]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - arch/x86/kvm/vmx.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: Use kvm_get_rflags() and kvm_set_rflags() instead of the raw versions
[mirror_ubuntu-bionic-kernel.git] / arch / x86 / kvm / vmx.c
CommitLineData
6aa8b732
AK		 1/*
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.
9611c187 8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
6aa8b732
AK		 9 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
85f455f7 19#include "irq.h"
1d737c8a 20#include "mmu.h"
e495606d 21
edf88417 22#include <linux/kvm_host.h>
6aa8b732 23#include <linux/module.h>
9d8f549d 24#include <linux/kernel.h>
6aa8b732
AK		 25#include <linux/mm.h>
26#include <linux/highmem.h>
e8edc6e0 27#include <linux/sched.h>
c7addb90 28#include <linux/moduleparam.h>
229456fc 29#include <linux/ftrace_event.h>
5a0e3ad6 30#include <linux/slab.h>
cafd6659 31#include <linux/tboot.h>
5fdbf976 32#include "kvm_cache_regs.h"
35920a35 33#include "x86.h"
e495606d 34
6aa8b732 35#include <asm/io.h>
3b3be0d1 36#include <asm/desc.h>
13673a90 37#include <asm/vmx.h>
6210e37b 38#include <asm/virtext.h>
a0861c02 39#include <asm/mce.h>
2acf923e
DC		 40#include <asm/i387.h>
41#include <asm/xcr.h>
6aa8b732 42
229456fc
MT		 43#include "trace.h"
44
4ecac3fd
AK		 45#define __ex(x) __kvm_handle_fault_on_reboot(x)
46
6aa8b732
AK		 47MODULE_AUTHOR("Qumranet");
48MODULE_LICENSE("GPL");
49
4462d21a 50static int __read_mostly bypass_guest_pf = 1;
c1f8bc04 51module_param(bypass_guest_pf, bool, S_IRUGO);
c7addb90 52
4462d21a 53static int __read_mostly enable_vpid = 1;
736caefe 54module_param_named(vpid, enable_vpid, bool, 0444);
2384d2b3 55
4462d21a 56static int __read_mostly flexpriority_enabled = 1;
736caefe 57module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
4c9fc8ef 58
4462d21a 59static int __read_mostly enable_ept = 1;
736caefe 60module_param_named(ept, enable_ept, bool, S_IRUGO);
d56f546d 61
3a624e29
NK		 62static int __read_mostly enable_unrestricted_guest = 1;
63module_param_named(unrestricted_guest,
64 enable_unrestricted_guest, bool, S_IRUGO);
65
4462d21a 66static int __read_mostly emulate_invalid_guest_state = 0;
c1f8bc04 67module_param(emulate_invalid_guest_state, bool, S_IRUGO);
04fa4d32 68
b923e62e
DX		 69static int __read_mostly vmm_exclusive = 1;
70module_param(vmm_exclusive, bool, S_IRUGO);
71
443381a8
AL		 72static int __read_mostly yield_on_hlt = 1;
73module_param(yield_on_hlt, bool, S_IRUGO);
74
cdc0e244
AK		 75#define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST \
76 (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
77#define KVM_GUEST_CR0_MASK \
78 (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
79#define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST \
81231c69 80 (X86_CR0_WP | X86_CR0_NE)
cdc0e244
AK		 81#define KVM_VM_CR0_ALWAYS_ON \
82 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
4c38609a
AK		 83#define KVM_CR4_GUEST_OWNED_BITS \
84 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
85 | X86_CR4_OSXMMEXCPT)
86
cdc0e244
AK		 87#define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
88#define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
89
78ac8b47
AK		 90#define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
91
4b8d54f9
ZE		 92/*
93 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
94 * ple_gap: upper bound on the amount of time between two successive
95 * executions of PAUSE in a loop. Also indicate if ple enabled.
00c25bce 96 * According to test, this time is usually smaller than 128 cycles.
4b8d54f9
ZE		 97 * ple_window: upper bound on the amount of time a guest is allowed to execute
98 * in a PAUSE loop. Tests indicate that most spinlocks are held for
99 * less than 2^12 cycles
100 * Time is measured based on a counter that runs at the same rate as the TSC,
101 * refer SDM volume 3b section 21.6.13 & 22.1.3.
102 */
00c25bce 103#define KVM_VMX_DEFAULT_PLE_GAP 128
4b8d54f9
ZE		 104#define KVM_VMX_DEFAULT_PLE_WINDOW 4096
105static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
106module_param(ple_gap, int, S_IRUGO);
107
108static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
109module_param(ple_window, int, S_IRUGO);
110
61d2ef2c
AK		 111#define NR_AUTOLOAD_MSRS 1
112
a2fa3e9f
GH		 113struct vmcs {
114 u32 revision_id;
115 u32 abort;
116 char data[0];
117};
118
26bb0981
AK		 119struct shared_msr_entry {
120 unsigned index;
121 u64 data;
d5696725 122 u64 mask;
26bb0981
AK		 123};
124
a2fa3e9f 125struct vcpu_vmx {
fb3f0f51 126 struct kvm_vcpu vcpu;
543e4243 127 struct list_head local_vcpus_link;
313dbd49 128 unsigned long host_rsp;
a2fa3e9f 129 int launched;
29bd8a78 130 u8 fail;
51aa01d1 131 u32 exit_intr_info;
1155f76a 132 u32 idt_vectoring_info;
26bb0981 133 struct shared_msr_entry *guest_msrs;
a2fa3e9f
GH		 134 int nmsrs;
135 int save_nmsrs;
a2fa3e9f 136#ifdef CONFIG_X86_64
44ea2b17
AK		 137 u64 msr_host_kernel_gs_base;
138 u64 msr_guest_kernel_gs_base;
a2fa3e9f
GH		 139#endif
140 struct vmcs *vmcs;
61d2ef2c
AK		 141 struct msr_autoload {
142 unsigned nr;
143 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
144 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
145 } msr_autoload;
a2fa3e9f
GH		 146 struct {
147 int loaded;
148 u16 fs_sel, gs_sel, ldt_sel;
152d3f2f
LV		 149 int gs_ldt_reload_needed;
150 int fs_reload_needed;
d77c26fc 151 } host_state;
9c8cba37 152 struct {
7ffd92c5 153 int vm86_active;
78ac8b47 154 ulong save_rflags;
7ffd92c5
AK		 155 struct kvm_save_segment {
156 u16 selector;
157 unsigned long base;
158 u32 limit;
159 u32 ar;
160 } tr, es, ds, fs, gs;
9c8cba37 161 } rmode;
2384d2b3 162 int vpid;
04fa4d32 163 bool emulation_required;
3b86cd99
JK		 164
165 /* Support for vnmi-less CPUs */
166 int soft_vnmi_blocked;
167 ktime_t entry_time;
168 s64 vnmi_blocked_time;
a0861c02 169 u32 exit_reason;
4e47c7a6
SY		 170
171 bool rdtscp_enabled;
a2fa3e9f
GH		 172};
173
174static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
175{
fb3f0f51 176 return container_of(vcpu, struct vcpu_vmx, vcpu);
a2fa3e9f
GH		 177}
178
4e1096d2 179static u64 construct_eptp(unsigned long root_hpa);
4610c9cc
DX		 180static void kvm_cpu_vmxon(u64 addr);
181static void kvm_cpu_vmxoff(void);
aff48baa 182static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
776e58ea 183static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
75880a01 184
6aa8b732
AK		 185static DEFINE_PER_CPU(struct vmcs *, vmxarea);
186static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
543e4243 187static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
3444d7da 188static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
6aa8b732 189
3e7c73e9
AK		 190static unsigned long *vmx_io_bitmap_a;
191static unsigned long *vmx_io_bitmap_b;
5897297b
AK		 192static unsigned long *vmx_msr_bitmap_legacy;
193static unsigned long *vmx_msr_bitmap_longmode;
fdef3ad1 194
110312c8
AK		 195static bool cpu_has_load_ia32_efer;
196
2384d2b3
SY		 197static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
198static DEFINE_SPINLOCK(vmx_vpid_lock);
199
1c3d14fe 200static struct vmcs_config {
6aa8b732
AK		 201 int size;
202 int order;
203 u32 revision_id;
1c3d14fe
YS		 204 u32 pin_based_exec_ctrl;
205 u32 cpu_based_exec_ctrl;
f78e0e2e 206 u32 cpu_based_2nd_exec_ctrl;
1c3d14fe
YS		 207 u32 vmexit_ctrl;
208 u32 vmentry_ctrl;
209} vmcs_config;
6aa8b732 210
efff9e53 211static struct vmx_capability {
d56f546d
SY		 212 u32 ept;
213 u32 vpid;
214} vmx_capability;
215
6aa8b732
AK		 216#define VMX_SEGMENT_FIELD(seg) \
217 [VCPU_SREG_##seg] = { \
218 .selector = GUEST_##seg##_SELECTOR, \
219 .base = GUEST_##seg##_BASE, \
220 .limit = GUEST_##seg##_LIMIT, \
221 .ar_bytes = GUEST_##seg##_AR_BYTES, \
222 }
223
224static struct kvm_vmx_segment_field {
225 unsigned selector;
226 unsigned base;
227 unsigned limit;
228 unsigned ar_bytes;
229} kvm_vmx_segment_fields[] = {
230 VMX_SEGMENT_FIELD(CS),
231 VMX_SEGMENT_FIELD(DS),
232 VMX_SEGMENT_FIELD(ES),
233 VMX_SEGMENT_FIELD(FS),
234 VMX_SEGMENT_FIELD(GS),
235 VMX_SEGMENT_FIELD(SS),
236 VMX_SEGMENT_FIELD(TR),
237 VMX_SEGMENT_FIELD(LDTR),
238};
239
26bb0981
AK		 240static u64 host_efer;
241
6de4f3ad
AK		 242static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
243
4d56c8a7 244/*
8c06585d 245 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
4d56c8a7
AK		 246 * away by decrementing the array size.
247 */
6aa8b732 248static const u32 vmx_msr_index[] = {
05b3e0c2 249#ifdef CONFIG_X86_64
44ea2b17 250 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
6aa8b732 251#endif
8c06585d 252 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
6aa8b732 253};
9d8f549d 254#define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
6aa8b732 255
31299944 256static inline bool is_page_fault(u32 intr_info)
6aa8b732
AK		 257{
258 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
259 INTR_INFO_VALID_MASK)) ==
8ab2d2e2 260 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
6aa8b732
AK		 261}
262
31299944 263static inline bool is_no_device(u32 intr_info)
2ab455cc
AL		 264{
265 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
266 INTR_INFO_VALID_MASK)) ==
8ab2d2e2 267 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
2ab455cc
AL		 268}
269
31299944 270static inline bool is_invalid_opcode(u32 intr_info)
7aa81cc0
AL		 271{
272 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
273 INTR_INFO_VALID_MASK)) ==
8ab2d2e2 274 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
7aa81cc0
AL		 275}
276
31299944 277static inline bool is_external_interrupt(u32 intr_info)
6aa8b732
AK		 278{
279 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
280 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
281}
282
31299944 283static inline bool is_machine_check(u32 intr_info)
a0861c02
AK		 284{
285 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
286 INTR_INFO_VALID_MASK)) ==
287 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
288}
289
31299944 290static inline bool cpu_has_vmx_msr_bitmap(void)
25c5f225 291{
04547156 292 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
25c5f225
SY		 293}
294
31299944 295static inline bool cpu_has_vmx_tpr_shadow(void)
6e5d865c 296{
04547156 297 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
6e5d865c
YS		 298}
299
31299944 300static inline bool vm_need_tpr_shadow(struct kvm *kvm)
6e5d865c 301{
04547156 302 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
6e5d865c
YS		 303}
304
31299944 305static inline bool cpu_has_secondary_exec_ctrls(void)
f78e0e2e 306{
04547156
SY		 307 return vmcs_config.cpu_based_exec_ctrl &
308 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
f78e0e2e
SY		 309}
310
774ead3a 311static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
f78e0e2e 312{
04547156
SY		 313 return vmcs_config.cpu_based_2nd_exec_ctrl &
314 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
315}
316
317static inline bool cpu_has_vmx_flexpriority(void)
318{
319 return cpu_has_vmx_tpr_shadow() &&
320 cpu_has_vmx_virtualize_apic_accesses();
f78e0e2e
SY		 321}
322
e799794e
MT		 323static inline bool cpu_has_vmx_ept_execute_only(void)
324{
31299944 325 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
e799794e
MT		 326}
327
328static inline bool cpu_has_vmx_eptp_uncacheable(void)
329{
31299944 330 return vmx_capability.ept & VMX_EPTP_UC_BIT;
e799794e
MT		 331}
332
333static inline bool cpu_has_vmx_eptp_writeback(void)
334{
31299944 335 return vmx_capability.ept & VMX_EPTP_WB_BIT;
e799794e
MT		 336}
337
338static inline bool cpu_has_vmx_ept_2m_page(void)
339{
31299944 340 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
e799794e
MT		 341}
342
878403b7
SY		 343static inline bool cpu_has_vmx_ept_1g_page(void)
344{
31299944 345 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
878403b7
SY		 346}
347
4bc9b982
SY		 348static inline bool cpu_has_vmx_ept_4levels(void)
349{
350 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
351}
352
31299944 353static inline bool cpu_has_vmx_invept_individual_addr(void)
d56f546d 354{
31299944 355 return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
d56f546d
SY		 356}
357
31299944 358static inline bool cpu_has_vmx_invept_context(void)
d56f546d 359{
31299944 360 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
d56f546d
SY		 361}
362
31299944 363static inline bool cpu_has_vmx_invept_global(void)
d56f546d 364{
31299944 365 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
d56f546d
SY		 366}
367
518c8aee
GJ		 368static inline bool cpu_has_vmx_invvpid_single(void)
369{
370 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
371}
372
b9d762fa
GJ		 373static inline bool cpu_has_vmx_invvpid_global(void)
374{
375 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
376}
377
31299944 378static inline bool cpu_has_vmx_ept(void)
d56f546d 379{
04547156
SY		 380 return vmcs_config.cpu_based_2nd_exec_ctrl &
381 SECONDARY_EXEC_ENABLE_EPT;
d56f546d
SY		 382}
383
31299944 384static inline bool cpu_has_vmx_unrestricted_guest(void)
3a624e29
NK		 385{
386 return vmcs_config.cpu_based_2nd_exec_ctrl &
387 SECONDARY_EXEC_UNRESTRICTED_GUEST;
388}
389
31299944 390static inline bool cpu_has_vmx_ple(void)
4b8d54f9
ZE		 391{
392 return vmcs_config.cpu_based_2nd_exec_ctrl &
393 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
394}
395
31299944 396static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
f78e0e2e 397{
6d3e435e 398 return flexpriority_enabled && irqchip_in_kernel(kvm);
f78e0e2e
SY		 399}
400
31299944 401static inline bool cpu_has_vmx_vpid(void)
2384d2b3 402{
04547156
SY		 403 return vmcs_config.cpu_based_2nd_exec_ctrl &
404 SECONDARY_EXEC_ENABLE_VPID;
2384d2b3
SY		 405}
406
31299944 407static inline bool cpu_has_vmx_rdtscp(void)
4e47c7a6
SY		 408{
409 return vmcs_config.cpu_based_2nd_exec_ctrl &
410 SECONDARY_EXEC_RDTSCP;
411}
412
31299944 413static inline bool cpu_has_virtual_nmis(void)
f08864b4
SY		 414{
415 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
416}
417
f5f48ee1
SY		 418static inline bool cpu_has_vmx_wbinvd_exit(void)
419{
420 return vmcs_config.cpu_based_2nd_exec_ctrl &
421 SECONDARY_EXEC_WBINVD_EXITING;
422}
423
04547156
SY		 424static inline bool report_flexpriority(void)
425{
426 return flexpriority_enabled;
427}
428
8b9cf98c 429static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
7725f0ba
AK		 430{
431 int i;
432
a2fa3e9f 433 for (i = 0; i < vmx->nmsrs; ++i)
26bb0981 434 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
a75beee6
ED		 435 return i;
436 return -1;
437}
438
2384d2b3
SY		 439static inline void __invvpid(int ext, u16 vpid, gva_t gva)
440{
441 struct {
442 u64 vpid : 16;
443 u64 rsvd : 48;
444 u64 gva;
445 } operand = { vpid, 0, gva };
446
4ecac3fd 447 asm volatile (__ex(ASM_VMX_INVVPID)
2384d2b3
SY		 448 /* CF==1 or ZF==1 --> rc = -1 */
449 "; ja 1f ; ud2 ; 1:"
450 : : "a"(&operand), "c"(ext) : "cc", "memory");
451}
452
1439442c
SY		 453static inline void __invept(int ext, u64 eptp, gpa_t gpa)
454{
455 struct {
456 u64 eptp, gpa;
457 } operand = {eptp, gpa};
458
4ecac3fd 459 asm volatile (__ex(ASM_VMX_INVEPT)
1439442c
SY		 460 /* CF==1 or ZF==1 --> rc = -1 */
461 "; ja 1f ; ud2 ; 1:\n"
462 : : "a" (&operand), "c" (ext) : "cc", "memory");
463}
464
26bb0981 465static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
a75beee6
ED		 466{
467 int i;
468
8b9cf98c 469 i = __find_msr_index(vmx, msr);
a75beee6 470 if (i >= 0)
a2fa3e9f 471 return &vmx->guest_msrs[i];
8b6d44c7 472 return NULL;
7725f0ba
AK		 473}
474
6aa8b732
AK		 475static void vmcs_clear(struct vmcs *vmcs)
476{
477 u64 phys_addr = __pa(vmcs);
478 u8 error;
479
4ecac3fd 480 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
16d8f72f 481 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
6aa8b732
AK		 482 : "cc", "memory");
483 if (error)
484 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
485 vmcs, phys_addr);
486}
487
7725b894
DX		 488static void vmcs_load(struct vmcs *vmcs)
489{
490 u64 phys_addr = __pa(vmcs);
491 u8 error;
492
493 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
16d8f72f 494 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
7725b894
DX		 495 : "cc", "memory");
496 if (error)
497 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
498 vmcs, phys_addr);
499}
500
6aa8b732
AK		 501static void __vcpu_clear(void *arg)
502{
8b9cf98c 503 struct vcpu_vmx *vmx = arg;
d3b2c338 504 int cpu = raw_smp_processor_id();
6aa8b732 505
8b9cf98c 506 if (vmx->vcpu.cpu == cpu)
a2fa3e9f
GH		 507 vmcs_clear(vmx->vmcs);
508 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
6aa8b732 509 per_cpu(current_vmcs, cpu) = NULL;
543e4243
AK		 510 list_del(&vmx->local_vcpus_link);
511 vmx->vcpu.cpu = -1;
512 vmx->launched = 0;
6aa8b732
AK		 513}
514
8b9cf98c 515static void vcpu_clear(struct vcpu_vmx *vmx)
8d0be2b3 516{
eae5ecb5
AK		 517 if (vmx->vcpu.cpu == -1)
518 return;
8691e5a8 519 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
8d0be2b3
AK		 520}
521
1760dd49 522static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
2384d2b3
SY		 523{
524 if (vmx->vpid == 0)
525 return;
526
518c8aee
GJ		 527 if (cpu_has_vmx_invvpid_single())
528 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
2384d2b3
SY		 529}
530
b9d762fa
GJ		 531static inline void vpid_sync_vcpu_global(void)
532{
533 if (cpu_has_vmx_invvpid_global())
534 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
535}
536
537static inline void vpid_sync_context(struct vcpu_vmx *vmx)
538{
539 if (cpu_has_vmx_invvpid_single())
1760dd49 540 vpid_sync_vcpu_single(vmx);
b9d762fa
GJ		 541 else
542 vpid_sync_vcpu_global();
543}
544
1439442c
SY		 545static inline void ept_sync_global(void)
546{
547 if (cpu_has_vmx_invept_global())
548 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
549}
550
551static inline void ept_sync_context(u64 eptp)
552{
089d034e 553 if (enable_ept) {
1439442c
SY		 554 if (cpu_has_vmx_invept_context())
555 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
556 else
557 ept_sync_global();
558 }
559}
560
561static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
562{
089d034e 563 if (enable_ept) {
1439442c
SY		 564 if (cpu_has_vmx_invept_individual_addr())
565 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
566 eptp, gpa);
567 else
568 ept_sync_context(eptp);
569 }
570}
571
6aa8b732
AK		 572static unsigned long vmcs_readl(unsigned long field)
573{
a295673a 574 unsigned long value = 0;
6aa8b732 575
4ecac3fd 576 asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
a295673a 577 : "+a"(value) : "d"(field) : "cc");
6aa8b732
AK		 578 return value;
579}
580
581static u16 vmcs_read16(unsigned long field)
582{
583 return vmcs_readl(field);
584}
585
586static u32 vmcs_read32(unsigned long field)
587{
588 return vmcs_readl(field);
589}
590
591static u64 vmcs_read64(unsigned long field)
592{
05b3e0c2 593#ifdef CONFIG_X86_64
6aa8b732
AK		 594 return vmcs_readl(field);
595#else
596 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
597#endif
598}
599
e52de1b8
AK		 600static noinline void vmwrite_error(unsigned long field, unsigned long value)
601{
602 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
603 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
604 dump_stack();
605}
606
6aa8b732
AK		 607static void vmcs_writel(unsigned long field, unsigned long value)
608{
609 u8 error;
610
4ecac3fd 611 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
d77c26fc 612 : "=q"(error) : "a"(value), "d"(field) : "cc");
e52de1b8
AK		 613 if (unlikely(error))
614 vmwrite_error(field, value);
6aa8b732
AK		 615}
616
617static void vmcs_write16(unsigned long field, u16 value)
618{
619 vmcs_writel(field, value);
620}
621
622static void vmcs_write32(unsigned long field, u32 value)
623{
624 vmcs_writel(field, value);
625}
626
627static void vmcs_write64(unsigned long field, u64 value)
628{
6aa8b732 629 vmcs_writel(field, value);
7682f2d0 630#ifndef CONFIG_X86_64
6aa8b732
AK		 631 asm volatile ("");
632 vmcs_writel(field+1, value >> 32);
633#endif
634}
635
2ab455cc
AL		 636static void vmcs_clear_bits(unsigned long field, u32 mask)
637{
638 vmcs_writel(field, vmcs_readl(field) & ~mask);
639}
640
641static void vmcs_set_bits(unsigned long field, u32 mask)
642{
643 vmcs_writel(field, vmcs_readl(field) | mask);
644}
645
abd3f2d6
AK		 646static void update_exception_bitmap(struct kvm_vcpu *vcpu)
647{
648 u32 eb;
649
fd7373cc
JK		 650 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
651 (1u << NM_VECTOR) | (1u << DB_VECTOR);
652 if ((vcpu->guest_debug &
653 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
654 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
655 eb |= 1u << BP_VECTOR;
7ffd92c5 656 if (to_vmx(vcpu)->rmode.vm86_active)
abd3f2d6 657 eb = ~0;
089d034e 658 if (enable_ept)
1439442c 659 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
02daab21
AK		 660 if (vcpu->fpu_active)
661 eb &= ~(1u << NM_VECTOR);
abd3f2d6
AK		 662 vmcs_write32(EXCEPTION_BITMAP, eb);
663}
664
61d2ef2c
AK		 665static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
666{
667 unsigned i;
668 struct msr_autoload *m = &vmx->msr_autoload;
669
110312c8
AK		 670 if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
671 vmcs_clear_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
672 vmcs_clear_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
673 return;
674 }
675
61d2ef2c
AK		 676 for (i = 0; i < m->nr; ++i)
677 if (m->guest[i].index == msr)
678 break;
679
680 if (i == m->nr)
681 return;
682 --m->nr;
683 m->guest[i] = m->guest[m->nr];
684 m->host[i] = m->host[m->nr];
685 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
686 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
687}
688
689static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
690 u64 guest_val, u64 host_val)
691{
692 unsigned i;
693 struct msr_autoload *m = &vmx->msr_autoload;
694
110312c8
AK		 695 if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
696 vmcs_write64(GUEST_IA32_EFER, guest_val);
697 vmcs_write64(HOST_IA32_EFER, host_val);
698 vmcs_set_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
699 vmcs_set_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
700 return;
701 }
702
61d2ef2c
AK		 703 for (i = 0; i < m->nr; ++i)
704 if (m->guest[i].index == msr)
705 break;
706
707 if (i == m->nr) {
708 ++m->nr;
709 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
710 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
711 }
712
713 m->guest[i].index = msr;
714 m->guest[i].value = guest_val;
715 m->host[i].index = msr;
716 m->host[i].value = host_val;
717}
718
33ed6329
AK		 719static void reload_tss(void)
720{
33ed6329
AK		 721 /*
722 * VT restores TR but not its size. Useless.
723 */
d359192f 724 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
a5f61300 725 struct desc_struct *descs;
33ed6329 726
d359192f 727 descs = (void *)gdt->address;
33ed6329
AK		 728 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
729 load_TR_desc();
33ed6329
AK		 730}
731
92c0d900 732static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
2cc51560 733{
3a34a881 734 u64 guest_efer;
51c6cf66
AK		 735 u64 ignore_bits;
736
f6801dff 737 guest_efer = vmx->vcpu.arch.efer;
3a34a881 738
51c6cf66
AK		 739 /*
740 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
741 * outside long mode
742 */
743 ignore_bits = EFER_NX | EFER_SCE;
744#ifdef CONFIG_X86_64
745 ignore_bits |= EFER_LMA | EFER_LME;
746 /* SCE is meaningful only in long mode on Intel */
747 if (guest_efer & EFER_LMA)
748 ignore_bits &= ~(u64)EFER_SCE;
749#endif
51c6cf66
AK		 750 guest_efer &= ~ignore_bits;
751 guest_efer |= host_efer & ignore_bits;
26bb0981 752 vmx->guest_msrs[efer_offset].data = guest_efer;
d5696725 753 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
84ad33ef
AK		 754
755 clear_atomic_switch_msr(vmx, MSR_EFER);
756 /* On ept, can't emulate nx, and must switch nx atomically */
757 if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
758 guest_efer = vmx->vcpu.arch.efer;
759 if (!(guest_efer & EFER_LMA))
760 guest_efer &= ~EFER_LME;
761 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
762 return false;
763 }
764
26bb0981 765 return true;
51c6cf66
AK		 766}
767
2d49ec72
GN		 768static unsigned long segment_base(u16 selector)
769{
d359192f 770 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
2d49ec72
GN		 771 struct desc_struct *d;
772 unsigned long table_base;
773 unsigned long v;
774
775 if (!(selector & ~3))
776 return 0;
777
d359192f 778 table_base = gdt->address;
2d49ec72
GN		 779
780 if (selector & 4) { /* from ldt */
781 u16 ldt_selector = kvm_read_ldt();
782
783 if (!(ldt_selector & ~3))
784 return 0;
785
786 table_base = segment_base(ldt_selector);
787 }
788 d = (struct desc_struct *)(table_base + (selector & ~7));
789 v = get_desc_base(d);
790#ifdef CONFIG_X86_64
791 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
792 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
793#endif
794 return v;
795}
796
797static inline unsigned long kvm_read_tr_base(void)
798{
799 u16 tr;
800 asm("str %0" : "=g"(tr));
801 return segment_base(tr);
802}
803
04d2cc77 804static void vmx_save_host_state(struct kvm_vcpu *vcpu)
33ed6329 805{
04d2cc77 806 struct vcpu_vmx *vmx = to_vmx(vcpu);
26bb0981 807 int i;
04d2cc77 808
a2fa3e9f 809 if (vmx->host_state.loaded)
33ed6329
AK		 810 return;
811
a2fa3e9f 812 vmx->host_state.loaded = 1;
33ed6329
AK		 813 /*
814 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
815 * allow segment selectors with cpl > 0 or ti == 1.
816 */
d6e88aec 817 vmx->host_state.ldt_sel = kvm_read_ldt();
152d3f2f 818 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
9581d442 819 savesegment(fs, vmx->host_state.fs_sel);
152d3f2f 820 if (!(vmx->host_state.fs_sel & 7)) {
a2fa3e9f 821 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
152d3f2f
LV		 822 vmx->host_state.fs_reload_needed = 0;
823 } else {
33ed6329 824 vmcs_write16(HOST_FS_SELECTOR, 0);
152d3f2f 825 vmx->host_state.fs_reload_needed = 1;
33ed6329 826 }
9581d442 827 savesegment(gs, vmx->host_state.gs_sel);
a2fa3e9f
GH		 828 if (!(vmx->host_state.gs_sel & 7))
829 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
33ed6329
AK		 830 else {
831 vmcs_write16(HOST_GS_SELECTOR, 0);
152d3f2f 832 vmx->host_state.gs_ldt_reload_needed = 1;
33ed6329
AK		 833 }
834
835#ifdef CONFIG_X86_64
836 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
837 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
838#else
a2fa3e9f
GH		 839 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
840 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
33ed6329 841#endif
707c0874
AK		 842
843#ifdef CONFIG_X86_64
c8770e7b
AK		 844 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
845 if (is_long_mode(&vmx->vcpu))
44ea2b17 846 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
707c0874 847#endif
26bb0981
AK		 848 for (i = 0; i < vmx->save_nmsrs; ++i)
849 kvm_set_shared_msr(vmx->guest_msrs[i].index,
d5696725
AK		 850 vmx->guest_msrs[i].data,
851 vmx->guest_msrs[i].mask);
33ed6329
AK		 852}
853
a9b21b62 854static void __vmx_load_host_state(struct vcpu_vmx *vmx)
33ed6329 855{
a2fa3e9f 856 if (!vmx->host_state.loaded)
33ed6329
AK		 857 return;
858
e1beb1d3 859 ++vmx->vcpu.stat.host_state_reload;
a2fa3e9f 860 vmx->host_state.loaded = 0;
c8770e7b
AK		 861#ifdef CONFIG_X86_64
862 if (is_long_mode(&vmx->vcpu))
863 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
864#endif
152d3f2f 865 if (vmx->host_state.gs_ldt_reload_needed) {
d6e88aec 866 kvm_load_ldt(vmx->host_state.ldt_sel);
33ed6329 867#ifdef CONFIG_X86_64
9581d442 868 load_gs_index(vmx->host_state.gs_sel);
9581d442
AK		 869#else
870 loadsegment(gs, vmx->host_state.gs_sel);
33ed6329 871#endif
33ed6329 872 }
0a77fe4c
AK		 873 if (vmx->host_state.fs_reload_needed)
874 loadsegment(fs, vmx->host_state.fs_sel);
152d3f2f 875 reload_tss();
44ea2b17 876#ifdef CONFIG_X86_64
c8770e7b 877 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
44ea2b17 878#endif
1c11e713
AK		 879 if (current_thread_info()->status & TS_USEDFPU)
880 clts();
3444d7da 881 load_gdt(&__get_cpu_var(host_gdt));
33ed6329
AK		 882}
883
a9b21b62
AK		 884static void vmx_load_host_state(struct vcpu_vmx *vmx)
885{
886 preempt_disable();
887 __vmx_load_host_state(vmx);
888 preempt_enable();
889}
890
6aa8b732
AK		 891/*
892 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
893 * vcpu mutex is already taken.
894 */
15ad7146 895static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
6aa8b732 896{
a2fa3e9f 897 struct vcpu_vmx *vmx = to_vmx(vcpu);
4610c9cc 898 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
6aa8b732 899
4610c9cc
DX		 900 if (!vmm_exclusive)
901 kvm_cpu_vmxon(phys_addr);
902 else if (vcpu->cpu != cpu)
8b9cf98c 903 vcpu_clear(vmx);
6aa8b732 904
a2fa3e9f 905 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
a2fa3e9f 906 per_cpu(current_vmcs, cpu) = vmx->vmcs;
7725b894 907 vmcs_load(vmx->vmcs);
6aa8b732
AK		 908 }
909
910 if (vcpu->cpu != cpu) {
d359192f 911 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
6aa8b732
AK		 912 unsigned long sysenter_esp;
913
a8eeb04a 914 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
92fe13be
DX		 915 local_irq_disable();
916 list_add(&vmx->local_vcpus_link,
917 &per_cpu(vcpus_on_cpu, cpu));
918 local_irq_enable();
919
6aa8b732
AK		 920 /*
921 * Linux uses per-cpu TSS and GDT, so set these when switching
922 * processors.
923 */
d6e88aec 924 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
d359192f 925 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
6aa8b732
AK		 926
927 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
928 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
929 }
6aa8b732
AK		 930}
931
932static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
933{
a9b21b62 934 __vmx_load_host_state(to_vmx(vcpu));
4610c9cc 935 if (!vmm_exclusive) {
b923e62e 936 __vcpu_clear(to_vmx(vcpu));
4610c9cc
DX		 937 kvm_cpu_vmxoff();
938 }
6aa8b732
AK		 939}
940
5fd86fcf
AK		 941static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
942{
81231c69
AK		 943 ulong cr0;
944
5fd86fcf
AK		 945 if (vcpu->fpu_active)
946 return;
947 vcpu->fpu_active = 1;
81231c69
AK		 948 cr0 = vmcs_readl(GUEST_CR0);
949 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
950 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
951 vmcs_writel(GUEST_CR0, cr0);
5fd86fcf 952 update_exception_bitmap(vcpu);
edcafe3c
AK		 953 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
954 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
5fd86fcf
AK		 955}
956
edcafe3c
AK		 957static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
958
5fd86fcf
AK		 959static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
960{
edcafe3c 961 vmx_decache_cr0_guest_bits(vcpu);
81231c69 962 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
5fd86fcf 963 update_exception_bitmap(vcpu);
edcafe3c
AK		 964 vcpu->arch.cr0_guest_owned_bits = 0;
965 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
966 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
5fd86fcf
AK		 967}
968
6aa8b732
AK		 969static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
970{
78ac8b47 971 unsigned long rflags, save_rflags;
345dcaa8
AK		 972
973 rflags = vmcs_readl(GUEST_RFLAGS);
78ac8b47
AK		 974 if (to_vmx(vcpu)->rmode.vm86_active) {
975 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
976 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
977 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
978 }
345dcaa8 979 return rflags;
6aa8b732
AK		 980}
981
982static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
983{
78ac8b47
AK		 984 if (to_vmx(vcpu)->rmode.vm86_active) {
985 to_vmx(vcpu)->rmode.save_rflags = rflags;
053de044 986 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
78ac8b47 987 }
6aa8b732
AK		 988 vmcs_writel(GUEST_RFLAGS, rflags);
989}
990
2809f5d2
GC		 991static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
992{
993 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
994 int ret = 0;
995
996 if (interruptibility & GUEST_INTR_STATE_STI)
48005f64 997 ret |= KVM_X86_SHADOW_INT_STI;
2809f5d2 998 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
48005f64 999 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2809f5d2
GC		 1000
1001 return ret & mask;
1002}
1003
1004static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1005{
1006 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1007 u32 interruptibility = interruptibility_old;
1008
1009 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1010
48005f64 1011 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2809f5d2 1012 interruptibility |= GUEST_INTR_STATE_MOV_SS;
48005f64 1013 else if (mask & KVM_X86_SHADOW_INT_STI)
2809f5d2
GC		 1014 interruptibility |= GUEST_INTR_STATE_STI;
1015
1016 if ((interruptibility != interruptibility_old))
1017 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1018}
1019
6aa8b732
AK		 1020static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1021{
1022 unsigned long rip;
6aa8b732 1023
5fdbf976 1024 rip = kvm_rip_read(vcpu);
6aa8b732 1025 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5fdbf976 1026 kvm_rip_write(vcpu, rip);
6aa8b732 1027
2809f5d2
GC		 1028 /* skipping an emulated instruction also counts */
1029 vmx_set_interrupt_shadow(vcpu, 0);
6aa8b732
AK		 1030}
1031
443381a8
AL		 1032static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1033{
1034 /* Ensure that we clear the HLT state in the VMCS. We don't need to
1035 * explicitly skip the instruction because if the HLT state is set, then
1036 * the instruction is already executing and RIP has already been
1037 * advanced. */
1038 if (!yield_on_hlt &&
1039 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1040 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1041}
1042
298101da 1043static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
ce7ddec4
JR		 1044 bool has_error_code, u32 error_code,
1045 bool reinject)
298101da 1046{
77ab6db0 1047 struct vcpu_vmx *vmx = to_vmx(vcpu);
8ab2d2e2 1048 u32 intr_info = nr | INTR_INFO_VALID_MASK;
77ab6db0 1049
8ab2d2e2 1050 if (has_error_code) {
77ab6db0 1051 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
8ab2d2e2
JK		 1052 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1053 }
77ab6db0 1054
7ffd92c5 1055 if (vmx->rmode.vm86_active) {
a92601bb
MG		 1056 if (kvm_inject_realmode_interrupt(vcpu, nr) != EMULATE_DONE)
1057 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
77ab6db0
JK		 1058 return;
1059 }
1060
66fd3f7f
GN		 1061 if (kvm_exception_is_soft(nr)) {
1062 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1063 vmx->vcpu.arch.event_exit_inst_len);
8ab2d2e2
JK		 1064 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1065 } else
1066 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1067
1068 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
443381a8 1069 vmx_clear_hlt(vcpu);
298101da
AK		 1070}
1071
4e47c7a6
SY		 1072static bool vmx_rdtscp_supported(void)
1073{
1074 return cpu_has_vmx_rdtscp();
1075}
1076
a75beee6
ED		 1077/*
1078 * Swap MSR entry in host/guest MSR entry array.
1079 */
8b9cf98c 1080static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
a75beee6 1081{
26bb0981 1082 struct shared_msr_entry tmp;
a2fa3e9f
GH		 1083
1084 tmp = vmx->guest_msrs[to];
1085 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1086 vmx->guest_msrs[from] = tmp;
a75beee6
ED		 1087}
1088
e38aea3e
AK		 1089/*
1090 * Set up the vmcs to automatically save and restore system
1091 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1092 * mode, as fiddling with msrs is very expensive.
1093 */
8b9cf98c 1094static void setup_msrs(struct vcpu_vmx *vmx)
e38aea3e 1095{
26bb0981 1096 int save_nmsrs, index;
5897297b 1097 unsigned long *msr_bitmap;
e38aea3e 1098
33f9c505 1099 vmx_load_host_state(vmx);
a75beee6
ED		 1100 save_nmsrs = 0;
1101#ifdef CONFIG_X86_64
8b9cf98c 1102 if (is_long_mode(&vmx->vcpu)) {
8b9cf98c 1103 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
a75beee6 1104 if (index >= 0)
8b9cf98c
RR		 1105 move_msr_up(vmx, index, save_nmsrs++);
1106 index = __find_msr_index(vmx, MSR_LSTAR);
a75beee6 1107 if (index >= 0)
8b9cf98c
RR		 1108 move_msr_up(vmx, index, save_nmsrs++);
1109 index = __find_msr_index(vmx, MSR_CSTAR);
a75beee6 1110 if (index >= 0)
8b9cf98c 1111 move_msr_up(vmx, index, save_nmsrs++);
4e47c7a6
SY		 1112 index = __find_msr_index(vmx, MSR_TSC_AUX);
1113 if (index >= 0 && vmx->rdtscp_enabled)
1114 move_msr_up(vmx, index, save_nmsrs++);
a75beee6 1115 /*
8c06585d 1116 * MSR_STAR is only needed on long mode guests, and only
a75beee6
ED		 1117 * if efer.sce is enabled.
1118 */
8c06585d 1119 index = __find_msr_index(vmx, MSR_STAR);
f6801dff 1120 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
8b9cf98c 1121 move_msr_up(vmx, index, save_nmsrs++);
a75beee6
ED		 1122 }
1123#endif
92c0d900
AK		 1124 index = __find_msr_index(vmx, MSR_EFER);
1125 if (index >= 0 && update_transition_efer(vmx, index))
26bb0981 1126 move_msr_up(vmx, index, save_nmsrs++);
e38aea3e 1127
26bb0981 1128 vmx->save_nmsrs = save_nmsrs;
5897297b
AK		 1129
1130 if (cpu_has_vmx_msr_bitmap()) {
1131 if (is_long_mode(&vmx->vcpu))
1132 msr_bitmap = vmx_msr_bitmap_longmode;
1133 else
1134 msr_bitmap = vmx_msr_bitmap_legacy;
1135
1136 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1137 }
e38aea3e
AK		 1138}
1139
6aa8b732
AK		 1140/*
1141 * reads and returns guest's timestamp counter "register"
1142 * guest_tsc = host_tsc + tsc_offset -- 21.3
1143 */
1144static u64 guest_read_tsc(void)
1145{
1146 u64 host_tsc, tsc_offset;
1147
1148 rdtscll(host_tsc);
1149 tsc_offset = vmcs_read64(TSC_OFFSET);
1150 return host_tsc + tsc_offset;
1151}
1152
1153/*
99e3e30a 1154 * writes 'offset' into guest's timestamp counter offset register
6aa8b732 1155 */
99e3e30a 1156static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
6aa8b732 1157{
f4e1b3c8 1158 vmcs_write64(TSC_OFFSET, offset);
6aa8b732
AK		 1159}
1160
e48672fa
ZA		 1161static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
1162{
1163 u64 offset = vmcs_read64(TSC_OFFSET);
1164 vmcs_write64(TSC_OFFSET, offset + adjustment);
1165}
1166
6aa8b732
AK		 1167/*
1168 * Reads an msr value (of 'msr_index') into 'pdata'.
1169 * Returns 0 on success, non-0 otherwise.
1170 * Assumes vcpu_load() was already called.
1171 */
1172static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1173{
1174 u64 data;
26bb0981 1175 struct shared_msr_entry *msr;
6aa8b732
AK		 1176
1177 if (!pdata) {
1178 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
1179 return -EINVAL;
1180 }
1181
1182 switch (msr_index) {
05b3e0c2 1183#ifdef CONFIG_X86_64
6aa8b732
AK		 1184 case MSR_FS_BASE:
1185 data = vmcs_readl(GUEST_FS_BASE);
1186 break;
1187 case MSR_GS_BASE:
1188 data = vmcs_readl(GUEST_GS_BASE);
1189 break;
44ea2b17
AK		 1190 case MSR_KERNEL_GS_BASE:
1191 vmx_load_host_state(to_vmx(vcpu));
1192 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
1193 break;
26bb0981 1194#endif
6aa8b732 1195 case MSR_EFER:
3bab1f5d 1196 return kvm_get_msr_common(vcpu, msr_index, pdata);
af24a4e4 1197 case MSR_IA32_TSC:
6aa8b732
AK		 1198 data = guest_read_tsc();
1199 break;
1200 case MSR_IA32_SYSENTER_CS:
1201 data = vmcs_read32(GUEST_SYSENTER_CS);
1202 break;
1203 case MSR_IA32_SYSENTER_EIP:
f5b42c33 1204 data = vmcs_readl(GUEST_SYSENTER_EIP);
6aa8b732
AK		 1205 break;
1206 case MSR_IA32_SYSENTER_ESP:
f5b42c33 1207 data = vmcs_readl(GUEST_SYSENTER_ESP);
6aa8b732 1208 break;
4e47c7a6
SY		 1209 case MSR_TSC_AUX:
1210 if (!to_vmx(vcpu)->rdtscp_enabled)
1211 return 1;
1212 /* Otherwise falls through */
6aa8b732 1213 default:
26bb0981 1214 vmx_load_host_state(to_vmx(vcpu));
8b9cf98c 1215 msr = find_msr_entry(to_vmx(vcpu), msr_index);
3bab1f5d 1216 if (msr) {
542423b0 1217 vmx_load_host_state(to_vmx(vcpu));
3bab1f5d
AK		 1218 data = msr->data;
1219 break;
6aa8b732 1220 }
3bab1f5d 1221 return kvm_get_msr_common(vcpu, msr_index, pdata);
6aa8b732
AK		 1222 }
1223
1224 *pdata = data;
1225 return 0;
1226}
1227
1228/*
1229 * Writes msr value into into the appropriate "register".
1230 * Returns 0 on success, non-0 otherwise.
1231 * Assumes vcpu_load() was already called.
1232 */
1233static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1234{
a2fa3e9f 1235 struct vcpu_vmx *vmx = to_vmx(vcpu);
26bb0981 1236 struct shared_msr_entry *msr;
2cc51560
ED		 1237 int ret = 0;
1238
6aa8b732 1239 switch (msr_index) {
3bab1f5d 1240 case MSR_EFER:
a9b21b62 1241 vmx_load_host_state(vmx);
2cc51560 1242 ret = kvm_set_msr_common(vcpu, msr_index, data);
2cc51560 1243 break;
16175a79 1244#ifdef CONFIG_X86_64
6aa8b732
AK		 1245 case MSR_FS_BASE:
1246 vmcs_writel(GUEST_FS_BASE, data);
1247 break;
1248 case MSR_GS_BASE:
1249 vmcs_writel(GUEST_GS_BASE, data);
1250 break;
44ea2b17
AK		 1251 case MSR_KERNEL_GS_BASE:
1252 vmx_load_host_state(vmx);
1253 vmx->msr_guest_kernel_gs_base = data;
1254 break;
6aa8b732
AK		 1255#endif
1256 case MSR_IA32_SYSENTER_CS:
1257 vmcs_write32(GUEST_SYSENTER_CS, data);
1258 break;
1259 case MSR_IA32_SYSENTER_EIP:
f5b42c33 1260 vmcs_writel(GUEST_SYSENTER_EIP, data);
6aa8b732
AK		 1261 break;
1262 case MSR_IA32_SYSENTER_ESP:
f5b42c33 1263 vmcs_writel(GUEST_SYSENTER_ESP, data);
6aa8b732 1264 break;
af24a4e4 1265 case MSR_IA32_TSC:
99e3e30a 1266 kvm_write_tsc(vcpu, data);
6aa8b732 1267 break;
468d472f
SY		 1268 case MSR_IA32_CR_PAT:
1269 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1270 vmcs_write64(GUEST_IA32_PAT, data);
1271 vcpu->arch.pat = data;
1272 break;
1273 }
4e47c7a6
SY		 1274 ret = kvm_set_msr_common(vcpu, msr_index, data);
1275 break;
1276 case MSR_TSC_AUX:
1277 if (!vmx->rdtscp_enabled)
1278 return 1;
1279 /* Check reserved bit, higher 32 bits should be zero */
1280 if ((data >> 32) != 0)
1281 return 1;
1282 /* Otherwise falls through */
6aa8b732 1283 default:
8b9cf98c 1284 msr = find_msr_entry(vmx, msr_index);
3bab1f5d 1285 if (msr) {
542423b0 1286 vmx_load_host_state(vmx);
3bab1f5d
AK		 1287 msr->data = data;
1288 break;
6aa8b732 1289 }
2cc51560 1290 ret = kvm_set_msr_common(vcpu, msr_index, data);
6aa8b732
AK		 1291 }
1292
2cc51560 1293 return ret;
6aa8b732
AK		 1294}
1295
5fdbf976 1296static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
6aa8b732 1297{
5fdbf976
MT		 1298 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1299 switch (reg) {
1300 case VCPU_REGS_RSP:
1301 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1302 break;
1303 case VCPU_REGS_RIP:
1304 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1305 break;
6de4f3ad
AK		 1306 case VCPU_EXREG_PDPTR:
1307 if (enable_ept)
1308 ept_save_pdptrs(vcpu);
1309 break;
5fdbf976
MT		 1310 default:
1311 break;
1312 }
6aa8b732
AK		 1313}
1314
355be0b9 1315static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
6aa8b732 1316{
ae675ef0
JK		 1317 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1318 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1319 else
1320 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1321
abd3f2d6 1322 update_exception_bitmap(vcpu);
6aa8b732
AK		 1323}
1324
1325static __init int cpu_has_kvm_support(void)
1326{
6210e37b 1327 return cpu_has_vmx();
6aa8b732
AK		 1328}
1329
1330static __init int vmx_disabled_by_bios(void)
1331{
1332 u64 msr;
1333
1334 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
cafd6659 1335 if (msr & FEATURE_CONTROL_LOCKED) {
23f3e991 1336 /* launched w/ TXT and VMX disabled */
cafd6659
SW		 1337 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
1338 && tboot_enabled())
1339 return 1;
23f3e991 1340 /* launched w/o TXT and VMX only enabled w/ TXT */
cafd6659 1341 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
23f3e991 1342 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
f9335afe
SW		 1343 && !tboot_enabled()) {
1344 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
23f3e991 1345 "activate TXT before enabling KVM\n");
cafd6659 1346 return 1;
f9335afe 1347 }
23f3e991
JC		 1348 /* launched w/o TXT and VMX disabled */
1349 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
1350 && !tboot_enabled())
1351 return 1;
cafd6659
SW		 1352 }
1353
1354 return 0;
6aa8b732
AK		 1355}
1356
7725b894
DX		 1357static void kvm_cpu_vmxon(u64 addr)
1358{
1359 asm volatile (ASM_VMX_VMXON_RAX
1360 : : "a"(&addr), "m"(addr)
1361 : "memory", "cc");
1362}
1363
10474ae8 1364static int hardware_enable(void *garbage)
6aa8b732
AK		 1365{
1366 int cpu = raw_smp_processor_id();
1367 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
cafd6659 1368 u64 old, test_bits;
6aa8b732 1369
10474ae8
AG		 1370 if (read_cr4() & X86_CR4_VMXE)
1371 return -EBUSY;
1372
543e4243 1373 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
6aa8b732 1374 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
cafd6659
SW		 1375
1376 test_bits = FEATURE_CONTROL_LOCKED;
1377 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
1378 if (tboot_enabled())
1379 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
1380
1381 if ((old & test_bits) != test_bits) {
6aa8b732 1382 /* enable and lock */
cafd6659
SW		 1383 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
1384 }
66aee91a 1385 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
10474ae8 1386
4610c9cc
DX		 1387 if (vmm_exclusive) {
1388 kvm_cpu_vmxon(phys_addr);
1389 ept_sync_global();
1390 }
10474ae8 1391
3444d7da
AK		 1392 store_gdt(&__get_cpu_var(host_gdt));
1393
10474ae8 1394 return 0;
6aa8b732
AK		 1395}
1396
543e4243
AK		 1397static void vmclear_local_vcpus(void)
1398{
1399 int cpu = raw_smp_processor_id();
1400 struct vcpu_vmx *vmx, *n;
1401
1402 list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1403 local_vcpus_link)
1404 __vcpu_clear(vmx);
1405}
1406
710ff4a8
EH		 1407
1408/* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1409 * tricks.
1410 */
1411static void kvm_cpu_vmxoff(void)
6aa8b732 1412{
4ecac3fd 1413 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
6aa8b732
AK		 1414}
1415
710ff4a8
EH		 1416static void hardware_disable(void *garbage)
1417{
4610c9cc
DX		 1418 if (vmm_exclusive) {
1419 vmclear_local_vcpus();
1420 kvm_cpu_vmxoff();
1421 }
7725b894 1422 write_cr4(read_cr4() & ~X86_CR4_VMXE);
710ff4a8
EH		 1423}
1424
1c3d14fe 1425static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
d77c26fc 1426 u32 msr, u32 *result)
1c3d14fe
YS		 1427{
1428 u32 vmx_msr_low, vmx_msr_high;
1429 u32 ctl = ctl_min | ctl_opt;
1430
1431 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1432
1433 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1434 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
1435
1436 /* Ensure minimum (required) set of control bits are supported. */
1437 if (ctl_min & ~ctl)
002c7f7c 1438 return -EIO;
1c3d14fe
YS		 1439
1440 *result = ctl;
1441 return 0;
1442}
1443
110312c8
AK		 1444static __init bool allow_1_setting(u32 msr, u32 ctl)
1445{
1446 u32 vmx_msr_low, vmx_msr_high;
1447
1448 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1449 return vmx_msr_high & ctl;
1450}
1451
002c7f7c 1452static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
6aa8b732
AK		 1453{
1454 u32 vmx_msr_low, vmx_msr_high;
d56f546d 1455 u32 min, opt, min2, opt2;
1c3d14fe
YS		 1456 u32 _pin_based_exec_control = 0;
1457 u32 _cpu_based_exec_control = 0;
f78e0e2e 1458 u32 _cpu_based_2nd_exec_control = 0;
1c3d14fe
YS		 1459 u32 _vmexit_control = 0;
1460 u32 _vmentry_control = 0;
1461
1462 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
f08864b4 1463 opt = PIN_BASED_VIRTUAL_NMIS;
1c3d14fe
YS		 1464 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1465 &_pin_based_exec_control) < 0)
002c7f7c 1466 return -EIO;
1c3d14fe 1467
443381a8 1468 min =
1c3d14fe
YS		 1469#ifdef CONFIG_X86_64
1470 CPU_BASED_CR8_LOAD_EXITING |
1471 CPU_BASED_CR8_STORE_EXITING |
1472#endif
d56f546d
SY		 1473 CPU_BASED_CR3_LOAD_EXITING |
1474 CPU_BASED_CR3_STORE_EXITING |
1c3d14fe
YS		 1475 CPU_BASED_USE_IO_BITMAPS |
1476 CPU_BASED_MOV_DR_EXITING |
a7052897 1477 CPU_BASED_USE_TSC_OFFSETING |
59708670
SY		 1478 CPU_BASED_MWAIT_EXITING |
1479 CPU_BASED_MONITOR_EXITING |
a7052897 1480 CPU_BASED_INVLPG_EXITING;
443381a8
AL		 1481
1482 if (yield_on_hlt)
1483 min |= CPU_BASED_HLT_EXITING;
1484
f78e0e2e 1485 opt = CPU_BASED_TPR_SHADOW |
25c5f225 1486 CPU_BASED_USE_MSR_BITMAPS |
f78e0e2e 1487 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1c3d14fe
YS		 1488 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1489 &_cpu_based_exec_control) < 0)
002c7f7c 1490 return -EIO;
6e5d865c
YS		 1491#ifdef CONFIG_X86_64
1492 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1493 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1494 ~CPU_BASED_CR8_STORE_EXITING;
1495#endif
f78e0e2e 1496 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
d56f546d
SY		 1497 min2 = 0;
1498 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2384d2b3 1499 SECONDARY_EXEC_WBINVD_EXITING |
d56f546d 1500 SECONDARY_EXEC_ENABLE_VPID |
3a624e29 1501 SECONDARY_EXEC_ENABLE_EPT |
4b8d54f9 1502 SECONDARY_EXEC_UNRESTRICTED_GUEST |
4e47c7a6
SY		 1503 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
1504 SECONDARY_EXEC_RDTSCP;
d56f546d
SY		 1505 if (adjust_vmx_controls(min2, opt2,
1506 MSR_IA32_VMX_PROCBASED_CTLS2,
f78e0e2e
SY		 1507 &_cpu_based_2nd_exec_control) < 0)
1508 return -EIO;
1509 }
1510#ifndef CONFIG_X86_64
1511 if (!(_cpu_based_2nd_exec_control &
1512 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1513 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1514#endif
d56f546d 1515 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
a7052897
MT		 1516 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1517 enabled */
5fff7d27
GN		 1518 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
1519 CPU_BASED_CR3_STORE_EXITING |
1520 CPU_BASED_INVLPG_EXITING);
d56f546d
SY		 1521 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1522 vmx_capability.ept, vmx_capability.vpid);
1523 }
1c3d14fe
YS		 1524
1525 min = 0;
1526#ifdef CONFIG_X86_64
1527 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1528#endif
468d472f 1529 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1c3d14fe
YS		 1530 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1531 &_vmexit_control) < 0)
002c7f7c 1532 return -EIO;
1c3d14fe 1533
468d472f
SY		 1534 min = 0;
1535 opt = VM_ENTRY_LOAD_IA32_PAT;
1c3d14fe
YS		 1536 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1537 &_vmentry_control) < 0)
002c7f7c 1538 return -EIO;
6aa8b732 1539
c68876fd 1540 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1c3d14fe
YS		 1541
1542 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1543 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
002c7f7c 1544 return -EIO;
1c3d14fe
YS		 1545
1546#ifdef CONFIG_X86_64
1547 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1548 if (vmx_msr_high & (1u<<16))
002c7f7c 1549 return -EIO;
1c3d14fe
YS		 1550#endif
1551
1552 /* Require Write-Back (WB) memory type for VMCS accesses. */
1553 if (((vmx_msr_high >> 18) & 15) != 6)
002c7f7c 1554 return -EIO;
1c3d14fe 1555
002c7f7c
YS		 1556 vmcs_conf->size = vmx_msr_high & 0x1fff;
1557 vmcs_conf->order = get_order(vmcs_config.size);
1558 vmcs_conf->revision_id = vmx_msr_low;
1c3d14fe 1559
002c7f7c
YS		 1560 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1561 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
f78e0e2e 1562 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
002c7f7c
YS		 1563 vmcs_conf->vmexit_ctrl = _vmexit_control;
1564 vmcs_conf->vmentry_ctrl = _vmentry_control;
1c3d14fe 1565
110312c8
AK		 1566 cpu_has_load_ia32_efer =
1567 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
1568 VM_ENTRY_LOAD_IA32_EFER)
1569 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
1570 VM_EXIT_LOAD_IA32_EFER);
1571
1c3d14fe 1572 return 0;
c68876fd 1573}
6aa8b732
AK		 1574
1575static struct vmcs *alloc_vmcs_cpu(int cpu)
1576{
1577 int node = cpu_to_node(cpu);
1578 struct page *pages;
1579 struct vmcs *vmcs;
1580
6484eb3e 1581 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
6aa8b732
AK		 1582 if (!pages)
1583 return NULL;
1584 vmcs = page_address(pages);
1c3d14fe
YS		 1585 memset(vmcs, 0, vmcs_config.size);
1586 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
6aa8b732
AK		 1587 return vmcs;
1588}
1589
1590static struct vmcs *alloc_vmcs(void)
1591{
d3b2c338 1592 return alloc_vmcs_cpu(raw_smp_processor_id());
6aa8b732
AK		 1593}
1594
1595static void free_vmcs(struct vmcs *vmcs)
1596{
1c3d14fe 1597 free_pages((unsigned long)vmcs, vmcs_config.order);
6aa8b732
AK		 1598}
1599
39959588 1600static void free_kvm_area(void)
6aa8b732
AK		 1601{
1602 int cpu;
1603
3230bb47 1604 for_each_possible_cpu(cpu) {
6aa8b732 1605 free_vmcs(per_cpu(vmxarea, cpu));
3230bb47
ZA		 1606 per_cpu(vmxarea, cpu) = NULL;
1607 }
6aa8b732
AK		 1608}
1609
6aa8b732
AK		 1610static __init int alloc_kvm_area(void)
1611{
1612 int cpu;
1613
3230bb47 1614 for_each_possible_cpu(cpu) {
6aa8b732
AK		 1615 struct vmcs *vmcs;
1616
1617 vmcs = alloc_vmcs_cpu(cpu);
1618 if (!vmcs) {
1619 free_kvm_area();
1620 return -ENOMEM;
1621 }
1622
1623 per_cpu(vmxarea, cpu) = vmcs;
1624 }
1625 return 0;
1626}
1627
1628static __init int hardware_setup(void)
1629{
002c7f7c
YS		 1630 if (setup_vmcs_config(&vmcs_config) < 0)
1631 return -EIO;
50a37eb4
JR		 1632
1633 if (boot_cpu_has(X86_FEATURE_NX))
1634 kvm_enable_efer_bits(EFER_NX);
1635
93ba03c2
SY		 1636 if (!cpu_has_vmx_vpid())
1637 enable_vpid = 0;
1638
4bc9b982
SY		 1639 if (!cpu_has_vmx_ept() ||
1640 !cpu_has_vmx_ept_4levels()) {
93ba03c2 1641 enable_ept = 0;
3a624e29
NK		 1642 enable_unrestricted_guest = 0;
1643 }
1644
1645 if (!cpu_has_vmx_unrestricted_guest())
1646 enable_unrestricted_guest = 0;
93ba03c2
SY		 1647
1648 if (!cpu_has_vmx_flexpriority())
1649 flexpriority_enabled = 0;
1650
95ba8273
GN		 1651 if (!cpu_has_vmx_tpr_shadow())
1652 kvm_x86_ops->update_cr8_intercept = NULL;
1653
54dee993
MT		 1654 if (enable_ept && !cpu_has_vmx_ept_2m_page())
1655 kvm_disable_largepages();
1656
4b8d54f9
ZE		 1657 if (!cpu_has_vmx_ple())
1658 ple_gap = 0;
1659
6aa8b732
AK		 1660 return alloc_kvm_area();
1661}
1662
1663static __exit void hardware_unsetup(void)
1664{
1665 free_kvm_area();
1666}
1667
6aa8b732
AK		 1668static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1669{
1670 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1671
6af11b9e 1672 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
6aa8b732
AK		 1673 vmcs_write16(sf->selector, save->selector);
1674 vmcs_writel(sf->base, save->base);
1675 vmcs_write32(sf->limit, save->limit);
1676 vmcs_write32(sf->ar_bytes, save->ar);
1677 } else {
1678 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1679 << AR_DPL_SHIFT;
1680 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1681 }
1682}
1683
1684static void enter_pmode(struct kvm_vcpu *vcpu)
1685{
1686 unsigned long flags;
a89a8fb9 1687 struct vcpu_vmx *vmx = to_vmx(vcpu);
6aa8b732 1688
a89a8fb9 1689 vmx->emulation_required = 1;
7ffd92c5 1690 vmx->rmode.vm86_active = 0;
6aa8b732 1691
d0ba64f9 1692 vmcs_write16(GUEST_TR_SELECTOR, vmx->rmode.tr.selector);
7ffd92c5
AK		 1693 vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1694 vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1695 vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
6aa8b732
AK		 1696
1697 flags = vmcs_readl(GUEST_RFLAGS);
78ac8b47
AK		 1698 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1699 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
6aa8b732
AK		 1700 vmcs_writel(GUEST_RFLAGS, flags);
1701
66aee91a
RR		 1702 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1703 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
6aa8b732
AK		 1704
1705 update_exception_bitmap(vcpu);
1706
a89a8fb9
MG		 1707 if (emulate_invalid_guest_state)
1708 return;
1709
7ffd92c5
AK		 1710 fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1711 fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1712 fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1713 fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
6aa8b732
AK		 1714
1715 vmcs_write16(GUEST_SS_SELECTOR, 0);
1716 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1717
1718 vmcs_write16(GUEST_CS_SELECTOR,
1719 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1720 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1721}
1722
d77c26fc 1723static gva_t rmode_tss_base(struct kvm *kvm)
6aa8b732 1724{
bfc6d222 1725 if (!kvm->arch.tss_addr) {
bc6678a3
MT		 1726 struct kvm_memslots *slots;
1727 gfn_t base_gfn;
1728
90d83dc3 1729 slots = kvm_memslots(kvm);
f495c6e5 1730 base_gfn = slots->memslots[0].base_gfn +
46a26bf5 1731 kvm->memslots->memslots[0].npages - 3;
cbc94022
IE		 1732 return base_gfn << PAGE_SHIFT;
1733 }
bfc6d222 1734 return kvm->arch.tss_addr;
6aa8b732
AK		 1735}
1736
1737static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1738{
1739 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1740
1741 save->selector = vmcs_read16(sf->selector);
1742 save->base = vmcs_readl(sf->base);
1743 save->limit = vmcs_read32(sf->limit);
1744 save->ar = vmcs_read32(sf->ar_bytes);
15b00f32 1745 vmcs_write16(sf->selector, save->base >> 4);
444e863d 1746 vmcs_write32(sf->base, save->base & 0xffff0);
6aa8b732
AK		 1747 vmcs_write32(sf->limit, 0xffff);
1748 vmcs_write32(sf->ar_bytes, 0xf3);
444e863d
GN		 1749 if (save->base & 0xf)
1750 printk_once(KERN_WARNING "kvm: segment base is not paragraph"
1751 " aligned when entering protected mode (seg=%d)",
1752 seg);
6aa8b732
AK		 1753}
1754
1755static void enter_rmode(struct kvm_vcpu *vcpu)
1756{
1757 unsigned long flags;
a89a8fb9 1758 struct vcpu_vmx *vmx = to_vmx(vcpu);
6aa8b732 1759
3a624e29
NK		 1760 if (enable_unrestricted_guest)
1761 return;
1762
a89a8fb9 1763 vmx->emulation_required = 1;
7ffd92c5 1764 vmx->rmode.vm86_active = 1;
6aa8b732 1765
776e58ea
GN		 1766 /*
1767 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
1768 * vcpu. Call it here with phys address pointing 16M below 4G.
1769 */
1770 if (!vcpu->kvm->arch.tss_addr) {
1771 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
1772 "called before entering vcpu\n");
1773 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1774 vmx_set_tss_addr(vcpu->kvm, 0xfeffd000);
1775 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1776 }
1777
d0ba64f9 1778 vmx->rmode.tr.selector = vmcs_read16(GUEST_TR_SELECTOR);
7ffd92c5 1779 vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
6aa8b732
AK		 1780 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1781
7ffd92c5 1782 vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
6aa8b732
AK		 1783 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1784
7ffd92c5 1785 vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
6aa8b732
AK		 1786 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1787
1788 flags = vmcs_readl(GUEST_RFLAGS);
78ac8b47 1789 vmx->rmode.save_rflags = flags;
6aa8b732 1790
053de044 1791 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
6aa8b732
AK		 1792
1793 vmcs_writel(GUEST_RFLAGS, flags);
66aee91a 1794 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
6aa8b732
AK		 1795 update_exception_bitmap(vcpu);
1796
a89a8fb9
MG		 1797 if (emulate_invalid_guest_state)
1798 goto continue_rmode;
1799
6aa8b732
AK		 1800 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1801 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1802 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1803
1804 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
abacf8df 1805 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
8cb5b033
AK		 1806 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1807 vmcs_writel(GUEST_CS_BASE, 0xf0000);
6aa8b732
AK		 1808 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1809
7ffd92c5
AK		 1810 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1811 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1812 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1813 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
75880a01 1814
a89a8fb9 1815continue_rmode:
8668a3c4 1816 kvm_mmu_reset_context(vcpu);
6aa8b732
AK		 1817}
1818
401d10de
AS		 1819static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1820{
1821 struct vcpu_vmx *vmx = to_vmx(vcpu);
26bb0981
AK		 1822 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1823
1824 if (!msr)
1825 return;
401d10de 1826
44ea2b17
AK		 1827 /*
1828 * Force kernel_gs_base reloading before EFER changes, as control
1829 * of this msr depends on is_long_mode().
1830 */
1831 vmx_load_host_state(to_vmx(vcpu));
f6801dff 1832 vcpu->arch.efer = efer;
401d10de
AS		 1833 if (efer & EFER_LMA) {
1834 vmcs_write32(VM_ENTRY_CONTROLS,
1835 vmcs_read32(VM_ENTRY_CONTROLS) |
1836 VM_ENTRY_IA32E_MODE);
1837 msr->data = efer;
1838 } else {
1839 vmcs_write32(VM_ENTRY_CONTROLS,
1840 vmcs_read32(VM_ENTRY_CONTROLS) &
1841 ~VM_ENTRY_IA32E_MODE);
1842
1843 msr->data = efer & ~EFER_LME;
1844 }
1845 setup_msrs(vmx);
1846}
1847
05b3e0c2 1848#ifdef CONFIG_X86_64
6aa8b732
AK		 1849
1850static void enter_lmode(struct kvm_vcpu *vcpu)
1851{
1852 u32 guest_tr_ar;
1853
1854 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1855 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1856 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
b8688d51 1857 __func__);
6aa8b732
AK		 1858 vmcs_write32(GUEST_TR_AR_BYTES,
1859 (guest_tr_ar & ~AR_TYPE_MASK)
1860 | AR_TYPE_BUSY_64_TSS);
1861 }
da38f438 1862 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
6aa8b732
AK		 1863}
1864
1865static void exit_lmode(struct kvm_vcpu *vcpu)
1866{
6aa8b732
AK		 1867 vmcs_write32(VM_ENTRY_CONTROLS,
1868 vmcs_read32(VM_ENTRY_CONTROLS)
1e4e6e00 1869 & ~VM_ENTRY_IA32E_MODE);
da38f438 1870 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
6aa8b732
AK		 1871}
1872
1873#endif
1874
2384d2b3
SY		 1875static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1876{
b9d762fa 1877 vpid_sync_context(to_vmx(vcpu));
dd180b3e
XG		 1878 if (enable_ept) {
1879 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
1880 return;
4e1096d2 1881 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
dd180b3e 1882 }
2384d2b3
SY		 1883}
1884
e8467fda
AK		 1885static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1886{
1887 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
1888
1889 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
1890 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
1891}
1892
aff48baa
AK		 1893static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
1894{
1895 if (enable_ept && is_paging(vcpu))
1896 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
1897 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
1898}
1899
25c4c276 1900static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
399badf3 1901{
fc78f519
AK		 1902 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
1903
1904 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
1905 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
399badf3
AK		 1906}
1907
1439442c
SY		 1908static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1909{
6de4f3ad
AK		 1910 if (!test_bit(VCPU_EXREG_PDPTR,
1911 (unsigned long *)&vcpu->arch.regs_dirty))
1912 return;
1913
1439442c 1914 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
ff03a073
JR		 1915 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
1916 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
1917 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
1918 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
1439442c
SY		 1919 }
1920}
1921
8f5d549f
AK		 1922static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1923{
1924 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
ff03a073
JR		 1925 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1926 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1927 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1928 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
8f5d549f 1929 }
6de4f3ad
AK		 1930
1931 __set_bit(VCPU_EXREG_PDPTR,
1932 (unsigned long *)&vcpu->arch.regs_avail);
1933 __set_bit(VCPU_EXREG_PDPTR,
1934 (unsigned long *)&vcpu->arch.regs_dirty);
8f5d549f
AK		 1935}
1936
1439442c
SY		 1937static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1938
1939static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1940 unsigned long cr0,
1941 struct kvm_vcpu *vcpu)
1942{
aff48baa 1943 vmx_decache_cr3(vcpu);
1439442c
SY		 1944 if (!(cr0 & X86_CR0_PG)) {
1945 /* From paging/starting to nonpaging */
1946 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
65267ea1 1947 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1439442c
SY		 1948 (CPU_BASED_CR3_LOAD_EXITING |
1949 CPU_BASED_CR3_STORE_EXITING));
1950 vcpu->arch.cr0 = cr0;
fc78f519 1951 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1439442c
SY		 1952 } else if (!is_paging(vcpu)) {
1953 /* From nonpaging to paging */
1954 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
65267ea1 1955 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1439442c
SY		 1956 ~(CPU_BASED_CR3_LOAD_EXITING |
1957 CPU_BASED_CR3_STORE_EXITING));
1958 vcpu->arch.cr0 = cr0;
fc78f519 1959 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1439442c 1960 }
95eb84a7
SY		 1961
1962 if (!(cr0 & X86_CR0_WP))
1963 *hw_cr0 &= ~X86_CR0_WP;
1439442c
SY		 1964}
1965
6aa8b732
AK		 1966static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1967{
7ffd92c5 1968 struct vcpu_vmx *vmx = to_vmx(vcpu);
3a624e29
NK		 1969 unsigned long hw_cr0;
1970
1971 if (enable_unrestricted_guest)
1972 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
1973 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
1974 else
1975 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
1439442c 1976
7ffd92c5 1977 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
6aa8b732
AK		 1978 enter_pmode(vcpu);
1979
7ffd92c5 1980 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
6aa8b732
AK		 1981 enter_rmode(vcpu);
1982
05b3e0c2 1983#ifdef CONFIG_X86_64
f6801dff 1984 if (vcpu->arch.efer & EFER_LME) {
707d92fa 1985 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
6aa8b732 1986 enter_lmode(vcpu);
707d92fa 1987 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
6aa8b732
AK		 1988 exit_lmode(vcpu);
1989 }
1990#endif
1991
089d034e 1992 if (enable_ept)
1439442c
SY		 1993 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1994
02daab21 1995 if (!vcpu->fpu_active)
81231c69 1996 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
02daab21 1997
6aa8b732 1998 vmcs_writel(CR0_READ_SHADOW, cr0);
1439442c 1999 vmcs_writel(GUEST_CR0, hw_cr0);
ad312c7c 2000 vcpu->arch.cr0 = cr0;
6aa8b732
AK		 2001}
2002
1439442c
SY		 2003static u64 construct_eptp(unsigned long root_hpa)
2004{
2005 u64 eptp;
2006
2007 /* TODO write the value reading from MSR */
2008 eptp = VMX_EPT_DEFAULT_MT |
2009 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
2010 eptp |= (root_hpa & PAGE_MASK);
2011
2012 return eptp;
2013}
2014
6aa8b732
AK		 2015static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2016{
1439442c
SY		 2017 unsigned long guest_cr3;
2018 u64 eptp;
2019
2020 guest_cr3 = cr3;
089d034e 2021 if (enable_ept) {
1439442c
SY		 2022 eptp = construct_eptp(cr3);
2023 vmcs_write64(EPT_POINTER, eptp);
9f8fe504 2024 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
b927a3ce 2025 vcpu->kvm->arch.ept_identity_map_addr;
7c93be44 2026 ept_load_pdptrs(vcpu);
1439442c
SY		 2027 }
2028
2384d2b3 2029 vmx_flush_tlb(vcpu);
1439442c 2030 vmcs_writel(GUEST_CR3, guest_cr3);
6aa8b732
AK		 2031}
2032
2033static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2034{
7ffd92c5 2035 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
1439442c
SY		 2036 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
2037
ad312c7c 2038 vcpu->arch.cr4 = cr4;
bc23008b
AK		 2039 if (enable_ept) {
2040 if (!is_paging(vcpu)) {
2041 hw_cr4 &= ~X86_CR4_PAE;
2042 hw_cr4 |= X86_CR4_PSE;
2043 } else if (!(cr4 & X86_CR4_PAE)) {
2044 hw_cr4 &= ~X86_CR4_PAE;
2045 }
2046 }
1439442c
SY		 2047
2048 vmcs_writel(CR4_READ_SHADOW, cr4);
2049 vmcs_writel(GUEST_CR4, hw_cr4);
6aa8b732
AK		 2050}
2051
6aa8b732
AK		 2052static void vmx_get_segment(struct kvm_vcpu *vcpu,
2053 struct kvm_segment *var, int seg)
2054{
a9179499 2055 struct vcpu_vmx *vmx = to_vmx(vcpu);
6aa8b732 2056 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
a9179499 2057 struct kvm_save_segment *save;
6aa8b732
AK		 2058 u32 ar;
2059
a9179499
AK		 2060 if (vmx->rmode.vm86_active
2061 && (seg == VCPU_SREG_TR || seg == VCPU_SREG_ES
2062 || seg == VCPU_SREG_DS || seg == VCPU_SREG_FS
2063 || seg == VCPU_SREG_GS)
2064 && !emulate_invalid_guest_state) {
2065 switch (seg) {
2066 case VCPU_SREG_TR: save = &vmx->rmode.tr; break;
2067 case VCPU_SREG_ES: save = &vmx->rmode.es; break;
2068 case VCPU_SREG_DS: save = &vmx->rmode.ds; break;
2069 case VCPU_SREG_FS: save = &vmx->rmode.fs; break;
2070 case VCPU_SREG_GS: save = &vmx->rmode.gs; break;
2071 default: BUG();
2072 }
2073 var->selector = save->selector;
2074 var->base = save->base;
2075 var->limit = save->limit;
2076 ar = save->ar;
2077 if (seg == VCPU_SREG_TR
2078 || var->selector == vmcs_read16(sf->selector))
2079 goto use_saved_rmode_seg;
2080 }
6aa8b732
AK		 2081 var->base = vmcs_readl(sf->base);
2082 var->limit = vmcs_read32(sf->limit);
2083 var->selector = vmcs_read16(sf->selector);
2084 ar = vmcs_read32(sf->ar_bytes);
a9179499 2085use_saved_rmode_seg:
9fd4a3b7 2086 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
6aa8b732
AK		 2087 ar = 0;
2088 var->type = ar & 15;
2089 var->s = (ar >> 4) & 1;
2090 var->dpl = (ar >> 5) & 3;
2091 var->present = (ar >> 7) & 1;
2092 var->avl = (ar >> 12) & 1;
2093 var->l = (ar >> 13) & 1;
2094 var->db = (ar >> 14) & 1;
2095 var->g = (ar >> 15) & 1;
2096 var->unusable = (ar >> 16) & 1;
2097}
2098
a9179499
AK		 2099static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2100{
2101 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2102 struct kvm_segment s;
2103
2104 if (to_vmx(vcpu)->rmode.vm86_active) {
2105 vmx_get_segment(vcpu, &s, seg);
2106 return s.base;
2107 }
2108 return vmcs_readl(sf->base);
2109}
2110
2e4d2653
IE		 2111static int vmx_get_cpl(struct kvm_vcpu *vcpu)
2112{
3eeb3288 2113 if (!is_protmode(vcpu))
2e4d2653
IE		 2114 return 0;
2115
f6e78475 2116 if (kvm_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
2e4d2653
IE		 2117 return 3;
2118
eab4b8aa 2119 return vmcs_read16(GUEST_CS_SELECTOR) & 3;
2e4d2653
IE		 2120}
2121
653e3108 2122static u32 vmx_segment_access_rights(struct kvm_segment *var)
6aa8b732 2123{
6aa8b732
AK		 2124 u32 ar;
2125
653e3108 2126 if (var->unusable)
6aa8b732
AK		 2127 ar = 1 << 16;
2128 else {
2129 ar = var->type & 15;
2130 ar |= (var->s & 1) << 4;
2131 ar |= (var->dpl & 3) << 5;
2132 ar |= (var->present & 1) << 7;
2133 ar |= (var->avl & 1) << 12;
2134 ar |= (var->l & 1) << 13;
2135 ar |= (var->db & 1) << 14;
2136 ar |= (var->g & 1) << 15;
2137 }
f7fbf1fd
UL		 2138 if (ar == 0) /* a 0 value means unusable */
2139 ar = AR_UNUSABLE_MASK;
653e3108
AK		 2140
2141 return ar;
2142}
2143
2144static void vmx_set_segment(struct kvm_vcpu *vcpu,
2145 struct kvm_segment *var, int seg)
2146{
7ffd92c5 2147 struct vcpu_vmx *vmx = to_vmx(vcpu);
653e3108
AK		 2148 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2149 u32 ar;
2150
7ffd92c5 2151 if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
a8ba6c26 2152 vmcs_write16(sf->selector, var->selector);
7ffd92c5
AK		 2153 vmx->rmode.tr.selector = var->selector;
2154 vmx->rmode.tr.base = var->base;
2155 vmx->rmode.tr.limit = var->limit;
2156 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
653e3108
AK		 2157 return;
2158 }
2159 vmcs_writel(sf->base, var->base);
2160 vmcs_write32(sf->limit, var->limit);
2161 vmcs_write16(sf->selector, var->selector);
7ffd92c5 2162 if (vmx->rmode.vm86_active && var->s) {
653e3108
AK		 2163 /*
2164 * Hack real-mode segments into vm86 compatibility.
2165 */
2166 if (var->base == 0xffff0000 && var->selector == 0xf000)
2167 vmcs_writel(sf->base, 0xf0000);
2168 ar = 0xf3;
2169 } else
2170 ar = vmx_segment_access_rights(var);
3a624e29
NK		 2171
2172 /*
2173 * Fix the "Accessed" bit in AR field of segment registers for older
2174 * qemu binaries.
2175 * IA32 arch specifies that at the time of processor reset the
2176 * "Accessed" bit in the AR field of segment registers is 1. And qemu
2177 * is setting it to 0 in the usedland code. This causes invalid guest
2178 * state vmexit when "unrestricted guest" mode is turned on.
2179 * Fix for this setup issue in cpu_reset is being pushed in the qemu
2180 * tree. Newer qemu binaries with that qemu fix would not need this
2181 * kvm hack.
2182 */
2183 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
2184 ar |= 0x1; /* Accessed */
2185
6aa8b732
AK		 2186 vmcs_write32(sf->ar_bytes, ar);
2187}
2188
6aa8b732
AK		 2189static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2190{
2191 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
2192
2193 *db = (ar >> 14) & 1;
2194 *l = (ar >> 13) & 1;
2195}
2196
89a27f4d 2197static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
6aa8b732 2198{
89a27f4d
GN		 2199 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
2200 dt->address = vmcs_readl(GUEST_IDTR_BASE);
6aa8b732
AK		 2201}
2202
89a27f4d 2203static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
6aa8b732 2204{
89a27f4d
GN		 2205 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
2206 vmcs_writel(GUEST_IDTR_BASE, dt->address);
6aa8b732
AK		 2207}
2208
89a27f4d 2209static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
6aa8b732 2210{
89a27f4d
GN		 2211 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
2212 dt->address = vmcs_readl(GUEST_GDTR_BASE);
6aa8b732
AK		 2213}
2214
89a27f4d 2215static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
6aa8b732 2216{
89a27f4d
GN		 2217 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
2218 vmcs_writel(GUEST_GDTR_BASE, dt->address);
6aa8b732
AK		 2219}
2220
648dfaa7
MG		 2221static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
2222{
2223 struct kvm_segment var;
2224 u32 ar;
2225
2226 vmx_get_segment(vcpu, &var, seg);
2227 ar = vmx_segment_access_rights(&var);
2228
2229 if (var.base != (var.selector << 4))
2230 return false;
2231 if (var.limit != 0xffff)
2232 return false;
2233 if (ar != 0xf3)
2234 return false;
2235
2236 return true;
2237}
2238
2239static bool code_segment_valid(struct kvm_vcpu *vcpu)
2240{
2241 struct kvm_segment cs;
2242 unsigned int cs_rpl;
2243
2244 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2245 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
2246
1872a3f4
AK		 2247 if (cs.unusable)
2248 return false;
648dfaa7
MG		 2249 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
2250 return false;
2251 if (!cs.s)
2252 return false;
1872a3f4 2253 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
648dfaa7
MG		 2254 if (cs.dpl > cs_rpl)
2255 return false;
1872a3f4 2256 } else {
648dfaa7
MG		 2257 if (cs.dpl != cs_rpl)
2258 return false;
2259 }
2260 if (!cs.present)
2261 return false;
2262
2263 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
2264 return true;
2265}
2266
2267static bool stack_segment_valid(struct kvm_vcpu *vcpu)
2268{
2269 struct kvm_segment ss;
2270 unsigned int ss_rpl;
2271
2272 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2273 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
2274
1872a3f4
AK		 2275 if (ss.unusable)
2276 return true;
2277 if (ss.type != 3 && ss.type != 7)
648dfaa7
MG		 2278 return false;
2279 if (!ss.s)
2280 return false;
2281 if (ss.dpl != ss_rpl) /* DPL != RPL */
2282 return false;
2283 if (!ss.present)
2284 return false;
2285
2286 return true;
2287}
2288
2289static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
2290{
2291 struct kvm_segment var;
2292 unsigned int rpl;
2293
2294 vmx_get_segment(vcpu, &var, seg);
2295 rpl = var.selector & SELECTOR_RPL_MASK;
2296
1872a3f4
AK		 2297 if (var.unusable)
2298 return true;
648dfaa7
MG		 2299 if (!var.s)
2300 return false;
2301 if (!var.present)
2302 return false;
2303 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
2304 if (var.dpl < rpl) /* DPL < RPL */
2305 return false;
2306 }
2307
2308 /* TODO: Add other members to kvm_segment_field to allow checking for other access
2309 * rights flags
2310 */
2311 return true;
2312}
2313
2314static bool tr_valid(struct kvm_vcpu *vcpu)
2315{
2316 struct kvm_segment tr;
2317
2318 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
2319
1872a3f4
AK		 2320 if (tr.unusable)
2321 return false;
648dfaa7
MG		 2322 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2323 return false;
1872a3f4 2324 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
648dfaa7
MG		 2325 return false;
2326 if (!tr.present)
2327 return false;
2328
2329 return true;
2330}
2331
2332static bool ldtr_valid(struct kvm_vcpu *vcpu)
2333{
2334 struct kvm_segment ldtr;
2335
2336 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
2337
1872a3f4
AK		 2338 if (ldtr.unusable)
2339 return true;
648dfaa7
MG		 2340 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2341 return false;
2342 if (ldtr.type != 2)
2343 return false;
2344 if (!ldtr.present)
2345 return false;
2346
2347 return true;
2348}
2349
2350static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2351{
2352 struct kvm_segment cs, ss;
2353
2354 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2355 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2356
2357 return ((cs.selector & SELECTOR_RPL_MASK) ==
2358 (ss.selector & SELECTOR_RPL_MASK));
2359}
2360
2361/*
2362 * Check if guest state is valid. Returns true if valid, false if
2363 * not.
2364 * We assume that registers are always usable
2365 */
2366static bool guest_state_valid(struct kvm_vcpu *vcpu)
2367{
2368 /* real mode guest state checks */
3eeb3288 2369 if (!is_protmode(vcpu)) {
648dfaa7
MG		 2370 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2371 return false;
2372 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2373 return false;
2374 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2375 return false;
2376 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2377 return false;
2378 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2379 return false;
2380 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2381 return false;
2382 } else {
2383 /* protected mode guest state checks */
2384 if (!cs_ss_rpl_check(vcpu))
2385 return false;
2386 if (!code_segment_valid(vcpu))
2387 return false;
2388 if (!stack_segment_valid(vcpu))
2389 return false;
2390 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2391 return false;
2392 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2393 return false;
2394 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2395 return false;
2396 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2397 return false;
2398 if (!tr_valid(vcpu))
2399 return false;
2400 if (!ldtr_valid(vcpu))
2401 return false;
2402 }
2403 /* TODO:
2404 * - Add checks on RIP
2405 * - Add checks on RFLAGS
2406 */
2407
2408 return true;
2409}
2410
d77c26fc 2411static int init_rmode_tss(struct kvm *kvm)
6aa8b732 2412{
40dcaa9f 2413 gfn_t fn;
195aefde 2414 u16 data = 0;
40dcaa9f 2415 int r, idx, ret = 0;
6aa8b732 2416
40dcaa9f
XG		 2417 idx = srcu_read_lock(&kvm->srcu);
2418 fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
195aefde
IE		 2419 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2420 if (r < 0)
10589a46 2421 goto out;
195aefde 2422 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
464d17c8
SY		 2423 r = kvm_write_guest_page(kvm, fn++, &data,
2424 TSS_IOPB_BASE_OFFSET, sizeof(u16));
195aefde 2425 if (r < 0)
10589a46 2426 goto out;
195aefde
IE		 2427 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2428 if (r < 0)
10589a46 2429 goto out;
195aefde
IE		 2430 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2431 if (r < 0)
10589a46 2432 goto out;
195aefde 2433 data = ~0;
10589a46
MT		 2434 r = kvm_write_guest_page(kvm, fn, &data,
2435 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2436 sizeof(u8));
195aefde 2437 if (r < 0)
10589a46
MT		 2438 goto out;
2439
2440 ret = 1;
2441out:
40dcaa9f 2442 srcu_read_unlock(&kvm->srcu, idx);
10589a46 2443 return ret;
6aa8b732
AK		 2444}
2445
b7ebfb05
SY		 2446static int init_rmode_identity_map(struct kvm *kvm)
2447{
40dcaa9f 2448 int i, idx, r, ret;
b7ebfb05
SY		 2449 pfn_t identity_map_pfn;
2450 u32 tmp;
2451
089d034e 2452 if (!enable_ept)
b7ebfb05
SY		 2453 return 1;
2454 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2455 printk(KERN_ERR "EPT: identity-mapping pagetable "
2456 "haven't been allocated!\n");
2457 return 0;
2458 }
2459 if (likely(kvm->arch.ept_identity_pagetable_done))
2460 return 1;
2461 ret = 0;
b927a3ce 2462 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
40dcaa9f 2463 idx = srcu_read_lock(&kvm->srcu);
b7ebfb05
SY		 2464 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2465 if (r < 0)
2466 goto out;
2467 /* Set up identity-mapping pagetable for EPT in real mode */
2468 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2469 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2470 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2471 r = kvm_write_guest_page(kvm, identity_map_pfn,
2472 &tmp, i * sizeof(tmp), sizeof(tmp));
2473 if (r < 0)
2474 goto out;
2475 }
2476 kvm->arch.ept_identity_pagetable_done = true;
2477 ret = 1;
2478out:
40dcaa9f 2479 srcu_read_unlock(&kvm->srcu, idx);
b7ebfb05
SY		 2480 return ret;
2481}
2482
6aa8b732
AK		 2483static void seg_setup(int seg)
2484{
2485 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3a624e29 2486 unsigned int ar;
6aa8b732
AK		 2487
2488 vmcs_write16(sf->selector, 0);
2489 vmcs_writel(sf->base, 0);
2490 vmcs_write32(sf->limit, 0xffff);
3a624e29
NK		 2491 if (enable_unrestricted_guest) {
2492 ar = 0x93;
2493 if (seg == VCPU_SREG_CS)
2494 ar |= 0x08; /* code segment */
2495 } else
2496 ar = 0xf3;
2497
2498 vmcs_write32(sf->ar_bytes, ar);
6aa8b732
AK		 2499}
2500
f78e0e2e
SY		 2501static int alloc_apic_access_page(struct kvm *kvm)
2502{
2503 struct kvm_userspace_memory_region kvm_userspace_mem;
2504 int r = 0;
2505
79fac95e 2506 mutex_lock(&kvm->slots_lock);
bfc6d222 2507 if (kvm->arch.apic_access_page)
f78e0e2e
SY		 2508 goto out;
2509 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2510 kvm_userspace_mem.flags = 0;
2511 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2512 kvm_userspace_mem.memory_size = PAGE_SIZE;
2513 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2514 if (r)
2515 goto out;
72dc67a6 2516
bfc6d222 2517 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
f78e0e2e 2518out:
79fac95e 2519 mutex_unlock(&kvm->slots_lock);
f78e0e2e
SY		 2520 return r;
2521}
2522
b7ebfb05
SY		 2523static int alloc_identity_pagetable(struct kvm *kvm)
2524{
2525 struct kvm_userspace_memory_region kvm_userspace_mem;
2526 int r = 0;
2527
79fac95e 2528 mutex_lock(&kvm->slots_lock);
b7ebfb05
SY		 2529 if (kvm->arch.ept_identity_pagetable)
2530 goto out;
2531 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2532 kvm_userspace_mem.flags = 0;
b927a3ce
SY		 2533 kvm_userspace_mem.guest_phys_addr =
2534 kvm->arch.ept_identity_map_addr;
b7ebfb05
SY		 2535 kvm_userspace_mem.memory_size = PAGE_SIZE;
2536 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2537 if (r)
2538 goto out;
2539
b7ebfb05 2540 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
b927a3ce 2541 kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
b7ebfb05 2542out:
79fac95e 2543 mutex_unlock(&kvm->slots_lock);
b7ebfb05
SY		 2544 return r;
2545}
2546
2384d2b3
SY		 2547static void allocate_vpid(struct vcpu_vmx *vmx)
2548{
2549 int vpid;
2550
2551 vmx->vpid = 0;
919818ab 2552 if (!enable_vpid)
2384d2b3
SY		 2553 return;
2554 spin_lock(&vmx_vpid_lock);
2555 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2556 if (vpid < VMX_NR_VPIDS) {
2557 vmx->vpid = vpid;
2558 __set_bit(vpid, vmx_vpid_bitmap);
2559 }
2560 spin_unlock(&vmx_vpid_lock);
2561}
2562
cdbecfc3
LJ		 2563static void free_vpid(struct vcpu_vmx *vmx)
2564{
2565 if (!enable_vpid)
2566 return;
2567 spin_lock(&vmx_vpid_lock);
2568 if (vmx->vpid != 0)
2569 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
2570 spin_unlock(&vmx_vpid_lock);
2571}
2572
5897297b 2573static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
25c5f225 2574{
3e7c73e9 2575 int f = sizeof(unsigned long);
25c5f225
SY		 2576
2577 if (!cpu_has_vmx_msr_bitmap())
2578 return;
2579
2580 /*
2581 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2582 * have the write-low and read-high bitmap offsets the wrong way round.
2583 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2584 */
25c5f225 2585 if (msr <= 0x1fff) {
3e7c73e9
AK		 2586 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2587 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
25c5f225
SY		 2588 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2589 msr &= 0x1fff;
3e7c73e9
AK		 2590 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2591 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
25c5f225 2592 }
25c5f225
SY		 2593}
2594
5897297b
AK		 2595static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2596{
2597 if (!longmode_only)
2598 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2599 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2600}
2601
6aa8b732
AK		 2602/*
2603 * Sets up the vmcs for emulated real mode.
2604 */
8b9cf98c 2605static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
6aa8b732 2606{
468d472f 2607 u32 host_sysenter_cs, msr_low, msr_high;
6aa8b732 2608 u32 junk;
f4e1b3c8 2609 u64 host_pat;
6aa8b732 2610 unsigned long a;
89a27f4d 2611 struct desc_ptr dt;
6aa8b732 2612 int i;
cd2276a7 2613 unsigned long kvm_vmx_return;
6e5d865c 2614 u32 exec_control;
6aa8b732 2615
6aa8b732 2616 /* I/O */
3e7c73e9
AK		 2617 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2618 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
6aa8b732 2619
25c5f225 2620 if (cpu_has_vmx_msr_bitmap())
5897297b 2621 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
25c5f225 2622
6aa8b732
AK		 2623 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2624
6aa8b732 2625 /* Control */
1c3d14fe
YS		 2626 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2627 vmcs_config.pin_based_exec_ctrl);
6e5d865c
YS		 2628
2629 exec_control = vmcs_config.cpu_based_exec_ctrl;
2630 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2631 exec_control &= ~CPU_BASED_TPR_SHADOW;
2632#ifdef CONFIG_X86_64
2633 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2634 CPU_BASED_CR8_LOAD_EXITING;
2635#endif
2636 }
089d034e 2637 if (!enable_ept)
d56f546d 2638 exec_control |= CPU_BASED_CR3_STORE_EXITING |
83dbc83a
MT		 2639 CPU_BASED_CR3_LOAD_EXITING |
2640 CPU_BASED_INVLPG_EXITING;
6e5d865c 2641 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
6aa8b732 2642
83ff3b9d
SY		 2643 if (cpu_has_secondary_exec_ctrls()) {
2644 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2645 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2646 exec_control &=
2647 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2384d2b3
SY		 2648 if (vmx->vpid == 0)
2649 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
046d8710 2650 if (!enable_ept) {
d56f546d 2651 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
046d8710
SY		 2652 enable_unrestricted_guest = 0;
2653 }
3a624e29
NK		 2654 if (!enable_unrestricted_guest)
2655 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4b8d54f9
ZE		 2656 if (!ple_gap)
2657 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
83ff3b9d
SY		 2658 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2659 }
f78e0e2e 2660
4b8d54f9
ZE		 2661 if (ple_gap) {
2662 vmcs_write32(PLE_GAP, ple_gap);
2663 vmcs_write32(PLE_WINDOW, ple_window);
2664 }
2665
c7addb90
AK		 2666 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2667 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
6aa8b732
AK		 2668 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
2669
1c11e713 2670 vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS); /* 22.2.3 */
6aa8b732
AK		 2671 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
2672 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2673
2674 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
2675 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2676 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
9581d442
AK		 2677 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
2678 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
6aa8b732 2679 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
05b3e0c2 2680#ifdef CONFIG_X86_64
6aa8b732
AK		 2681 rdmsrl(MSR_FS_BASE, a);
2682 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2683 rdmsrl(MSR_GS_BASE, a);
2684 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2685#else
2686 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2687 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2688#endif
2689
2690 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
2691
ec68798c 2692 native_store_idt(&dt);
89a27f4d 2693 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
6aa8b732 2694
d77c26fc 2695 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
cd2276a7 2696 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2cc51560
ED		 2697 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2698 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
61d2ef2c 2699 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
2cc51560 2700 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
61d2ef2c 2701 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
6aa8b732
AK		 2702
2703 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2704 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2705 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2706 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
2707 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2708 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2709
468d472f
SY		 2710 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2711 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2712 host_pat = msr_low | ((u64) msr_high << 32);
2713 vmcs_write64(HOST_IA32_PAT, host_pat);
2714 }
2715 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2716 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2717 host_pat = msr_low | ((u64) msr_high << 32);
2718 /* Write the default value follow host pat */
2719 vmcs_write64(GUEST_IA32_PAT, host_pat);
2720 /* Keep arch.pat sync with GUEST_IA32_PAT */
2721 vmx->vcpu.arch.pat = host_pat;
2722 }
2723
6aa8b732
AK		 2724 for (i = 0; i < NR_VMX_MSR; ++i) {
2725 u32 index = vmx_msr_index[i];
2726 u32 data_low, data_high;
a2fa3e9f 2727 int j = vmx->nmsrs;
6aa8b732
AK		 2728
2729 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2730 continue;
432bd6cb
AK		 2731 if (wrmsr_safe(index, data_low, data_high) < 0)
2732 continue;
26bb0981
AK		 2733 vmx->guest_msrs[j].index = i;
2734 vmx->guest_msrs[j].data = 0;
d5696725 2735 vmx->guest_msrs[j].mask = -1ull;
a2fa3e9f 2736 ++vmx->nmsrs;
6aa8b732 2737 }
6aa8b732 2738
1c3d14fe 2739 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
6aa8b732
AK		 2740
2741 /* 22.2.1, 20.8.1 */
1c3d14fe
YS		 2742 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2743
e00c8cf2 2744 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4c38609a 2745 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
ce03e4f2
AK		 2746 if (enable_ept)
2747 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4c38609a 2748 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
e00c8cf2 2749
99e3e30a 2750 kvm_write_tsc(&vmx->vcpu, 0);
f78e0e2e 2751
e00c8cf2
AK		 2752 return 0;
2753}
2754
2755static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2756{
2757 struct vcpu_vmx *vmx = to_vmx(vcpu);
2758 u64 msr;
4b9d3a04 2759 int ret;
e00c8cf2 2760
5fdbf976 2761 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
e00c8cf2 2762
7ffd92c5 2763 vmx->rmode.vm86_active = 0;
e00c8cf2 2764
3b86cd99
JK		 2765 vmx->soft_vnmi_blocked = 0;
2766
ad312c7c 2767 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2d3ad1f4 2768 kvm_set_cr8(&vmx->vcpu, 0);
e00c8cf2 2769 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
c5af89b6 2770 if (kvm_vcpu_is_bsp(&vmx->vcpu))
e00c8cf2
AK		 2771 msr |= MSR_IA32_APICBASE_BSP;
2772 kvm_set_apic_base(&vmx->vcpu, msr);
2773
10ab25cd
JK		 2774 ret = fx_init(&vmx->vcpu);
2775 if (ret != 0)
2776 goto out;
e00c8cf2 2777
5706be0d 2778 seg_setup(VCPU_SREG_CS);
e00c8cf2
AK		 2779 /*
2780 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2781 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2782 */
c5af89b6 2783 if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
e00c8cf2
AK		 2784 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2785 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2786 } else {
ad312c7c
ZX		 2787 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2788 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
e00c8cf2 2789 }
e00c8cf2
AK		 2790
2791 seg_setup(VCPU_SREG_DS);
2792 seg_setup(VCPU_SREG_ES);
2793 seg_setup(VCPU_SREG_FS);
2794 seg_setup(VCPU_SREG_GS);
2795 seg_setup(VCPU_SREG_SS);
2796
2797 vmcs_write16(GUEST_TR_SELECTOR, 0);
2798 vmcs_writel(GUEST_TR_BASE, 0);
2799 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2800 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2801
2802 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2803 vmcs_writel(GUEST_LDTR_BASE, 0);
2804 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2805 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2806
2807 vmcs_write32(GUEST_SYSENTER_CS, 0);
2808 vmcs_writel(GUEST_SYSENTER_ESP, 0);
2809 vmcs_writel(GUEST_SYSENTER_EIP, 0);
2810
2811 vmcs_writel(GUEST_RFLAGS, 0x02);
c5af89b6 2812 if (kvm_vcpu_is_bsp(&vmx->vcpu))
5fdbf976 2813 kvm_rip_write(vcpu, 0xfff0);
e00c8cf2 2814 else
5fdbf976
MT		 2815 kvm_rip_write(vcpu, 0);
2816 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
e00c8cf2 2817
e00c8cf2
AK		 2818 vmcs_writel(GUEST_DR7, 0x400);
2819
2820 vmcs_writel(GUEST_GDTR_BASE, 0);
2821 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2822
2823 vmcs_writel(GUEST_IDTR_BASE, 0);
2824 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2825
443381a8 2826 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
e00c8cf2
AK		 2827 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2828 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2829
e00c8cf2
AK		 2830 /* Special registers */
2831 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2832
2833 setup_msrs(vmx);
2834
6aa8b732
AK		 2835 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
2836
f78e0e2e
SY		 2837 if (cpu_has_vmx_tpr_shadow()) {
2838 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2839 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2840 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
afc20184 2841 __pa(vmx->vcpu.arch.apic->regs));
f78e0e2e
SY		 2842 vmcs_write32(TPR_THRESHOLD, 0);
2843 }
2844
2845 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2846 vmcs_write64(APIC_ACCESS_ADDR,
bfc6d222 2847 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
6aa8b732 2848
2384d2b3
SY		 2849 if (vmx->vpid != 0)
2850 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2851
fa40052c 2852 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4d4ec087 2853 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
8b9cf98c 2854 vmx_set_cr4(&vmx->vcpu, 0);
8b9cf98c 2855 vmx_set_efer(&vmx->vcpu, 0);
8b9cf98c
RR		 2856 vmx_fpu_activate(&vmx->vcpu);
2857 update_exception_bitmap(&vmx->vcpu);
6aa8b732 2858
b9d762fa 2859 vpid_sync_context(vmx);
2384d2b3 2860
3200f405 2861 ret = 0;
6aa8b732 2862
a89a8fb9
MG		 2863 /* HACK: Don't enable emulation on guest boot/reset */
2864 vmx->emulation_required = 0;
2865
6aa8b732
AK		 2866out:
2867 return ret;
2868}
2869
3b86cd99
JK		 2870static void enable_irq_window(struct kvm_vcpu *vcpu)
2871{
2872 u32 cpu_based_vm_exec_control;
2873
2874 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2875 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2876 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2877}
2878
2879static void enable_nmi_window(struct kvm_vcpu *vcpu)
2880{
2881 u32 cpu_based_vm_exec_control;
2882
2883 if (!cpu_has_virtual_nmis()) {
2884 enable_irq_window(vcpu);
2885 return;
2886 }
2887
30bd0c4c
AK		 2888 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
2889 enable_irq_window(vcpu);
2890 return;
2891 }
3b86cd99
JK		 2892 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2893 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2894 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2895}
2896
66fd3f7f 2897static void vmx_inject_irq(struct kvm_vcpu *vcpu)
85f455f7 2898{
9c8cba37 2899 struct vcpu_vmx *vmx = to_vmx(vcpu);
66fd3f7f
GN		 2900 uint32_t intr;
2901 int irq = vcpu->arch.interrupt.nr;
9c8cba37 2902
229456fc 2903 trace_kvm_inj_virq(irq);
2714d1d3 2904
fa89a817 2905 ++vcpu->stat.irq_injections;
7ffd92c5 2906 if (vmx->rmode.vm86_active) {
a92601bb
MG		 2907 if (kvm_inject_realmode_interrupt(vcpu, irq) != EMULATE_DONE)
2908 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
85f455f7
ED		 2909 return;
2910 }
66fd3f7f
GN		 2911 intr = irq | INTR_INFO_VALID_MASK;
2912 if (vcpu->arch.interrupt.soft) {
2913 intr |= INTR_TYPE_SOFT_INTR;
2914 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2915 vmx->vcpu.arch.event_exit_inst_len);
2916 } else
2917 intr |= INTR_TYPE_EXT_INTR;
2918 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
443381a8 2919 vmx_clear_hlt(vcpu);
85f455f7
ED		 2920}
2921
f08864b4
SY		 2922static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2923{
66a5a347
JK		 2924 struct vcpu_vmx *vmx = to_vmx(vcpu);
2925
3b86cd99
JK		 2926 if (!cpu_has_virtual_nmis()) {
2927 /*
2928 * Tracking the NMI-blocked state in software is built upon
2929 * finding the next open IRQ window. This, in turn, depends on
2930 * well-behaving guests: They have to keep IRQs disabled at
2931 * least as long as the NMI handler runs. Otherwise we may
2932 * cause NMI nesting, maybe breaking the guest. But as this is
2933 * highly unlikely, we can live with the residual risk.
2934 */
2935 vmx->soft_vnmi_blocked = 1;
2936 vmx->vnmi_blocked_time = 0;
2937 }
2938
487b391d 2939 ++vcpu->stat.nmi_injections;
7ffd92c5 2940 if (vmx->rmode.vm86_active) {
a92601bb
MG		 2941 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR) != EMULATE_DONE)
2942 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
66a5a347
JK		 2943 return;
2944 }
f08864b4
SY		 2945 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2946 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
443381a8 2947 vmx_clear_hlt(vcpu);
f08864b4
SY		 2948}
2949
c4282df9 2950static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
33f089ca 2951{
3b86cd99 2952 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
c4282df9 2953 return 0;
33f089ca 2954
c4282df9 2955 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
30bd0c4c
AK		 2956 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
2957 | GUEST_INTR_STATE_NMI));
33f089ca
JK		 2958}
2959
3cfc3092
JK		 2960static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
2961{
2962 if (!cpu_has_virtual_nmis())
2963 return to_vmx(vcpu)->soft_vnmi_blocked;
c332c83a 2964 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
3cfc3092
JK		 2965}
2966
2967static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
2968{
2969 struct vcpu_vmx *vmx = to_vmx(vcpu);
2970
2971 if (!cpu_has_virtual_nmis()) {
2972 if (vmx->soft_vnmi_blocked != masked) {
2973 vmx->soft_vnmi_blocked = masked;
2974 vmx->vnmi_blocked_time = 0;
2975 }
2976 } else {
2977 if (masked)
2978 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
2979 GUEST_INTR_STATE_NMI);
2980 else
2981 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
2982 GUEST_INTR_STATE_NMI);
2983 }
2984}
2985
78646121
GN		 2986static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2987{
c4282df9
GN		 2988 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2989 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2990 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
78646121
GN		 2991}
2992
cbc94022
IE		 2993static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2994{
2995 int ret;
2996 struct kvm_userspace_memory_region tss_mem = {
6fe63979 2997 .slot = TSS_PRIVATE_MEMSLOT,
cbc94022
IE		 2998 .guest_phys_addr = addr,
2999 .memory_size = PAGE_SIZE * 3,
3000 .flags = 0,
3001 };
3002
3003 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
3004 if (ret)
3005 return ret;
bfc6d222 3006 kvm->arch.tss_addr = addr;
93ea5388
GN		 3007 if (!init_rmode_tss(kvm))
3008 return -ENOMEM;
3009
cbc94022
IE		 3010 return 0;
3011}
3012
6aa8b732
AK		 3013static int handle_rmode_exception(struct kvm_vcpu *vcpu,
3014 int vec, u32 err_code)
3015{
b3f37707
NK		 3016 /*
3017 * Instruction with address size override prefix opcode 0x67
3018 * Cause the #SS fault with 0 error code in VM86 mode.
3019 */
3020 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
51d8b661 3021 if (emulate_instruction(vcpu, 0) == EMULATE_DONE)
6aa8b732 3022 return 1;
77ab6db0
JK		 3023 /*
3024 * Forward all other exceptions that are valid in real mode.
3025 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
3026 * the required debugging infrastructure rework.
3027 */
3028 switch (vec) {
77ab6db0 3029 case DB_VECTOR:
d0bfb940
JK		 3030 if (vcpu->guest_debug &
3031 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
3032 return 0;
3033 kvm_queue_exception(vcpu, vec);
3034 return 1;
77ab6db0 3035 case BP_VECTOR:
c573cd22
JK		 3036 /*
3037 * Update instruction length as we may reinject the exception
3038 * from user space while in guest debugging mode.
3039 */
3040 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
3041 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
d0bfb940
JK		 3042 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
3043 return 0;
3044 /* fall through */
3045 case DE_VECTOR:
77ab6db0
JK		 3046 case OF_VECTOR:
3047 case BR_VECTOR:
3048 case UD_VECTOR:
3049 case DF_VECTOR:
3050 case SS_VECTOR:
3051 case GP_VECTOR:
3052 case MF_VECTOR:
3053 kvm_queue_exception(vcpu, vec);
3054 return 1;
3055 }
6aa8b732
AK		 3056 return 0;
3057}
3058
a0861c02
AK		 3059/*
3060 * Trigger machine check on the host. We assume all the MSRs are already set up
3061 * by the CPU and that we still run on the same CPU as the MCE occurred on.
3062 * We pass a fake environment to the machine check handler because we want
3063 * the guest to be always treated like user space, no matter what context
3064 * it used internally.
3065 */
3066static void kvm_machine_check(void)
3067{
3068#if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
3069 struct pt_regs regs = {
3070 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
3071 .flags = X86_EFLAGS_IF,
3072 };
3073
3074 do_machine_check(&regs, 0);
3075#endif
3076}
3077
851ba692 3078static int handle_machine_check(struct kvm_vcpu *vcpu)
a0861c02
AK		 3079{
3080 /* already handled by vcpu_run */
3081 return 1;
3082}
3083
851ba692 3084static int handle_exception(struct kvm_vcpu *vcpu)
6aa8b732 3085{
1155f76a 3086 struct vcpu_vmx *vmx = to_vmx(vcpu);
851ba692 3087 struct kvm_run *kvm_run = vcpu->run;
d0bfb940 3088 u32 intr_info, ex_no, error_code;
42dbaa5a 3089 unsigned long cr2, rip, dr6;
6aa8b732
AK		 3090 u32 vect_info;
3091 enum emulation_result er;
3092
1155f76a 3093 vect_info = vmx->idt_vectoring_info;
6aa8b732
AK		 3094 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3095
a0861c02 3096 if (is_machine_check(intr_info))
851ba692 3097 return handle_machine_check(vcpu);
a0861c02 3098
6aa8b732 3099 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
65ac7264
AK		 3100 !is_page_fault(intr_info)) {
3101 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3102 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
3103 vcpu->run->internal.ndata = 2;
3104 vcpu->run->internal.data[0] = vect_info;
3105 vcpu->run->internal.data[1] = intr_info;
3106 return 0;
3107 }
6aa8b732 3108
e4a41889 3109 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
1b6269db 3110 return 1; /* already handled by vmx_vcpu_run() */
2ab455cc
AL		 3111
3112 if (is_no_device(intr_info)) {
5fd86fcf 3113 vmx_fpu_activate(vcpu);
2ab455cc
AL		 3114 return 1;
3115 }
3116
7aa81cc0 3117 if (is_invalid_opcode(intr_info)) {
51d8b661 3118 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
7aa81cc0 3119 if (er != EMULATE_DONE)
7ee5d940 3120 kvm_queue_exception(vcpu, UD_VECTOR);
7aa81cc0
AL		 3121 return 1;
3122 }
3123
6aa8b732 3124 error_code = 0;
5fdbf976 3125 rip = kvm_rip_read(vcpu);
2e11384c 3126 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
6aa8b732
AK		 3127 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
3128 if (is_page_fault(intr_info)) {
1439442c 3129 /* EPT won't cause page fault directly */
089d034e 3130 if (enable_ept)
1439442c 3131 BUG();
6aa8b732 3132 cr2 = vmcs_readl(EXIT_QUALIFICATION);
229456fc
MT		 3133 trace_kvm_page_fault(cr2, error_code);
3134
3298b75c 3135 if (kvm_event_needs_reinjection(vcpu))
577bdc49 3136 kvm_mmu_unprotect_page_virt(vcpu, cr2);
dc25e89e 3137 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
6aa8b732
AK		 3138 }
3139
7ffd92c5 3140 if (vmx->rmode.vm86_active &&
6aa8b732 3141 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
72d6e5a0 3142 error_code)) {
ad312c7c
ZX		 3143 if (vcpu->arch.halt_request) {
3144 vcpu->arch.halt_request = 0;
72d6e5a0
AK		 3145 return kvm_emulate_halt(vcpu);
3146 }
6aa8b732 3147 return 1;
72d6e5a0 3148 }
6aa8b732 3149
d0bfb940 3150 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
42dbaa5a
JK		 3151 switch (ex_no) {
3152 case DB_VECTOR:
3153 dr6 = vmcs_readl(EXIT_QUALIFICATION);
3154 if (!(vcpu->guest_debug &
3155 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
3156 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
3157 kvm_queue_exception(vcpu, DB_VECTOR);
3158 return 1;
3159 }
3160 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
3161 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
3162 /* fall through */
3163 case BP_VECTOR:
c573cd22
JK		 3164 /*
3165 * Update instruction length as we may reinject #BP from
3166 * user space while in guest debugging mode. Reading it for
3167 * #DB as well causes no harm, it is not used in that case.
3168 */
3169 vmx->vcpu.arch.event_exit_inst_len =
3170 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6aa8b732 3171 kvm_run->exit_reason = KVM_EXIT_DEBUG;
d0bfb940
JK		 3172 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
3173 kvm_run->debug.arch.exception = ex_no;
42dbaa5a
JK		 3174 break;
3175 default:
d0bfb940
JK		 3176 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
3177 kvm_run->ex.exception = ex_no;
3178 kvm_run->ex.error_code = error_code;
42dbaa5a 3179 break;
6aa8b732 3180 }
6aa8b732
AK		 3181 return 0;
3182}
3183
851ba692 3184static int handle_external_interrupt(struct kvm_vcpu *vcpu)
6aa8b732 3185{
1165f5fe 3186 ++vcpu->stat.irq_exits;
6aa8b732
AK		 3187 return 1;
3188}
3189
851ba692 3190static int handle_triple_fault(struct kvm_vcpu *vcpu)
988ad74f 3191{
851ba692 3192 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
988ad74f
AK		 3193 return 0;
3194}
6aa8b732 3195
851ba692 3196static int handle_io(struct kvm_vcpu *vcpu)
6aa8b732 3197{
bfdaab09 3198 unsigned long exit_qualification;
34c33d16 3199 int size, in, string;
039576c0 3200 unsigned port;
6aa8b732 3201
bfdaab09 3202 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
039576c0 3203 string = (exit_qualification & 16) != 0;
cf8f70bf 3204 in = (exit_qualification & 8) != 0;
e70669ab 3205
cf8f70bf 3206 ++vcpu->stat.io_exits;
e70669ab 3207
cf8f70bf 3208 if (string || in)
51d8b661 3209 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
e70669ab 3210
cf8f70bf
GN		 3211 port = exit_qualification >> 16;
3212 size = (exit_qualification & 7) + 1;
e93f36bc 3213 skip_emulated_instruction(vcpu);
cf8f70bf
GN		 3214
3215 return kvm_fast_pio_out(vcpu, size, port);
6aa8b732
AK		 3216}
3217
102d8325
IM		 3218static void
3219vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3220{
3221 /*
3222 * Patch in the VMCALL instruction:
3223 */
3224 hypercall[0] = 0x0f;
3225 hypercall[1] = 0x01;
3226 hypercall[2] = 0xc1;
102d8325
IM		 3227}
3228
851ba692 3229static int handle_cr(struct kvm_vcpu *vcpu)
6aa8b732 3230{
229456fc 3231 unsigned long exit_qualification, val;
6aa8b732
AK		 3232 int cr;
3233 int reg;
49a9b07e 3234 int err;
6aa8b732 3235
bfdaab09 3236 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6aa8b732
AK		 3237 cr = exit_qualification & 15;
3238 reg = (exit_qualification >> 8) & 15;
3239 switch ((exit_qualification >> 4) & 3) {
3240 case 0: /* mov to cr */
229456fc
MT		 3241 val = kvm_register_read(vcpu, reg);
3242 trace_kvm_cr_write(cr, val);
6aa8b732
AK		 3243 switch (cr) {
3244 case 0:
49a9b07e 3245 err = kvm_set_cr0(vcpu, val);
db8fcefa 3246 kvm_complete_insn_gp(vcpu, err);
6aa8b732
AK		 3247 return 1;
3248 case 3:
2390218b 3249 err = kvm_set_cr3(vcpu, val);
db8fcefa 3250 kvm_complete_insn_gp(vcpu, err);
6aa8b732
AK		 3251 return 1;
3252 case 4:
a83b29c6 3253 err = kvm_set_cr4(vcpu, val);
db8fcefa 3254 kvm_complete_insn_gp(vcpu, err);
6aa8b732 3255 return 1;
0a5fff19
GN		 3256 case 8: {
3257 u8 cr8_prev = kvm_get_cr8(vcpu);
3258 u8 cr8 = kvm_register_read(vcpu, reg);
eea1cff9 3259 err = kvm_set_cr8(vcpu, cr8);
db8fcefa 3260 kvm_complete_insn_gp(vcpu, err);
0a5fff19
GN		 3261 if (irqchip_in_kernel(vcpu->kvm))
3262 return 1;
3263 if (cr8_prev <= cr8)
3264 return 1;
851ba692 3265 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
0a5fff19
GN		 3266 return 0;
3267 }
6aa8b732
AK		 3268 };
3269 break;
25c4c276 3270 case 2: /* clts */
edcafe3c 3271 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4d4ec087 3272 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
25c4c276 3273 skip_emulated_instruction(vcpu);
6b52d186 3274 vmx_fpu_activate(vcpu);
25c4c276 3275 return 1;
6aa8b732
AK		 3276 case 1: /*mov from cr*/
3277 switch (cr) {
3278 case 3:
9f8fe504
AK		 3279 val = kvm_read_cr3(vcpu);
3280 kvm_register_write(vcpu, reg, val);
3281 trace_kvm_cr_read(cr, val);
6aa8b732
AK		 3282 skip_emulated_instruction(vcpu);
3283 return 1;
3284 case 8:
229456fc
MT		 3285 val = kvm_get_cr8(vcpu);
3286 kvm_register_write(vcpu, reg, val);
3287 trace_kvm_cr_read(cr, val);
6aa8b732
AK		 3288 skip_emulated_instruction(vcpu);
3289 return 1;
3290 }
3291 break;
3292 case 3: /* lmsw */
a1f83a74 3293 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4d4ec087 3294 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
a1f83a74 3295 kvm_lmsw(vcpu, val);
6aa8b732
AK		 3296
3297 skip_emulated_instruction(vcpu);
3298 return 1;
3299 default:
3300 break;
3301 }
851ba692 3302 vcpu->run->exit_reason = 0;
f0242478 3303 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
6aa8b732
AK		 3304 (int)(exit_qualification >> 4) & 3, cr);
3305 return 0;
3306}
3307
851ba692 3308static int handle_dr(struct kvm_vcpu *vcpu)
6aa8b732 3309{
bfdaab09 3310 unsigned long exit_qualification;
6aa8b732
AK		 3311 int dr, reg;
3312
f2483415 3313 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
0a79b009
AK		 3314 if (!kvm_require_cpl(vcpu, 0))
3315 return 1;
42dbaa5a
JK		 3316 dr = vmcs_readl(GUEST_DR7);
3317 if (dr & DR7_GD) {
3318 /*
3319 * As the vm-exit takes precedence over the debug trap, we
3320 * need to emulate the latter, either for the host or the
3321 * guest debugging itself.
3322 */
3323 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
851ba692
AK		 3324 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
3325 vcpu->run->debug.arch.dr7 = dr;
3326 vcpu->run->debug.arch.pc =
42dbaa5a
JK		 3327 vmcs_readl(GUEST_CS_BASE) +
3328 vmcs_readl(GUEST_RIP);
851ba692
AK		 3329 vcpu->run->debug.arch.exception = DB_VECTOR;
3330 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
42dbaa5a
JK		 3331 return 0;
3332 } else {
3333 vcpu->arch.dr7 &= ~DR7_GD;
3334 vcpu->arch.dr6 |= DR6_BD;
3335 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3336 kvm_queue_exception(vcpu, DB_VECTOR);
3337 return 1;
3338 }
3339 }
3340
bfdaab09 3341 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
42dbaa5a
JK		 3342 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
3343 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
3344 if (exit_qualification & TYPE_MOV_FROM_DR) {
020df079
GN		 3345 unsigned long val;
3346 if (!kvm_get_dr(vcpu, dr, &val))
3347 kvm_register_write(vcpu, reg, val);
3348 } else
3349 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
6aa8b732
AK		 3350 skip_emulated_instruction(vcpu);
3351 return 1;
3352}
3353
020df079
GN		 3354static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
3355{
3356 vmcs_writel(GUEST_DR7, val);
3357}
3358
851ba692 3359static int handle_cpuid(struct kvm_vcpu *vcpu)
6aa8b732 3360{
06465c5a
AK		 3361 kvm_emulate_cpuid(vcpu);
3362 return 1;
6aa8b732
AK		 3363}
3364
851ba692 3365static int handle_rdmsr(struct kvm_vcpu *vcpu)
6aa8b732 3366{
ad312c7c 3367 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6aa8b732
AK		 3368 u64 data;
3369
3370 if (vmx_get_msr(vcpu, ecx, &data)) {
59200273 3371 trace_kvm_msr_read_ex(ecx);
c1a5d4f9 3372 kvm_inject_gp(vcpu, 0);
6aa8b732
AK		 3373 return 1;
3374 }
3375
229456fc 3376 trace_kvm_msr_read(ecx, data);
2714d1d3 3377
6aa8b732 3378 /* FIXME: handling of bits 32:63 of rax, rdx */
ad312c7c
ZX		 3379 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3380 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
6aa8b732
AK		 3381 skip_emulated_instruction(vcpu);
3382 return 1;
3383}
3384
851ba692 3385static int handle_wrmsr(struct kvm_vcpu *vcpu)
6aa8b732 3386{
ad312c7c
ZX		 3387 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3388 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3389 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6aa8b732
AK		 3390
3391 if (vmx_set_msr(vcpu, ecx, data) != 0) {
59200273 3392 trace_kvm_msr_write_ex(ecx, data);
c1a5d4f9 3393 kvm_inject_gp(vcpu, 0);
6aa8b732
AK		 3394 return 1;
3395 }
3396
59200273 3397 trace_kvm_msr_write(ecx, data);
6aa8b732
AK		 3398 skip_emulated_instruction(vcpu);
3399 return 1;
3400}
3401
851ba692 3402static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6e5d865c 3403{
3842d135 3404 kvm_make_request(KVM_REQ_EVENT, vcpu);
6e5d865c
YS		 3405 return 1;
3406}
3407
851ba692 3408static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6aa8b732 3409{
85f455f7
ED		 3410 u32 cpu_based_vm_exec_control;
3411
3412 /* clear pending irq */
3413 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3414 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3415 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2714d1d3 3416
3842d135
AK		 3417 kvm_make_request(KVM_REQ_EVENT, vcpu);
3418
a26bf12a 3419 ++vcpu->stat.irq_window_exits;
2714d1d3 3420
c1150d8c
DL		 3421 /*
3422 * If the user space waits to inject interrupts, exit as soon as
3423 * possible
3424 */
8061823a 3425 if (!irqchip_in_kernel(vcpu->kvm) &&
851ba692 3426 vcpu->run->request_interrupt_window &&
8061823a 3427 !kvm_cpu_has_interrupt(vcpu)) {
851ba692 3428 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
c1150d8c
DL		 3429 return 0;
3430 }
6aa8b732
AK		 3431 return 1;
3432}
3433
851ba692 3434static int handle_halt(struct kvm_vcpu *vcpu)
6aa8b732
AK		 3435{
3436 skip_emulated_instruction(vcpu);
d3bef15f 3437 return kvm_emulate_halt(vcpu);
6aa8b732
AK		 3438}
3439
851ba692 3440static int handle_vmcall(struct kvm_vcpu *vcpu)
c21415e8 3441{
510043da 3442 skip_emulated_instruction(vcpu);
7aa81cc0
AL		 3443 kvm_emulate_hypercall(vcpu);
3444 return 1;
c21415e8
IM		 3445}
3446
851ba692 3447static int handle_vmx_insn(struct kvm_vcpu *vcpu)
e3c7cb6a
AK		 3448{
3449 kvm_queue_exception(vcpu, UD_VECTOR);
3450 return 1;
3451}
3452
ec25d5e6
GN		 3453static int handle_invd(struct kvm_vcpu *vcpu)
3454{
51d8b661 3455 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
ec25d5e6
GN		 3456}
3457
851ba692 3458static int handle_invlpg(struct kvm_vcpu *vcpu)
a7052897 3459{
f9c617f6 3460 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
a7052897
MT		 3461
3462 kvm_mmu_invlpg(vcpu, exit_qualification);
3463 skip_emulated_instruction(vcpu);
3464 return 1;
3465}
3466
851ba692 3467static int handle_wbinvd(struct kvm_vcpu *vcpu)
e5edaa01
ED		 3468{
3469 skip_emulated_instruction(vcpu);
f5f48ee1 3470 kvm_emulate_wbinvd(vcpu);
e5edaa01
ED		 3471 return 1;
3472}
3473
2acf923e
DC		 3474static int handle_xsetbv(struct kvm_vcpu *vcpu)
3475{
3476 u64 new_bv = kvm_read_edx_eax(vcpu);
3477 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3478
3479 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
3480 skip_emulated_instruction(vcpu);
3481 return 1;
3482}
3483
851ba692 3484static int handle_apic_access(struct kvm_vcpu *vcpu)
f78e0e2e 3485{
51d8b661 3486 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
f78e0e2e
SY		 3487}
3488
851ba692 3489static int handle_task_switch(struct kvm_vcpu *vcpu)
37817f29 3490{
60637aac 3491 struct vcpu_vmx *vmx = to_vmx(vcpu);
37817f29 3492 unsigned long exit_qualification;
e269fb21
JK		 3493 bool has_error_code = false;
3494 u32 error_code = 0;
37817f29 3495 u16 tss_selector;
64a7ec06
GN		 3496 int reason, type, idt_v;
3497
3498 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3499 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
37817f29
IE		 3500
3501 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3502
3503 reason = (u32)exit_qualification >> 30;
64a7ec06
GN		 3504 if (reason == TASK_SWITCH_GATE && idt_v) {
3505 switch (type) {
3506 case INTR_TYPE_NMI_INTR:
3507 vcpu->arch.nmi_injected = false;
3508 if (cpu_has_virtual_nmis())
3509 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3510 GUEST_INTR_STATE_NMI);
3511 break;
3512 case INTR_TYPE_EXT_INTR:
66fd3f7f 3513 case INTR_TYPE_SOFT_INTR:
64a7ec06
GN		 3514 kvm_clear_interrupt_queue(vcpu);
3515 break;
3516 case INTR_TYPE_HARD_EXCEPTION:
e269fb21
JK		 3517 if (vmx->idt_vectoring_info &
3518 VECTORING_INFO_DELIVER_CODE_MASK) {
3519 has_error_code = true;
3520 error_code =
3521 vmcs_read32(IDT_VECTORING_ERROR_CODE);
3522 }
3523 /* fall through */
64a7ec06
GN		 3524 case INTR_TYPE_SOFT_EXCEPTION:
3525 kvm_clear_exception_queue(vcpu);
3526 break;
3527 default:
3528 break;
3529 }
60637aac 3530 }
37817f29
IE		 3531 tss_selector = exit_qualification;
3532
64a7ec06
GN		 3533 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3534 type != INTR_TYPE_EXT_INTR &&
3535 type != INTR_TYPE_NMI_INTR))
3536 skip_emulated_instruction(vcpu);
3537
acb54517
GN		 3538 if (kvm_task_switch(vcpu, tss_selector, reason,
3539 has_error_code, error_code) == EMULATE_FAIL) {
3540 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3541 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3542 vcpu->run->internal.ndata = 0;
42dbaa5a 3543 return 0;
acb54517 3544 }
42dbaa5a
JK		 3545
3546 /* clear all local breakpoint enable flags */
3547 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3548
3549 /*
3550 * TODO: What about debug traps on tss switch?
3551 * Are we supposed to inject them and update dr6?
3552 */
3553
3554 return 1;
37817f29
IE		 3555}
3556
851ba692 3557static int handle_ept_violation(struct kvm_vcpu *vcpu)
1439442c 3558{
f9c617f6 3559 unsigned long exit_qualification;
1439442c 3560 gpa_t gpa;
1439442c 3561 int gla_validity;
1439442c 3562
f9c617f6 3563 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
1439442c
SY		 3564
3565 if (exit_qualification & (1 << 6)) {
3566 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
7f582ab6 3567 return -EINVAL;
1439442c
SY		 3568 }
3569
3570 gla_validity = (exit_qualification >> 7) & 0x3;
3571 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3572 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3573 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3574 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
f9c617f6 3575 vmcs_readl(GUEST_LINEAR_ADDRESS));
1439442c
SY		 3576 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3577 (long unsigned int)exit_qualification);
851ba692
AK		 3578 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3579 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
596ae895 3580 return 0;
1439442c
SY		 3581 }
3582
3583 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
229456fc 3584 trace_kvm_page_fault(gpa, exit_qualification);
dc25e89e 3585 return kvm_mmu_page_fault(vcpu, gpa, exit_qualification & 0x3, NULL, 0);
1439442c
SY		 3586}
3587
68f89400
MT		 3588static u64 ept_rsvd_mask(u64 spte, int level)
3589{
3590 int i;
3591 u64 mask = 0;
3592
3593 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3594 mask |= (1ULL << i);
3595
3596 if (level > 2)
3597 /* bits 7:3 reserved */
3598 mask |= 0xf8;
3599 else if (level == 2) {
3600 if (spte & (1ULL << 7))
3601 /* 2MB ref, bits 20:12 reserved */
3602 mask |= 0x1ff000;
3603 else
3604 /* bits 6:3 reserved */
3605 mask |= 0x78;
3606 }
3607
3608 return mask;
3609}
3610
3611static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3612 int level)
3613{
3614 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3615
3616 /* 010b (write-only) */
3617 WARN_ON((spte & 0x7) == 0x2);
3618
3619 /* 110b (write/execute) */
3620 WARN_ON((spte & 0x7) == 0x6);
3621
3622 /* 100b (execute-only) and value not supported by logical processor */
3623 if (!cpu_has_vmx_ept_execute_only())
3624 WARN_ON((spte & 0x7) == 0x4);
3625
3626 /* not 000b */
3627 if ((spte & 0x7)) {
3628 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3629
3630 if (rsvd_bits != 0) {
3631 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3632 __func__, rsvd_bits);
3633 WARN_ON(1);
3634 }
3635
3636 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3637 u64 ept_mem_type = (spte & 0x38) >> 3;
3638
3639 if (ept_mem_type == 2 || ept_mem_type == 3 ||
3640 ept_mem_type == 7) {
3641 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3642 __func__, ept_mem_type);
3643 WARN_ON(1);
3644 }
3645 }
3646 }
3647}
3648
851ba692 3649static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
68f89400
MT		 3650{
3651 u64 sptes[4];
3652 int nr_sptes, i;
3653 gpa_t gpa;
3654
3655 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3656
3657 printk(KERN_ERR "EPT: Misconfiguration.\n");
3658 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3659
3660 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3661
3662 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3663 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3664
851ba692
AK		 3665 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3666 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
68f89400
MT		 3667
3668 return 0;
3669}
3670
851ba692 3671static int handle_nmi_window(struct kvm_vcpu *vcpu)
f08864b4
SY		 3672{
3673 u32 cpu_based_vm_exec_control;
3674
3675 /* clear pending NMI */
3676 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3677 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3678 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3679 ++vcpu->stat.nmi_window_exits;
3842d135 3680 kvm_make_request(KVM_REQ_EVENT, vcpu);
f08864b4
SY		 3681
3682 return 1;
3683}
3684
80ced186 3685static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
ea953ef0 3686{
8b3079a5
AK		 3687 struct vcpu_vmx *vmx = to_vmx(vcpu);
3688 enum emulation_result err = EMULATE_DONE;
80ced186 3689 int ret = 1;
49e9d557
AK		 3690 u32 cpu_exec_ctrl;
3691 bool intr_window_requested;
3692
3693 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3694 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
ea953ef0
MG		 3695
3696 while (!guest_state_valid(vcpu)) {
49e9d557
AK		 3697 if (intr_window_requested
3698 && (kvm_get_rflags(&vmx->vcpu) & X86_EFLAGS_IF))
3699 return handle_interrupt_window(&vmx->vcpu);
3700
51d8b661 3701 err = emulate_instruction(vcpu, 0);
ea953ef0 3702
80ced186
MG		 3703 if (err == EMULATE_DO_MMIO) {
3704 ret = 0;
3705 goto out;
3706 }
1d5a4d9b 3707
6d77dbfc
GN		 3708 if (err != EMULATE_DONE)
3709 return 0;
ea953ef0
MG		 3710
3711 if (signal_pending(current))
80ced186 3712 goto out;
ea953ef0
MG		 3713 if (need_resched())
3714 schedule();
3715 }
3716
80ced186
MG		 3717 vmx->emulation_required = 0;
3718out:
3719 return ret;
ea953ef0
MG		 3720}
3721
4b8d54f9
ZE		 3722/*
3723 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3724 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3725 */
9fb41ba8 3726static int handle_pause(struct kvm_vcpu *vcpu)
4b8d54f9
ZE		 3727{
3728 skip_emulated_instruction(vcpu);
3729 kvm_vcpu_on_spin(vcpu);
3730
3731 return 1;
3732}
3733
59708670
SY		 3734static int handle_invalid_op(struct kvm_vcpu *vcpu)
3735{
3736 kvm_queue_exception(vcpu, UD_VECTOR);
3737 return 1;
3738}
3739
6aa8b732
AK		 3740/*
3741 * The exit handlers return 1 if the exit was handled fully and guest execution
3742 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
3743 * to be done to userspace and return 0.
3744 */
851ba692 3745static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
6aa8b732
AK		 3746 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
3747 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
988ad74f 3748 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
f08864b4 3749 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
6aa8b732 3750 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
6aa8b732
AK		 3751 [EXIT_REASON_CR_ACCESS] = handle_cr,
3752 [EXIT_REASON_DR_ACCESS] = handle_dr,
3753 [EXIT_REASON_CPUID] = handle_cpuid,
3754 [EXIT_REASON_MSR_READ] = handle_rdmsr,
3755 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
3756 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
3757 [EXIT_REASON_HLT] = handle_halt,
ec25d5e6 3758 [EXIT_REASON_INVD] = handle_invd,
a7052897 3759 [EXIT_REASON_INVLPG] = handle_invlpg,
c21415e8 3760 [EXIT_REASON_VMCALL] = handle_vmcall,
e3c7cb6a
AK		 3761 [EXIT_REASON_VMCLEAR] = handle_vmx_insn,
3762 [EXIT_REASON_VMLAUNCH] = handle_vmx_insn,
3763 [EXIT_REASON_VMPTRLD] = handle_vmx_insn,
3764 [EXIT_REASON_VMPTRST] = handle_vmx_insn,
3765 [EXIT_REASON_VMREAD] = handle_vmx_insn,
3766 [EXIT_REASON_VMRESUME] = handle_vmx_insn,
3767 [EXIT_REASON_VMWRITE] = handle_vmx_insn,
3768 [EXIT_REASON_VMOFF] = handle_vmx_insn,
3769 [EXIT_REASON_VMON] = handle_vmx_insn,
f78e0e2e
SY		 3770 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
3771 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
e5edaa01 3772 [EXIT_REASON_WBINVD] = handle_wbinvd,
2acf923e 3773 [EXIT_REASON_XSETBV] = handle_xsetbv,
37817f29 3774 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
a0861c02 3775 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
68f89400
MT		 3776 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
3777 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
4b8d54f9 3778 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
59708670
SY		 3779 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_invalid_op,
3780 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_invalid_op,
6aa8b732
AK		 3781};
3782
3783static const int kvm_vmx_max_exit_handlers =
50a3485c 3784 ARRAY_SIZE(kvm_vmx_exit_handlers);
6aa8b732 3785
586f9607
AK		 3786static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3787{
3788 *info1 = vmcs_readl(EXIT_QUALIFICATION);
3789 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
3790}
3791
6aa8b732
AK		 3792/*
3793 * The guest has exited. See if we can fix it or if we need userspace
3794 * assistance.
3795 */
851ba692 3796static int vmx_handle_exit(struct kvm_vcpu *vcpu)
6aa8b732 3797{
29bd8a78 3798 struct vcpu_vmx *vmx = to_vmx(vcpu);
a0861c02 3799 u32 exit_reason = vmx->exit_reason;
1155f76a 3800 u32 vectoring_info = vmx->idt_vectoring_info;
29bd8a78 3801
aa17911e 3802 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
2714d1d3 3803
80ced186
MG		 3804 /* If guest state is invalid, start emulating */
3805 if (vmx->emulation_required && emulate_invalid_guest_state)
3806 return handle_invalid_guest_state(vcpu);
1d5a4d9b 3807
5120702e
MG		 3808 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
3809 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3810 vcpu->run->fail_entry.hardware_entry_failure_reason
3811 = exit_reason;
3812 return 0;
3813 }
3814
29bd8a78 3815 if (unlikely(vmx->fail)) {
851ba692
AK		 3816 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3817 vcpu->run->fail_entry.hardware_entry_failure_reason
29bd8a78
AK		 3818 = vmcs_read32(VM_INSTRUCTION_ERROR);
3819 return 0;
3820 }
6aa8b732 3821
d77c26fc 3822 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
1439442c 3823 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
60637aac
JK		 3824 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3825 exit_reason != EXIT_REASON_TASK_SWITCH))
3826 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3827 "(0x%x) and exit reason is 0x%x\n",
3828 __func__, vectoring_info, exit_reason);
3b86cd99
JK		 3829
3830 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
c4282df9 3831 if (vmx_interrupt_allowed(vcpu)) {
3b86cd99 3832 vmx->soft_vnmi_blocked = 0;
3b86cd99 3833 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
4531220b 3834 vcpu->arch.nmi_pending) {
3b86cd99
JK		 3835 /*
3836 * This CPU don't support us in finding the end of an
3837 * NMI-blocked window if the guest runs with IRQs
3838 * disabled. So we pull the trigger after 1 s of
3839 * futile waiting, but inform the user about this.
3840 */
3841 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3842 "state on VCPU %d after 1 s timeout\n",
3843 __func__, vcpu->vcpu_id);
3844 vmx->soft_vnmi_blocked = 0;
3b86cd99 3845 }
3b86cd99
JK		 3846 }
3847
6aa8b732
AK		 3848 if (exit_reason < kvm_vmx_max_exit_handlers
3849 && kvm_vmx_exit_handlers[exit_reason])
851ba692 3850 return kvm_vmx_exit_handlers[exit_reason](vcpu);
6aa8b732 3851 else {
851ba692
AK		 3852 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3853 vcpu->run->hw.hardware_exit_reason = exit_reason;
6aa8b732
AK		 3854 }
3855 return 0;
3856}
3857
95ba8273 3858static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6e5d865c 3859{
95ba8273 3860 if (irr == -1 || tpr < irr) {
6e5d865c
YS		 3861 vmcs_write32(TPR_THRESHOLD, 0);
3862 return;
3863 }
3864
95ba8273 3865 vmcs_write32(TPR_THRESHOLD, irr);
6e5d865c
YS		 3866}
3867
51aa01d1 3868static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
cf393f75 3869{
51aa01d1 3870 u32 exit_intr_info = vmx->exit_intr_info;
a0861c02
AK		 3871
3872 /* Handle machine checks before interrupts are enabled */
3873 if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3874 || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3875 && is_machine_check(exit_intr_info)))
3876 kvm_machine_check();
3877
20f65983
GN		 3878 /* We need to handle NMIs before interrupts are enabled */
3879 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
ff9d07a0
ZY		 3880 (exit_intr_info & INTR_INFO_VALID_MASK)) {
3881 kvm_before_handle_nmi(&vmx->vcpu);
20f65983 3882 asm("int $2");
ff9d07a0
ZY		 3883 kvm_after_handle_nmi(&vmx->vcpu);
3884 }
51aa01d1 3885}
20f65983 3886
51aa01d1
AK		 3887static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
3888{
3889 u32 exit_intr_info = vmx->exit_intr_info;
3890 bool unblock_nmi;
3891 u8 vector;
3892 bool idtv_info_valid;
3893
3894 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
20f65983 3895
cf393f75
AK		 3896 if (cpu_has_virtual_nmis()) {
3897 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3898 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3899 /*
7b4a25cb 3900 * SDM 3: 27.7.1.2 (September 2008)
cf393f75
AK		 3901 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3902 * a guest IRET fault.
7b4a25cb
GN		 3903 * SDM 3: 23.2.2 (September 2008)
3904 * Bit 12 is undefined in any of the following cases:
3905 * If the VM exit sets the valid bit in the IDT-vectoring
3906 * information field.
3907 * If the VM exit is due to a double fault.
cf393f75 3908 */
7b4a25cb
GN		 3909 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3910 vector != DF_VECTOR && !idtv_info_valid)
cf393f75
AK		 3911 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3912 GUEST_INTR_STATE_NMI);
3b86cd99
JK		 3913 } else if (unlikely(vmx->soft_vnmi_blocked))
3914 vmx->vnmi_blocked_time +=
3915 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
51aa01d1
AK		 3916}
3917
83422e17
AK		 3918static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
3919 u32 idt_vectoring_info,
3920 int instr_len_field,
3921 int error_code_field)
51aa01d1 3922{
51aa01d1
AK		 3923 u8 vector;
3924 int type;
3925 bool idtv_info_valid;
3926
3927 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
668f612f 3928
37b96e98
GN		 3929 vmx->vcpu.arch.nmi_injected = false;
3930 kvm_clear_exception_queue(&vmx->vcpu);
3931 kvm_clear_interrupt_queue(&vmx->vcpu);
3932
3933 if (!idtv_info_valid)
3934 return;
3935
3842d135
AK		 3936 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
3937
668f612f
AK		 3938 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3939 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
37b96e98 3940
64a7ec06 3941 switch (type) {
37b96e98
GN		 3942 case INTR_TYPE_NMI_INTR:
3943 vmx->vcpu.arch.nmi_injected = true;
668f612f 3944 /*
7b4a25cb 3945 * SDM 3: 27.7.1.2 (September 2008)
37b96e98
GN		 3946 * Clear bit "block by NMI" before VM entry if a NMI
3947 * delivery faulted.
668f612f 3948 */
37b96e98
GN		 3949 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3950 GUEST_INTR_STATE_NMI);
3951 break;
37b96e98 3952 case INTR_TYPE_SOFT_EXCEPTION:
66fd3f7f 3953 vmx->vcpu.arch.event_exit_inst_len =
83422e17 3954 vmcs_read32(instr_len_field);
66fd3f7f
GN		 3955 /* fall through */
3956 case INTR_TYPE_HARD_EXCEPTION:
35920a35 3957 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
83422e17 3958 u32 err = vmcs_read32(error_code_field);
37b96e98 3959 kvm_queue_exception_e(&vmx->vcpu, vector, err);
35920a35
AK		 3960 } else
3961 kvm_queue_exception(&vmx->vcpu, vector);
37b96e98 3962 break;
66fd3f7f
GN		 3963 case INTR_TYPE_SOFT_INTR:
3964 vmx->vcpu.arch.event_exit_inst_len =
83422e17 3965 vmcs_read32(instr_len_field);
66fd3f7f 3966 /* fall through */
37b96e98 3967 case INTR_TYPE_EXT_INTR:
66fd3f7f
GN		 3968 kvm_queue_interrupt(&vmx->vcpu, vector,
3969 type == INTR_TYPE_SOFT_INTR);
37b96e98
GN		 3970 break;
3971 default:
3972 break;
f7d9238f 3973 }
cf393f75
AK		 3974}
3975
83422e17
AK		 3976static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3977{
3978 __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
3979 VM_EXIT_INSTRUCTION_LEN,
3980 IDT_VECTORING_ERROR_CODE);
3981}
3982
b463a6f7
AK		 3983static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
3984{
3985 __vmx_complete_interrupts(to_vmx(vcpu),
3986 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
3987 VM_ENTRY_INSTRUCTION_LEN,
3988 VM_ENTRY_EXCEPTION_ERROR_CODE);
3989
3990 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
3991}
3992
c801949d
AK		 3993#ifdef CONFIG_X86_64
3994#define R "r"
3995#define Q "q"
3996#else
3997#define R "e"
3998#define Q "l"
3999#endif
4000
a3b5ba49 4001static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
6aa8b732 4002{
a2fa3e9f 4003 struct vcpu_vmx *vmx = to_vmx(vcpu);
104f226b
AK		 4004
4005 /* Record the guest's net vcpu time for enforced NMI injections. */
4006 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
4007 vmx->entry_time = ktime_get();
4008
4009 /* Don't enter VMX if guest state is invalid, let the exit handler
4010 start emulation until we arrive back to a valid state */
4011 if (vmx->emulation_required && emulate_invalid_guest_state)
4012 return;
4013
4014 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
4015 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
4016 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
4017 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
4018
4019 /* When single-stepping over STI and MOV SS, we must clear the
4020 * corresponding interruptibility bits in the guest state. Otherwise
4021 * vmentry fails as it then expects bit 14 (BS) in pending debug
4022 * exceptions being set, but that's not correct for the guest debugging
4023 * case. */
4024 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
4025 vmx_set_interrupt_shadow(vcpu, 0);
4026
4027 asm(
6aa8b732 4028 /* Store host registers */
c801949d 4029 "push %%"R"dx; push %%"R"bp;"
40712fae 4030 "push %%"R"cx \n\t" /* placeholder for guest rcx */
c801949d 4031 "push %%"R"cx \n\t"
313dbd49
AK		 4032 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
4033 "je 1f \n\t"
4034 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
4ecac3fd 4035 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
313dbd49 4036 "1: \n\t"
d3edefc0
AK		 4037 /* Reload cr2 if changed */
4038 "mov %c[cr2](%0), %%"R"ax \n\t"
4039 "mov %%cr2, %%"R"dx \n\t"
4040 "cmp %%"R"ax, %%"R"dx \n\t"
4041 "je 2f \n\t"
4042 "mov %%"R"ax, %%cr2 \n\t"
4043 "2: \n\t"
6aa8b732 4044 /* Check if vmlaunch of vmresume is needed */
e08aa78a 4045 "cmpl $0, %c[launched](%0) \n\t"
6aa8b732 4046 /* Load guest registers. Don't clobber flags. */
c801949d
AK		 4047 "mov %c[rax](%0), %%"R"ax \n\t"
4048 "mov %c[rbx](%0), %%"R"bx \n\t"
4049 "mov %c[rdx](%0), %%"R"dx \n\t"
4050 "mov %c[rsi](%0), %%"R"si \n\t"
4051 "mov %c[rdi](%0), %%"R"di \n\t"
4052 "mov %c[rbp](%0), %%"R"bp \n\t"
05b3e0c2 4053#ifdef CONFIG_X86_64
e08aa78a
AK		 4054 "mov %c[r8](%0), %%r8 \n\t"
4055 "mov %c[r9](%0), %%r9 \n\t"
4056 "mov %c[r10](%0), %%r10 \n\t"
4057 "mov %c[r11](%0), %%r11 \n\t"
4058 "mov %c[r12](%0), %%r12 \n\t"
4059 "mov %c[r13](%0), %%r13 \n\t"
4060 "mov %c[r14](%0), %%r14 \n\t"
4061 "mov %c[r15](%0), %%r15 \n\t"
6aa8b732 4062#endif
c801949d
AK		 4063 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
4064
6aa8b732 4065 /* Enter guest mode */
cd2276a7 4066 "jne .Llaunched \n\t"
4ecac3fd 4067 __ex(ASM_VMX_VMLAUNCH) "\n\t"
cd2276a7 4068 "jmp .Lkvm_vmx_return \n\t"
4ecac3fd 4069 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
cd2276a7 4070 ".Lkvm_vmx_return: "
6aa8b732 4071 /* Save guest registers, load host registers, keep flags */
40712fae
AK		 4072 "mov %0, %c[wordsize](%%"R"sp) \n\t"
4073 "pop %0 \n\t"
c801949d
AK		 4074 "mov %%"R"ax, %c[rax](%0) \n\t"
4075 "mov %%"R"bx, %c[rbx](%0) \n\t"
1c696d0e 4076 "pop"Q" %c[rcx](%0) \n\t"
c801949d
AK		 4077 "mov %%"R"dx, %c[rdx](%0) \n\t"
4078 "mov %%"R"si, %c[rsi](%0) \n\t"
4079 "mov %%"R"di, %c[rdi](%0) \n\t"
4080 "mov %%"R"bp, %c[rbp](%0) \n\t"
05b3e0c2 4081#ifdef CONFIG_X86_64
e08aa78a
AK		 4082 "mov %%r8, %c[r8](%0) \n\t"
4083 "mov %%r9, %c[r9](%0) \n\t"
4084 "mov %%r10, %c[r10](%0) \n\t"
4085 "mov %%r11, %c[r11](%0) \n\t"
4086 "mov %%r12, %c[r12](%0) \n\t"
4087 "mov %%r13, %c[r13](%0) \n\t"
4088 "mov %%r14, %c[r14](%0) \n\t"
4089 "mov %%r15, %c[r15](%0) \n\t"
6aa8b732 4090#endif
c801949d
AK		 4091 "mov %%cr2, %%"R"ax \n\t"
4092 "mov %%"R"ax, %c[cr2](%0) \n\t"
4093
1c696d0e 4094 "pop %%"R"bp; pop %%"R"dx \n\t"
e08aa78a
AK		 4095 "setbe %c[fail](%0) \n\t"
4096 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
4097 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
4098 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
313dbd49 4099 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
ad312c7c
ZX		 4100 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
4101 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
4102 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
4103 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
4104 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
4105 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
4106 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
05b3e0c2 4107#ifdef CONFIG_X86_64
ad312c7c
ZX		 4108 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
4109 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
4110 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
4111 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
4112 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
4113 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
4114 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
4115 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
6aa8b732 4116#endif
40712fae
AK		 4117 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
4118 [wordsize]"i"(sizeof(ulong))
c2036300 4119 : "cc", "memory"
07d6f555 4120 , R"ax", R"bx", R"di", R"si"
c2036300 4121#ifdef CONFIG_X86_64
c2036300
LV		 4122 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
4123#endif
4124 );
6aa8b732 4125
6de4f3ad 4126 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
aff48baa
AK		 4127 | (1 << VCPU_EXREG_PDPTR)
4128 | (1 << VCPU_EXREG_CR3));
5fdbf976
MT		 4129 vcpu->arch.regs_dirty = 0;
4130
1155f76a
AK		 4131 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
4132
d77c26fc 4133 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
15ad7146 4134 vmx->launched = 1;
1b6269db 4135
51aa01d1
AK		 4136 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
4137 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
4138
4139 vmx_complete_atomic_exit(vmx);
4140 vmx_recover_nmi_blocking(vmx);
cf393f75 4141 vmx_complete_interrupts(vmx);
6aa8b732
AK		 4142}
4143
c801949d
AK		 4144#undef R
4145#undef Q
4146
6aa8b732
AK		 4147static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
4148{
a2fa3e9f
GH		 4149 struct vcpu_vmx *vmx = to_vmx(vcpu);
4150
4151 if (vmx->vmcs) {
543e4243 4152 vcpu_clear(vmx);
a2fa3e9f
GH		 4153 free_vmcs(vmx->vmcs);
4154 vmx->vmcs = NULL;
6aa8b732
AK		 4155 }
4156}
4157
4158static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
4159{
fb3f0f51
RR		 4160 struct vcpu_vmx *vmx = to_vmx(vcpu);
4161
cdbecfc3 4162 free_vpid(vmx);
6aa8b732 4163 vmx_free_vmcs(vcpu);
fb3f0f51
RR		 4164 kfree(vmx->guest_msrs);
4165 kvm_vcpu_uninit(vcpu);
a4770347 4166 kmem_cache_free(kvm_vcpu_cache, vmx);
6aa8b732
AK		 4167}
4168
4610c9cc
DX		 4169static inline void vmcs_init(struct vmcs *vmcs)
4170{
4171 u64 phys_addr = __pa(per_cpu(vmxarea, raw_smp_processor_id()));
4172
4173 if (!vmm_exclusive)
4174 kvm_cpu_vmxon(phys_addr);
4175
4176 vmcs_clear(vmcs);
4177
4178 if (!vmm_exclusive)
4179 kvm_cpu_vmxoff();
4180}
4181
fb3f0f51 4182static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6aa8b732 4183{
fb3f0f51 4184 int err;
c16f862d 4185 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
15ad7146 4186 int cpu;
6aa8b732 4187
a2fa3e9f 4188 if (!vmx)
fb3f0f51
RR		 4189 return ERR_PTR(-ENOMEM);
4190
2384d2b3
SY		 4191 allocate_vpid(vmx);
4192
fb3f0f51
RR		 4193 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
4194 if (err)
4195 goto free_vcpu;
965b58a5 4196
a2fa3e9f 4197 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
fb3f0f51
RR		 4198 if (!vmx->guest_msrs) {
4199 err = -ENOMEM;
4200 goto uninit_vcpu;
4201 }
965b58a5 4202
a2fa3e9f
GH		 4203 vmx->vmcs = alloc_vmcs();
4204 if (!vmx->vmcs)
fb3f0f51 4205 goto free_msrs;
a2fa3e9f 4206
4610c9cc 4207 vmcs_init(vmx->vmcs);
a2fa3e9f 4208
15ad7146
AK		 4209 cpu = get_cpu();
4210 vmx_vcpu_load(&vmx->vcpu, cpu);
e48672fa 4211 vmx->vcpu.cpu = cpu;
8b9cf98c 4212 err = vmx_vcpu_setup(vmx);
fb3f0f51 4213 vmx_vcpu_put(&vmx->vcpu);
15ad7146 4214 put_cpu();
fb3f0f51
RR		 4215 if (err)
4216 goto free_vmcs;
5e4a0b3c
MT		 4217 if (vm_need_virtualize_apic_accesses(kvm))
4218 if (alloc_apic_access_page(kvm) != 0)
4219 goto free_vmcs;
fb3f0f51 4220
b927a3ce
SY		 4221 if (enable_ept) {
4222 if (!kvm->arch.ept_identity_map_addr)
4223 kvm->arch.ept_identity_map_addr =
4224 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
93ea5388 4225 err = -ENOMEM;
b7ebfb05
SY		 4226 if (alloc_identity_pagetable(kvm) != 0)
4227 goto free_vmcs;
93ea5388
GN		 4228 if (!init_rmode_identity_map(kvm))
4229 goto free_vmcs;
b927a3ce 4230 }
b7ebfb05 4231
fb3f0f51
RR		 4232 return &vmx->vcpu;
4233
4234free_vmcs:
4235 free_vmcs(vmx->vmcs);
4236free_msrs:
fb3f0f51
RR		 4237 kfree(vmx->guest_msrs);
4238uninit_vcpu:
4239 kvm_vcpu_uninit(&vmx->vcpu);
4240free_vcpu:
cdbecfc3 4241 free_vpid(vmx);
a4770347 4242 kmem_cache_free(kvm_vcpu_cache, vmx);
fb3f0f51 4243 return ERR_PTR(err);
6aa8b732
AK		 4244}
4245
002c7f7c
YS		 4246static void __init vmx_check_processor_compat(void *rtn)
4247{
4248 struct vmcs_config vmcs_conf;
4249
4250 *(int *)rtn = 0;
4251 if (setup_vmcs_config(&vmcs_conf) < 0)
4252 *(int *)rtn = -EIO;
4253 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
4254 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
4255 smp_processor_id());
4256 *(int *)rtn = -EIO;
4257 }
4258}
4259
67253af5
SY		 4260static int get_ept_level(void)
4261{
4262 return VMX_EPT_DEFAULT_GAW + 1;
4263}
4264
4b12f0de 4265static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
64d4d521 4266{
4b12f0de
SY		 4267 u64 ret;
4268
522c68c4
SY		 4269 /* For VT-d and EPT combination
4270 * 1. MMIO: always map as UC
4271 * 2. EPT with VT-d:
4272 * a. VT-d without snooping control feature: can't guarantee the
4273 * result, try to trust guest.
4274 * b. VT-d with snooping control feature: snooping control feature of
4275 * VT-d engine can guarantee the cache correctness. Just set it
4276 * to WB to keep consistent with host. So the same as item 3.
a19a6d11 4277 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
522c68c4
SY		 4278 * consistent with host MTRR
4279 */
4b12f0de
SY		 4280 if (is_mmio)
4281 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
522c68c4
SY		 4282 else if (vcpu->kvm->arch.iommu_domain &&
4283 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
4284 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
4285 VMX_EPT_MT_EPTE_SHIFT;
4b12f0de 4286 else
522c68c4 4287 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
a19a6d11 4288 | VMX_EPT_IPAT_BIT;
4b12f0de
SY		 4289
4290 return ret;
64d4d521
SY		 4291}
4292
f4c9e87c
AK		 4293#define _ER(x) { EXIT_REASON_##x, #x }
4294
229456fc 4295static const struct trace_print_flags vmx_exit_reasons_str[] = {
f4c9e87c
AK		 4296 _ER(EXCEPTION_NMI),
4297 _ER(EXTERNAL_INTERRUPT),
4298 _ER(TRIPLE_FAULT),
4299 _ER(PENDING_INTERRUPT),
4300 _ER(NMI_WINDOW),
4301 _ER(TASK_SWITCH),
4302 _ER(CPUID),
4303 _ER(HLT),
4304 _ER(INVLPG),
4305 _ER(RDPMC),
4306 _ER(RDTSC),
4307 _ER(VMCALL),
4308 _ER(VMCLEAR),
4309 _ER(VMLAUNCH),
4310 _ER(VMPTRLD),
4311 _ER(VMPTRST),
4312 _ER(VMREAD),
4313 _ER(VMRESUME),
4314 _ER(VMWRITE),
4315 _ER(VMOFF),
4316 _ER(VMON),
4317 _ER(CR_ACCESS),
4318 _ER(DR_ACCESS),
4319 _ER(IO_INSTRUCTION),
4320 _ER(MSR_READ),
4321 _ER(MSR_WRITE),
4322 _ER(MWAIT_INSTRUCTION),
4323 _ER(MONITOR_INSTRUCTION),
4324 _ER(PAUSE_INSTRUCTION),
4325 _ER(MCE_DURING_VMENTRY),
4326 _ER(TPR_BELOW_THRESHOLD),
4327 _ER(APIC_ACCESS),
4328 _ER(EPT_VIOLATION),
4329 _ER(EPT_MISCONFIG),
4330 _ER(WBINVD),
229456fc
MT		 4331 { -1, NULL }
4332};
4333
f4c9e87c
AK		 4334#undef _ER
4335
17cc3935 4336static int vmx_get_lpage_level(void)
344f414f 4337{
878403b7
SY		 4338 if (enable_ept && !cpu_has_vmx_ept_1g_page())
4339 return PT_DIRECTORY_LEVEL;
4340 else
4341 /* For shadow and EPT supported 1GB page */
4342 return PT_PDPE_LEVEL;
344f414f
JR		 4343}
4344
0e851880
SY		 4345static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
4346{
4e47c7a6
SY		 4347 struct kvm_cpuid_entry2 *best;
4348 struct vcpu_vmx *vmx = to_vmx(vcpu);
4349 u32 exec_control;
4350
4351 vmx->rdtscp_enabled = false;
4352 if (vmx_rdtscp_supported()) {
4353 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
4354 if (exec_control & SECONDARY_EXEC_RDTSCP) {
4355 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
4356 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
4357 vmx->rdtscp_enabled = true;
4358 else {
4359 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4360 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4361 exec_control);
4362 }
4363 }
4364 }
0e851880
SY		 4365}
4366
d4330ef2
JR		 4367static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4368{
4369}
4370
cbdd1bea 4371static struct kvm_x86_ops vmx_x86_ops = {
6aa8b732
AK		 4372 .cpu_has_kvm_support = cpu_has_kvm_support,
4373 .disabled_by_bios = vmx_disabled_by_bios,
4374 .hardware_setup = hardware_setup,
4375 .hardware_unsetup = hardware_unsetup,
002c7f7c 4376 .check_processor_compatibility = vmx_check_processor_compat,
6aa8b732
AK		 4377 .hardware_enable = hardware_enable,
4378 .hardware_disable = hardware_disable,
04547156 4379 .cpu_has_accelerated_tpr = report_flexpriority,
6aa8b732
AK		 4380
4381 .vcpu_create = vmx_create_vcpu,
4382 .vcpu_free = vmx_free_vcpu,
04d2cc77 4383 .vcpu_reset = vmx_vcpu_reset,
6aa8b732 4384
04d2cc77 4385 .prepare_guest_switch = vmx_save_host_state,
6aa8b732
AK		 4386 .vcpu_load = vmx_vcpu_load,
4387 .vcpu_put = vmx_vcpu_put,
4388
4389 .set_guest_debug = set_guest_debug,
4390 .get_msr = vmx_get_msr,
4391 .set_msr = vmx_set_msr,
4392 .get_segment_base = vmx_get_segment_base,
4393 .get_segment = vmx_get_segment,
4394 .set_segment = vmx_set_segment,
2e4d2653 4395 .get_cpl = vmx_get_cpl,
6aa8b732 4396 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
e8467fda 4397 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
aff48baa 4398 .decache_cr3 = vmx_decache_cr3,
25c4c276 4399 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
6aa8b732 4400 .set_cr0 = vmx_set_cr0,
6aa8b732
AK		 4401 .set_cr3 = vmx_set_cr3,
4402 .set_cr4 = vmx_set_cr4,
6aa8b732 4403 .set_efer = vmx_set_efer,
6aa8b732
AK		 4404 .get_idt = vmx_get_idt,
4405 .set_idt = vmx_set_idt,
4406 .get_gdt = vmx_get_gdt,
4407 .set_gdt = vmx_set_gdt,
020df079 4408 .set_dr7 = vmx_set_dr7,
5fdbf976 4409 .cache_reg = vmx_cache_reg,
6aa8b732
AK		 4410 .get_rflags = vmx_get_rflags,
4411 .set_rflags = vmx_set_rflags,
ebcbab4c 4412 .fpu_activate = vmx_fpu_activate,
02daab21 4413 .fpu_deactivate = vmx_fpu_deactivate,
6aa8b732
AK		 4414
4415 .tlb_flush = vmx_flush_tlb,
6aa8b732 4416
6aa8b732 4417 .run = vmx_vcpu_run,
6062d012 4418 .handle_exit = vmx_handle_exit,
6aa8b732 4419 .skip_emulated_instruction = skip_emulated_instruction,
2809f5d2
GC		 4420 .set_interrupt_shadow = vmx_set_interrupt_shadow,
4421 .get_interrupt_shadow = vmx_get_interrupt_shadow,
102d8325 4422 .patch_hypercall = vmx_patch_hypercall,
2a8067f1 4423 .set_irq = vmx_inject_irq,
95ba8273 4424 .set_nmi = vmx_inject_nmi,
298101da 4425 .queue_exception = vmx_queue_exception,
b463a6f7 4426 .cancel_injection = vmx_cancel_injection,
78646121 4427 .interrupt_allowed = vmx_interrupt_allowed,
95ba8273 4428 .nmi_allowed = vmx_nmi_allowed,
3cfc3092
JK		 4429 .get_nmi_mask = vmx_get_nmi_mask,
4430 .set_nmi_mask = vmx_set_nmi_mask,
95ba8273
GN		 4431 .enable_nmi_window = enable_nmi_window,
4432 .enable_irq_window = enable_irq_window,
4433 .update_cr8_intercept = update_cr8_intercept,
95ba8273 4434
cbc94022 4435 .set_tss_addr = vmx_set_tss_addr,
67253af5 4436 .get_tdp_level = get_ept_level,
4b12f0de 4437 .get_mt_mask = vmx_get_mt_mask,
229456fc 4438
586f9607 4439 .get_exit_info = vmx_get_exit_info,
229456fc 4440 .exit_reasons_str = vmx_exit_reasons_str,
586f9607 4441
17cc3935 4442 .get_lpage_level = vmx_get_lpage_level,
0e851880
SY		 4443
4444 .cpuid_update = vmx_cpuid_update,
4e47c7a6
SY		 4445
4446 .rdtscp_supported = vmx_rdtscp_supported,
d4330ef2
JR		 4447
4448 .set_supported_cpuid = vmx_set_supported_cpuid,
f5f48ee1
SY		 4449
4450 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
99e3e30a
ZA		 4451
4452 .write_tsc_offset = vmx_write_tsc_offset,
e48672fa 4453 .adjust_tsc_offset = vmx_adjust_tsc_offset,
1c97f0a0
JR		 4454
4455 .set_tdp_cr3 = vmx_set_cr3,
6aa8b732
AK		 4456};
4457
4458static int __init vmx_init(void)
4459{
26bb0981
AK		 4460 int r, i;
4461
4462 rdmsrl_safe(MSR_EFER, &host_efer);
4463
4464 for (i = 0; i < NR_VMX_MSR; ++i)
4465 kvm_define_shared_msr(i, vmx_msr_index[i]);
fdef3ad1 4466
3e7c73e9 4467 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
fdef3ad1
HQ		 4468 if (!vmx_io_bitmap_a)
4469 return -ENOMEM;
4470
3e7c73e9 4471 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
fdef3ad1
HQ		 4472 if (!vmx_io_bitmap_b) {
4473 r = -ENOMEM;
4474 goto out;
4475 }
4476
5897297b
AK		 4477 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
4478 if (!vmx_msr_bitmap_legacy) {
25c5f225
SY		 4479 r = -ENOMEM;
4480 goto out1;
4481 }
4482
5897297b
AK		 4483 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
4484 if (!vmx_msr_bitmap_longmode) {
4485 r = -ENOMEM;
4486 goto out2;
4487 }
4488
fdef3ad1
HQ		 4489 /*
4490 * Allow direct access to the PC debug port (it is often used for I/O
4491 * delays, but the vmexits simply slow things down).
4492 */
3e7c73e9
AK		 4493 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4494 clear_bit(0x80, vmx_io_bitmap_a);
fdef3ad1 4495
3e7c73e9 4496 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
fdef3ad1 4497
5897297b
AK		 4498 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4499 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
25c5f225 4500
2384d2b3
SY		 4501 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4502
0ee75bea
AK		 4503 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
4504 __alignof__(struct vcpu_vmx), THIS_MODULE);
fdef3ad1 4505 if (r)
5897297b 4506 goto out3;
25c5f225 4507
5897297b
AK		 4508 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4509 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4510 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4511 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4512 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4513 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
fdef3ad1 4514
089d034e 4515 if (enable_ept) {
1439442c 4516 bypass_guest_pf = 0;
534e38b4 4517 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4b12f0de 4518 VMX_EPT_EXECUTABLE_MASK);
5fdbcb9d
SY		 4519 kvm_enable_tdp();
4520 } else
4521 kvm_disable_tdp();
1439442c 4522
c7addb90
AK		 4523 if (bypass_guest_pf)
4524 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4525
fdef3ad1
HQ		 4526 return 0;
4527
5897297b
AK		 4528out3:
4529 free_page((unsigned long)vmx_msr_bitmap_longmode);
25c5f225 4530out2:
5897297b 4531 free_page((unsigned long)vmx_msr_bitmap_legacy);
fdef3ad1 4532out1:
3e7c73e9 4533 free_page((unsigned long)vmx_io_bitmap_b);
fdef3ad1 4534out:
3e7c73e9 4535 free_page((unsigned long)vmx_io_bitmap_a);
fdef3ad1 4536 return r;
6aa8b732
AK		 4537}
4538
4539static void __exit vmx_exit(void)
4540{
5897297b
AK		 4541 free_page((unsigned long)vmx_msr_bitmap_legacy);
4542 free_page((unsigned long)vmx_msr_bitmap_longmode);
3e7c73e9
AK		 4543 free_page((unsigned long)vmx_io_bitmap_b);
4544 free_page((unsigned long)vmx_io_bitmap_a);
fdef3ad1 4545
cb498ea2 4546 kvm_exit();
6aa8b732
AK		 4547}
4548
4549module_init(vmx_init)
4550module_exit(vmx_exit)
ubuntu-bionic-kernel mirror