9cf982007-07-30 16:31:43 +1000848static void setup_msrs(struct vcpu_vmx *vmx)
Avi Kivitye38aea32007-04-19 13:22:48 +0300849{
Eddie Dong2cc51562007-05-21 07:28:09 +0300850 int save_nmsrs;
Avi Kivity58972972009-02-24 22:26:47 +0200851 unsigned long *msr_bitmap;
Avi Kivitye38aea32007-04-19 13:22:48 +0300852
Avi Kivity33f9c502008-02-27 16:06:57 +0200853 vmx_load_host_state(vmx);
Eddie Donga75beee2007-05-17 18:55:15 +0300854 save_nmsrs = 0;
Avi Kivity4d56c8a2007-04-19 14:28:44 +0300855#ifdef CONFIG_X86_64
Rusty Russell8b9cf982007-07-30 16:31:43 +1000856 if (is_long_mode(&vmx->vcpu)) {
Eddie Dong2cc51562007-05-21 07:28:09 +0300857 int index;
858
Rusty Russell8b9cf982007-07-30 16:31:43 +1000859 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
Eddie Donga75beee2007-05-17 18:55:15 +0300860 if (index >= 0)
Rusty Russell8b9cf982007-07-30 16:31:43 +1000861 move_msr_up(vmx, index, save_nmsrs++);
862 index = __find_msr_index(vmx, MSR_LSTAR);
Eddie Donga75beee2007-05-17 18:55:15 +0300863 if (index >= 0)
Rusty Russell8b9cf982007-07-30 16:31:43 +1000864 move_msr_up(vmx, index, save_nmsrs++);
865 index = __find_msr_index(vmx, MSR_CSTAR);
Eddie Donga75beee2007-05-17 18:55:15 +0300866 if (index >= 0)
Rusty Russell8b9cf982007-07-30 16:31:43 +1000867 move_msr_up(vmx, index, save_nmsrs++);
868 index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
Eddie Donga75beee2007-05-17 18:55:15 +0300869 if (index >= 0)
Rusty Russell8b9cf982007-07-30 16:31:43 +1000870 move_msr_up(vmx, index, save_nmsrs++);
Eddie Donga75beee2007-05-17 18:55:15 +0300871 /*
872 * MSR_K6_STAR is only needed on long mode guests, and only
873 * if efer.sce is enabled.
874 */
Rusty Russell8b9cf982007-07-30 16:31:43 +1000875 index = __find_msr_index(vmx, MSR_K6_STAR);
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800876 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
Rusty Russell8b9cf982007-07-30 16:31:43 +1000877 move_msr_up(vmx, index, save_nmsrs++);
Avi Kivity4d56c8a2007-04-19 14:28:44 +0300878 }
Eddie Donga75beee2007-05-17 18:55:15 +0300879#endif
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400880 vmx->save_nmsrs = save_nmsrs;
Avi Kivity4d56c8a2007-04-19 14:28:44 +0300881
Eddie Donga75beee2007-05-17 18:55:15 +0300882#ifdef CONFIG_X86_64
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400883 vmx->msr_offset_kernel_gs_base =
Rusty Russell8b9cf982007-07-30 16:31:43 +1000884 __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
Eddie Donga75beee2007-05-17 18:55:15 +0300885#endif
Rusty Russell8b9cf982007-07-30 16:31:43 +1000886 vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER);
Avi Kivity58972972009-02-24 22:26:47 +0200887
888 if (cpu_has_vmx_msr_bitmap()) {
889 if (is_long_mode(&vmx->vcpu))
890 msr_bitmap = vmx_msr_bitmap_longmode;
891 else
892 msr_bitmap = vmx_msr_bitmap_legacy;
893
894 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
895 }
Avi Kivitye38aea32007-04-19 13:22:48 +0300896}
897
898/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800899 * reads and returns guest's timestamp counter "register"
900 * guest_tsc = host_tsc + tsc_offset -- 21.3
901 */
902static u64 guest_read_tsc(void)
903{
904 u64 host_tsc, tsc_offset;
905
906 rdtscll(host_tsc);
907 tsc_offset = vmcs_read64(TSC_OFFSET);
908 return host_tsc + tsc_offset;
909}
910
911/*
912 * writes 'guest_tsc' into guest's timestamp counter "register"
913 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
914 */
Marcelo Tosatti53f658b32008-12-11 20:45:05 +0100915static void guest_write_tsc(u64 guest_tsc, u64 host_tsc)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800916{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800917 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
918}
919
Avi Kivity6aa8b732006-12-10 02:21:36 -0800920/*
921 * Reads an msr value (of 'msr_index') into 'pdata'.
922 * Returns 0 on success, non-0 otherwise.
923 * Assumes vcpu_load() was already called.
924 */
925static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
926{
927 u64 data;
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400928 struct kvm_msr_entry *msr;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800929
930 if (!pdata) {
931 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
932 return -EINVAL;
933 }
934
935 switch (msr_index) {
Avi Kivity05b3e0c2006-12-13 00:33:45 -0800936#ifdef CONFIG_X86_64
Avi Kivity6aa8b732006-12-10 02:21:36 -0800937 case MSR_FS_BASE:
938 data = vmcs_readl(GUEST_FS_BASE);
939 break;
940 case MSR_GS_BASE:
941 data = vmcs_readl(GUEST_GS_BASE);
942 break;
943 case MSR_EFER:
Avi Kivity3bab1f52006-12-29 16:49:48 -0800944 return kvm_get_msr_common(vcpu, msr_index, pdata);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800945#endif
Jaswinder Singh Rajputaf24a4e2009-05-15 18:42:05 +0530946 case MSR_IA32_TSC:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800947 data = guest_read_tsc();
948 break;
949 case MSR_IA32_SYSENTER_CS:
950 data = vmcs_read32(GUEST_SYSENTER_CS);
951 break;
952 case MSR_IA32_SYSENTER_EIP:
Avi Kivityf5b42c32007-03-06 12:05:53 +0200953 data = vmcs_readl(GUEST_SYSENTER_EIP);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800954 break;
955 case MSR_IA32_SYSENTER_ESP:
Avi Kivityf5b42c32007-03-06 12:05:53 +0200956 data = vmcs_readl(GUEST_SYSENTER_ESP);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800957 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800958 default:
Avi Kivity516a1a72009-02-15 02:32:07 +0200959 vmx_load_host_state(to_vmx(vcpu));
Rusty Russell8b9cf982007-07-30 16:31:43 +1000960 msr = find_msr_entry(to_vmx(vcpu), msr_index);
Avi Kivity3bab1f52006-12-29 16:49:48 -0800961 if (msr) {
962 data = msr->data;
963 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800964 }
Avi Kivity3bab1f52006-12-29 16:49:48 -0800965 return kvm_get_msr_common(vcpu, msr_index, pdata);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800966 }
967
968 *pdata = data;
969 return 0;
970}
971
972/*
973 * Writes msr value into into the appropriate "register".
974 * Returns 0 on success, non-0 otherwise.
975 * Assumes vcpu_load() was already called.
976 */
977static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
978{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400979 struct vcpu_vmx *vmx = to_vmx(vcpu);
980 struct kvm_msr_entry *msr;
Marcelo Tosatti53f658b32008-12-11 20:45:05 +0100981 u64 host_tsc;
Eddie Dong2cc51562007-05-21 07:28:09 +0300982 int ret = 0;
983
Avi Kivity6aa8b732006-12-10 02:21:36 -0800984 switch (msr_index) {
Avi Kivity3bab1f52006-12-29 16:49:48 -0800985 case MSR_EFER:
Avi Kivitya9b21b62008-06-24 11:48:49 +0300986 vmx_load_host_state(vmx);
Eddie Dong2cc51562007-05-21 07:28:09 +0300987 ret = kvm_set_msr_common(vcpu, msr_index, data);
Eddie Dong2cc51562007-05-21 07:28:09 +0300988 break;
Avi Kivity16175a72009-03-23 22:13:44 +0200989#ifdef CONFIG_X86_64
Avi Kivity6aa8b732006-12-10 02:21:36 -0800990 case MSR_FS_BASE:
991 vmcs_writel(GUEST_FS_BASE, data);
992 break;
993 case MSR_GS_BASE:
994 vmcs_writel(GUEST_GS_BASE, data);
995 break;
996#endif
997 case MSR_IA32_SYSENTER_CS:
998 vmcs_write32(GUEST_SYSENTER_CS, data);
999 break;
1000 case MSR_IA32_SYSENTER_EIP:
Avi Kivityf5b42c32007-03-06 12:05:53 +02001001 vmcs_writel(GUEST_SYSENTER_EIP, data);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001002 break;
1003 case MSR_IA32_SYSENTER_ESP:
Avi Kivityf5b42c32007-03-06 12:05:53 +02001004 vmcs_writel(GUEST_SYSENTER_ESP, data);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001005 break;
Jaswinder Singh Rajputaf24a4e2009-05-15 18:42:05 +05301006 case MSR_IA32_TSC:
Marcelo Tosatti53f658b32008-12-11 20:45:05 +01001007 rdtscll(host_tsc);
1008 guest_write_tsc(data, host_tsc);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001009 break;
Chris Lalancetteefa67e02008-06-20 09:51:30 +02001010 case MSR_P6_PERFCTR0:
1011 case MSR_P6_PERFCTR1:
1012 case MSR_P6_EVNTSEL0:
1013 case MSR_P6_EVNTSEL1:
1014 /*
1015 * Just discard all writes to the performance counters; this
1016 * should keep both older linux and windows 64-bit guests
1017 * happy
1018 */
1019 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", msr_index, data);
1020
1021 break;
Sheng Yang468d4722008-10-09 16:01:55 +08001022 case MSR_IA32_CR_PAT:
1023 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1024 vmcs_write64(GUEST_IA32_PAT, data);
1025 vcpu->arch.pat = data;
1026 break;
1027 }
1028 /* Otherwise falls through to kvm_set_msr_common */
Avi Kivity6aa8b732006-12-10 02:21:36 -08001029 default:
Avi Kivitya9b21b62008-06-24 11:48:49 +03001030 vmx_load_host_state(vmx);
Rusty Russell8b9cf982007-07-30 16:31:43 +10001031 msr = find_msr_entry(vmx, msr_index);
Avi Kivity3bab1f52006-12-29 16:49:48 -08001032 if (msr) {
1033 msr->data = data;
1034 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001035 }
Eddie Dong2cc51562007-05-21 07:28:09 +03001036 ret = kvm_set_msr_common(vcpu, msr_index, data);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001037 }
1038
Eddie Dong2cc51562007-05-21 07:28:09 +03001039 return ret;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001040}
1041
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03001042static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001043{
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03001044 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1045 switch (reg) {
1046 case VCPU_REGS_RSP:
1047 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1048 break;
1049 case VCPU_REGS_RIP:
1050 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1051 break;
Avi Kivity6de4f3a2009-05-31 22:58:47 +03001052 case VCPU_EXREG_PDPTR:
1053 if (enable_ept)
1054 ept_save_pdptrs(vcpu);
1055 break;
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03001056 default:
1057 break;
1058 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08001059}
1060
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001061static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001062{
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001063 int old_debug = vcpu->guest_debug;
1064 unsigned long flags;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001065
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001066 vcpu->guest_debug = dbg->control;
1067 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
1068 vcpu->guest_debug = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001069
Jan Kiszkaae675ef2008-12-15 13:52:10 +01001070 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1071 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1072 else
1073 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1074
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001075 flags = vmcs_readl(GUEST_RFLAGS);
1076 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
1077 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1078 else if (old_debug & KVM_GUESTDBG_SINGLESTEP)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001079 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001080 vmcs_writel(GUEST_RFLAGS, flags);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001081
Avi Kivityabd3f2d2007-05-02 17:57:40 +03001082 update_exception_bitmap(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001083
1084 return 0;
1085}
1086
1087static __init int cpu_has_kvm_support(void)
1088{
Eduardo Habkost6210e372008-11-17 19:03:16 -02001089 return cpu_has_vmx();
Avi Kivity6aa8b732006-12-10 02:21:36 -08001090}
1091
1092static __init int vmx_disabled_by_bios(void)
1093{
1094 u64 msr;
1095
1096 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
Sheng Yang9ea542f2008-09-11 15:27:49 +08001097 return (msr & (FEATURE_CONTROL_LOCKED |
1098 FEATURE_CONTROL_VMXON_ENABLED))
1099 == FEATURE_CONTROL_LOCKED;
Yang, Sheng62b3ffb2007-07-25 12:17:06 +03001100 /* locked but not enabled */
Avi Kivity6aa8b732006-12-10 02:21:36 -08001101}
1102
Avi Kivity774c47f2007-02-12 00:54:47 -08001103static void hardware_enable(void *garbage)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001104{
1105 int cpu = raw_smp_processor_id();
1106 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1107 u64 old;
1108
Avi Kivity543e4242008-05-13 16:22:47 +03001109 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
Avi Kivity6aa8b732006-12-10 02:21:36 -08001110 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
Sheng Yang9ea542f2008-09-11 15:27:49 +08001111 if ((old & (FEATURE_CONTROL_LOCKED |
1112 FEATURE_CONTROL_VMXON_ENABLED))
1113 != (FEATURE_CONTROL_LOCKED |
1114 FEATURE_CONTROL_VMXON_ENABLED))
Avi Kivity6aa8b732006-12-10 02:21:36 -08001115 /* enable and lock */
Yang, Sheng62b3ffb2007-07-25 12:17:06 +03001116 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
Sheng Yang9ea542f2008-09-11 15:27:49 +08001117 FEATURE_CONTROL_LOCKED |
1118 FEATURE_CONTROL_VMXON_ENABLED);
Rusty Russell66aee912007-07-17 23:34:16 +10001119 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
Avi Kivity4ecac3f2008-05-13 13:23:38 +03001120 asm volatile (ASM_VMX_VMXON_RAX
1121 : : "a"(&phys_addr), "m"(phys_addr)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001122 : "memory", "cc");
1123}
1124
Avi Kivity543e4242008-05-13 16:22:47 +03001125static void vmclear_local_vcpus(void)
1126{
1127 int cpu = raw_smp_processor_id();
1128 struct vcpu_vmx *vmx, *n;
1129
1130 list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1131 local_vcpus_link)
1132 __vcpu_clear(vmx);
1133}
1134
Eduardo Habkost710ff4a2008-11-17 19:03:18 -02001135
1136/* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1137 * tricks.
1138 */
1139static void kvm_cpu_vmxoff(void)
1140{
1141 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1142 write_cr4(read_cr4() & ~X86_CR4_VMXE);
1143}
1144
Avi Kivity6aa8b732006-12-10 02:21:36 -08001145static void hardware_disable(void *garbage)
1146{
Avi Kivity543e4242008-05-13 16:22:47 +03001147 vmclear_local_vcpus();
Eduardo Habkost710ff4a2008-11-17 19:03:18 -02001148 kvm_cpu_vmxoff();
Avi Kivity6aa8b732006-12-10 02:21:36 -08001149}
1150
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001151static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
Mike Dayd77c26f2007-10-08 09:02:08 -04001152 u32 msr, u32 *result)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001153{
1154 u32 vmx_msr_low, vmx_msr_high;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001155 u32 ctl = ctl_min | ctl_opt;
1156
1157 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1158
1159 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1160 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
1161
1162 /* Ensure minimum (required) set of control bits are supported. */
1163 if (ctl_min & ~ctl)
Yang, Sheng002c7f72007-07-31 14:23:01 +03001164 return -EIO;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001165
1166 *result = ctl;
1167 return 0;
1168}
1169
Yang, Sheng002c7f72007-07-31 14:23:01 +03001170static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001171{
1172 u32 vmx_msr_low, vmx_msr_high;
Sheng Yangd56f5462008-04-25 10:13:16 +08001173 u32 min, opt, min2, opt2;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001174 u32 _pin_based_exec_control = 0;
1175 u32 _cpu_based_exec_control = 0;
Sheng Yangf78e0e22007-10-29 09:40:42 +08001176 u32 _cpu_based_2nd_exec_control = 0;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001177 u32 _vmexit_control = 0;
1178 u32 _vmentry_control = 0;
1179
1180 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
Sheng Yangf08864b2008-05-15 18:23:25 +08001181 opt = PIN_BASED_VIRTUAL_NMIS;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001182 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1183 &_pin_based_exec_control) < 0)
Yang, Sheng002c7f72007-07-31 14:23:01 +03001184 return -EIO;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001185
1186 min = CPU_BASED_HLT_EXITING |
1187#ifdef CONFIG_X86_64
1188 CPU_BASED_CR8_LOAD_EXITING |
1189 CPU_BASED_CR8_STORE_EXITING |
1190#endif
Sheng Yangd56f5462008-04-25 10:13:16 +08001191 CPU_BASED_CR3_LOAD_EXITING |
1192 CPU_BASED_CR3_STORE_EXITING |
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001193 CPU_BASED_USE_IO_BITMAPS |
1194 CPU_BASED_MOV_DR_EXITING |
Marcelo Tosattia7052892008-09-23 13:18:35 -03001195 CPU_BASED_USE_TSC_OFFSETING |
1196 CPU_BASED_INVLPG_EXITING;
Sheng Yangf78e0e22007-10-29 09:40:42 +08001197 opt = CPU_BASED_TPR_SHADOW |
Sheng Yang25c5f222008-03-28 13:18:56 +08001198 CPU_BASED_USE_MSR_BITMAPS |
Sheng Yangf78e0e22007-10-29 09:40:42 +08001199 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001200 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1201 &_cpu_based_exec_control) < 0)
Yang, Sheng002c7f72007-07-31 14:23:01 +03001202 return -EIO;
Yang, Sheng6e5d8652007-09-12 18:03:11 +08001203#ifdef CONFIG_X86_64
1204 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1205 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1206 ~CPU_BASED_CR8_STORE_EXITING;
1207#endif
Sheng Yangf78e0e22007-10-29 09:40:42 +08001208 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
Sheng Yangd56f5462008-04-25 10:13:16 +08001209 min2 = 0;
1210 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
Sheng Yang2384d2b2008-01-17 15:14:33 +08001211 SECONDARY_EXEC_WBINVD_EXITING |
Sheng Yangd56f5462008-04-25 10:13:16 +08001212 SECONDARY_EXEC_ENABLE_VPID |
1213 SECONDARY_EXEC_ENABLE_EPT;
1214 if (adjust_vmx_controls(min2, opt2,
1215 MSR_IA32_VMX_PROCBASED_CTLS2,
Sheng Yangf78e0e22007-10-29 09:40:42 +08001216 &_cpu_based_2nd_exec_control) < 0)
1217 return -EIO;
1218 }
1219#ifndef CONFIG_X86_64
1220 if (!(_cpu_based_2nd_exec_control &
1221 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1222 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1223#endif
Sheng Yangd56f5462008-04-25 10:13:16 +08001224 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
Marcelo Tosattia7052892008-09-23 13:18:35 -03001225 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1226 enabled */
Sheng Yangd56f5462008-04-25 10:13:16 +08001227 min &= ~(CPU_BASED_CR3_LOAD_EXITING |
Marcelo Tosattia7052892008-09-23 13:18:35 -03001228 CPU_BASED_CR3_STORE_EXITING |
1229 CPU_BASED_INVLPG_EXITING);
Sheng Yangd56f5462008-04-25 10:13:16 +08001230 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1231 &_cpu_based_exec_control) < 0)
1232 return -EIO;
1233 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1234 vmx_capability.ept, vmx_capability.vpid);
1235 }
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001236
1237 min = 0;
1238#ifdef CONFIG_X86_64
1239 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1240#endif
Sheng Yang468d4722008-10-09 16:01:55 +08001241 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001242 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1243 &_vmexit_control) < 0)
Yang, Sheng002c7f72007-07-31 14:23:01 +03001244 return -EIO;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001245
Sheng Yang468d4722008-10-09 16:01:55 +08001246 min = 0;
1247 opt = VM_ENTRY_LOAD_IA32_PAT;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001248 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1249 &_vmentry_control) < 0)
Yang, Sheng002c7f72007-07-31 14:23:01 +03001250 return -EIO;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001251
Nguyen Anh Quynhc68876f2006-12-29 16:49:54 -08001252 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001253
1254 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1255 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
Yang, Sheng002c7f72007-07-31 14:23:01 +03001256 return -EIO;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001257
1258#ifdef CONFIG_X86_64
1259 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1260 if (vmx_msr_high & (1u<<16))
Yang, Sheng002c7f72007-07-31 14:23:01 +03001261 return -EIO;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001262#endif
1263
1264 /* Require Write-Back (WB) memory type for VMCS accesses. */
1265 if (((vmx_msr_high >> 18) & 15) != 6)
Yang, Sheng002c7f72007-07-31 14:23:01 +03001266 return -EIO;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001267
Yang, Sheng002c7f72007-07-31 14:23:01 +03001268 vmcs_conf->size = vmx_msr_high & 0x1fff;
1269 vmcs_conf->order = get_order(vmcs_config.size);
1270 vmcs_conf->revision_id = vmx_msr_low;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001271
Yang, Sheng002c7f72007-07-31 14:23:01 +03001272 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1273 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
Sheng Yangf78e0e22007-10-29 09:40:42 +08001274 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
Yang, Sheng002c7f72007-07-31 14:23:01 +03001275 vmcs_conf->vmexit_ctrl = _vmexit_control;
1276 vmcs_conf->vmentry_ctrl = _vmentry_control;
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001277
1278 return 0;
Nguyen Anh Quynhc68876f2006-12-29 16:49:54 -08001279}
Avi Kivity6aa8b732006-12-10 02:21:36 -08001280
1281static struct vmcs *alloc_vmcs_cpu(int cpu)
1282{
1283 int node = cpu_to_node(cpu);
1284 struct page *pages;
1285 struct vmcs *vmcs;
1286
Mel Gorman6484eb32009-06-16 15:31:54 -07001287 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001288 if (!pages)
1289 return NULL;
1290 vmcs = page_address(pages);
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001291 memset(vmcs, 0, vmcs_config.size);
1292 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
Avi Kivity6aa8b732006-12-10 02:21:36 -08001293 return vmcs;
1294}
1295
1296static struct vmcs *alloc_vmcs(void)
1297{
Ingo Molnard3b2c332007-01-05 16:36:23 -08001298 return alloc_vmcs_cpu(raw_smp_processor_id());
Avi Kivity6aa8b732006-12-10 02:21:36 -08001299}
1300
1301static void free_vmcs(struct vmcs *vmcs)
1302{
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03001303 free_pages((unsigned long)vmcs, vmcs_config.order);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001304}
1305
Sam Ravnborg39959582007-06-01 00:47:13 -07001306static void free_kvm_area(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001307{
1308 int cpu;
1309
1310 for_each_online_cpu(cpu)
1311 free_vmcs(per_cpu(vmxarea, cpu));
1312}
1313
Avi Kivity6aa8b732006-12-10 02:21:36 -08001314static __init int alloc_kvm_area(void)
1315{
1316 int cpu;
1317
1318 for_each_online_cpu(cpu) {
1319 struct vmcs *vmcs;
1320
1321 vmcs = alloc_vmcs_cpu(cpu);
1322 if (!vmcs) {
1323 free_kvm_area();
1324 return -ENOMEM;
1325 }
1326
1327 per_cpu(vmxarea, cpu) = vmcs;
1328 }
1329 return 0;
1330}
1331
1332static __init int hardware_setup(void)
1333{
Yang, Sheng002c7f72007-07-31 14:23:01 +03001334 if (setup_vmcs_config(&vmcs_config) < 0)
1335 return -EIO;
Joerg Roedel50a37eb2008-01-31 14:57:38 +01001336
1337 if (boot_cpu_has(X86_FEATURE_NX))
1338 kvm_enable_efer_bits(EFER_NX);
1339
Sheng Yang93ba03c2009-04-01 15:52:32 +08001340 if (!cpu_has_vmx_vpid())
1341 enable_vpid = 0;
1342
1343 if (!cpu_has_vmx_ept())
1344 enable_ept = 0;
1345
1346 if (!cpu_has_vmx_flexpriority())
1347 flexpriority_enabled = 0;
1348
Gleb Natapov95ba8273132009-04-21 17:45:08 +03001349 if (!cpu_has_vmx_tpr_shadow())
1350 kvm_x86_ops->update_cr8_intercept = NULL;
1351
Avi Kivity6aa8b732006-12-10 02:21:36 -08001352 return alloc_kvm_area();
1353}
1354
1355static __exit void hardware_unsetup(void)
1356{
1357 free_kvm_area();
1358}
1359
Avi Kivity6aa8b732006-12-10 02:21:36 -08001360static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1361{
1362 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1363
Avi Kivity6af11b92007-03-19 13:18:10 +02001364 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
Avi Kivity6aa8b732006-12-10 02:21:36 -08001365 vmcs_write16(sf->selector, save->selector);
1366 vmcs_writel(sf->base, save->base);
1367 vmcs_write32(sf->limit, save->limit);
1368 vmcs_write32(sf->ar_bytes, save->ar);
1369 } else {
1370 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1371 << AR_DPL_SHIFT;
1372 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1373 }
1374}
1375
1376static void enter_pmode(struct kvm_vcpu *vcpu)
1377{
1378 unsigned long flags;
Mohammed Gamala89a8fb2008-08-17 16:42:16 +03001379 struct vcpu_vmx *vmx = to_vmx(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001380
Mohammed Gamala89a8fb2008-08-17 16:42:16 +03001381 vmx->emulation_required = 1;
Nitin A Kamble56b237e2009-06-04 15:04:08 -07001382 vcpu->arch.rmode.vm86_active = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001383
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001384 vmcs_writel(GUEST_TR_BASE, vcpu->arch.rmode.tr.base);
1385 vmcs_write32(GUEST_TR_LIMIT, vcpu->arch.rmode.tr.limit);
1386 vmcs_write32(GUEST_TR_AR_BYTES, vcpu->arch.rmode.tr.ar);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001387
1388 flags = vmcs_readl(GUEST_RFLAGS);
Glauber de Oliveira Costa053de042008-01-30 13:31:27 +01001389 flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001390 flags |= (vcpu->arch.rmode.save_iopl << IOPL_SHIFT);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001391 vmcs_writel(GUEST_RFLAGS, flags);
1392
Rusty Russell66aee912007-07-17 23:34:16 +10001393 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1394 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
Avi Kivity6aa8b732006-12-10 02:21:36 -08001395
1396 update_exception_bitmap(vcpu);
1397
Mohammed Gamala89a8fb2008-08-17 16:42:16 +03001398 if (emulate_invalid_guest_state)
1399 return;
1400
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001401 fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1402 fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1403 fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1404 fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001405
1406 vmcs_write16(GUEST_SS_SELECTOR, 0);
1407 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1408
1409 vmcs_write16(GUEST_CS_SELECTOR,
1410 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1411 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1412}
1413
Mike Dayd77c26f2007-10-08 09:02:08 -04001414static gva_t rmode_tss_base(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001415{
Zhang Xiantaobfc6d222007-12-14 10:20:16 +08001416 if (!kvm->arch.tss_addr) {
Izik Eiduscbc94022007-10-25 00:29:55 +02001417 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1418 kvm->memslots[0].npages - 3;
1419 return base_gfn << PAGE_SHIFT;
1420 }
Zhang Xiantaobfc6d222007-12-14 10:20:16 +08001421 return kvm->arch.tss_addr;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001422}
1423
1424static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1425{
1426 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1427
1428 save->selector = vmcs_read16(sf->selector);
1429 save->base = vmcs_readl(sf->base);
1430 save->limit = vmcs_read32(sf->limit);
1431 save->ar = vmcs_read32(sf->ar_bytes);
Jan Kiszka15b00f32007-11-19 10:21:45 +01001432 vmcs_write16(sf->selector, save->base >> 4);
1433 vmcs_write32(sf->base, save->base & 0xfffff);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001434 vmcs_write32(sf->limit, 0xffff);
1435 vmcs_write32(sf->ar_bytes, 0xf3);
1436}
1437
1438static void enter_rmode(struct kvm_vcpu *vcpu)
1439{
1440 unsigned long flags;
Mohammed Gamala89a8fb2008-08-17 16:42:16 +03001441 struct vcpu_vmx *vmx = to_vmx(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001442
Mohammed Gamala89a8fb2008-08-17 16:42:16 +03001443 vmx->emulation_required = 1;
Nitin A Kamble56b237e2009-06-04 15:04:08 -07001444 vcpu->arch.rmode.vm86_active = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001445
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001446 vcpu->arch.rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001447 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1448
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001449 vcpu->arch.rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001450 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1451
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001452 vcpu->arch.rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001453 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1454
1455 flags = vmcs_readl(GUEST_RFLAGS);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001456 vcpu->arch.rmode.save_iopl
1457 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001458
Glauber de Oliveira Costa053de042008-01-30 13:31:27 +01001459 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001460
1461 vmcs_writel(GUEST_RFLAGS, flags);
Rusty Russell66aee912007-07-17 23:34:16 +10001462 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001463 update_exception_bitmap(vcpu);
1464
Mohammed Gamala89a8fb2008-08-17 16:42:16 +03001465 if (emulate_invalid_guest_state)
1466 goto continue_rmode;
1467
Avi Kivity6aa8b732006-12-10 02:21:36 -08001468 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1469 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1470 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1471
1472 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
Michael Riepeabacf8d2006-12-22 01:05:45 -08001473 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
Avi Kivity8cb5b032007-03-20 18:40:40 +02001474 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1475 vmcs_writel(GUEST_CS_BASE, 0xf0000);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001476 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1477
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001478 fix_rmode_seg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1479 fix_rmode_seg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1480 fix_rmode_seg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1481 fix_rmode_seg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
Avi Kivity75880a02007-06-20 11:20:04 +03001482
Mohammed Gamala89a8fb2008-08-17 16:42:16 +03001483continue_rmode:
Eddie Dong8668a3c2007-10-10 14:26:45 +08001484 kvm_mmu_reset_context(vcpu);
Sheng Yangb7ebfb02008-04-25 21:44:52 +08001485 init_rmode(vcpu->kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001486}
1487
Amit Shah401d10d2009-02-20 22:53:37 +05301488static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1489{
1490 struct vcpu_vmx *vmx = to_vmx(vcpu);
1491 struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1492
1493 vcpu->arch.shadow_efer = efer;
1494 if (!msr)
1495 return;
1496 if (efer & EFER_LMA) {
1497 vmcs_write32(VM_ENTRY_CONTROLS,
1498 vmcs_read32(VM_ENTRY_CONTROLS) |
1499 VM_ENTRY_IA32E_MODE);
1500 msr->data = efer;
1501 } else {
1502 vmcs_write32(VM_ENTRY_CONTROLS,
1503 vmcs_read32(VM_ENTRY_CONTROLS) &
1504 ~VM_ENTRY_IA32E_MODE);
1505
1506 msr->data = efer & ~EFER_LME;
1507 }
1508 setup_msrs(vmx);
1509}
1510
Avi Kivity05b3e0c2006-12-13 00:33:45 -08001511#ifdef CONFIG_X86_64
Avi Kivity6aa8b732006-12-10 02:21:36 -08001512
1513static void enter_lmode(struct kvm_vcpu *vcpu)
1514{
1515 u32 guest_tr_ar;
1516
1517 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1518 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1519 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
Harvey Harrisonb8688d52008-03-03 12:59:56 -08001520 __func__);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001521 vmcs_write32(GUEST_TR_AR_BYTES,
1522 (guest_tr_ar & ~AR_TYPE_MASK)
1523 | AR_TYPE_BUSY_64_TSS);
1524 }
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001525 vcpu->arch.shadow_efer |= EFER_LMA;
Amit Shah401d10d2009-02-20 22:53:37 +05301526 vmx_set_efer(vcpu, vcpu->arch.shadow_efer);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001527}
1528
1529static void exit_lmode(struct kvm_vcpu *vcpu)
1530{
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001531 vcpu->arch.shadow_efer &= ~EFER_LMA;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001532
1533 vmcs_write32(VM_ENTRY_CONTROLS,
1534 vmcs_read32(VM_ENTRY_CONTROLS)
Li, Xin B1e4e6e02007-08-01 21:49:10 +03001535 & ~VM_ENTRY_IA32E_MODE);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001536}
1537
1538#endif
1539
Sheng Yang2384d2b2008-01-17 15:14:33 +08001540static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1541{
1542 vpid_sync_vcpu_all(to_vmx(vcpu));
Avi Kivity089d0342009-03-23 18:26:32 +02001543 if (enable_ept)
Sheng Yang4e1096d2008-07-06 19:16:51 +08001544 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
Sheng Yang2384d2b2008-01-17 15:14:33 +08001545}
1546
Anthony Liguori25c4c272007-04-27 09:29:21 +03001547static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
Avi Kivity399badf2007-01-05 16:36:38 -08001548{
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001549 vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1550 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
Avi Kivity399badf2007-01-05 16:36:38 -08001551}
1552
Sheng Yang14394422008-04-28 12:24:45 +08001553static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1554{
Avi Kivity6de4f3a2009-05-31 22:58:47 +03001555 if (!test_bit(VCPU_EXREG_PDPTR,
1556 (unsigned long *)&vcpu->arch.regs_dirty))
1557 return;
1558
Sheng Yang14394422008-04-28 12:24:45 +08001559 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
Sheng Yang14394422008-04-28 12:24:45 +08001560 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1561 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1562 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1563 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1564 }
1565}
1566
Avi Kivity8f5d5492009-05-31 18:41:29 +03001567static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1568{
1569 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1570 vcpu->arch.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1571 vcpu->arch.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1572 vcpu->arch.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1573 vcpu->arch.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
1574 }
Avi Kivity6de4f3a2009-05-31 22:58:47 +03001575
1576 __set_bit(VCPU_EXREG_PDPTR,
1577 (unsigned long *)&vcpu->arch.regs_avail);
1578 __set_bit(VCPU_EXREG_PDPTR,
1579 (unsigned long *)&vcpu->arch.regs_dirty);
Avi Kivity8f5d5492009-05-31 18:41:29 +03001580}
1581
Sheng Yang14394422008-04-28 12:24:45 +08001582static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1583
1584static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1585 unsigned long cr0,
1586 struct kvm_vcpu *vcpu)
1587{
1588 if (!(cr0 & X86_CR0_PG)) {
1589 /* From paging/starting to nonpaging */
1590 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
Sheng Yang65267ea2008-06-18 14:43:38 +08001591 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
Sheng Yang14394422008-04-28 12:24:45 +08001592 (CPU_BASED_CR3_LOAD_EXITING |
1593 CPU_BASED_CR3_STORE_EXITING));
1594 vcpu->arch.cr0 = cr0;
1595 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1596 *hw_cr0 |= X86_CR0_PE | X86_CR0_PG;
1597 *hw_cr0 &= ~X86_CR0_WP;
1598 } else if (!is_paging(vcpu)) {
1599 /* From nonpaging to paging */
1600 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
Sheng Yang65267ea2008-06-18 14:43:38 +08001601 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
Sheng Yang14394422008-04-28 12:24:45 +08001602 ~(CPU_BASED_CR3_LOAD_EXITING |
1603 CPU_BASED_CR3_STORE_EXITING));
1604 vcpu->arch.cr0 = cr0;
1605 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1606 if (!(vcpu->arch.cr0 & X86_CR0_WP))
1607 *hw_cr0 &= ~X86_CR0_WP;
1608 }
1609}
1610
1611static void ept_update_paging_mode_cr4(unsigned long *hw_cr4,
1612 struct kvm_vcpu *vcpu)
1613{
1614 if (!is_paging(vcpu)) {
1615 *hw_cr4 &= ~X86_CR4_PAE;
1616 *hw_cr4 |= X86_CR4_PSE;
1617 } else if (!(vcpu->arch.cr4 & X86_CR4_PAE))
1618 *hw_cr4 &= ~X86_CR4_PAE;
1619}
1620
Avi Kivity6aa8b732006-12-10 02:21:36 -08001621static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1622{
Sheng Yang14394422008-04-28 12:24:45 +08001623 unsigned long hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) |
1624 KVM_VM_CR0_ALWAYS_ON;
1625
Avi Kivity5fd86fc2007-05-02 20:40:00 +03001626 vmx_fpu_deactivate(vcpu);
1627
Nitin A Kamble56b237e2009-06-04 15:04:08 -07001628 if (vcpu->arch.rmode.vm86_active && (cr0 & X86_CR0_PE))
Avi Kivity6aa8b732006-12-10 02:21:36 -08001629 enter_pmode(vcpu);
1630
Nitin A Kamble56b237e2009-06-04 15:04:08 -07001631 if (!vcpu->arch.rmode.vm86_active && !(cr0 & X86_CR0_PE))
Avi Kivity6aa8b732006-12-10 02:21:36 -08001632 enter_rmode(vcpu);
1633
Avi Kivity05b3e0c2006-12-13 00:33:45 -08001634#ifdef CONFIG_X86_64
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001635 if (vcpu->arch.shadow_efer & EFER_LME) {
Rusty Russell707d92fa2007-07-17 23:19:08 +10001636 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
Avi Kivity6aa8b732006-12-10 02:21:36 -08001637 enter_lmode(vcpu);
Rusty Russell707d92fa2007-07-17 23:19:08 +10001638 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
Avi Kivity6aa8b732006-12-10 02:21:36 -08001639 exit_lmode(vcpu);
1640 }
1641#endif
1642
Avi Kivity089d0342009-03-23 18:26:32 +02001643 if (enable_ept)
Sheng Yang14394422008-04-28 12:24:45 +08001644 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1645
Avi Kivity6aa8b732006-12-10 02:21:36 -08001646 vmcs_writel(CR0_READ_SHADOW, cr0);
Sheng Yang14394422008-04-28 12:24:45 +08001647 vmcs_writel(GUEST_CR0, hw_cr0);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001648 vcpu->arch.cr0 = cr0;
Avi Kivity5fd86fc2007-05-02 20:40:00 +03001649
Rusty Russell707d92fa2007-07-17 23:19:08 +10001650 if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
Avi Kivity5fd86fc2007-05-02 20:40:00 +03001651 vmx_fpu_activate(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001652}
1653
Sheng Yang14394422008-04-28 12:24:45 +08001654static u64 construct_eptp(unsigned long root_hpa)
1655{
1656 u64 eptp;
1657
1658 /* TODO write the value reading from MSR */
1659 eptp = VMX_EPT_DEFAULT_MT |
1660 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1661 eptp |= (root_hpa & PAGE_MASK);
1662
1663 return eptp;
1664}
1665
Avi Kivity6aa8b732006-12-10 02:21:36 -08001666static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1667{
Sheng Yang14394422008-04-28 12:24:45 +08001668 unsigned long guest_cr3;
1669 u64 eptp;
1670
1671 guest_cr3 = cr3;
Avi Kivity089d0342009-03-23 18:26:32 +02001672 if (enable_ept) {
Sheng Yang14394422008-04-28 12:24:45 +08001673 eptp = construct_eptp(cr3);
1674 vmcs_write64(EPT_POINTER, eptp);
Sheng Yang14394422008-04-28 12:24:45 +08001675 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1676 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
1677 }
1678
Sheng Yang2384d2b2008-01-17 15:14:33 +08001679 vmx_flush_tlb(vcpu);
Sheng Yang14394422008-04-28 12:24:45 +08001680 vmcs_writel(GUEST_CR3, guest_cr3);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001681 if (vcpu->arch.cr0 & X86_CR0_PE)
Avi Kivity5fd86fc2007-05-02 20:40:00 +03001682 vmx_fpu_deactivate(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001683}
1684
1685static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1686{
Nitin A Kamble56b237e2009-06-04 15:04:08 -07001687 unsigned long hw_cr4 = cr4 | (vcpu->arch.rmode.vm86_active ?
Sheng Yang14394422008-04-28 12:24:45 +08001688 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1689
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001690 vcpu->arch.cr4 = cr4;
Avi Kivity089d0342009-03-23 18:26:32 +02001691 if (enable_ept)
Sheng Yang14394422008-04-28 12:24:45 +08001692 ept_update_paging_mode_cr4(&hw_cr4, vcpu);
1693
1694 vmcs_writel(CR4_READ_SHADOW, cr4);
1695 vmcs_writel(GUEST_CR4, hw_cr4);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001696}
1697
Avi Kivity6aa8b732006-12-10 02:21:36 -08001698static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1699{
1700 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1701
1702 return vmcs_readl(sf->base);
1703}
1704
1705static void vmx_get_segment(struct kvm_vcpu *vcpu,
1706 struct kvm_segment *var, int seg)
1707{
1708 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1709 u32 ar;
1710
1711 var->base = vmcs_readl(sf->base);
1712 var->limit = vmcs_read32(sf->limit);
1713 var->selector = vmcs_read16(sf->selector);
1714 ar = vmcs_read32(sf->ar_bytes);
Avi Kivity9fd4a3b2009-01-04 23:43:42 +02001715 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001716 ar = 0;
1717 var->type = ar & 15;
1718 var->s = (ar >> 4) & 1;
1719 var->dpl = (ar >> 5) & 3;
1720 var->present = (ar >> 7) & 1;
1721 var->avl = (ar >> 12) & 1;
1722 var->l = (ar >> 13) & 1;
1723 var->db = (ar >> 14) & 1;
1724 var->g = (ar >> 15) & 1;
1725 var->unusable = (ar >> 16) & 1;
1726}
1727
Izik Eidus2e4d2652008-03-24 19:38:34 +02001728static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1729{
1730 struct kvm_segment kvm_seg;
1731
1732 if (!(vcpu->arch.cr0 & X86_CR0_PE)) /* if real mode */
1733 return 0;
1734
1735 if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1736 return 3;
1737
1738 vmx_get_segment(vcpu, &kvm_seg, VCPU_SREG_CS);
1739 return kvm_seg.selector & 3;
1740}
1741
Avi Kivity653e3102007-05-07 10:55:37 +03001742static u32 vmx_segment_access_rights(struct kvm_segment *var)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001743{
Avi Kivity6aa8b732006-12-10 02:21:36 -08001744 u32 ar;
1745
Avi Kivity653e3102007-05-07 10:55:37 +03001746 if (var->unusable)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001747 ar = 1 << 16;
1748 else {
1749 ar = var->type & 15;
1750 ar |= (var->s & 1) << 4;
1751 ar |= (var->dpl & 3) << 5;
1752 ar |= (var->present & 1) << 7;
1753 ar |= (var->avl & 1) << 12;
1754 ar |= (var->l & 1) << 13;
1755 ar |= (var->db & 1) << 14;
1756 ar |= (var->g & 1) << 15;
1757 }
Uri Lublinf7fbf1f2006-12-13 00:34:00 -08001758 if (ar == 0) /* a 0 value means unusable */
1759 ar = AR_UNUSABLE_MASK;
Avi Kivity653e3102007-05-07 10:55:37 +03001760
1761 return ar;
1762}
1763
1764static void vmx_set_segment(struct kvm_vcpu *vcpu,
1765 struct kvm_segment *var, int seg)
1766{
1767 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1768 u32 ar;
1769
Nitin A Kamble56b237e2009-06-04 15:04:08 -07001770 if (vcpu->arch.rmode.vm86_active && seg == VCPU_SREG_TR) {
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001771 vcpu->arch.rmode.tr.selector = var->selector;
1772 vcpu->arch.rmode.tr.base = var->base;
1773 vcpu->arch.rmode.tr.limit = var->limit;
1774 vcpu->arch.rmode.tr.ar = vmx_segment_access_rights(var);
Avi Kivity653e3102007-05-07 10:55:37 +03001775 return;
1776 }
1777 vmcs_writel(sf->base, var->base);
1778 vmcs_write32(sf->limit, var->limit);
1779 vmcs_write16(sf->selector, var->selector);
Nitin A Kamble56b237e2009-06-04 15:04:08 -07001780 if (vcpu->arch.rmode.vm86_active && var->s) {
Avi Kivity653e3102007-05-07 10:55:37 +03001781 /*
1782 * Hack real-mode segments into vm86 compatibility.
1783 */
1784 if (var->base == 0xffff0000 && var->selector == 0xf000)
1785 vmcs_writel(sf->base, 0xf0000);
1786 ar = 0xf3;
1787 } else
1788 ar = vmx_segment_access_rights(var);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001789 vmcs_write32(sf->ar_bytes, ar);
1790}
1791
Avi Kivity6aa8b732006-12-10 02:21:36 -08001792static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1793{
1794 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1795
1796 *db = (ar >> 14) & 1;
1797 *l = (ar >> 13) & 1;
1798}
1799
1800static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1801{
1802 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1803 dt->base = vmcs_readl(GUEST_IDTR_BASE);
1804}
1805
1806static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1807{
1808 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1809 vmcs_writel(GUEST_IDTR_BASE, dt->base);
1810}
1811
1812static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1813{
1814 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1815 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1816}
1817
1818static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1819{
1820 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1821 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1822}
1823
Mohammed Gamal648dfaa2008-08-17 16:38:32 +03001824static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
1825{
1826 struct kvm_segment var;
1827 u32 ar;
1828
1829 vmx_get_segment(vcpu, &var, seg);
1830 ar = vmx_segment_access_rights(&var);
1831
1832 if (var.base != (var.selector << 4))
1833 return false;
1834 if (var.limit != 0xffff)
1835 return false;
1836 if (ar != 0xf3)
1837 return false;
1838
1839 return true;
1840}
1841
1842static bool code_segment_valid(struct kvm_vcpu *vcpu)
1843{
1844 struct kvm_segment cs;
1845 unsigned int cs_rpl;
1846
1847 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1848 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
1849
Avi Kivity1872a3f2009-01-04 23:26:52 +02001850 if (cs.unusable)
1851 return false;
Mohammed Gamal648dfaa2008-08-17 16:38:32 +03001852 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
1853 return false;
1854 if (!cs.s)
1855 return false;
Avi Kivity1872a3f2009-01-04 23:26:52 +02001856 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
Mohammed Gamal648dfaa2008-08-17 16:38:32 +03001857 if (cs.dpl > cs_rpl)
1858 return false;
Avi Kivity1872a3f2009-01-04 23:26:52 +02001859 } else {
Mohammed Gamal648dfaa2008-08-17 16:38:32 +03001860 if (cs.dpl != cs_rpl)
1861 return false;
1862 }
1863 if (!cs.present)
1864 return false;
1865
1866 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
1867 return true;
1868}
1869
1870static bool stack_segment_valid(struct kvm_vcpu *vcpu)
1871{
1872 struct kvm_segment ss;
1873 unsigned int ss_rpl;
1874
1875 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1876 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
1877
Avi Kivity1872a3f2009-01-04 23:26:52 +02001878 if (ss.unusable)
1879 return true;
1880 if (ss.type != 3 && ss.type != 7)
Mohammed Gamal648dfaa2008-08-17 16:38:32 +03001881 return false;
1882 if (!ss.s)
1883 return false;
1884 if (ss.dpl != ss_rpl) /* DPL != RPL */
1885 return false;
1886 if (!ss.present)
1887 return false;
1888
1889 return true;
1890}
1891
1892static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
1893{
1894 struct kvm_segment var;
1895 unsigned int rpl;
1896
1897 vmx_get_segment(vcpu, &var, seg);
1898 rpl = var.selector & SELECTOR_RPL_MASK;
1899
Avi Kivity1872a3f2009-01-04 23:26:52 +02001900 if (var.unusable)
1901 return true;
Mohammed Gamal648dfaa2008-08-17 16:38:32 +03001902 if (!var.s)
1903 return false;
1904 if (!var.present)
1905 return false;
1906 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
1907 if (var.dpl < rpl) /* DPL < RPL */
1908 return false;
1909 }
1910
1911 /* TODO: Add other members to kvm_segment_field to allow checking for other access
1912 * rights flags
1913 */
1914 return true;
1915}
1916
1917static bool tr_valid(struct kvm_vcpu *vcpu)
1918{
1919 struct kvm_segment tr;
1920
1921 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
1922
Avi Kivity1872a3f2009-01-04 23:26:52 +02001923 if (tr.unusable)
1924 return false;
Mohammed Gamal648dfaa2008-08-17 16:38:32 +03001925 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
1926 return false;
Avi Kivity1872a3f2009-01-04 23:26:52 +02001927 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
Mohammed Gamal648dfaa2008-08-17 16:38:32 +03001928 return false;
1929 if (!tr.present)
1930 return false;
1931
1932 return true;
1933}
1934
1935static bool ldtr_valid(struct kvm_vcpu *vcpu)
1936{
1937 struct kvm_segment ldtr;
1938
1939 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
1940
Avi Kivity1872a3f2009-01-04 23:26:52 +02001941 if (ldtr.unusable)
1942 return true;
Mohammed Gamal648dfaa2008-08-17 16:38:32 +03001943 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
1944 return false;
1945 if (ldtr.type != 2)
1946 return false;
1947 if (!ldtr.present)
1948 return false;
1949
1950 return true;
1951}
1952
1953static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
1954{
1955 struct kvm_segment cs, ss;
1956
1957 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1958 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1959
1960 return ((cs.selector & SELECTOR_RPL_MASK) ==
1961 (ss.selector & SELECTOR_RPL_MASK));
1962}
1963
1964/*
1965 * Check if guest state is valid. Returns true if valid, false if
1966 * not.
1967 * We assume that registers are always usable
1968 */
1969static bool guest_state_valid(struct kvm_vcpu *vcpu)
1970{
1971 /* real mode guest state checks */
1972 if (!(vcpu->arch.cr0 & X86_CR0_PE)) {
1973 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
1974 return false;
1975 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
1976 return false;
1977 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
1978 return false;
1979 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
1980 return false;
1981 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
1982 return false;
1983 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
1984 return false;
1985 } else {
1986 /* protected mode guest state checks */
1987 if (!cs_ss_rpl_check(vcpu))
1988 return false;
1989 if (!code_segment_valid(vcpu))
1990 return false;
1991 if (!stack_segment_valid(vcpu))
1992 return false;
1993 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
1994 return false;
1995 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
1996 return false;
1997 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
1998 return false;
1999 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2000 return false;
2001 if (!tr_valid(vcpu))
2002 return false;
2003 if (!ldtr_valid(vcpu))
2004 return false;
2005 }
2006 /* TODO:
2007 * - Add checks on RIP
2008 * - Add checks on RFLAGS
2009 */
2010
2011 return true;
2012}
2013
Mike Dayd77c26f2007-10-08 09:02:08 -04002014static int init_rmode_tss(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002015{
Avi Kivity6aa8b732006-12-10 02:21:36 -08002016 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
Izik Eidus195aefd2007-10-01 22:14:18 +02002017 u16 data = 0;
Marcelo Tosatti10589a42007-12-20 19:18:22 -05002018 int ret = 0;
Izik Eidus195aefd2007-10-01 22:14:18 +02002019 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002020
Izik Eidus195aefd2007-10-01 22:14:18 +02002021 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2022 if (r < 0)
Marcelo Tosatti10589a42007-12-20 19:18:22 -05002023 goto out;
Izik Eidus195aefd2007-10-01 22:14:18 +02002024 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
Sheng Yang464d17c2008-08-13 14:10:33 +08002025 r = kvm_write_guest_page(kvm, fn++, &data,
2026 TSS_IOPB_BASE_OFFSET, sizeof(u16));
Izik Eidus195aefd2007-10-01 22:14:18 +02002027 if (r < 0)
Marcelo Tosatti10589a42007-12-20 19:18:22 -05002028 goto out;
Izik Eidus195aefd2007-10-01 22:14:18 +02002029 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2030 if (r < 0)
Marcelo Tosatti10589a42007-12-20 19:18:22 -05002031 goto out;
Izik Eidus195aefd2007-10-01 22:14:18 +02002032 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2033 if (r < 0)
Marcelo Tosatti10589a42007-12-20 19:18:22 -05002034 goto out;
Izik Eidus195aefd2007-10-01 22:14:18 +02002035 data = ~0;
Marcelo Tosatti10589a42007-12-20 19:18:22 -05002036 r = kvm_write_guest_page(kvm, fn, &data,
2037 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2038 sizeof(u8));
Izik Eidus195aefd2007-10-01 22:14:18 +02002039 if (r < 0)
Marcelo Tosatti10589a42007-12-20 19:18:22 -05002040 goto out;
2041
2042 ret = 1;
2043out:
Marcelo Tosatti10589a42007-12-20 19:18:22 -05002044 return ret;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002045}
2046
Sheng Yangb7ebfb02008-04-25 21:44:52 +08002047static int init_rmode_identity_map(struct kvm *kvm)
2048{
2049 int i, r, ret;
2050 pfn_t identity_map_pfn;
2051 u32 tmp;
2052
Avi Kivity089d0342009-03-23 18:26:32 +02002053 if (!enable_ept)
Sheng Yangb7ebfb02008-04-25 21:44:52 +08002054 return 1;
2055 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2056 printk(KERN_ERR "EPT: identity-mapping pagetable "
2057 "haven't been allocated!\n");
2058 return 0;
2059 }
2060 if (likely(kvm->arch.ept_identity_pagetable_done))
2061 return 1;
2062 ret = 0;
2063 identity_map_pfn = VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT;
2064 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2065 if (r < 0)
2066 goto out;
2067 /* Set up identity-mapping pagetable for EPT in real mode */
2068 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2069 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2070 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2071 r = kvm_write_guest_page(kvm, identity_map_pfn,
2072 &tmp, i * sizeof(tmp), sizeof(tmp));
2073 if (r < 0)
2074 goto out;
2075 }
2076 kvm->arch.ept_identity_pagetable_done = true;
2077 ret = 1;
2078out:
2079 return ret;
2080}
2081
Avi Kivity6aa8b732006-12-10 02:21:36 -08002082static void seg_setup(int seg)
2083{
2084 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2085
2086 vmcs_write16(sf->selector, 0);
2087 vmcs_writel(sf->base, 0);
2088 vmcs_write32(sf->limit, 0xffff);
Avi Kivitya16b20d2008-08-20 15:48:27 +03002089 vmcs_write32(sf->ar_bytes, 0xf3);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002090}
2091
Sheng Yangf78e0e22007-10-29 09:40:42 +08002092static int alloc_apic_access_page(struct kvm *kvm)
2093{
2094 struct kvm_userspace_memory_region kvm_userspace_mem;
2095 int r = 0;
2096
Izik Eidus72dc67a2008-02-10 18:04:15 +02002097 down_write(&kvm->slots_lock);
Zhang Xiantaobfc6d222007-12-14 10:20:16 +08002098 if (kvm->arch.apic_access_page)
Sheng Yangf78e0e22007-10-29 09:40:42 +08002099 goto out;
2100 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2101 kvm_userspace_mem.flags = 0;
2102 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2103 kvm_userspace_mem.memory_size = PAGE_SIZE;
2104 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2105 if (r)
2106 goto out;
Izik Eidus72dc67a2008-02-10 18:04:15 +02002107
Zhang Xiantaobfc6d222007-12-14 10:20:16 +08002108 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
Sheng Yangf78e0e22007-10-29 09:40:42 +08002109out:
Izik Eidus72dc67a2008-02-10 18:04:15 +02002110 up_write(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +08002111 return r;
2112}
2113
Sheng Yangb7ebfb02008-04-25 21:44:52 +08002114static int alloc_identity_pagetable(struct kvm *kvm)
2115{
2116 struct kvm_userspace_memory_region kvm_userspace_mem;
2117 int r = 0;
2118
2119 down_write(&kvm->slots_lock);
2120 if (kvm->arch.ept_identity_pagetable)
2121 goto out;
2122 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2123 kvm_userspace_mem.flags = 0;
2124 kvm_userspace_mem.guest_phys_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
2125 kvm_userspace_mem.memory_size = PAGE_SIZE;
2126 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2127 if (r)
2128 goto out;
2129
Sheng Yangb7ebfb02008-04-25 21:44:52 +08002130 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2131 VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT);
Sheng Yangb7ebfb02008-04-25 21:44:52 +08002132out:
2133 up_write(&kvm->slots_lock);
2134 return r;
2135}
2136
Sheng Yang2384d2b2008-01-17 15:14:33 +08002137static void allocate_vpid(struct vcpu_vmx *vmx)
2138{
2139 int vpid;
2140
2141 vmx->vpid = 0;
Avi Kivity919818a2009-03-23 18:01:29 +02002142 if (!enable_vpid)
Sheng Yang2384d2b2008-01-17 15:14:33 +08002143 return;
2144 spin_lock(&vmx_vpid_lock);
2145 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2146 if (vpid < VMX_NR_VPIDS) {
2147 vmx->vpid = vpid;
2148 __set_bit(vpid, vmx_vpid_bitmap);
2149 }
2150 spin_unlock(&vmx_vpid_lock);
2151}
2152
Avi Kivity58972972009-02-24 22:26:47 +02002153static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
Sheng Yang25c5f222008-03-28 13:18:56 +08002154{
Avi Kivity3e7c73e2009-02-24 21:46:19 +02002155 int f = sizeof(unsigned long);
Sheng Yang25c5f222008-03-28 13:18:56 +08002156
2157 if (!cpu_has_vmx_msr_bitmap())
2158 return;
2159
2160 /*
2161 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2162 * have the write-low and read-high bitmap offsets the wrong way round.
2163 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2164 */
Sheng Yang25c5f222008-03-28 13:18:56 +08002165 if (msr <= 0x1fff) {
Avi Kivity3e7c73e2009-02-24 21:46:19 +02002166 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2167 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
Sheng Yang25c5f222008-03-28 13:18:56 +08002168 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2169 msr &= 0x1fff;
Avi Kivity3e7c73e2009-02-24 21:46:19 +02002170 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2171 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
Sheng Yang25c5f222008-03-28 13:18:56 +08002172 }
Sheng Yang25c5f222008-03-28 13:18:56 +08002173}
2174
Avi Kivity58972972009-02-24 22:26:47 +02002175static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2176{
2177 if (!longmode_only)
2178 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2179 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2180}
2181
Avi Kivity6aa8b732006-12-10 02:21:36 -08002182/*
2183 * Sets up the vmcs for emulated real mode.
2184 */
Rusty Russell8b9cf982007-07-30 16:31:43 +10002185static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002186{
Sheng Yang468d4722008-10-09 16:01:55 +08002187 u32 host_sysenter_cs, msr_low, msr_high;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002188 u32 junk;
Marcelo Tosatti53f658b32008-12-11 20:45:05 +01002189 u64 host_pat, tsc_this, tsc_base;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002190 unsigned long a;
2191 struct descriptor_table dt;
2192 int i;
Avi Kivitycd2276a2007-05-14 20:41:13 +03002193 unsigned long kvm_vmx_return;
Yang, Sheng6e5d8652007-09-12 18:03:11 +08002194 u32 exec_control;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002195
Avi Kivity6aa8b732006-12-10 02:21:36 -08002196 /* I/O */
Avi Kivity3e7c73e2009-02-24 21:46:19 +02002197 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2198 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
Avi Kivity6aa8b732006-12-10 02:21:36 -08002199
Sheng Yang25c5f222008-03-28 13:18:56 +08002200 if (cpu_has_vmx_msr_bitmap())
Avi Kivity58972972009-02-24 22:26:47 +02002201 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
Sheng Yang25c5f222008-03-28 13:18:56 +08002202
Avi Kivity6aa8b732006-12-10 02:21:36 -08002203 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2204
Avi Kivity6aa8b732006-12-10 02:21:36 -08002205 /* Control */
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03002206 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2207 vmcs_config.pin_based_exec_ctrl);
Yang, Sheng6e5d8652007-09-12 18:03:11 +08002208
2209 exec_control = vmcs_config.cpu_based_exec_ctrl;
2210 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2211 exec_control &= ~CPU_BASED_TPR_SHADOW;
2212#ifdef CONFIG_X86_64
2213 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2214 CPU_BASED_CR8_LOAD_EXITING;
2215#endif
2216 }
Avi Kivity089d0342009-03-23 18:26:32 +02002217 if (!enable_ept)
Sheng Yangd56f5462008-04-25 10:13:16 +08002218 exec_control |= CPU_BASED_CR3_STORE_EXITING |
Marcelo Tosatti83dbc832008-10-07 17:01:27 -03002219 CPU_BASED_CR3_LOAD_EXITING |
2220 CPU_BASED_INVLPG_EXITING;
Yang, Sheng6e5d8652007-09-12 18:03:11 +08002221 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002222
Sheng Yang83ff3b92007-11-21 14:33:25 +08002223 if (cpu_has_secondary_exec_ctrls()) {
2224 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2225 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2226 exec_control &=
2227 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
Sheng Yang2384d2b2008-01-17 15:14:33 +08002228 if (vmx->vpid == 0)
2229 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
Avi Kivity089d0342009-03-23 18:26:32 +02002230 if (!enable_ept)
Sheng Yangd56f5462008-04-25 10:13:16 +08002231 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
Sheng Yang83ff3b92007-11-21 14:33:25 +08002232 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2233 }
Sheng Yangf78e0e22007-10-29 09:40:42 +08002234
Avi Kivityc7addb92007-09-16 18:58:32 +02002235 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2236 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002237 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
2238
2239 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
2240 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
2241 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2242
2243 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
2244 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2245 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
Avi Kivityd6e88ae2008-07-10 16:53:33 +03002246 vmcs_write16(HOST_FS_SELECTOR, kvm_read_fs()); /* 22.2.4 */
2247 vmcs_write16(HOST_GS_SELECTOR, kvm_read_gs()); /* 22.2.4 */
Avi Kivity6aa8b732006-12-10 02:21:36 -08002248 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
Avi Kivity05b3e0c2006-12-13 00:33:45 -08002249#ifdef CONFIG_X86_64
Avi Kivity6aa8b732006-12-10 02:21:36 -08002250 rdmsrl(MSR_FS_BASE, a);
2251 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2252 rdmsrl(MSR_GS_BASE, a);
2253 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2254#else
2255 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2256 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2257#endif
2258
2259 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
2260
Avi Kivityd6e88ae2008-07-10 16:53:33 +03002261 kvm_get_idt(&dt);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002262 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
2263
Mike Dayd77c26f2007-10-08 09:02:08 -04002264 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
Avi Kivitycd2276a2007-05-14 20:41:13 +03002265 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
Eddie Dong2cc51562007-05-21 07:28:09 +03002266 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2267 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2268 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002269
2270 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2271 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2272 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2273 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
2274 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2275 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2276
Sheng Yang468d4722008-10-09 16:01:55 +08002277 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2278 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2279 host_pat = msr_low | ((u64) msr_high << 32);
2280 vmcs_write64(HOST_IA32_PAT, host_pat);
2281 }
2282 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2283 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2284 host_pat = msr_low | ((u64) msr_high << 32);
2285 /* Write the default value follow host pat */
2286 vmcs_write64(GUEST_IA32_PAT, host_pat);
2287 /* Keep arch.pat sync with GUEST_IA32_PAT */
2288 vmx->vcpu.arch.pat = host_pat;
2289 }
2290
Avi Kivity6aa8b732006-12-10 02:21:36 -08002291 for (i = 0; i < NR_VMX_MSR; ++i) {
2292 u32 index = vmx_msr_index[i];
2293 u32 data_low, data_high;
2294 u64 data;
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04002295 int j = vmx->nmsrs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002296
2297 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2298 continue;
Avi Kivity432bd6c2007-01-31 23:48:13 -08002299 if (wrmsr_safe(index, data_low, data_high) < 0)
2300 continue;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002301 data = data_low | ((u64)data_high << 32);
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04002302 vmx->host_msrs[j].index = index;
2303 vmx->host_msrs[j].reserved = 0;
2304 vmx->host_msrs[j].data = data;
2305 vmx->guest_msrs[j] = vmx->host_msrs[j];
2306 ++vmx->nmsrs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002307 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002308
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03002309 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002310
2311 /* 22.2.1, 20.8.1 */
Yang, Sheng1c3d14fe2007-07-29 11:07:42 +03002312 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2313
Avi Kivitye00c8cf2007-10-21 11:00:39 +02002314 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2315 vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
2316
Marcelo Tosatti53f658b32008-12-11 20:45:05 +01002317 tsc_base = vmx->vcpu.kvm->arch.vm_init_tsc;
2318 rdtscll(tsc_this);
2319 if (tsc_this < vmx->vcpu.kvm->arch.vm_init_tsc)
2320 tsc_base = tsc_this;
2321
2322 guest_write_tsc(0, tsc_base);
Sheng Yangf78e0e22007-10-29 09:40:42 +08002323
Avi Kivitye00c8cf2007-10-21 11:00:39 +02002324 return 0;
2325}
2326
Sheng Yangb7ebfb02008-04-25 21:44:52 +08002327static int init_rmode(struct kvm *kvm)
2328{
2329 if (!init_rmode_tss(kvm))
2330 return 0;
2331 if (!init_rmode_identity_map(kvm))
2332 return 0;
2333 return 1;
2334}
2335
Avi Kivitye00c8cf2007-10-21 11:00:39 +02002336static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2337{
2338 struct vcpu_vmx *vmx = to_vmx(vcpu);
2339 u64 msr;
2340 int ret;
2341
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002342 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
Marcelo Tosatti3200f402008-03-29 20:17:59 -03002343 down_read(&vcpu->kvm->slots_lock);
Sheng Yangb7ebfb02008-04-25 21:44:52 +08002344 if (!init_rmode(vmx->vcpu.kvm)) {
Avi Kivitye00c8cf2007-10-21 11:00:39 +02002345 ret = -ENOMEM;
2346 goto out;
2347 }
2348
Nitin A Kamble56b237e2009-06-04 15:04:08 -07002349 vmx->vcpu.arch.rmode.vm86_active = 0;
Avi Kivitye00c8cf2007-10-21 11:00:39 +02002350
Jan Kiszka3b86cd92008-09-26 09:30:57 +02002351 vmx->soft_vnmi_blocked = 0;
2352
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002353 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
Avi Kivity2d3ad1f2008-02-24 11:20:43 +02002354 kvm_set_cr8(&vmx->vcpu, 0);
Avi Kivitye00c8cf2007-10-21 11:00:39 +02002355 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2356 if (vmx->vcpu.vcpu_id == 0)
2357 msr |= MSR_IA32_APICBASE_BSP;
2358 kvm_set_apic_base(&vmx->vcpu, msr);
2359
2360 fx_init(&vmx->vcpu);
2361
Avi Kivity5706be02008-08-20 15:07:31 +03002362 seg_setup(VCPU_SREG_CS);
Avi Kivitye00c8cf2007-10-21 11:00:39 +02002363 /*
2364 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2365 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2366 */
2367 if (vmx->vcpu.vcpu_id == 0) {
2368 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2369 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2370 } else {
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002371 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2372 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
Avi Kivitye00c8cf2007-10-21 11:00:39 +02002373 }
Avi Kivitye00c8cf2007-10-21 11:00:39 +02002374
2375 seg_setup(VCPU_SREG_DS);
2376 seg_setup(VCPU_SREG_ES);
2377 seg_setup(VCPU_SREG_FS);
2378 seg_setup(VCPU_SREG_GS);
2379 seg_setup(VCPU_SREG_SS);
2380
2381 vmcs_write16(GUEST_TR_SELECTOR, 0);
2382 vmcs_writel(GUEST_TR_BASE, 0);
2383 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2384 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2385
2386 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2387 vmcs_writel(GUEST_LDTR_BASE, 0);
2388 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2389 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2390
2391 vmcs_write32(GUEST_SYSENTER_CS, 0);
2392 vmcs_writel(GUEST_SYSENTER_ESP, 0);
2393 vmcs_writel(GUEST_SYSENTER_EIP, 0);
2394
2395 vmcs_writel(GUEST_RFLAGS, 0x02);
2396 if (vmx->vcpu.vcpu_id == 0)
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002397 kvm_rip_write(vcpu, 0xfff0);
Avi Kivitye00c8cf2007-10-21 11:00:39 +02002398 else
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002399 kvm_rip_write(vcpu, 0);
2400 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
Avi Kivitye00c8cf2007-10-21 11:00:39 +02002401
Avi Kivitye00c8cf2007-10-21 11:00:39 +02002402 vmcs_writel(GUEST_DR7, 0x400);
2403
2404 vmcs_writel(GUEST_GDTR_BASE, 0);
2405 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2406
2407 vmcs_writel(GUEST_IDTR_BASE, 0);
2408 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2409
2410 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2411 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2412 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2413
Avi Kivitye00c8cf2007-10-21 11:00:39 +02002414 /* Special registers */
2415 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2416
2417 setup_msrs(vmx);
2418
Avi Kivity6aa8b732006-12-10 02:21:36 -08002419 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
2420
Sheng Yangf78e0e22007-10-29 09:40:42 +08002421 if (cpu_has_vmx_tpr_shadow()) {
2422 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2423 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2424 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002425 page_to_phys(vmx->vcpu.arch.apic->regs_page));
Sheng Yangf78e0e22007-10-29 09:40:42 +08002426 vmcs_write32(TPR_THRESHOLD, 0);
2427 }
2428
2429 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2430 vmcs_write64(APIC_ACCESS_ADDR,
Zhang Xiantaobfc6d222007-12-14 10:20:16 +08002431 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
Avi Kivity6aa8b732006-12-10 02:21:36 -08002432
Sheng Yang2384d2b2008-01-17 15:14:33 +08002433 if (vmx->vpid != 0)
2434 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2435
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002436 vmx->vcpu.arch.cr0 = 0x60000010;
2437 vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
Rusty Russell8b9cf982007-07-30 16:31:43 +10002438 vmx_set_cr4(&vmx->vcpu, 0);
Rusty Russell8b9cf982007-07-30 16:31:43 +10002439 vmx_set_efer(&vmx->vcpu, 0);
Rusty Russell8b9cf982007-07-30 16:31:43 +10002440 vmx_fpu_activate(&vmx->vcpu);
2441 update_exception_bitmap(&vmx->vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002442
Sheng Yang2384d2b2008-01-17 15:14:33 +08002443 vpid_sync_vcpu_all(vmx);
2444
Marcelo Tosatti3200f402008-03-29 20:17:59 -03002445 ret = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002446
Mohammed Gamala89a8fb2008-08-17 16:42:16 +03002447 /* HACK: Don't enable emulation on guest boot/reset */
2448 vmx->emulation_required = 0;
2449
Avi Kivity6aa8b732006-12-10 02:21:36 -08002450out:
Marcelo Tosatti3200f402008-03-29 20:17:59 -03002451 up_read(&vcpu->kvm->slots_lock);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002452 return ret;
2453}
2454
Jan Kiszka3b86cd92008-09-26 09:30:57 +02002455static void enable_irq_window(struct kvm_vcpu *vcpu)
2456{
2457 u32 cpu_based_vm_exec_control;
2458
2459 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2460 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2461 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2462}
2463
2464static void enable_nmi_window(struct kvm_vcpu *vcpu)
2465{
2466 u32 cpu_based_vm_exec_control;
2467
2468 if (!cpu_has_virtual_nmis()) {
2469 enable_irq_window(vcpu);
2470 return;
2471 }
2472
2473 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2474 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2475 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2476}
2477
Gleb Natapov66fd3f72009-05-11 13:35:50 +03002478static void vmx_inject_irq(struct kvm_vcpu *vcpu)
Eddie Dong85f455f2007-07-06 12:20:49 +03002479{
Avi Kivity9c8cba32007-11-22 11:42:59 +02002480 struct vcpu_vmx *vmx = to_vmx(vcpu);
Gleb Natapov66fd3f72009-05-11 13:35:50 +03002481 uint32_t intr;
2482 int irq = vcpu->arch.interrupt.nr;
Avi Kivity9c8cba32007-11-22 11:42:59 +02002483
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04002484 KVMTRACE_1D(INJ_VIRQ, vcpu, (u32)irq, handler);
2485
Avi Kivityfa89a812008-09-01 15:57:51 +03002486 ++vcpu->stat.irq_injections;
Nitin A Kamble56b237e2009-06-04 15:04:08 -07002487 if (vcpu->arch.rmode.vm86_active) {
Avi Kivity9c8cba32007-11-22 11:42:59 +02002488 vmx->rmode.irq.pending = true;
2489 vmx->rmode.irq.vector = irq;
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002490 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
Gleb Natapovae0bb3e2009-05-19 11:07:10 +03002491 if (vcpu->arch.interrupt.soft)
2492 vmx->rmode.irq.rip +=
2493 vmx->vcpu.arch.event_exit_inst_len;
Avi Kivity9c5623e2007-11-08 18:19:20 +02002494 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2495 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2496 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002497 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
Eddie Dong85f455f2007-07-06 12:20:49 +03002498 return;
2499 }
Gleb Natapov66fd3f72009-05-11 13:35:50 +03002500 intr = irq | INTR_INFO_VALID_MASK;
2501 if (vcpu->arch.interrupt.soft) {
2502 intr |= INTR_TYPE_SOFT_INTR;
2503 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2504 vmx->vcpu.arch.event_exit_inst_len);
2505 } else
2506 intr |= INTR_TYPE_EXT_INTR;
2507 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
Eddie Dong85f455f2007-07-06 12:20:49 +03002508}
2509
Sheng Yangf08864b2008-05-15 18:23:25 +08002510static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2511{
Jan Kiszka66a5a342008-09-26 09:30:51 +02002512 struct vcpu_vmx *vmx = to_vmx(vcpu);
2513
Jan Kiszka3b86cd92008-09-26 09:30:57 +02002514 if (!cpu_has_virtual_nmis()) {
2515 /*
2516 * Tracking the NMI-blocked state in software is built upon
2517 * finding the next open IRQ window. This, in turn, depends on
2518 * well-behaving guests: They have to keep IRQs disabled at
2519 * least as long as the NMI handler runs. Otherwise we may
2520 * cause NMI nesting, maybe breaking the guest. But as this is
2521 * highly unlikely, we can live with the residual risk.
2522 */
2523 vmx->soft_vnmi_blocked = 1;
2524 vmx->vnmi_blocked_time = 0;
2525 }
2526
Jan Kiszka487b3912008-09-26 09:30:56 +02002527 ++vcpu->stat.nmi_injections;
Nitin A Kamble56b237e2009-06-04 15:04:08 -07002528 if (vcpu->arch.rmode.vm86_active) {
Jan Kiszka66a5a342008-09-26 09:30:51 +02002529 vmx->rmode.irq.pending = true;
2530 vmx->rmode.irq.vector = NMI_VECTOR;
2531 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2532 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2533 NMI_VECTOR | INTR_TYPE_SOFT_INTR |
2534 INTR_INFO_VALID_MASK);
2535 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2536 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2537 return;
2538 }
Sheng Yangf08864b2008-05-15 18:23:25 +08002539 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2540 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
Sheng Yangf08864b2008-05-15 18:23:25 +08002541}
2542
Gleb Natapovc4282df2009-04-21 17:45:07 +03002543static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
Jan Kiszka33f089c2008-09-26 09:30:49 +02002544{
Jan Kiszka3b86cd92008-09-26 09:30:57 +02002545 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
Gleb Natapovc4282df2009-04-21 17:45:07 +03002546 return 0;
Jan Kiszka33f089c2008-09-26 09:30:49 +02002547
Gleb Natapovc4282df2009-04-21 17:45:07 +03002548 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2549 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS |
2550 GUEST_INTR_STATE_NMI));
Jan Kiszka33f089c2008-09-26 09:30:49 +02002551}
2552
Gleb Natapov78646122009-03-23 12:12:11 +02002553static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2554{
Gleb Natapovc4282df2009-04-21 17:45:07 +03002555 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2556 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2557 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
Gleb Natapov78646122009-03-23 12:12:11 +02002558}
2559
Izik Eiduscbc94022007-10-25 00:29:55 +02002560static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2561{
2562 int ret;
2563 struct kvm_userspace_memory_region tss_mem = {
Sheng Yang6fe63972008-10-16 17:30:58 +08002564 .slot = TSS_PRIVATE_MEMSLOT,
Izik Eiduscbc94022007-10-25 00:29:55 +02002565 .guest_phys_addr = addr,
2566 .memory_size = PAGE_SIZE * 3,
2567 .flags = 0,
2568 };
2569
2570 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2571 if (ret)
2572 return ret;
Zhang Xiantaobfc6d222007-12-14 10:20:16 +08002573 kvm->arch.tss_addr = addr;
Izik Eiduscbc94022007-10-25 00:29:55 +02002574 return 0;
2575}
2576
Avi Kivity6aa8b732006-12-10 02:21:36 -08002577static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2578 int vec, u32 err_code)
2579{
Nitin A Kambleb3f37702007-05-17 15:50:34 +03002580 /*
2581 * Instruction with address size override prefix opcode 0x67
2582 * Cause the #SS fault with 0 error code in VM86 mode.
2583 */
2584 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
Laurent Vivier34273182007-09-18 11:27:37 +02002585 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002586 return 1;
Jan Kiszka77ab6db2008-07-14 12:28:51 +02002587 /*
2588 * Forward all other exceptions that are valid in real mode.
2589 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2590 * the required debugging infrastructure rework.
2591 */
2592 switch (vec) {
Jan Kiszka77ab6db2008-07-14 12:28:51 +02002593 case DB_VECTOR:
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002594 if (vcpu->guest_debug &
2595 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2596 return 0;
2597 kvm_queue_exception(vcpu, vec);
2598 return 1;
Jan Kiszka77ab6db2008-07-14 12:28:51 +02002599 case BP_VECTOR:
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002600 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2601 return 0;
2602 /* fall through */
2603 case DE_VECTOR:
Jan Kiszka77ab6db2008-07-14 12:28:51 +02002604 case OF_VECTOR:
2605 case BR_VECTOR:
2606 case UD_VECTOR:
2607 case DF_VECTOR:
2608 case SS_VECTOR:
2609 case GP_VECTOR:
2610 case MF_VECTOR:
2611 kvm_queue_exception(vcpu, vec);
2612 return 1;
2613 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002614 return 0;
2615}
2616
Andi Kleena0861c02009-06-08 17:37:09 +08002617/*
2618 * Trigger machine check on the host. We assume all the MSRs are already set up
2619 * by the CPU and that we still run on the same CPU as the MCE occurred on.
2620 * We pass a fake environment to the machine check handler because we want
2621 * the guest to be always treated like user space, no matter what context
2622 * it used internally.
2623 */
2624static void kvm_machine_check(void)
2625{
2626#if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
2627 struct pt_regs regs = {
2628 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
2629 .flags = X86_EFLAGS_IF,
2630 };
2631
2632 do_machine_check(&regs, 0);
2633#endif
2634}
2635
2636static int handle_machine_check(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2637{
2638 /* already handled by vcpu_run */
2639 return 1;
2640}
2641
Avi Kivity6aa8b732006-12-10 02:21:36 -08002642static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2643{
Avi Kivity1155f762007-11-22 11:30:47 +02002644 struct vcpu_vmx *vmx = to_vmx(vcpu);
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002645 u32 intr_info, ex_no, error_code;
Jan Kiszka42dbaa52008-12-15 13:52:10 +01002646 unsigned long cr2, rip, dr6;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002647 u32 vect_info;
2648 enum emulation_result er;
2649
Avi Kivity1155f762007-11-22 11:30:47 +02002650 vect_info = vmx->idt_vectoring_info;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002651 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2652
Andi Kleena0861c02009-06-08 17:37:09 +08002653 if (is_machine_check(intr_info))
2654 return handle_machine_check(vcpu, kvm_run);
2655
Avi Kivity6aa8b732006-12-10 02:21:36 -08002656 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
Mike Dayd77c26f2007-10-08 09:02:08 -04002657 !is_page_fault(intr_info))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002658 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
Harvey Harrisonb8688d52008-03-03 12:59:56 -08002659 "intr info 0x%x\n", __func__, vect_info, intr_info);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002660
Jan Kiszkae4a41882008-09-26 09:30:46 +02002661 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
Avi Kivity1b6269d2007-10-09 12:12:19 +02002662 return 1; /* already handled by vmx_vcpu_run() */
Anthony Liguori2ab455c2007-04-27 09:29:49 +03002663
2664 if (is_no_device(intr_info)) {
Avi Kivity5fd86fc2007-05-02 20:40:00 +03002665 vmx_fpu_activate(vcpu);
Anthony Liguori2ab455c2007-04-27 09:29:49 +03002666 return 1;
2667 }
2668
Anthony Liguori7aa81cc2007-09-17 14:57:50 -05002669 if (is_invalid_opcode(intr_info)) {
Sheng Yang571008d2008-01-02 14:49:22 +08002670 er = emulate_instruction(vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
Anthony Liguori7aa81cc2007-09-17 14:57:50 -05002671 if (er != EMULATE_DONE)
Avi Kivity7ee5d9402007-11-25 15:22:50 +02002672 kvm_queue_exception(vcpu, UD_VECTOR);
Anthony Liguori7aa81cc2007-09-17 14:57:50 -05002673 return 1;
2674 }
2675
Avi Kivity6aa8b732006-12-10 02:21:36 -08002676 error_code = 0;
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002677 rip = kvm_rip_read(vcpu);
Ryan Harper2e113842008-02-11 10:26:38 -06002678 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002679 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2680 if (is_page_fault(intr_info)) {
Sheng Yang14394422008-04-28 12:24:45 +08002681 /* EPT won't cause page fault directly */
Avi Kivity089d0342009-03-23 18:26:32 +02002682 if (enable_ept)
Sheng Yang14394422008-04-28 12:24:45 +08002683 BUG();
Avi Kivity6aa8b732006-12-10 02:21:36 -08002684 cr2 = vmcs_readl(EXIT_QUALIFICATION);
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04002685 KVMTRACE_3D(PAGE_FAULT, vcpu, error_code, (u32)cr2,
2686 (u32)((u64)cr2 >> 32), handler);
Gleb Natapov3298b752009-05-11 13:35:46 +03002687 if (kvm_event_needs_reinjection(vcpu))
Avi Kivity577bdc42008-07-19 08:57:05 +03002688 kvm_mmu_unprotect_page_virt(vcpu, cr2);
Avi Kivity30677142007-10-28 18:48:59 +02002689 return kvm_mmu_page_fault(vcpu, cr2, error_code);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002690 }
2691
Nitin A Kamble56b237e2009-06-04 15:04:08 -07002692 if (vcpu->arch.rmode.vm86_active &&
Avi Kivity6aa8b732006-12-10 02:21:36 -08002693 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
Avi Kivity72d6e5a2007-06-05 16:15:51 +03002694 error_code)) {
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002695 if (vcpu->arch.halt_request) {
2696 vcpu->arch.halt_request = 0;
Avi Kivity72d6e5a2007-06-05 16:15:51 +03002697 return kvm_emulate_halt(vcpu);
2698 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002699 return 1;
Avi Kivity72d6e5a2007-06-05 16:15:51 +03002700 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002701
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002702 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
Jan Kiszka42dbaa52008-12-15 13:52:10 +01002703 switch (ex_no) {
2704 case DB_VECTOR:
2705 dr6 = vmcs_readl(EXIT_QUALIFICATION);
2706 if (!(vcpu->guest_debug &
2707 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
2708 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
2709 kvm_queue_exception(vcpu, DB_VECTOR);
2710 return 1;
2711 }
2712 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
2713 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
2714 /* fall through */
2715 case BP_VECTOR:
Avi Kivity6aa8b732006-12-10 02:21:36 -08002716 kvm_run->exit_reason = KVM_EXIT_DEBUG;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002717 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
2718 kvm_run->debug.arch.exception = ex_no;
Jan Kiszka42dbaa52008-12-15 13:52:10 +01002719 break;
2720 default:
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002721 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
2722 kvm_run->ex.exception = ex_no;
2723 kvm_run->ex.error_code = error_code;
Jan Kiszka42dbaa52008-12-15 13:52:10 +01002724 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002725 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002726 return 0;
2727}
2728
2729static int handle_external_interrupt(struct kvm_vcpu *vcpu,
2730 struct kvm_run *kvm_run)
2731{
Avi Kivity1165f5f2007-04-19 17:27:43 +03002732 ++vcpu->stat.irq_exits;
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04002733 KVMTRACE_1D(INTR, vcpu, vmcs_read32(VM_EXIT_INTR_INFO), handler);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002734 return 1;
2735}
2736
Avi Kivity988ad742007-02-12 00:54:36 -08002737static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2738{
2739 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2740 return 0;
2741}
Avi Kivity6aa8b732006-12-10 02:21:36 -08002742
Avi Kivity6aa8b732006-12-10 02:21:36 -08002743static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2744{
He, Qingbfdaab02007-09-12 14:18:28 +08002745 unsigned long exit_qualification;
Jan Kiszka34c33d12009-02-08 13:28:15 +01002746 int size, in, string;
Avi Kivity039576c2007-03-20 12:46:50 +02002747 unsigned port;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002748
Avi Kivity1165f5f2007-04-19 17:27:43 +03002749 ++vcpu->stat.io_exits;
He, Qingbfdaab02007-09-12 14:18:28 +08002750 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
Avi Kivity039576c2007-03-20 12:46:50 +02002751 string = (exit_qualification & 16) != 0;
Laurent Viviere70669a2007-08-05 10:36:40 +03002752
2753 if (string) {
Laurent Vivier34273182007-09-18 11:27:37 +02002754 if (emulate_instruction(vcpu,
2755 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
Laurent Viviere70669a2007-08-05 10:36:40 +03002756 return 0;
2757 return 1;
2758 }
2759
2760 size = (exit_qualification & 7) + 1;
2761 in = (exit_qualification & 8) != 0;
Avi Kivity039576c2007-03-20 12:46:50 +02002762 port = exit_qualification >> 16;
Laurent Viviere70669a2007-08-05 10:36:40 +03002763
Guillaume Thouvenine93f36b2008-10-28 10:51:30 +01002764 skip_emulated_instruction(vcpu);
Laurent Vivier3090dd72007-08-05 10:43:32 +03002765 return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002766}
2767
Ingo Molnar102d8322007-02-19 14:37:47 +02002768static void
2769vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2770{
2771 /*
2772 * Patch in the VMCALL instruction:
2773 */
2774 hypercall[0] = 0x0f;
2775 hypercall[1] = 0x01;
2776 hypercall[2] = 0xc1;
Ingo Molnar102d8322007-02-19 14:37:47 +02002777}
2778
Avi Kivity6aa8b732006-12-10 02:21:36 -08002779static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2780{
He, Qingbfdaab02007-09-12 14:18:28 +08002781 unsigned long exit_qualification;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002782 int cr;
2783 int reg;
2784
He, Qingbfdaab02007-09-12 14:18:28 +08002785 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002786 cr = exit_qualification & 15;
2787 reg = (exit_qualification >> 8) & 15;
2788 switch ((exit_qualification >> 4) & 3) {
2789 case 0: /* mov to cr */
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002790 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr,
2791 (u32)kvm_register_read(vcpu, reg),
2792 (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2793 handler);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002794 switch (cr) {
2795 case 0:
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002796 kvm_set_cr0(vcpu, kvm_register_read(vcpu, reg));
Avi Kivity6aa8b732006-12-10 02:21:36 -08002797 skip_emulated_instruction(vcpu);
2798 return 1;
2799 case 3:
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002800 kvm_set_cr3(vcpu, kvm_register_read(vcpu, reg));
Avi Kivity6aa8b732006-12-10 02:21:36 -08002801 skip_emulated_instruction(vcpu);
2802 return 1;
2803 case 4:
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002804 kvm_set_cr4(vcpu, kvm_register_read(vcpu, reg));
Avi Kivity6aa8b732006-12-10 02:21:36 -08002805 skip_emulated_instruction(vcpu);
2806 return 1;
Gleb Natapov0a5fff192009-04-21 17:45:06 +03002807 case 8: {
2808 u8 cr8_prev = kvm_get_cr8(vcpu);
2809 u8 cr8 = kvm_register_read(vcpu, reg);
2810 kvm_set_cr8(vcpu, cr8);
2811 skip_emulated_instruction(vcpu);
2812 if (irqchip_in_kernel(vcpu->kvm))
2813 return 1;
2814 if (cr8_prev <= cr8)
2815 return 1;
2816 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2817 return 0;
2818 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002819 };
2820 break;
Anthony Liguori25c4c272007-04-27 09:29:21 +03002821 case 2: /* clts */
Avi Kivity5fd86fc2007-05-02 20:40:00 +03002822 vmx_fpu_deactivate(vcpu);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002823 vcpu->arch.cr0 &= ~X86_CR0_TS;
2824 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
Avi Kivity5fd86fc2007-05-02 20:40:00 +03002825 vmx_fpu_activate(vcpu);
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04002826 KVMTRACE_0D(CLTS, vcpu, handler);
Anthony Liguori25c4c272007-04-27 09:29:21 +03002827 skip_emulated_instruction(vcpu);
2828 return 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002829 case 1: /*mov from cr*/
2830 switch (cr) {
2831 case 3:
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002832 kvm_register_write(vcpu, reg, vcpu->arch.cr3);
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04002833 KVMTRACE_3D(CR_READ, vcpu, (u32)cr,
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002834 (u32)kvm_register_read(vcpu, reg),
2835 (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04002836 handler);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002837 skip_emulated_instruction(vcpu);
2838 return 1;
2839 case 8:
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002840 kvm_register_write(vcpu, reg, kvm_get_cr8(vcpu));
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04002841 KVMTRACE_2D(CR_READ, vcpu, (u32)cr,
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002842 (u32)kvm_register_read(vcpu, reg), handler);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002843 skip_emulated_instruction(vcpu);
2844 return 1;
2845 }
2846 break;
2847 case 3: /* lmsw */
Avi Kivity2d3ad1f2008-02-24 11:20:43 +02002848 kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002849
2850 skip_emulated_instruction(vcpu);
2851 return 1;
2852 default:
2853 break;
2854 }
2855 kvm_run->exit_reason = 0;
Rusty Russellf0242472007-08-01 10:48:02 +10002856 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
Avi Kivity6aa8b732006-12-10 02:21:36 -08002857 (int)(exit_qualification >> 4) & 3, cr);
2858 return 0;
2859}
2860
2861static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2862{
He, Qingbfdaab02007-09-12 14:18:28 +08002863 unsigned long exit_qualification;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002864 unsigned long val;
2865 int dr, reg;
2866
Jan Kiszka42dbaa52008-12-15 13:52:10 +01002867 dr = vmcs_readl(GUEST_DR7);
2868 if (dr & DR7_GD) {
2869 /*
2870 * As the vm-exit takes precedence over the debug trap, we
2871 * need to emulate the latter, either for the host or the
2872 * guest debugging itself.
2873 */
2874 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
2875 kvm_run->debug.arch.dr6 = vcpu->arch.dr6;
2876 kvm_run->debug.arch.dr7 = dr;
2877 kvm_run->debug.arch.pc =
2878 vmcs_readl(GUEST_CS_BASE) +
2879 vmcs_readl(GUEST_RIP);
2880 kvm_run->debug.arch.exception = DB_VECTOR;
2881 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2882 return 0;
2883 } else {
2884 vcpu->arch.dr7 &= ~DR7_GD;
2885 vcpu->arch.dr6 |= DR6_BD;
2886 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2887 kvm_queue_exception(vcpu, DB_VECTOR);
2888 return 1;
2889 }
2890 }
2891
He, Qingbfdaab02007-09-12 14:18:28 +08002892 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
Jan Kiszka42dbaa52008-12-15 13:52:10 +01002893 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
2894 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
2895 if (exit_qualification & TYPE_MOV_FROM_DR) {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002896 switch (dr) {
Jan Kiszka42dbaa52008-12-15 13:52:10 +01002897 case 0 ... 3:
2898 val = vcpu->arch.db[dr];
2899 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002900 case 6:
Jan Kiszka42dbaa52008-12-15 13:52:10 +01002901 val = vcpu->arch.dr6;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002902 break;
2903 case 7:
Jan Kiszka42dbaa52008-12-15 13:52:10 +01002904 val = vcpu->arch.dr7;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002905 break;
2906 default:
2907 val = 0;
2908 }
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03002909 kvm_register_write(vcpu, reg, val);
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04002910 KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002911 } else {
Jan Kiszka42dbaa52008-12-15 13:52:10 +01002912 val = vcpu->arch.regs[reg];
2913 switch (dr) {
2914 case 0 ... 3:
2915 vcpu->arch.db[dr] = val;
2916 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
2917 vcpu->arch.eff_db[dr] = val;
2918 break;
2919 case 4 ... 5:
2920 if (vcpu->arch.cr4 & X86_CR4_DE)
2921 kvm_queue_exception(vcpu, UD_VECTOR);
2922 break;
2923 case 6:
2924 if (val & 0xffffffff00000000ULL) {
2925 kvm_queue_exception(vcpu, GP_VECTOR);
2926 break;
2927 }
2928 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
2929 break;
2930 case 7:
2931 if (val & 0xffffffff00000000ULL) {
2932 kvm_queue_exception(vcpu, GP_VECTOR);
2933 break;
2934 }
2935 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
2936 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
2937 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2938 vcpu->arch.switch_db_regs =
2939 (val & DR7_BP_EN_MASK);
2940 }
2941 break;
2942 }
2943 KVMTRACE_2D(DR_WRITE, vcpu, (u32)dr, (u32)val, handler);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002944 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002945 skip_emulated_instruction(vcpu);
2946 return 1;
2947}
2948
2949static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2950{
Avi Kivity06465c52007-02-28 20:46:53 +02002951 kvm_emulate_cpuid(vcpu);
2952 return 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002953}
2954
2955static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2956{
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002957 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
Avi Kivity6aa8b732006-12-10 02:21:36 -08002958 u64 data;
2959
2960 if (vmx_get_msr(vcpu, ecx, &data)) {
Avi Kivityc1a5d4f2007-11-25 14:12:03 +02002961 kvm_inject_gp(vcpu, 0);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002962 return 1;
2963 }
2964
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04002965 KVMTRACE_3D(MSR_READ, vcpu, ecx, (u32)data, (u32)(data >> 32),
2966 handler);
2967
Avi Kivity6aa8b732006-12-10 02:21:36 -08002968 /* FIXME: handling of bits 32:63 of rax, rdx */
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002969 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
2970 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002971 skip_emulated_instruction(vcpu);
2972 return 1;
2973}
2974
2975static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2976{
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002977 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2978 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
2979 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002980
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04002981 KVMTRACE_3D(MSR_WRITE, vcpu, ecx, (u32)data, (u32)(data >> 32),
2982 handler);
2983
Avi Kivity6aa8b732006-12-10 02:21:36 -08002984 if (vmx_set_msr(vcpu, ecx, data) != 0) {
Avi Kivityc1a5d4f2007-11-25 14:12:03 +02002985 kvm_inject_gp(vcpu, 0);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002986 return 1;
2987 }
2988
2989 skip_emulated_instruction(vcpu);
2990 return 1;
2991}
2992
Yang, Sheng6e5d8652007-09-12 18:03:11 +08002993static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu,
2994 struct kvm_run *kvm_run)
2995{
2996 return 1;
2997}
2998
Avi Kivity6aa8b732006-12-10 02:21:36 -08002999static int handle_interrupt_window(struct kvm_vcpu *vcpu,
3000 struct kvm_run *kvm_run)
3001{
Eddie Dong85f455f2007-07-06 12:20:49 +03003002 u32 cpu_based_vm_exec_control;
3003
3004 /* clear pending irq */
3005 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3006 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3007 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04003008
3009 KVMTRACE_0D(PEND_INTR, vcpu, handler);
Jan Kiszkaa26bf122008-09-26 09:30:45 +02003010 ++vcpu->stat.irq_window_exits;
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04003011
Dor Laorc1150d82007-01-05 16:36:24 -08003012 /*
3013 * If the user space waits to inject interrupts, exit as soon as
3014 * possible
3015 */
Gleb Natapov80618232009-04-21 17:44:56 +03003016 if (!irqchip_in_kernel(vcpu->kvm) &&
3017 kvm_run->request_interrupt_window &&
3018 !kvm_cpu_has_interrupt(vcpu)) {
Dor Laorc1150d82007-01-05 16:36:24 -08003019 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
Dor Laorc1150d82007-01-05 16:36:24 -08003020 return 0;
3021 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08003022 return 1;
3023}
3024
3025static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3026{
3027 skip_emulated_instruction(vcpu);
Avi Kivityd3bef152007-06-05 15:53:05 +03003028 return kvm_emulate_halt(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003029}
3030
Ingo Molnarc21415e2007-02-19 14:37:47 +02003031static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3032{
Dor Laor510043d2007-02-19 18:25:43 +02003033 skip_emulated_instruction(vcpu);
Anthony Liguori7aa81cc2007-09-17 14:57:50 -05003034 kvm_emulate_hypercall(vcpu);
3035 return 1;
Ingo Molnarc21415e2007-02-19 14:37:47 +02003036}
3037
Avi Kivitye3c7cb62009-06-16 14:19:52 +03003038static int handle_vmx_insn(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3039{
3040 kvm_queue_exception(vcpu, UD_VECTOR);
3041 return 1;
3042}
3043
Marcelo Tosattia7052892008-09-23 13:18:35 -03003044static int handle_invlpg(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3045{
Sheng Yangf9c617f2009-03-25 10:08:52 +08003046 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
Marcelo Tosattia7052892008-09-23 13:18:35 -03003047
3048 kvm_mmu_invlpg(vcpu, exit_qualification);
3049 skip_emulated_instruction(vcpu);
3050 return 1;
3051}
3052
Eddie Donge5edaa02007-11-11 12:28:35 +02003053static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3054{
3055 skip_emulated_instruction(vcpu);
3056 /* TODO: Add support for VT-d/pass-through device */
3057 return 1;
3058}
3059
Sheng Yangf78e0e22007-10-29 09:40:42 +08003060static int handle_apic_access(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3061{
Sheng Yangf9c617f2009-03-25 10:08:52 +08003062 unsigned long exit_qualification;
Sheng Yangf78e0e22007-10-29 09:40:42 +08003063 enum emulation_result er;
3064 unsigned long offset;
3065
Sheng Yangf9c617f2009-03-25 10:08:52 +08003066 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
Sheng Yangf78e0e22007-10-29 09:40:42 +08003067 offset = exit_qualification & 0xffful;
3068
3069 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
3070
3071 if (er != EMULATE_DONE) {
3072 printk(KERN_ERR
3073 "Fail to handle apic access vmexit! Offset is 0x%lx\n",
3074 offset);
3075 return -ENOTSUPP;
3076 }
3077 return 1;
3078}
3079
Izik Eidus37817f22008-03-24 23:14:53 +02003080static int handle_task_switch(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3081{
Jan Kiszka60637aa2008-09-26 09:30:47 +02003082 struct vcpu_vmx *vmx = to_vmx(vcpu);
Izik Eidus37817f22008-03-24 23:14:53 +02003083 unsigned long exit_qualification;
3084 u16 tss_selector;
Gleb Natapov64a7ec02009-03-30 16:03:29 +03003085 int reason, type, idt_v;
3086
3087 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3088 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
Izik Eidus37817f22008-03-24 23:14:53 +02003089
3090 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3091
3092 reason = (u32)exit_qualification >> 30;
Gleb Natapov64a7ec02009-03-30 16:03:29 +03003093 if (reason == TASK_SWITCH_GATE && idt_v) {
3094 switch (type) {
3095 case INTR_TYPE_NMI_INTR:
3096 vcpu->arch.nmi_injected = false;
3097 if (cpu_has_virtual_nmis())
3098 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3099 GUEST_INTR_STATE_NMI);
3100 break;
3101 case INTR_TYPE_EXT_INTR:
Gleb Natapov66fd3f72009-05-11 13:35:50 +03003102 case INTR_TYPE_SOFT_INTR:
Gleb Natapov64a7ec02009-03-30 16:03:29 +03003103 kvm_clear_interrupt_queue(vcpu);
3104 break;
3105 case INTR_TYPE_HARD_EXCEPTION:
3106 case INTR_TYPE_SOFT_EXCEPTION:
3107 kvm_clear_exception_queue(vcpu);
3108 break;
3109 default:
3110 break;
3111 }
Jan Kiszka60637aa2008-09-26 09:30:47 +02003112 }
Izik Eidus37817f22008-03-24 23:14:53 +02003113 tss_selector = exit_qualification;
3114
Gleb Natapov64a7ec02009-03-30 16:03:29 +03003115 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3116 type != INTR_TYPE_EXT_INTR &&
3117 type != INTR_TYPE_NMI_INTR))
3118 skip_emulated_instruction(vcpu);
3119
Jan Kiszka42dbaa52008-12-15 13:52:10 +01003120 if (!kvm_task_switch(vcpu, tss_selector, reason))
3121 return 0;
3122
3123 /* clear all local breakpoint enable flags */
3124 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3125
3126 /*
3127 * TODO: What about debug traps on tss switch?
3128 * Are we supposed to inject them and update dr6?
3129 */
3130
3131 return 1;
Izik Eidus37817f22008-03-24 23:14:53 +02003132}
3133
Sheng Yang14394422008-04-28 12:24:45 +08003134static int handle_ept_violation(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3135{
Sheng Yangf9c617f2009-03-25 10:08:52 +08003136 unsigned long exit_qualification;
Sheng Yang14394422008-04-28 12:24:45 +08003137 gpa_t gpa;
Sheng Yang14394422008-04-28 12:24:45 +08003138 int gla_validity;
Sheng Yang14394422008-04-28 12:24:45 +08003139
Sheng Yangf9c617f2009-03-25 10:08:52 +08003140 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
Sheng Yang14394422008-04-28 12:24:45 +08003141
3142 if (exit_qualification & (1 << 6)) {
3143 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3144 return -ENOTSUPP;
3145 }
3146
3147 gla_validity = (exit_qualification >> 7) & 0x3;
3148 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3149 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3150 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3151 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
Sheng Yangf9c617f2009-03-25 10:08:52 +08003152 vmcs_readl(GUEST_LINEAR_ADDRESS));
Sheng Yang14394422008-04-28 12:24:45 +08003153 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3154 (long unsigned int)exit_qualification);
3155 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
Avi Kivity596ae892009-06-03 14:12:10 +03003156 kvm_run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
3157 return 0;
Sheng Yang14394422008-04-28 12:24:45 +08003158 }
3159
3160 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
Sheng Yang49cd7d22009-02-11 13:50:40 +08003161 return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
Sheng Yang14394422008-04-28 12:24:45 +08003162}
3163
Sheng Yangf08864b2008-05-15 18:23:25 +08003164static int handle_nmi_window(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3165{
3166 u32 cpu_based_vm_exec_control;
3167
3168 /* clear pending NMI */
3169 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3170 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3171 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3172 ++vcpu->stat.nmi_window_exits;
3173
3174 return 1;
3175}
3176
Mohammed Gamalea953ef2008-08-17 16:47:05 +03003177static void handle_invalid_guest_state(struct kvm_vcpu *vcpu,
3178 struct kvm_run *kvm_run)
3179{
Avi Kivity8b3079a2009-01-05 12:10:54 +02003180 struct vcpu_vmx *vmx = to_vmx(vcpu);
3181 enum emulation_result err = EMULATE_DONE;
Mohammed Gamalea953ef2008-08-17 16:47:05 +03003182
Mohammed Gamalea953ef2008-08-17 16:47:05 +03003183 local_irq_enable();
Jan Kiszka34f0c1a2009-07-22 23:53:26 +02003184 preempt_enable();
Mohammed Gamalea953ef2008-08-17 16:47:05 +03003185
3186 while (!guest_state_valid(vcpu)) {
3187 err = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
3188
Guillaume Thouvenin1d5a4d92008-10-29 09:39:42 +01003189 if (err == EMULATE_DO_MMIO)
3190 break;
3191
3192 if (err != EMULATE_DONE) {
3193 kvm_report_emulation_failure(vcpu, "emulation failure");
Jan Kiszka263799a2009-07-21 10:43:07 +02003194 break;
Mohammed Gamalea953ef2008-08-17 16:47:05 +03003195 }
3196
3197 if (signal_pending(current))
3198 break;
3199 if (need_resched())
3200 schedule();
3201 }
3202
Mohammed Gamalea953ef2008-08-17 16:47:05 +03003203 preempt_disable();
Jan Kiszka34f0c1a2009-07-22 23:53:26 +02003204 local_irq_disable();
Avi Kivity8b3079a2009-01-05 12:10:54 +02003205
3206 vmx->invalid_state_emulation_result = err;
Mohammed Gamalea953ef2008-08-17 16:47:05 +03003207}
3208
Avi Kivity6aa8b732006-12-10 02:21:36 -08003209/*
3210 * The exit handlers return 1 if the exit was handled fully and guest execution
3211 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
3212 * to be done to userspace and return 0.
3213 */
3214static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
3215 struct kvm_run *kvm_run) = {
3216 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
3217 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
Avi Kivity988ad742007-02-12 00:54:36 -08003218 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
Sheng Yangf08864b2008-05-15 18:23:25 +08003219 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003220 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003221 [EXIT_REASON_CR_ACCESS] = handle_cr,
3222 [EXIT_REASON_DR_ACCESS] = handle_dr,
3223 [EXIT_REASON_CPUID] = handle_cpuid,
3224 [EXIT_REASON_MSR_READ] = handle_rdmsr,
3225 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
3226 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
3227 [EXIT_REASON_HLT] = handle_halt,
Marcelo Tosattia7052892008-09-23 13:18:35 -03003228 [EXIT_REASON_INVLPG] = handle_invlpg,
Ingo Molnarc21415e2007-02-19 14:37:47 +02003229 [EXIT_REASON_VMCALL] = handle_vmcall,
Avi Kivitye3c7cb62009-06-16 14:19:52 +03003230 [EXIT_REASON_VMCLEAR] = handle_vmx_insn,
3231 [EXIT_REASON_VMLAUNCH] = handle_vmx_insn,
3232 [EXIT_REASON_VMPTRLD] = handle_vmx_insn,
3233 [EXIT_REASON_VMPTRST] = handle_vmx_insn,
3234 [EXIT_REASON_VMREAD] = handle_vmx_insn,
3235 [EXIT_REASON_VMRESUME] = handle_vmx_insn,
3236 [EXIT_REASON_VMWRITE] = handle_vmx_insn,
3237 [EXIT_REASON_VMOFF] = handle_vmx_insn,
3238 [EXIT_REASON_VMON] = handle_vmx_insn,
Sheng Yangf78e0e22007-10-29 09:40:42 +08003239 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
3240 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
Eddie Donge5edaa02007-11-11 12:28:35 +02003241 [EXIT_REASON_WBINVD] = handle_wbinvd,
Izik Eidus37817f22008-03-24 23:14:53 +02003242 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
Sheng Yang14394422008-04-28 12:24:45 +08003243 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
Andi Kleena0861c02009-06-08 17:37:09 +08003244 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003245};
3246
3247static const int kvm_vmx_max_exit_handlers =
Robert P. J. Day50a34852007-06-03 13:35:29 -04003248 ARRAY_SIZE(kvm_vmx_exit_handlers);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003249
3250/*
3251 * The guest has exited. See if we can fix it or if we need userspace
3252 * assistance.
3253 */
Avi Kivity6062d012009-03-23 17:35:17 +02003254static int vmx_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003255{
Avi Kivity29bd8a72007-09-10 17:27:03 +03003256 struct vcpu_vmx *vmx = to_vmx(vcpu);
Andi Kleena0861c02009-06-08 17:37:09 +08003257 u32 exit_reason = vmx->exit_reason;
Avi Kivity1155f762007-11-22 11:30:47 +02003258 u32 vectoring_info = vmx->idt_vectoring_info;
Avi Kivity29bd8a72007-09-10 17:27:03 +03003259
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03003260 KVMTRACE_3D(VMEXIT, vcpu, exit_reason, (u32)kvm_rip_read(vcpu),
3261 (u32)((u64)kvm_rip_read(vcpu) >> 32), entryexit);
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04003262
Guillaume Thouvenin1d5a4d92008-10-29 09:39:42 +01003263 /* If we need to emulate an MMIO from handle_invalid_guest_state
3264 * we just return 0 */
Avi Kivity10f32d82009-01-05 00:53:19 +02003265 if (vmx->emulation_required && emulate_invalid_guest_state) {
3266 if (guest_state_valid(vcpu))
3267 vmx->emulation_required = 0;
Avi Kivity8b3079a2009-01-05 12:10:54 +02003268 return vmx->invalid_state_emulation_result != EMULATE_DO_MMIO;
Avi Kivity10f32d82009-01-05 00:53:19 +02003269 }
Guillaume Thouvenin1d5a4d92008-10-29 09:39:42 +01003270
Sheng Yang14394422008-04-28 12:24:45 +08003271 /* Access CR3 don't cause VMExit in paging mode, so we need
3272 * to sync with guest real CR3. */
Avi Kivity6de4f3a2009-05-31 22:58:47 +03003273 if (enable_ept && is_paging(vcpu))
Sheng Yang14394422008-04-28 12:24:45 +08003274 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
Sheng Yang14394422008-04-28 12:24:45 +08003275
Avi Kivity29bd8a72007-09-10 17:27:03 +03003276 if (unlikely(vmx->fail)) {
3277 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3278 kvm_run->fail_entry.hardware_entry_failure_reason
3279 = vmcs_read32(VM_INSTRUCTION_ERROR);
3280 return 0;
3281 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08003282
Mike Dayd77c26f2007-10-08 09:02:08 -04003283 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
Sheng Yang14394422008-04-28 12:24:45 +08003284 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
Jan Kiszka60637aa2008-09-26 09:30:47 +02003285 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3286 exit_reason != EXIT_REASON_TASK_SWITCH))
3287 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3288 "(0x%x) and exit reason is 0x%x\n",
3289 __func__, vectoring_info, exit_reason);
Jan Kiszka3b86cd92008-09-26 09:30:57 +02003290
3291 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
Gleb Natapovc4282df2009-04-21 17:45:07 +03003292 if (vmx_interrupt_allowed(vcpu)) {
Jan Kiszka3b86cd92008-09-26 09:30:57 +02003293 vmx->soft_vnmi_blocked = 0;
Jan Kiszka3b86cd92008-09-26 09:30:57 +02003294 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
Jan Kiszka45312202008-12-11 16:54:54 +01003295 vcpu->arch.nmi_pending) {
Jan Kiszka3b86cd92008-09-26 09:30:57 +02003296 /*
3297 * This CPU don't support us in finding the end of an
3298 * NMI-blocked window if the guest runs with IRQs
3299 * disabled. So we pull the trigger after 1 s of
3300 * futile waiting, but inform the user about this.
3301 */
3302 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3303 "state on VCPU %d after 1 s timeout\n",
3304 __func__, vcpu->vcpu_id);
3305 vmx->soft_vnmi_blocked = 0;
Jan Kiszka3b86cd92008-09-26 09:30:57 +02003306 }
Jan Kiszka3b86cd92008-09-26 09:30:57 +02003307 }
3308
Avi Kivity6aa8b732006-12-10 02:21:36 -08003309 if (exit_reason < kvm_vmx_max_exit_handlers
3310 && kvm_vmx_exit_handlers[exit_reason])
3311 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
3312 else {
3313 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3314 kvm_run->hw.hardware_exit_reason = exit_reason;
3315 }
3316 return 0;
3317}
3318
Gleb Natapov95ba8273132009-04-21 17:45:08 +03003319static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
Yang, Sheng6e5d8652007-09-12 18:03:11 +08003320{
Gleb Natapov95ba8273132009-04-21 17:45:08 +03003321 if (irr == -1 || tpr < irr) {
Yang, Sheng6e5d8652007-09-12 18:03:11 +08003322 vmcs_write32(TPR_THRESHOLD, 0);
3323 return;
3324 }
3325
Gleb Natapov95ba8273132009-04-21 17:45:08 +03003326 vmcs_write32(TPR_THRESHOLD, irr);
Yang, Sheng6e5d8652007-09-12 18:03:11 +08003327}
3328
Avi Kivitycf393f72008-07-01 16:20:21 +03003329static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3330{
3331 u32 exit_intr_info;
Gleb Natapov7b4a25c2009-03-30 16:03:08 +03003332 u32 idt_vectoring_info = vmx->idt_vectoring_info;
Avi Kivitycf393f72008-07-01 16:20:21 +03003333 bool unblock_nmi;
3334 u8 vector;
Avi Kivity668f6122008-07-02 09:28:55 +03003335 int type;
3336 bool idtv_info_valid;
Avi Kivitycf393f72008-07-01 16:20:21 +03003337
3338 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
Gleb Natapov20f65982009-05-11 13:35:55 +03003339
Andi Kleena0861c02009-06-08 17:37:09 +08003340 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
3341
3342 /* Handle machine checks before interrupts are enabled */
3343 if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3344 || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3345 && is_machine_check(exit_intr_info)))
3346 kvm_machine_check();
3347
Gleb Natapov20f65982009-05-11 13:35:55 +03003348 /* We need to handle NMIs before interrupts are enabled */
3349 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3350 (exit_intr_info & INTR_INFO_VALID_MASK)) {
3351 KVMTRACE_0D(NMI, &vmx->vcpu, handler);
3352 asm("int $2");
3353 }
3354
3355 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3356
Avi Kivitycf393f72008-07-01 16:20:21 +03003357 if (cpu_has_virtual_nmis()) {
3358 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3359 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3360 /*
Gleb Natapov7b4a25c2009-03-30 16:03:08 +03003361 * SDM 3: 27.7.1.2 (September 2008)
Avi Kivitycf393f72008-07-01 16:20:21 +03003362 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3363 * a guest IRET fault.
Gleb Natapov7b4a25c2009-03-30 16:03:08 +03003364 * SDM 3: 23.2.2 (September 2008)
3365 * Bit 12 is undefined in any of the following cases:
3366 * If the VM exit sets the valid bit in the IDT-vectoring
3367 * information field.
3368 * If the VM exit is due to a double fault.
Avi Kivitycf393f72008-07-01 16:20:21 +03003369 */
Gleb Natapov7b4a25c2009-03-30 16:03:08 +03003370 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3371 vector != DF_VECTOR && !idtv_info_valid)
Avi Kivitycf393f72008-07-01 16:20:21 +03003372 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3373 GUEST_INTR_STATE_NMI);
Jan Kiszka3b86cd92008-09-26 09:30:57 +02003374 } else if (unlikely(vmx->soft_vnmi_blocked))
3375 vmx->vnmi_blocked_time +=
3376 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
Avi Kivity668f6122008-07-02 09:28:55 +03003377
Gleb Natapov37b96e92009-03-30 16:03:13 +03003378 vmx->vcpu.arch.nmi_injected = false;
3379 kvm_clear_exception_queue(&vmx->vcpu);
3380 kvm_clear_interrupt_queue(&vmx->vcpu);
3381
3382 if (!idtv_info_valid)
3383 return;
3384
Avi Kivity668f6122008-07-02 09:28:55 +03003385 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3386 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
Gleb Natapov37b96e92009-03-30 16:03:13 +03003387
Gleb Natapov64a7ec02009-03-30 16:03:29 +03003388 switch (type) {
Gleb Natapov37b96e92009-03-30 16:03:13 +03003389 case INTR_TYPE_NMI_INTR:
3390 vmx->vcpu.arch.nmi_injected = true;
Avi Kivity668f6122008-07-02 09:28:55 +03003391 /*
Gleb Natapov7b4a25c2009-03-30 16:03:08 +03003392 * SDM 3: 27.7.1.2 (September 2008)
Gleb Natapov37b96e92009-03-30 16:03:13 +03003393 * Clear bit "block by NMI" before VM entry if a NMI
3394 * delivery faulted.
Avi Kivity668f6122008-07-02 09:28:55 +03003395 */
Gleb Natapov37b96e92009-03-30 16:03:13 +03003396 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3397 GUEST_INTR_STATE_NMI);
3398 break;
Gleb Natapov37b96e92009-03-30 16:03:13 +03003399 case INTR_TYPE_SOFT_EXCEPTION:
Gleb Natapov66fd3f72009-05-11 13:35:50 +03003400 vmx->vcpu.arch.event_exit_inst_len =
3401 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3402 /* fall through */
3403 case INTR_TYPE_HARD_EXCEPTION:
Avi Kivity35920a32008-07-03 14:50:12 +03003404 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
Gleb Natapov37b96e92009-03-30 16:03:13 +03003405 u32 err = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3406 kvm_queue_exception_e(&vmx->vcpu, vector, err);
Avi Kivity35920a32008-07-03 14:50:12 +03003407 } else
3408 kvm_queue_exception(&vmx->vcpu, vector);
Gleb Natapov37b96e92009-03-30 16:03:13 +03003409 break;
Gleb Natapov66fd3f72009-05-11 13:35:50 +03003410 case INTR_TYPE_SOFT_INTR:
3411 vmx->vcpu.arch.event_exit_inst_len =
3412 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3413 /* fall through */
Gleb Natapov37b96e92009-03-30 16:03:13 +03003414 case INTR_TYPE_EXT_INTR:
Gleb Natapov66fd3f72009-05-11 13:35:50 +03003415 kvm_queue_interrupt(&vmx->vcpu, vector,
3416 type == INTR_TYPE_SOFT_INTR);
Gleb Natapov37b96e92009-03-30 16:03:13 +03003417 break;
3418 default:
3419 break;
Avi Kivityf7d92382008-07-03 16:14:28 +03003420 }
Avi Kivitycf393f72008-07-01 16:20:21 +03003421}
3422
Avi Kivity9c8cba32007-11-22 11:42:59 +02003423/*
3424 * Failure to inject an interrupt should give us the information
3425 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
3426 * when fetching the interrupt redirection bitmap in the real-mode
3427 * tss, this doesn't happen. So we do it ourselves.
3428 */
3429static void fixup_rmode_irq(struct vcpu_vmx *vmx)
3430{
3431 vmx->rmode.irq.pending = 0;
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03003432 if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
Avi Kivity9c8cba32007-11-22 11:42:59 +02003433 return;
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03003434 kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
Avi Kivity9c8cba32007-11-22 11:42:59 +02003435 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
3436 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
3437 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
3438 return;
3439 }
3440 vmx->idt_vectoring_info =
3441 VECTORING_INFO_VALID_MASK
3442 | INTR_TYPE_EXT_INTR
3443 | vmx->rmode.irq.vector;
3444}
3445
Avi Kivityc8019492008-07-14 14:44:59 +03003446#ifdef CONFIG_X86_64
3447#define R "r"
3448#define Q "q"
3449#else
3450#define R "e"
3451#define Q "l"
3452#endif
3453
Avi Kivity04d2cc72007-09-10 18:10:54 +03003454static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003455{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04003456 struct vcpu_vmx *vmx = to_vmx(vcpu);
Avi Kivitye6adf282007-04-30 16:07:54 +03003457
Avi Kivity8f5d5492009-05-31 18:41:29 +03003458 if (enable_ept && is_paging(vcpu)) {
3459 vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3460 ept_load_pdptrs(vcpu);
3461 }
Jan Kiszka3b86cd92008-09-26 09:30:57 +02003462 /* Record the guest's net vcpu time for enforced NMI injections. */
3463 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3464 vmx->entry_time = ktime_get();
3465
Mohammed Gamala89a8fb2008-08-17 16:42:16 +03003466 /* Handle invalid guest state instead of entering VMX */
3467 if (vmx->emulation_required && emulate_invalid_guest_state) {
3468 handle_invalid_guest_state(vcpu, kvm_run);
3469 return;
3470 }
3471
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03003472 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3473 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3474 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3475 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3476
Gleb Natapov787ff732009-05-18 11:44:06 +03003477 /* When single-stepping over STI and MOV SS, we must clear the
3478 * corresponding interruptibility bits in the guest state. Otherwise
3479 * vmentry fails as it then expects bit 14 (BS) in pending debug
3480 * exceptions being set, but that's not correct for the guest debugging
3481 * case. */
3482 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3483 vmx_set_interrupt_shadow(vcpu, 0);
3484
Avi Kivitye6adf282007-04-30 16:07:54 +03003485 /*
3486 * Loading guest fpu may have cleared host cr0.ts
3487 */
3488 vmcs_writel(HOST_CR0, read_cr0());
3489
Jan Kiszka42dbaa52008-12-15 13:52:10 +01003490 set_debugreg(vcpu->arch.dr6, 6);
3491
Mike Dayd77c26f2007-10-08 09:02:08 -04003492 asm(
Avi Kivity6aa8b732006-12-10 02:21:36 -08003493 /* Store host registers */
Avi Kivityc8019492008-07-14 14:44:59 +03003494 "push %%"R"dx; push %%"R"bp;"
3495 "push %%"R"cx \n\t"
Avi Kivity313dbd42008-07-17 18:04:30 +03003496 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3497 "je 1f \n\t"
3498 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
Avi Kivity4ecac3f2008-05-13 13:23:38 +03003499 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
Avi Kivity313dbd42008-07-17 18:04:30 +03003500 "1: \n\t"
Avi Kivity6aa8b732006-12-10 02:21:36 -08003501 /* Check if vmlaunch of vmresume is needed */
Avi Kivitye08aa782007-11-15 18:06:18 +02003502 "cmpl $0, %c[launched](%0) \n\t"
Avi Kivity6aa8b732006-12-10 02:21:36 -08003503 /* Load guest registers. Don't clobber flags. */
Avi Kivityc8019492008-07-14 14:44:59 +03003504 "mov %c[cr2](%0), %%"R"ax \n\t"
3505 "mov %%"R"ax, %%cr2 \n\t"
3506 "mov %c[rax](%0), %%"R"ax \n\t"
3507 "mov %c[rbx](%0), %%"R"bx \n\t"
3508 "mov %c[rdx](%0), %%"R"dx \n\t"
3509 "mov %c[rsi](%0), %%"R"si \n\t"
3510 "mov %c[rdi](%0), %%"R"di \n\t"
3511 "mov %c[rbp](%0), %%"R"bp \n\t"
Avi Kivity05b3e0c2006-12-13 00:33:45 -08003512#ifdef CONFIG_X86_64
Avi Kivitye08aa782007-11-15 18:06:18 +02003513 "mov %c[r8](%0), %%r8 \n\t"
3514 "mov %c[r9](%0), %%r9 \n\t"
3515 "mov %c[r10](%0), %%r10 \n\t"
3516 "mov %c[r11](%0), %%r11 \n\t"
3517 "mov %c[r12](%0), %%r12 \n\t"
3518 "mov %c[r13](%0), %%r13 \n\t"
3519 "mov %c[r14](%0), %%r14 \n\t"
3520 "mov %c[r15](%0), %%r15 \n\t"
Avi Kivity6aa8b732006-12-10 02:21:36 -08003521#endif
Avi Kivityc8019492008-07-14 14:44:59 +03003522 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3523
Avi Kivity6aa8b732006-12-10 02:21:36 -08003524 /* Enter guest mode */
Avi Kivitycd2276a2007-05-14 20:41:13 +03003525 "jne .Llaunched \n\t"
Avi Kivity4ecac3f2008-05-13 13:23:38 +03003526 __ex(ASM_VMX_VMLAUNCH) "\n\t"
Avi Kivitycd2276a2007-05-14 20:41:13 +03003527 "jmp .Lkvm_vmx_return \n\t"
Avi Kivity4ecac3f2008-05-13 13:23:38 +03003528 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
Avi Kivitycd2276a2007-05-14 20:41:13 +03003529 ".Lkvm_vmx_return: "
Avi Kivity6aa8b732006-12-10 02:21:36 -08003530 /* Save guest registers, load host registers, keep flags */
Avi Kivityc8019492008-07-14 14:44:59 +03003531 "xchg %0, (%%"R"sp) \n\t"
3532 "mov %%"R"ax, %c[rax](%0) \n\t"
3533 "mov %%"R"bx, %c[rbx](%0) \n\t"
3534 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3535 "mov %%"R"dx, %c[rdx](%0) \n\t"
3536 "mov %%"R"si, %c[rsi](%0) \n\t"
3537 "mov %%"R"di, %c[rdi](%0) \n\t"
3538 "mov %%"R"bp, %c[rbp](%0) \n\t"
Avi Kivity05b3e0c2006-12-13 00:33:45 -08003539#ifdef CONFIG_X86_64
Avi Kivitye08aa782007-11-15 18:06:18 +02003540 "mov %%r8, %c[r8](%0) \n\t"
3541 "mov %%r9, %c[r9](%0) \n\t"
3542 "mov %%r10, %c[r10](%0) \n\t"
3543 "mov %%r11, %c[r11](%0) \n\t"
3544 "mov %%r12, %c[r12](%0) \n\t"
3545 "mov %%r13, %c[r13](%0) \n\t"
3546 "mov %%r14, %c[r14](%0) \n\t"
3547 "mov %%r15, %c[r15](%0) \n\t"
Avi Kivity6aa8b732006-12-10 02:21:36 -08003548#endif
Avi Kivityc8019492008-07-14 14:44:59 +03003549 "mov %%cr2, %%"R"ax \n\t"
3550 "mov %%"R"ax, %c[cr2](%0) \n\t"
3551
3552 "pop %%"R"bp; pop %%"R"bp; pop %%"R"dx \n\t"
Avi Kivitye08aa782007-11-15 18:06:18 +02003553 "setbe %c[fail](%0) \n\t"
3554 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
3555 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
3556 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
Avi Kivity313dbd42008-07-17 18:04:30 +03003557 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
Zhang Xiantaoad312c72007-12-13 23:50:52 +08003558 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
3559 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
3560 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
3561 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
3562 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
3563 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
3564 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
Avi Kivity05b3e0c2006-12-13 00:33:45 -08003565#ifdef CONFIG_X86_64
Zhang Xiantaoad312c72007-12-13 23:50:52 +08003566 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
3567 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
3568 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
3569 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
3570 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
3571 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
3572 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
3573 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
Avi Kivity6aa8b732006-12-10 02:21:36 -08003574#endif
Zhang Xiantaoad312c72007-12-13 23:50:52 +08003575 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
Laurent Vivierc2036302007-10-25 14:18:52 +02003576 : "cc", "memory"
Avi Kivityc8019492008-07-14 14:44:59 +03003577 , R"bx", R"di", R"si"
Laurent Vivierc2036302007-10-25 14:18:52 +02003578#ifdef CONFIG_X86_64
Laurent Vivierc2036302007-10-25 14:18:52 +02003579 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
3580#endif
3581 );
Avi Kivity6aa8b732006-12-10 02:21:36 -08003582
Avi Kivity6de4f3a2009-05-31 22:58:47 +03003583 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
3584 | (1 << VCPU_EXREG_PDPTR));
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03003585 vcpu->arch.regs_dirty = 0;
3586
Jan Kiszka42dbaa52008-12-15 13:52:10 +01003587 get_debugreg(vcpu->arch.dr6, 6);
3588
Avi Kivity1155f762007-11-22 11:30:47 +02003589 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
Avi Kivity9c8cba32007-11-22 11:42:59 +02003590 if (vmx->rmode.irq.pending)
3591 fixup_rmode_irq(vmx);
Avi Kivity1155f762007-11-22 11:30:47 +02003592
Mike Dayd77c26f2007-10-08 09:02:08 -04003593 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
Avi Kivity15ad7142007-07-11 18:17:21 +03003594 vmx->launched = 1;
Avi Kivity1b6269d2007-10-09 12:12:19 +02003595
Avi Kivitycf393f72008-07-01 16:20:21 +03003596 vmx_complete_interrupts(vmx);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003597}
3598
Avi Kivityc8019492008-07-14 14:44:59 +03003599#undef R
3600#undef Q
3601
Avi Kivity6aa8b732006-12-10 02:21:36 -08003602static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
3603{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04003604 struct vcpu_vmx *vmx = to_vmx(vcpu);
3605
3606 if (vmx->vmcs) {
Avi Kivity543e4242008-05-13 16:22:47 +03003607 vcpu_clear(vmx);
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04003608 free_vmcs(vmx->vmcs);
3609 vmx->vmcs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003610 }
3611}
3612
3613static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
3614{
Rusty Russellfb3f0f52007-07-27 17:16:56 +10003615 struct vcpu_vmx *vmx = to_vmx(vcpu);
3616
Sheng Yang2384d2b2008-01-17 15:14:33 +08003617 spin_lock(&vmx_vpid_lock);
3618 if (vmx->vpid != 0)
3619 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3620 spin_unlock(&vmx_vpid_lock);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003621 vmx_free_vmcs(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10003622 kfree(vmx->host_msrs);
3623 kfree(vmx->guest_msrs);
3624 kvm_vcpu_uninit(vcpu);
Rusty Russella4770342007-08-01 14:46:11 +10003625 kmem_cache_free(kvm_vcpu_cache, vmx);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003626}
3627
Rusty Russellfb3f0f52007-07-27 17:16:56 +10003628static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003629{
Rusty Russellfb3f0f52007-07-27 17:16:56 +10003630 int err;
Rusty Russellc16f8622007-07-30 21:12:19 +10003631 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
Avi Kivity15ad7142007-07-11 18:17:21 +03003632 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003633
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04003634 if (!vmx)
Rusty Russellfb3f0f52007-07-27 17:16:56 +10003635 return ERR_PTR(-ENOMEM);
3636
Sheng Yang2384d2b2008-01-17 15:14:33 +08003637 allocate_vpid(vmx);
3638
Rusty Russellfb3f0f52007-07-27 17:16:56 +10003639 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
3640 if (err)
3641 goto free_vcpu;
Ingo Molnar965b58a2007-01-05 16:36:23 -08003642
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04003643 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10003644 if (!vmx->guest_msrs) {
3645 err = -ENOMEM;
3646 goto uninit_vcpu;
3647 }
Ingo Molnar965b58a2007-01-05 16:36:23 -08003648
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04003649 vmx->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3650 if (!vmx->host_msrs)
Rusty Russellfb3f0f52007-07-27 17:16:56 +10003651 goto free_guest_msrs;
Ingo Molnar965b58a2007-01-05 16:36:23 -08003652
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04003653 vmx->vmcs = alloc_vmcs();
3654 if (!vmx->vmcs)
Rusty Russellfb3f0f52007-07-27 17:16:56 +10003655 goto free_msrs;
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04003656
3657 vmcs_clear(vmx->vmcs);
3658
Avi Kivity15ad7142007-07-11 18:17:21 +03003659 cpu = get_cpu();
3660 vmx_vcpu_load(&vmx->vcpu, cpu);
Rusty Russell8b9cf982007-07-30 16:31:43 +10003661 err = vmx_vcpu_setup(vmx);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10003662 vmx_vcpu_put(&vmx->vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003663 put_cpu();
Rusty Russellfb3f0f52007-07-27 17:16:56 +10003664 if (err)
3665 goto free_vmcs;
Marcelo Tosatti5e4a0b32008-02-14 21:21:43 -02003666 if (vm_need_virtualize_apic_accesses(kvm))
3667 if (alloc_apic_access_page(kvm) != 0)
3668 goto free_vmcs;
Ingo Molnar965b58a2007-01-05 16:36:23 -08003669
Avi Kivity089d0342009-03-23 18:26:32 +02003670 if (enable_ept)
Sheng Yangb7ebfb02008-04-25 21:44:52 +08003671 if (alloc_identity_pagetable(kvm) != 0)
3672 goto free_vmcs;
3673
Rusty Russellfb3f0f52007-07-27 17:16:56 +10003674 return &vmx->vcpu;
Ingo Molnar965b58a2007-01-05 16:36:23 -08003675
Rusty Russellfb3f0f52007-07-27 17:16:56 +10003676free_vmcs:
3677 free_vmcs(vmx->vmcs);
3678free_msrs:
3679 kfree(vmx->host_msrs);
3680free_guest_msrs:
3681 kfree(vmx->guest_msrs);
3682uninit_vcpu:
3683 kvm_vcpu_uninit(&vmx->vcpu);
3684free_vcpu:
Rusty Russella4770342007-08-01 14:46:11 +10003685 kmem_cache_free(kvm_vcpu_cache, vmx);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10003686 return ERR_PTR(err);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003687}
3688
Yang, Sheng002c7f72007-07-31 14:23:01 +03003689static void __init vmx_check_processor_compat(void *rtn)
3690{
3691 struct vmcs_config vmcs_conf;
3692
3693 *(int *)rtn = 0;
3694 if (setup_vmcs_config(&vmcs_conf) < 0)
3695 *(int *)rtn = -EIO;
3696 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
3697 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
3698 smp_processor_id());
3699 *(int *)rtn = -EIO;
3700 }
3701}
3702
Sheng Yang67253af2008-04-25 10:20:22 +08003703static int get_ept_level(void)
3704{
3705 return VMX_EPT_DEFAULT_GAW + 1;
3706}
3707
Sheng Yang4b12f0d2009-04-27 20:35:42 +08003708static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
Sheng Yang64d4d522008-10-09 16:01:57 +08003709{
Sheng Yang4b12f0d2009-04-27 20:35:42 +08003710 u64 ret;
3711
Sheng Yang522c68c2009-04-27 20:35:43 +08003712 /* For VT-d and EPT combination
3713 * 1. MMIO: always map as UC
3714 * 2. EPT with VT-d:
3715 * a. VT-d without snooping control feature: can't guarantee the
3716 * result, try to trust guest.
3717 * b. VT-d with snooping control feature: snooping control feature of
3718 * VT-d engine can guarantee the cache correctness. Just set it
3719 * to WB to keep consistent with host. So the same as item 3.
3720 * 3. EPT without VT-d: always map as WB and set IGMT=1 to keep
3721 * consistent with host MTRR
3722 */
Sheng Yang4b12f0d2009-04-27 20:35:42 +08003723 if (is_mmio)
3724 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
Sheng Yang522c68c2009-04-27 20:35:43 +08003725 else if (vcpu->kvm->arch.iommu_domain &&
3726 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
3727 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
3728 VMX_EPT_MT_EPTE_SHIFT;
Sheng Yang4b12f0d2009-04-27 20:35:42 +08003729 else
Sheng Yang522c68c2009-04-27 20:35:43 +08003730 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
3731 | VMX_EPT_IGMT_BIT;
Sheng Yang4b12f0d2009-04-27 20:35:42 +08003732
3733 return ret;
Sheng Yang64d4d522008-10-09 16:01:57 +08003734}
3735
Christian Ehrhardtcbdd1be2007-09-09 15:41:59 +03003736static struct kvm_x86_ops vmx_x86_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08003737 .cpu_has_kvm_support = cpu_has_kvm_support,
3738 .disabled_by_bios = vmx_disabled_by_bios,
3739 .hardware_setup = hardware_setup,
3740 .hardware_unsetup = hardware_unsetup,
Yang, Sheng002c7f72007-07-31 14:23:01 +03003741 .check_processor_compatibility = vmx_check_processor_compat,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003742 .hardware_enable = hardware_enable,
3743 .hardware_disable = hardware_disable,
Sheng Yang04547152009-04-01 15:52:31 +08003744 .cpu_has_accelerated_tpr = report_flexpriority,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003745
3746 .vcpu_create = vmx_create_vcpu,
3747 .vcpu_free = vmx_free_vcpu,
Avi Kivity04d2cc72007-09-10 18:10:54 +03003748 .vcpu_reset = vmx_vcpu_reset,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003749
Avi Kivity04d2cc72007-09-10 18:10:54 +03003750 .prepare_guest_switch = vmx_save_host_state,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003751 .vcpu_load = vmx_vcpu_load,
3752 .vcpu_put = vmx_vcpu_put,
3753
3754 .set_guest_debug = set_guest_debug,
3755 .get_msr = vmx_get_msr,
3756 .set_msr = vmx_set_msr,
3757 .get_segment_base = vmx_get_segment_base,
3758 .get_segment = vmx_get_segment,
3759 .set_segment = vmx_set_segment,
Izik Eidus2e4d2652008-03-24 19:38:34 +02003760 .get_cpl = vmx_get_cpl,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003761 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
Anthony Liguori25c4c272007-04-27 09:29:21 +03003762 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003763 .set_cr0 = vmx_set_cr0,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003764 .set_cr3 = vmx_set_cr3,
3765 .set_cr4 = vmx_set_cr4,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003766 .set_efer = vmx_set_efer,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003767 .get_idt = vmx_get_idt,
3768 .set_idt = vmx_set_idt,
3769 .get_gdt = vmx_get_gdt,
3770 .set_gdt = vmx_set_gdt,
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03003771 .cache_reg = vmx_cache_reg,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003772 .get_rflags = vmx_get_rflags,
3773 .set_rflags = vmx_set_rflags,
3774
3775 .tlb_flush = vmx_flush_tlb,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003776
Avi Kivity6aa8b732006-12-10 02:21:36 -08003777 .run = vmx_vcpu_run,
Avi Kivity6062d012009-03-23 17:35:17 +02003778 .handle_exit = vmx_handle_exit,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003779 .skip_emulated_instruction = skip_emulated_instruction,
Glauber Costa2809f5d2009-05-12 16:21:05 -04003780 .set_interrupt_shadow = vmx_set_interrupt_shadow,
3781 .get_interrupt_shadow = vmx_get_interrupt_shadow,
Ingo Molnar102d8322007-02-19 14:37:47 +02003782 .patch_hypercall = vmx_patch_hypercall,
Eddie Dong2a8067f2007-08-06 16:29:07 +03003783 .set_irq = vmx_inject_irq,
Gleb Natapov95ba8273132009-04-21 17:45:08 +03003784 .set_nmi = vmx_inject_nmi,
Avi Kivity298101d2007-11-25 13:41:11 +02003785 .queue_exception = vmx_queue_exception,
Gleb Natapov78646122009-03-23 12:12:11 +02003786 .interrupt_allowed = vmx_interrupt_allowed,
Gleb Natapov95ba8273132009-04-21 17:45:08 +03003787 .nmi_allowed = vmx_nmi_allowed,
3788 .enable_nmi_window = enable_nmi_window,
3789 .enable_irq_window = enable_irq_window,
3790 .update_cr8_intercept = update_cr8_intercept,
Gleb Natapov95ba8273132009-04-21 17:45:08 +03003791
Izik Eiduscbc94022007-10-25 00:29:55 +02003792 .set_tss_addr = vmx_set_tss_addr,
Sheng Yang67253af2008-04-25 10:20:22 +08003793 .get_tdp_level = get_ept_level,
Sheng Yang4b12f0d2009-04-27 20:35:42 +08003794 .get_mt_mask = vmx_get_mt_mask,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003795};
3796
3797static int __init vmx_init(void)
3798{
He, Qingfdef3ad2007-04-30 09:45:24 +03003799 int r;
3800
Avi Kivity3e7c73e2009-02-24 21:46:19 +02003801 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
He, Qingfdef3ad2007-04-30 09:45:24 +03003802 if (!vmx_io_bitmap_a)
3803 return -ENOMEM;
3804
Avi Kivity3e7c73e2009-02-24 21:46:19 +02003805 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
He, Qingfdef3ad2007-04-30 09:45:24 +03003806 if (!vmx_io_bitmap_b) {
3807 r = -ENOMEM;
3808 goto out;
3809 }
3810
Avi Kivity58972972009-02-24 22:26:47 +02003811 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
3812 if (!vmx_msr_bitmap_legacy) {
Sheng Yang25c5f222008-03-28 13:18:56 +08003813 r = -ENOMEM;
3814 goto out1;
3815 }
3816
Avi Kivity58972972009-02-24 22:26:47 +02003817 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
3818 if (!vmx_msr_bitmap_longmode) {
3819 r = -ENOMEM;
3820 goto out2;
3821 }
3822
He, Qingfdef3ad2007-04-30 09:45:24 +03003823 /*
3824 * Allow direct access to the PC debug port (it is often used for I/O
3825 * delays, but the vmexits simply slow things down).
3826 */
Avi Kivity3e7c73e2009-02-24 21:46:19 +02003827 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
3828 clear_bit(0x80, vmx_io_bitmap_a);
He, Qingfdef3ad2007-04-30 09:45:24 +03003829
Avi Kivity3e7c73e2009-02-24 21:46:19 +02003830 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
He, Qingfdef3ad2007-04-30 09:45:24 +03003831
Avi Kivity58972972009-02-24 22:26:47 +02003832 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
3833 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
Sheng Yang25c5f222008-03-28 13:18:56 +08003834
Sheng Yang2384d2b2008-01-17 15:14:33 +08003835 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
3836
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003837 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
He, Qingfdef3ad2007-04-30 09:45:24 +03003838 if (r)
Avi Kivity58972972009-02-24 22:26:47 +02003839 goto out3;
Sheng Yang25c5f222008-03-28 13:18:56 +08003840
Avi Kivity58972972009-02-24 22:26:47 +02003841 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
3842 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
3843 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
3844 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
3845 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
3846 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
He, Qingfdef3ad2007-04-30 09:45:24 +03003847
Avi Kivity089d0342009-03-23 18:26:32 +02003848 if (enable_ept) {
Sheng Yang14394422008-04-28 12:24:45 +08003849 bypass_guest_pf = 0;
Sheng Yang5fdbcb92008-07-16 09:25:40 +08003850 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
Sheng Yang2aaf69d2009-01-21 16:52:16 +08003851 VMX_EPT_WRITABLE_MASK);
Sheng Yang534e38b2008-09-08 15:12:30 +08003852 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
Sheng Yang4b12f0d2009-04-27 20:35:42 +08003853 VMX_EPT_EXECUTABLE_MASK);
Sheng Yang5fdbcb92008-07-16 09:25:40 +08003854 kvm_enable_tdp();
3855 } else
3856 kvm_disable_tdp();
Sheng Yang14394422008-04-28 12:24:45 +08003857
Avi Kivityc7addb92007-09-16 18:58:32 +02003858 if (bypass_guest_pf)
3859 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
3860
Sheng Yang14394422008-04-28 12:24:45 +08003861 ept_sync_global();
3862
He, Qingfdef3ad2007-04-30 09:45:24 +03003863 return 0;
3864
Avi Kivity58972972009-02-24 22:26:47 +02003865out3:
3866 free_page((unsigned long)vmx_msr_bitmap_longmode);
Sheng Yang25c5f222008-03-28 13:18:56 +08003867out2:
Avi Kivity58972972009-02-24 22:26:47 +02003868 free_page((unsigned long)vmx_msr_bitmap_legacy);
He, Qingfdef3ad2007-04-30 09:45:24 +03003869out1:
Avi Kivity3e7c73e2009-02-24 21:46:19 +02003870 free_page((unsigned long)vmx_io_bitmap_b);
He, Qingfdef3ad2007-04-30 09:45:24 +03003871out:
Avi Kivity3e7c73e2009-02-24 21:46:19 +02003872 free_page((unsigned long)vmx_io_bitmap_a);
He, Qingfdef3ad2007-04-30 09:45:24 +03003873 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003874}
3875
3876static void __exit vmx_exit(void)
3877{
Avi Kivity58972972009-02-24 22:26:47 +02003878 free_page((unsigned long)vmx_msr_bitmap_legacy);
3879 free_page((unsigned long)vmx_msr_bitmap_longmode);
Avi Kivity3e7c73e2009-02-24 21:46:19 +02003880 free_page((unsigned long)vmx_io_bitmap_b);
3881 free_page((unsigned long)vmx_io_bitmap_a);
He, Qingfdef3ad2007-04-30 09:45:24 +03003882
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003883 kvm_exit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003884}
3885
3886module_init(vmx_init)
3887module_exit(vmx_exit)