projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: Drop kvm_{load,put}_guest_fpu() exports
[GitHub/mt8127/android_kernel_alcatel_ttab.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.
8 *
9 * Authors:
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
15 *
16 */
17
85f455f7 18#include "irq.h"
1d737c8a 19#include "mmu.h"
e495606d 20
edf88417 21#include <linux/kvm_host.h>
6aa8b732 22#include <linux/module.h>
9d8f549d 23#include <linux/kernel.h>
6aa8b732
AK		 24#include <linux/mm.h>
25#include <linux/highmem.h>
e8edc6e0 26#include <linux/sched.h>
c7addb90 27#include <linux/moduleparam.h>
229456fc 28#include <linux/ftrace_event.h>
5fdbf976 29#include "kvm_cache_regs.h"
35920a35 30#include "x86.h"
e495606d 31
6aa8b732 32#include <asm/io.h>
3b3be0d1 33#include <asm/desc.h>
13673a90 34#include <asm/vmx.h>
6210e37b 35#include <asm/virtext.h>
a0861c02 36#include <asm/mce.h>
6aa8b732 37
229456fc
MT		 38#include "trace.h"
39
4ecac3fd
AK		 40#define __ex(x) __kvm_handle_fault_on_reboot(x)
41
6aa8b732
AK		 42MODULE_AUTHOR("Qumranet");
43MODULE_LICENSE("GPL");
44
4462d21a 45static int __read_mostly bypass_guest_pf = 1;
c1f8bc04 46module_param(bypass_guest_pf, bool, S_IRUGO);
c7addb90 47
4462d21a 48static int __read_mostly enable_vpid = 1;
736caefe 49module_param_named(vpid, enable_vpid, bool, 0444);
2384d2b3 50
4462d21a 51static int __read_mostly flexpriority_enabled = 1;
736caefe 52module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
4c9fc8ef 53
4462d21a 54static int __read_mostly enable_ept = 1;
736caefe 55module_param_named(ept, enable_ept, bool, S_IRUGO);
d56f546d 56
3a624e29
NK		 57static int __read_mostly enable_unrestricted_guest = 1;
58module_param_named(unrestricted_guest,
59 enable_unrestricted_guest, bool, S_IRUGO);
60
4462d21a 61static int __read_mostly emulate_invalid_guest_state = 0;
c1f8bc04 62module_param(emulate_invalid_guest_state, bool, S_IRUGO);
04fa4d32 63
cdc0e244
AK		 64#define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST \
65 (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
66#define KVM_GUEST_CR0_MASK \
67 (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
68#define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST \
02daab21 69 (X86_CR0_WP | X86_CR0_NE | X86_CR0_MP)
cdc0e244
AK		 70#define KVM_VM_CR0_ALWAYS_ON \
71 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
4c38609a
AK		 72#define KVM_CR4_GUEST_OWNED_BITS \
73 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
74 | X86_CR4_OSXMMEXCPT)
75
cdc0e244
AK		 76#define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
77#define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
78
4b8d54f9
ZE		 79/*
80 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
81 * ple_gap: upper bound on the amount of time between two successive
82 * executions of PAUSE in a loop. Also indicate if ple enabled.
83 * According to test, this time is usually small than 41 cycles.
84 * ple_window: upper bound on the amount of time a guest is allowed to execute
85 * in a PAUSE loop. Tests indicate that most spinlocks are held for
86 * less than 2^12 cycles
87 * Time is measured based on a counter that runs at the same rate as the TSC,
88 * refer SDM volume 3b section 21.6.13 & 22.1.3.
89 */
90#define KVM_VMX_DEFAULT_PLE_GAP 41
91#define KVM_VMX_DEFAULT_PLE_WINDOW 4096
92static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
93module_param(ple_gap, int, S_IRUGO);
94
95static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
96module_param(ple_window, int, S_IRUGO);
97
a2fa3e9f
GH		 98struct vmcs {
99 u32 revision_id;
100 u32 abort;
101 char data[0];
102};
103
26bb0981
AK		 104struct shared_msr_entry {
105 unsigned index;
106 u64 data;
d5696725 107 u64 mask;
26bb0981
AK		 108};
109
a2fa3e9f 110struct vcpu_vmx {
fb3f0f51 111 struct kvm_vcpu vcpu;
543e4243 112 struct list_head local_vcpus_link;
313dbd49 113 unsigned long host_rsp;
a2fa3e9f 114 int launched;
29bd8a78 115 u8 fail;
1155f76a 116 u32 idt_vectoring_info;
26bb0981 117 struct shared_msr_entry *guest_msrs;
a2fa3e9f
GH		 118 int nmsrs;
119 int save_nmsrs;
a2fa3e9f 120#ifdef CONFIG_X86_64
44ea2b17
AK		 121 u64 msr_host_kernel_gs_base;
122 u64 msr_guest_kernel_gs_base;
a2fa3e9f
GH		 123#endif
124 struct vmcs *vmcs;
125 struct {
126 int loaded;
127 u16 fs_sel, gs_sel, ldt_sel;
152d3f2f
LV		 128 int gs_ldt_reload_needed;
129 int fs_reload_needed;
d77c26fc 130 } host_state;
9c8cba37 131 struct {
7ffd92c5
AK		 132 int vm86_active;
133 u8 save_iopl;
134 struct kvm_save_segment {
135 u16 selector;
136 unsigned long base;
137 u32 limit;
138 u32 ar;
139 } tr, es, ds, fs, gs;
9c8cba37
AK		 140 struct {
141 bool pending;
142 u8 vector;
143 unsigned rip;
144 } irq;
145 } rmode;
2384d2b3 146 int vpid;
04fa4d32 147 bool emulation_required;
3b86cd99
JK		 148
149 /* Support for vnmi-less CPUs */
150 int soft_vnmi_blocked;
151 ktime_t entry_time;
152 s64 vnmi_blocked_time;
a0861c02 153 u32 exit_reason;
4e47c7a6
SY		 154
155 bool rdtscp_enabled;
a2fa3e9f
GH		 156};
157
158static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
159{
fb3f0f51 160 return container_of(vcpu, struct vcpu_vmx, vcpu);
a2fa3e9f
GH		 161}
162
b7ebfb05 163static int init_rmode(struct kvm *kvm);
4e1096d2 164static u64 construct_eptp(unsigned long root_hpa);
75880a01 165
6aa8b732
AK		 166static DEFINE_PER_CPU(struct vmcs *, vmxarea);
167static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
543e4243 168static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
6aa8b732 169
3e7c73e9
AK		 170static unsigned long *vmx_io_bitmap_a;
171static unsigned long *vmx_io_bitmap_b;
5897297b
AK		 172static unsigned long *vmx_msr_bitmap_legacy;
173static unsigned long *vmx_msr_bitmap_longmode;
fdef3ad1 174
2384d2b3
SY		 175static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
176static DEFINE_SPINLOCK(vmx_vpid_lock);
177
1c3d14fe 178static struct vmcs_config {
6aa8b732
AK		 179 int size;
180 int order;
181 u32 revision_id;
1c3d14fe
YS		 182 u32 pin_based_exec_ctrl;
183 u32 cpu_based_exec_ctrl;
f78e0e2e 184 u32 cpu_based_2nd_exec_ctrl;
1c3d14fe
YS		 185 u32 vmexit_ctrl;
186 u32 vmentry_ctrl;
187} vmcs_config;
6aa8b732 188
efff9e53 189static struct vmx_capability {
d56f546d
SY		 190 u32 ept;
191 u32 vpid;
192} vmx_capability;
193
6aa8b732
AK		 194#define VMX_SEGMENT_FIELD(seg) \
195 [VCPU_SREG_##seg] = { \
196 .selector = GUEST_##seg##_SELECTOR, \
197 .base = GUEST_##seg##_BASE, \
198 .limit = GUEST_##seg##_LIMIT, \
199 .ar_bytes = GUEST_##seg##_AR_BYTES, \
200 }
201
202static struct kvm_vmx_segment_field {
203 unsigned selector;
204 unsigned base;
205 unsigned limit;
206 unsigned ar_bytes;
207} kvm_vmx_segment_fields[] = {
208 VMX_SEGMENT_FIELD(CS),
209 VMX_SEGMENT_FIELD(DS),
210 VMX_SEGMENT_FIELD(ES),
211 VMX_SEGMENT_FIELD(FS),
212 VMX_SEGMENT_FIELD(GS),
213 VMX_SEGMENT_FIELD(SS),
214 VMX_SEGMENT_FIELD(TR),
215 VMX_SEGMENT_FIELD(LDTR),
216};
217
26bb0981
AK		 218static u64 host_efer;
219
6de4f3ad
AK		 220static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
221
4d56c8a7
AK		 222/*
223 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
224 * away by decrementing the array size.
225 */
6aa8b732 226static const u32 vmx_msr_index[] = {
05b3e0c2 227#ifdef CONFIG_X86_64
44ea2b17 228 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
6aa8b732 229#endif
4e47c7a6 230 MSR_EFER, MSR_TSC_AUX, MSR_K6_STAR,
6aa8b732 231};
9d8f549d 232#define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
6aa8b732 233
6aa8b732
AK		 234static inline int is_page_fault(u32 intr_info)
235{
236 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
237 INTR_INFO_VALID_MASK)) ==
8ab2d2e2 238 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
6aa8b732
AK		 239}
240
2ab455cc
AL		 241static inline int is_no_device(u32 intr_info)
242{
243 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
244 INTR_INFO_VALID_MASK)) ==
8ab2d2e2 245 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
2ab455cc
AL		 246}
247
7aa81cc0
AL		 248static inline int is_invalid_opcode(u32 intr_info)
249{
250 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
251 INTR_INFO_VALID_MASK)) ==
8ab2d2e2 252 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
7aa81cc0
AL		 253}
254
6aa8b732
AK		 255static inline int is_external_interrupt(u32 intr_info)
256{
257 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
258 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
259}
260
a0861c02
AK		 261static inline int is_machine_check(u32 intr_info)
262{
263 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
264 INTR_INFO_VALID_MASK)) ==
265 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
266}
267
25c5f225
SY		 268static inline int cpu_has_vmx_msr_bitmap(void)
269{
04547156 270 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
25c5f225
SY		 271}
272
6e5d865c
YS		 273static inline int cpu_has_vmx_tpr_shadow(void)
274{
04547156 275 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
6e5d865c
YS		 276}
277
278static inline int vm_need_tpr_shadow(struct kvm *kvm)
279{
04547156 280 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
6e5d865c
YS		 281}
282
f78e0e2e
SY		 283static inline int cpu_has_secondary_exec_ctrls(void)
284{
04547156
SY		 285 return vmcs_config.cpu_based_exec_ctrl &
286 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
f78e0e2e
SY		 287}
288
774ead3a 289static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
f78e0e2e 290{
04547156
SY		 291 return vmcs_config.cpu_based_2nd_exec_ctrl &
292 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
293}
294
295static inline bool cpu_has_vmx_flexpriority(void)
296{
297 return cpu_has_vmx_tpr_shadow() &&
298 cpu_has_vmx_virtualize_apic_accesses();
f78e0e2e
SY		 299}
300
e799794e
MT		 301static inline bool cpu_has_vmx_ept_execute_only(void)
302{
303 return !!(vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT);
304}
305
306static inline bool cpu_has_vmx_eptp_uncacheable(void)
307{
308 return !!(vmx_capability.ept & VMX_EPTP_UC_BIT);
309}
310
311static inline bool cpu_has_vmx_eptp_writeback(void)
312{
313 return !!(vmx_capability.ept & VMX_EPTP_WB_BIT);
314}
315
316static inline bool cpu_has_vmx_ept_2m_page(void)
317{
318 return !!(vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT);
319}
320
878403b7
SY		 321static inline bool cpu_has_vmx_ept_1g_page(void)
322{
323 return !!(vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT);
324}
325
d56f546d
SY		 326static inline int cpu_has_vmx_invept_individual_addr(void)
327{
04547156 328 return !!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT);
d56f546d
SY		 329}
330
331static inline int cpu_has_vmx_invept_context(void)
332{
04547156 333 return !!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT);
d56f546d
SY		 334}
335
336static inline int cpu_has_vmx_invept_global(void)
337{
04547156 338 return !!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT);
d56f546d
SY		 339}
340
341static inline int cpu_has_vmx_ept(void)
342{
04547156
SY		 343 return vmcs_config.cpu_based_2nd_exec_ctrl &
344 SECONDARY_EXEC_ENABLE_EPT;
d56f546d
SY		 345}
346
3a624e29
NK		 347static inline int cpu_has_vmx_unrestricted_guest(void)
348{
349 return vmcs_config.cpu_based_2nd_exec_ctrl &
350 SECONDARY_EXEC_UNRESTRICTED_GUEST;
351}
352
4b8d54f9
ZE		 353static inline int cpu_has_vmx_ple(void)
354{
355 return vmcs_config.cpu_based_2nd_exec_ctrl &
356 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
357}
358
f78e0e2e
SY		 359static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
360{
04547156
SY		 361 return flexpriority_enabled &&
362 (cpu_has_vmx_virtualize_apic_accesses()) &&
363 (irqchip_in_kernel(kvm));
f78e0e2e
SY		 364}
365
2384d2b3
SY		 366static inline int cpu_has_vmx_vpid(void)
367{
04547156
SY		 368 return vmcs_config.cpu_based_2nd_exec_ctrl &
369 SECONDARY_EXEC_ENABLE_VPID;
2384d2b3
SY		 370}
371
4e47c7a6
SY		 372static inline int cpu_has_vmx_rdtscp(void)
373{
374 return vmcs_config.cpu_based_2nd_exec_ctrl &
375 SECONDARY_EXEC_RDTSCP;
376}
377
f08864b4
SY		 378static inline int cpu_has_virtual_nmis(void)
379{
380 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
381}
382
04547156
SY		 383static inline bool report_flexpriority(void)
384{
385 return flexpriority_enabled;
386}
387
8b9cf98c 388static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
7725f0ba
AK		 389{
390 int i;
391
a2fa3e9f 392 for (i = 0; i < vmx->nmsrs; ++i)
26bb0981 393 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
a75beee6
ED		 394 return i;
395 return -1;
396}
397
2384d2b3
SY		 398static inline void __invvpid(int ext, u16 vpid, gva_t gva)
399{
400 struct {
401 u64 vpid : 16;
402 u64 rsvd : 48;
403 u64 gva;
404 } operand = { vpid, 0, gva };
405
4ecac3fd 406 asm volatile (__ex(ASM_VMX_INVVPID)
2384d2b3
SY		 407 /* CF==1 or ZF==1 --> rc = -1 */
408 "; ja 1f ; ud2 ; 1:"
409 : : "a"(&operand), "c"(ext) : "cc", "memory");
410}
411
1439442c
SY		 412static inline void __invept(int ext, u64 eptp, gpa_t gpa)
413{
414 struct {
415 u64 eptp, gpa;
416 } operand = {eptp, gpa};
417
4ecac3fd 418 asm volatile (__ex(ASM_VMX_INVEPT)
1439442c
SY		 419 /* CF==1 or ZF==1 --> rc = -1 */
420 "; ja 1f ; ud2 ; 1:\n"
421 : : "a" (&operand), "c" (ext) : "cc", "memory");
422}
423
26bb0981 424static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
a75beee6
ED		 425{
426 int i;
427
8b9cf98c 428 i = __find_msr_index(vmx, msr);
a75beee6 429 if (i >= 0)
a2fa3e9f 430 return &vmx->guest_msrs[i];
8b6d44c7 431 return NULL;
7725f0ba
AK		 432}
433
6aa8b732
AK		 434static void vmcs_clear(struct vmcs *vmcs)
435{
436 u64 phys_addr = __pa(vmcs);
437 u8 error;
438
4ecac3fd 439 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
6aa8b732
AK		 440 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
441 : "cc", "memory");
442 if (error)
443 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
444 vmcs, phys_addr);
445}
446
447static void __vcpu_clear(void *arg)
448{
8b9cf98c 449 struct vcpu_vmx *vmx = arg;
d3b2c338 450 int cpu = raw_smp_processor_id();
6aa8b732 451
8b9cf98c 452 if (vmx->vcpu.cpu == cpu)
a2fa3e9f
GH		 453 vmcs_clear(vmx->vmcs);
454 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
6aa8b732 455 per_cpu(current_vmcs, cpu) = NULL;
ad312c7c 456 rdtscll(vmx->vcpu.arch.host_tsc);
543e4243
AK		 457 list_del(&vmx->local_vcpus_link);
458 vmx->vcpu.cpu = -1;
459 vmx->launched = 0;
6aa8b732
AK		 460}
461
8b9cf98c 462static void vcpu_clear(struct vcpu_vmx *vmx)
8d0be2b3 463{
eae5ecb5
AK		 464 if (vmx->vcpu.cpu == -1)
465 return;
8691e5a8 466 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
8d0be2b3
AK		 467}
468
2384d2b3
SY		 469static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
470{
471 if (vmx->vpid == 0)
472 return;
473
474 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
475}
476
1439442c
SY		 477static inline void ept_sync_global(void)
478{
479 if (cpu_has_vmx_invept_global())
480 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
481}
482
483static inline void ept_sync_context(u64 eptp)
484{
089d034e 485 if (enable_ept) {
1439442c
SY		 486 if (cpu_has_vmx_invept_context())
487 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
488 else
489 ept_sync_global();
490 }
491}
492
493static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
494{
089d034e 495 if (enable_ept) {
1439442c
SY		 496 if (cpu_has_vmx_invept_individual_addr())
497 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
498 eptp, gpa);
499 else
500 ept_sync_context(eptp);
501 }
502}
503
6aa8b732
AK		 504static unsigned long vmcs_readl(unsigned long field)
505{
506 unsigned long value;
507
4ecac3fd 508 asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
6aa8b732
AK		 509 : "=a"(value) : "d"(field) : "cc");
510 return value;
511}
512
513static u16 vmcs_read16(unsigned long field)
514{
515 return vmcs_readl(field);
516}
517
518static u32 vmcs_read32(unsigned long field)
519{
520 return vmcs_readl(field);
521}
522
523static u64 vmcs_read64(unsigned long field)
524{
05b3e0c2 525#ifdef CONFIG_X86_64
6aa8b732
AK		 526 return vmcs_readl(field);
527#else
528 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
529#endif
530}
531
e52de1b8
AK		 532static noinline void vmwrite_error(unsigned long field, unsigned long value)
533{
534 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
535 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
536 dump_stack();
537}
538
6aa8b732
AK		 539static void vmcs_writel(unsigned long field, unsigned long value)
540{
541 u8 error;
542
4ecac3fd 543 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
d77c26fc 544 : "=q"(error) : "a"(value), "d"(field) : "cc");
e52de1b8
AK		 545 if (unlikely(error))
546 vmwrite_error(field, value);
6aa8b732
AK		 547}
548
549static void vmcs_write16(unsigned long field, u16 value)
550{
551 vmcs_writel(field, value);
552}
553
554static void vmcs_write32(unsigned long field, u32 value)
555{
556 vmcs_writel(field, value);
557}
558
559static void vmcs_write64(unsigned long field, u64 value)
560{
6aa8b732 561 vmcs_writel(field, value);
7682f2d0 562#ifndef CONFIG_X86_64
6aa8b732
AK		 563 asm volatile ("");
564 vmcs_writel(field+1, value >> 32);
565#endif
566}
567
2ab455cc
AL		 568static void vmcs_clear_bits(unsigned long field, u32 mask)
569{
570 vmcs_writel(field, vmcs_readl(field) & ~mask);
571}
572
573static void vmcs_set_bits(unsigned long field, u32 mask)
574{
575 vmcs_writel(field, vmcs_readl(field) | mask);
576}
577
abd3f2d6
AK		 578static void update_exception_bitmap(struct kvm_vcpu *vcpu)
579{
580 u32 eb;
581
fd7373cc
JK		 582 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
583 (1u << NM_VECTOR) | (1u << DB_VECTOR);
584 if ((vcpu->guest_debug &
585 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
586 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
587 eb |= 1u << BP_VECTOR;
7ffd92c5 588 if (to_vmx(vcpu)->rmode.vm86_active)
abd3f2d6 589 eb = ~0;
089d034e 590 if (enable_ept)
1439442c 591 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
02daab21
AK		 592 if (vcpu->fpu_active)
593 eb &= ~(1u << NM_VECTOR);
abd3f2d6
AK		 594 vmcs_write32(EXCEPTION_BITMAP, eb);
595}
596
33ed6329
AK		 597static void reload_tss(void)
598{
33ed6329
AK		 599 /*
600 * VT restores TR but not its size. Useless.
601 */
602 struct descriptor_table gdt;
a5f61300 603 struct desc_struct *descs;
33ed6329 604
d6e88aec 605 kvm_get_gdt(&gdt);
33ed6329
AK		 606 descs = (void *)gdt.base;
607 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
608 load_TR_desc();
33ed6329
AK		 609}
610
92c0d900 611static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
2cc51560 612{
3a34a881 613 u64 guest_efer;
51c6cf66
AK		 614 u64 ignore_bits;
615
26bb0981 616 guest_efer = vmx->vcpu.arch.shadow_efer;
3a34a881 617
51c6cf66
AK		 618 /*
619 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
620 * outside long mode
621 */
622 ignore_bits = EFER_NX | EFER_SCE;
623#ifdef CONFIG_X86_64
624 ignore_bits |= EFER_LMA | EFER_LME;
625 /* SCE is meaningful only in long mode on Intel */
626 if (guest_efer & EFER_LMA)
627 ignore_bits &= ~(u64)EFER_SCE;
628#endif
51c6cf66
AK		 629 guest_efer &= ~ignore_bits;
630 guest_efer |= host_efer & ignore_bits;
26bb0981 631 vmx->guest_msrs[efer_offset].data = guest_efer;
d5696725 632 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
26bb0981 633 return true;
51c6cf66
AK		 634}
635
04d2cc77 636static void vmx_save_host_state(struct kvm_vcpu *vcpu)
33ed6329 637{
04d2cc77 638 struct vcpu_vmx *vmx = to_vmx(vcpu);
26bb0981 639 int i;
04d2cc77 640
a2fa3e9f 641 if (vmx->host_state.loaded)
33ed6329
AK		 642 return;
643
a2fa3e9f 644 vmx->host_state.loaded = 1;
33ed6329
AK		 645 /*
646 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
647 * allow segment selectors with cpl > 0 or ti == 1.
648 */
d6e88aec 649 vmx->host_state.ldt_sel = kvm_read_ldt();
152d3f2f 650 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
d6e88aec 651 vmx->host_state.fs_sel = kvm_read_fs();
152d3f2f 652 if (!(vmx->host_state.fs_sel & 7)) {
a2fa3e9f 653 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
152d3f2f
LV		 654 vmx->host_state.fs_reload_needed = 0;
655 } else {
33ed6329 656 vmcs_write16(HOST_FS_SELECTOR, 0);
152d3f2f 657 vmx->host_state.fs_reload_needed = 1;
33ed6329 658 }
d6e88aec 659 vmx->host_state.gs_sel = kvm_read_gs();
a2fa3e9f
GH		 660 if (!(vmx->host_state.gs_sel & 7))
661 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
33ed6329
AK		 662 else {
663 vmcs_write16(HOST_GS_SELECTOR, 0);
152d3f2f 664 vmx->host_state.gs_ldt_reload_needed = 1;
33ed6329
AK		 665 }
666
667#ifdef CONFIG_X86_64
668 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
669 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
670#else
a2fa3e9f
GH		 671 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
672 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
33ed6329 673#endif
707c0874
AK		 674
675#ifdef CONFIG_X86_64
44ea2b17
AK		 676 if (is_long_mode(&vmx->vcpu)) {
677 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
678 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
679 }
707c0874 680#endif
26bb0981
AK		 681 for (i = 0; i < vmx->save_nmsrs; ++i)
682 kvm_set_shared_msr(vmx->guest_msrs[i].index,
d5696725
AK		 683 vmx->guest_msrs[i].data,
684 vmx->guest_msrs[i].mask);
33ed6329
AK		 685}
686
a9b21b62 687static void __vmx_load_host_state(struct vcpu_vmx *vmx)
33ed6329 688{
15ad7146 689 unsigned long flags;
33ed6329 690
a2fa3e9f 691 if (!vmx->host_state.loaded)
33ed6329
AK		 692 return;
693
e1beb1d3 694 ++vmx->vcpu.stat.host_state_reload;
a2fa3e9f 695 vmx->host_state.loaded = 0;
152d3f2f 696 if (vmx->host_state.fs_reload_needed)
d6e88aec 697 kvm_load_fs(vmx->host_state.fs_sel);
152d3f2f 698 if (vmx->host_state.gs_ldt_reload_needed) {
d6e88aec 699 kvm_load_ldt(vmx->host_state.ldt_sel);
33ed6329
AK		 700 /*
701 * If we have to reload gs, we must take care to
702 * preserve our gs base.
703 */
15ad7146 704 local_irq_save(flags);
d6e88aec 705 kvm_load_gs(vmx->host_state.gs_sel);
33ed6329
AK		 706#ifdef CONFIG_X86_64
707 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
708#endif
15ad7146 709 local_irq_restore(flags);
33ed6329 710 }
152d3f2f 711 reload_tss();
44ea2b17
AK		 712#ifdef CONFIG_X86_64
713 if (is_long_mode(&vmx->vcpu)) {
714 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
715 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
716 }
717#endif
33ed6329
AK		 718}
719
a9b21b62
AK		 720static void vmx_load_host_state(struct vcpu_vmx *vmx)
721{
722 preempt_disable();
723 __vmx_load_host_state(vmx);
724 preempt_enable();
725}
726
6aa8b732
AK		 727/*
728 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
729 * vcpu mutex is already taken.
730 */
15ad7146 731static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
6aa8b732 732{
a2fa3e9f
GH		 733 struct vcpu_vmx *vmx = to_vmx(vcpu);
734 u64 phys_addr = __pa(vmx->vmcs);
019960ae 735 u64 tsc_this, delta, new_offset;
6aa8b732 736
a3d7f85f 737 if (vcpu->cpu != cpu) {
8b9cf98c 738 vcpu_clear(vmx);
2f599714 739 kvm_migrate_timers(vcpu);
eb5109e3 740 set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests);
543e4243
AK		 741 local_irq_disable();
742 list_add(&vmx->local_vcpus_link,
743 &per_cpu(vcpus_on_cpu, cpu));
744 local_irq_enable();
a3d7f85f 745 }
6aa8b732 746
a2fa3e9f 747 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
6aa8b732
AK		 748 u8 error;
749
a2fa3e9f 750 per_cpu(current_vmcs, cpu) = vmx->vmcs;
4ecac3fd 751 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
6aa8b732
AK		 752 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
753 : "cc");
754 if (error)
755 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
a2fa3e9f 756 vmx->vmcs, phys_addr);
6aa8b732
AK		 757 }
758
759 if (vcpu->cpu != cpu) {
760 struct descriptor_table dt;
761 unsigned long sysenter_esp;
762
763 vcpu->cpu = cpu;
764 /*
765 * Linux uses per-cpu TSS and GDT, so set these when switching
766 * processors.
767 */
d6e88aec
AK		 768 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
769 kvm_get_gdt(&dt);
6aa8b732
AK		 770 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
771
772 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
773 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
7700270e
AK		 774
775 /*
776 * Make sure the time stamp counter is monotonous.
777 */
778 rdtscll(tsc_this);
019960ae
AK		 779 if (tsc_this < vcpu->arch.host_tsc) {
780 delta = vcpu->arch.host_tsc - tsc_this;
781 new_offset = vmcs_read64(TSC_OFFSET) + delta;
782 vmcs_write64(TSC_OFFSET, new_offset);
783 }
6aa8b732 784 }
6aa8b732
AK		 785}
786
787static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
788{
a9b21b62 789 __vmx_load_host_state(to_vmx(vcpu));
6aa8b732
AK		 790}
791
5fd86fcf
AK		 792static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
793{
794 if (vcpu->fpu_active)
795 return;
796 vcpu->fpu_active = 1;
707d92fa 797 vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
4d4ec087 798 if (kvm_read_cr0_bits(vcpu, X86_CR0_TS))
707d92fa 799 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
5fd86fcf 800 update_exception_bitmap(vcpu);
edcafe3c
AK		 801 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
802 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
5fd86fcf
AK		 803}
804
edcafe3c
AK		 805static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
806
5fd86fcf
AK		 807static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
808{
edcafe3c 809 vmx_decache_cr0_guest_bits(vcpu);
707d92fa 810 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
5fd86fcf 811 update_exception_bitmap(vcpu);
edcafe3c
AK		 812 vcpu->arch.cr0_guest_owned_bits = 0;
813 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
814 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
5fd86fcf
AK		 815}
816
6aa8b732
AK		 817static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
818{
345dcaa8
AK		 819 unsigned long rflags;
820
821 rflags = vmcs_readl(GUEST_RFLAGS);
822 if (to_vmx(vcpu)->rmode.vm86_active)
823 rflags &= ~(unsigned long)(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
824 return rflags;
6aa8b732
AK		 825}
826
827static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
828{
7ffd92c5 829 if (to_vmx(vcpu)->rmode.vm86_active)
053de044 830 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
6aa8b732
AK		 831 vmcs_writel(GUEST_RFLAGS, rflags);
832}
833
2809f5d2
GC		 834static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
835{
836 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
837 int ret = 0;
838
839 if (interruptibility & GUEST_INTR_STATE_STI)
840 ret |= X86_SHADOW_INT_STI;
841 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
842 ret |= X86_SHADOW_INT_MOV_SS;
843
844 return ret & mask;
845}
846
847static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
848{
849 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
850 u32 interruptibility = interruptibility_old;
851
852 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
853
854 if (mask & X86_SHADOW_INT_MOV_SS)
855 interruptibility |= GUEST_INTR_STATE_MOV_SS;
856 if (mask & X86_SHADOW_INT_STI)
857 interruptibility |= GUEST_INTR_STATE_STI;
858
859 if ((interruptibility != interruptibility_old))
860 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
861}
862
6aa8b732
AK		 863static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
864{
865 unsigned long rip;
6aa8b732 866
5fdbf976 867 rip = kvm_rip_read(vcpu);
6aa8b732 868 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5fdbf976 869 kvm_rip_write(vcpu, rip);
6aa8b732 870
2809f5d2
GC		 871 /* skipping an emulated instruction also counts */
872 vmx_set_interrupt_shadow(vcpu, 0);
6aa8b732
AK		 873}
874
298101da
AK		 875static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
876 bool has_error_code, u32 error_code)
877{
77ab6db0 878 struct vcpu_vmx *vmx = to_vmx(vcpu);
8ab2d2e2 879 u32 intr_info = nr | INTR_INFO_VALID_MASK;
77ab6db0 880
8ab2d2e2 881 if (has_error_code) {
77ab6db0 882 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
8ab2d2e2
JK		 883 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
884 }
77ab6db0 885
7ffd92c5 886 if (vmx->rmode.vm86_active) {
77ab6db0
JK		 887 vmx->rmode.irq.pending = true;
888 vmx->rmode.irq.vector = nr;
889 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
ae0bb3e0
GN		 890 if (kvm_exception_is_soft(nr))
891 vmx->rmode.irq.rip +=
892 vmx->vcpu.arch.event_exit_inst_len;
8ab2d2e2
JK		 893 intr_info |= INTR_TYPE_SOFT_INTR;
894 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
77ab6db0
JK		 895 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
896 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
897 return;
898 }
899
66fd3f7f
GN		 900 if (kvm_exception_is_soft(nr)) {
901 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
902 vmx->vcpu.arch.event_exit_inst_len);
8ab2d2e2
JK		 903 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
904 } else
905 intr_info |= INTR_TYPE_HARD_EXCEPTION;
906
907 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
298101da
AK		 908}
909
4e47c7a6
SY		 910static bool vmx_rdtscp_supported(void)
911{
912 return cpu_has_vmx_rdtscp();
913}
914
a75beee6
ED		 915/*
916 * Swap MSR entry in host/guest MSR entry array.
917 */
8b9cf98c 918static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
a75beee6 919{
26bb0981 920 struct shared_msr_entry tmp;
a2fa3e9f
GH		 921
922 tmp = vmx->guest_msrs[to];
923 vmx->guest_msrs[to] = vmx->guest_msrs[from];
924 vmx->guest_msrs[from] = tmp;
a75beee6
ED		 925}
926
e38aea3e
AK		 927/*
928 * Set up the vmcs to automatically save and restore system
929 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
930 * mode, as fiddling with msrs is very expensive.
931 */
8b9cf98c 932static void setup_msrs(struct vcpu_vmx *vmx)
e38aea3e 933{
26bb0981 934 int save_nmsrs, index;
5897297b 935 unsigned long *msr_bitmap;
e38aea3e 936
33f9c505 937 vmx_load_host_state(vmx);
a75beee6
ED		 938 save_nmsrs = 0;
939#ifdef CONFIG_X86_64
8b9cf98c 940 if (is_long_mode(&vmx->vcpu)) {
8b9cf98c 941 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
a75beee6 942 if (index >= 0)
8b9cf98c
RR		 943 move_msr_up(vmx, index, save_nmsrs++);
944 index = __find_msr_index(vmx, MSR_LSTAR);
a75beee6 945 if (index >= 0)
8b9cf98c
RR		 946 move_msr_up(vmx, index, save_nmsrs++);
947 index = __find_msr_index(vmx, MSR_CSTAR);
a75beee6 948 if (index >= 0)
8b9cf98c 949 move_msr_up(vmx, index, save_nmsrs++);
4e47c7a6
SY		 950 index = __find_msr_index(vmx, MSR_TSC_AUX);
951 if (index >= 0 && vmx->rdtscp_enabled)
952 move_msr_up(vmx, index, save_nmsrs++);
a75beee6
ED		 953 /*
954 * MSR_K6_STAR is only needed on long mode guests, and only
955 * if efer.sce is enabled.
956 */
8b9cf98c 957 index = __find_msr_index(vmx, MSR_K6_STAR);
ad312c7c 958 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
8b9cf98c 959 move_msr_up(vmx, index, save_nmsrs++);
a75beee6
ED		 960 }
961#endif
92c0d900
AK		 962 index = __find_msr_index(vmx, MSR_EFER);
963 if (index >= 0 && update_transition_efer(vmx, index))
26bb0981 964 move_msr_up(vmx, index, save_nmsrs++);
e38aea3e 965
26bb0981 966 vmx->save_nmsrs = save_nmsrs;
5897297b
AK		 967
968 if (cpu_has_vmx_msr_bitmap()) {
969 if (is_long_mode(&vmx->vcpu))
970 msr_bitmap = vmx_msr_bitmap_longmode;
971 else
972 msr_bitmap = vmx_msr_bitmap_legacy;
973
974 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
975 }
e38aea3e
AK		 976}
977
6aa8b732
AK		 978/*
979 * reads and returns guest's timestamp counter "register"
980 * guest_tsc = host_tsc + tsc_offset -- 21.3
981 */
982static u64 guest_read_tsc(void)
983{
984 u64 host_tsc, tsc_offset;
985
986 rdtscll(host_tsc);
987 tsc_offset = vmcs_read64(TSC_OFFSET);
988 return host_tsc + tsc_offset;
989}
990
991/*
992 * writes 'guest_tsc' into guest's timestamp counter "register"
993 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
994 */
53f658b3 995static void guest_write_tsc(u64 guest_tsc, u64 host_tsc)
6aa8b732 996{
6aa8b732
AK		 997 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
998}
999
6aa8b732
AK		 1000/*
1001 * Reads an msr value (of 'msr_index') into 'pdata'.
1002 * Returns 0 on success, non-0 otherwise.
1003 * Assumes vcpu_load() was already called.
1004 */
1005static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1006{
1007 u64 data;
26bb0981 1008 struct shared_msr_entry *msr;
6aa8b732
AK		 1009
1010 if (!pdata) {
1011 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
1012 return -EINVAL;
1013 }
1014
1015 switch (msr_index) {
05b3e0c2 1016#ifdef CONFIG_X86_64
6aa8b732
AK		 1017 case MSR_FS_BASE:
1018 data = vmcs_readl(GUEST_FS_BASE);
1019 break;
1020 case MSR_GS_BASE:
1021 data = vmcs_readl(GUEST_GS_BASE);
1022 break;
44ea2b17
AK		 1023 case MSR_KERNEL_GS_BASE:
1024 vmx_load_host_state(to_vmx(vcpu));
1025 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
1026 break;
26bb0981 1027#endif
6aa8b732 1028 case MSR_EFER:
3bab1f5d 1029 return kvm_get_msr_common(vcpu, msr_index, pdata);
af24a4e4 1030 case MSR_IA32_TSC:
6aa8b732
AK		 1031 data = guest_read_tsc();
1032 break;
1033 case MSR_IA32_SYSENTER_CS:
1034 data = vmcs_read32(GUEST_SYSENTER_CS);
1035 break;
1036 case MSR_IA32_SYSENTER_EIP:
f5b42c33 1037 data = vmcs_readl(GUEST_SYSENTER_EIP);
6aa8b732
AK		 1038 break;
1039 case MSR_IA32_SYSENTER_ESP:
f5b42c33 1040 data = vmcs_readl(GUEST_SYSENTER_ESP);
6aa8b732 1041 break;
4e47c7a6
SY		 1042 case MSR_TSC_AUX:
1043 if (!to_vmx(vcpu)->rdtscp_enabled)
1044 return 1;
1045 /* Otherwise falls through */
6aa8b732 1046 default:
26bb0981 1047 vmx_load_host_state(to_vmx(vcpu));
8b9cf98c 1048 msr = find_msr_entry(to_vmx(vcpu), msr_index);
3bab1f5d 1049 if (msr) {
542423b0 1050 vmx_load_host_state(to_vmx(vcpu));
3bab1f5d
AK		 1051 data = msr->data;
1052 break;
6aa8b732 1053 }
3bab1f5d 1054 return kvm_get_msr_common(vcpu, msr_index, pdata);
6aa8b732
AK		 1055 }
1056
1057 *pdata = data;
1058 return 0;
1059}
1060
1061/*
1062 * Writes msr value into into the appropriate "register".
1063 * Returns 0 on success, non-0 otherwise.
1064 * Assumes vcpu_load() was already called.
1065 */
1066static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1067{
a2fa3e9f 1068 struct vcpu_vmx *vmx = to_vmx(vcpu);
26bb0981 1069 struct shared_msr_entry *msr;
53f658b3 1070 u64 host_tsc;
2cc51560
ED		 1071 int ret = 0;
1072
6aa8b732 1073 switch (msr_index) {
3bab1f5d 1074 case MSR_EFER:
a9b21b62 1075 vmx_load_host_state(vmx);
2cc51560 1076 ret = kvm_set_msr_common(vcpu, msr_index, data);
2cc51560 1077 break;
16175a79 1078#ifdef CONFIG_X86_64
6aa8b732
AK		 1079 case MSR_FS_BASE:
1080 vmcs_writel(GUEST_FS_BASE, data);
1081 break;
1082 case MSR_GS_BASE:
1083 vmcs_writel(GUEST_GS_BASE, data);
1084 break;
44ea2b17
AK		 1085 case MSR_KERNEL_GS_BASE:
1086 vmx_load_host_state(vmx);
1087 vmx->msr_guest_kernel_gs_base = data;
1088 break;
6aa8b732
AK		 1089#endif
1090 case MSR_IA32_SYSENTER_CS:
1091 vmcs_write32(GUEST_SYSENTER_CS, data);
1092 break;
1093 case MSR_IA32_SYSENTER_EIP:
f5b42c33 1094 vmcs_writel(GUEST_SYSENTER_EIP, data);
6aa8b732
AK		 1095 break;
1096 case MSR_IA32_SYSENTER_ESP:
f5b42c33 1097 vmcs_writel(GUEST_SYSENTER_ESP, data);
6aa8b732 1098 break;
af24a4e4 1099 case MSR_IA32_TSC:
53f658b3
MT		 1100 rdtscll(host_tsc);
1101 guest_write_tsc(data, host_tsc);
6aa8b732 1102 break;
468d472f
SY		 1103 case MSR_IA32_CR_PAT:
1104 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1105 vmcs_write64(GUEST_IA32_PAT, data);
1106 vcpu->arch.pat = data;
1107 break;
1108 }
4e47c7a6
SY		 1109 ret = kvm_set_msr_common(vcpu, msr_index, data);
1110 break;
1111 case MSR_TSC_AUX:
1112 if (!vmx->rdtscp_enabled)
1113 return 1;
1114 /* Check reserved bit, higher 32 bits should be zero */
1115 if ((data >> 32) != 0)
1116 return 1;
1117 /* Otherwise falls through */
6aa8b732 1118 default:
8b9cf98c 1119 msr = find_msr_entry(vmx, msr_index);
3bab1f5d 1120 if (msr) {
542423b0 1121 vmx_load_host_state(vmx);
3bab1f5d
AK		 1122 msr->data = data;
1123 break;
6aa8b732 1124 }
2cc51560 1125 ret = kvm_set_msr_common(vcpu, msr_index, data);
6aa8b732
AK		 1126 }
1127
2cc51560 1128 return ret;
6aa8b732
AK		 1129}
1130
5fdbf976 1131static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
6aa8b732 1132{
5fdbf976
MT		 1133 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1134 switch (reg) {
1135 case VCPU_REGS_RSP:
1136 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1137 break;
1138 case VCPU_REGS_RIP:
1139 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1140 break;
6de4f3ad
AK		 1141 case VCPU_EXREG_PDPTR:
1142 if (enable_ept)
1143 ept_save_pdptrs(vcpu);
1144 break;
5fdbf976
MT		 1145 default:
1146 break;
1147 }
6aa8b732
AK		 1148}
1149
355be0b9 1150static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
6aa8b732 1151{
ae675ef0
JK		 1152 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1153 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1154 else
1155 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1156
abd3f2d6 1157 update_exception_bitmap(vcpu);
6aa8b732
AK		 1158}
1159
1160static __init int cpu_has_kvm_support(void)
1161{
6210e37b 1162 return cpu_has_vmx();
6aa8b732
AK		 1163}
1164
1165static __init int vmx_disabled_by_bios(void)
1166{
1167 u64 msr;
1168
1169 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
9ea542fa
SY		 1170 return (msr & (FEATURE_CONTROL_LOCKED |
1171 FEATURE_CONTROL_VMXON_ENABLED))
1172 == FEATURE_CONTROL_LOCKED;
62b3ffb8 1173 /* locked but not enabled */
6aa8b732
AK		 1174}
1175
10474ae8 1176static int hardware_enable(void *garbage)
6aa8b732
AK		 1177{
1178 int cpu = raw_smp_processor_id();
1179 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1180 u64 old;
1181
10474ae8
AG		 1182 if (read_cr4() & X86_CR4_VMXE)
1183 return -EBUSY;
1184
543e4243 1185 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
6aa8b732 1186 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
9ea542fa
SY		 1187 if ((old & (FEATURE_CONTROL_LOCKED |
1188 FEATURE_CONTROL_VMXON_ENABLED))
1189 != (FEATURE_CONTROL_LOCKED |
1190 FEATURE_CONTROL_VMXON_ENABLED))
6aa8b732 1191 /* enable and lock */
62b3ffb8 1192 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
9ea542fa
SY		 1193 FEATURE_CONTROL_LOCKED |
1194 FEATURE_CONTROL_VMXON_ENABLED);
66aee91a 1195 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
4ecac3fd
AK		 1196 asm volatile (ASM_VMX_VMXON_RAX
1197 : : "a"(&phys_addr), "m"(phys_addr)
6aa8b732 1198 : "memory", "cc");
10474ae8
AG		 1199
1200 ept_sync_global();
1201
1202 return 0;
6aa8b732
AK		 1203}
1204
543e4243
AK		 1205static void vmclear_local_vcpus(void)
1206{
1207 int cpu = raw_smp_processor_id();
1208 struct vcpu_vmx *vmx, *n;
1209
1210 list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1211 local_vcpus_link)
1212 __vcpu_clear(vmx);
1213}
1214
710ff4a8
EH		 1215
1216/* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1217 * tricks.
1218 */
1219static void kvm_cpu_vmxoff(void)
6aa8b732 1220{
4ecac3fd 1221 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
e693d71b 1222 write_cr4(read_cr4() & ~X86_CR4_VMXE);
6aa8b732
AK		 1223}
1224
710ff4a8
EH		 1225static void hardware_disable(void *garbage)
1226{
1227 vmclear_local_vcpus();
1228 kvm_cpu_vmxoff();
1229}
1230
1c3d14fe 1231static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
d77c26fc 1232 u32 msr, u32 *result)
1c3d14fe
YS		 1233{
1234 u32 vmx_msr_low, vmx_msr_high;
1235 u32 ctl = ctl_min | ctl_opt;
1236
1237 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1238
1239 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1240 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
1241
1242 /* Ensure minimum (required) set of control bits are supported. */
1243 if (ctl_min & ~ctl)
002c7f7c 1244 return -EIO;
1c3d14fe
YS		 1245
1246 *result = ctl;
1247 return 0;
1248}
1249
002c7f7c 1250static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
6aa8b732
AK		 1251{
1252 u32 vmx_msr_low, vmx_msr_high;
d56f546d 1253 u32 min, opt, min2, opt2;
1c3d14fe
YS		 1254 u32 _pin_based_exec_control = 0;
1255 u32 _cpu_based_exec_control = 0;
f78e0e2e 1256 u32 _cpu_based_2nd_exec_control = 0;
1c3d14fe
YS		 1257 u32 _vmexit_control = 0;
1258 u32 _vmentry_control = 0;
1259
1260 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
f08864b4 1261 opt = PIN_BASED_VIRTUAL_NMIS;
1c3d14fe
YS		 1262 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1263 &_pin_based_exec_control) < 0)
002c7f7c 1264 return -EIO;
1c3d14fe
YS		 1265
1266 min = CPU_BASED_HLT_EXITING |
1267#ifdef CONFIG_X86_64
1268 CPU_BASED_CR8_LOAD_EXITING |
1269 CPU_BASED_CR8_STORE_EXITING |
1270#endif
d56f546d
SY		 1271 CPU_BASED_CR3_LOAD_EXITING |
1272 CPU_BASED_CR3_STORE_EXITING |
1c3d14fe
YS		 1273 CPU_BASED_USE_IO_BITMAPS |
1274 CPU_BASED_MOV_DR_EXITING |
a7052897 1275 CPU_BASED_USE_TSC_OFFSETING |
59708670
SY		 1276 CPU_BASED_MWAIT_EXITING |
1277 CPU_BASED_MONITOR_EXITING |
a7052897 1278 CPU_BASED_INVLPG_EXITING;
f78e0e2e 1279 opt = CPU_BASED_TPR_SHADOW |
25c5f225 1280 CPU_BASED_USE_MSR_BITMAPS |
f78e0e2e 1281 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1c3d14fe
YS		 1282 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1283 &_cpu_based_exec_control) < 0)
002c7f7c 1284 return -EIO;
6e5d865c
YS		 1285#ifdef CONFIG_X86_64
1286 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1287 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1288 ~CPU_BASED_CR8_STORE_EXITING;
1289#endif
f78e0e2e 1290 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
d56f546d
SY		 1291 min2 = 0;
1292 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2384d2b3 1293 SECONDARY_EXEC_WBINVD_EXITING |
d56f546d 1294 SECONDARY_EXEC_ENABLE_VPID |
3a624e29 1295 SECONDARY_EXEC_ENABLE_EPT |
4b8d54f9 1296 SECONDARY_EXEC_UNRESTRICTED_GUEST |
4e47c7a6
SY		 1297 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
1298 SECONDARY_EXEC_RDTSCP;
d56f546d
SY		 1299 if (adjust_vmx_controls(min2, opt2,
1300 MSR_IA32_VMX_PROCBASED_CTLS2,
f78e0e2e
SY		 1301 &_cpu_based_2nd_exec_control) < 0)
1302 return -EIO;
1303 }
1304#ifndef CONFIG_X86_64
1305 if (!(_cpu_based_2nd_exec_control &
1306 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1307 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1308#endif
d56f546d 1309 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
a7052897
MT		 1310 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1311 enabled */
5fff7d27
GN		 1312 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
1313 CPU_BASED_CR3_STORE_EXITING |
1314 CPU_BASED_INVLPG_EXITING);
d56f546d
SY		 1315 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1316 vmx_capability.ept, vmx_capability.vpid);
1317 }
1c3d14fe
YS		 1318
1319 min = 0;
1320#ifdef CONFIG_X86_64
1321 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1322#endif
468d472f 1323 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1c3d14fe
YS		 1324 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1325 &_vmexit_control) < 0)
002c7f7c 1326 return -EIO;
1c3d14fe 1327
468d472f
SY		 1328 min = 0;
1329 opt = VM_ENTRY_LOAD_IA32_PAT;
1c3d14fe
YS		 1330 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1331 &_vmentry_control) < 0)
002c7f7c 1332 return -EIO;
6aa8b732 1333
c68876fd 1334 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1c3d14fe
YS		 1335
1336 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1337 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
002c7f7c 1338 return -EIO;
1c3d14fe
YS		 1339
1340#ifdef CONFIG_X86_64
1341 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1342 if (vmx_msr_high & (1u<<16))
002c7f7c 1343 return -EIO;
1c3d14fe
YS		 1344#endif
1345
1346 /* Require Write-Back (WB) memory type for VMCS accesses. */
1347 if (((vmx_msr_high >> 18) & 15) != 6)
002c7f7c 1348 return -EIO;
1c3d14fe 1349
002c7f7c
YS		 1350 vmcs_conf->size = vmx_msr_high & 0x1fff;
1351 vmcs_conf->order = get_order(vmcs_config.size);
1352 vmcs_conf->revision_id = vmx_msr_low;
1c3d14fe 1353
002c7f7c
YS		 1354 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1355 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
f78e0e2e 1356 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
002c7f7c
YS		 1357 vmcs_conf->vmexit_ctrl = _vmexit_control;
1358 vmcs_conf->vmentry_ctrl = _vmentry_control;
1c3d14fe
YS		 1359
1360 return 0;
c68876fd 1361}
6aa8b732
AK		 1362
1363static struct vmcs *alloc_vmcs_cpu(int cpu)
1364{
1365 int node = cpu_to_node(cpu);
1366 struct page *pages;
1367 struct vmcs *vmcs;
1368
6484eb3e 1369 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
6aa8b732
AK		 1370 if (!pages)
1371 return NULL;
1372 vmcs = page_address(pages);
1c3d14fe
YS		 1373 memset(vmcs, 0, vmcs_config.size);
1374 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
6aa8b732
AK		 1375 return vmcs;
1376}
1377
1378static struct vmcs *alloc_vmcs(void)
1379{
d3b2c338 1380 return alloc_vmcs_cpu(raw_smp_processor_id());
6aa8b732
AK		 1381}
1382
1383static void free_vmcs(struct vmcs *vmcs)
1384{
1c3d14fe 1385 free_pages((unsigned long)vmcs, vmcs_config.order);
6aa8b732
AK		 1386}
1387
39959588 1388static void free_kvm_area(void)
6aa8b732
AK		 1389{
1390 int cpu;
1391
3230bb47 1392 for_each_possible_cpu(cpu) {
6aa8b732 1393 free_vmcs(per_cpu(vmxarea, cpu));
3230bb47
ZA		 1394 per_cpu(vmxarea, cpu) = NULL;
1395 }
6aa8b732
AK		 1396}
1397
6aa8b732
AK		 1398static __init int alloc_kvm_area(void)
1399{
1400 int cpu;
1401
3230bb47 1402 for_each_possible_cpu(cpu) {
6aa8b732
AK		 1403 struct vmcs *vmcs;
1404
1405 vmcs = alloc_vmcs_cpu(cpu);
1406 if (!vmcs) {
1407 free_kvm_area();
1408 return -ENOMEM;
1409 }
1410
1411 per_cpu(vmxarea, cpu) = vmcs;
1412 }
1413 return 0;
1414}
1415
1416static __init int hardware_setup(void)
1417{
002c7f7c
YS		 1418 if (setup_vmcs_config(&vmcs_config) < 0)
1419 return -EIO;
50a37eb4
JR		 1420
1421 if (boot_cpu_has(X86_FEATURE_NX))
1422 kvm_enable_efer_bits(EFER_NX);
1423
93ba03c2
SY		 1424 if (!cpu_has_vmx_vpid())
1425 enable_vpid = 0;
1426
3a624e29 1427 if (!cpu_has_vmx_ept()) {
93ba03c2 1428 enable_ept = 0;
3a624e29
NK		 1429 enable_unrestricted_guest = 0;
1430 }
1431
1432 if (!cpu_has_vmx_unrestricted_guest())
1433 enable_unrestricted_guest = 0;
93ba03c2
SY		 1434
1435 if (!cpu_has_vmx_flexpriority())
1436 flexpriority_enabled = 0;
1437
95ba8273
GN		 1438 if (!cpu_has_vmx_tpr_shadow())
1439 kvm_x86_ops->update_cr8_intercept = NULL;
1440
54dee993
MT		 1441 if (enable_ept && !cpu_has_vmx_ept_2m_page())
1442 kvm_disable_largepages();
1443
4b8d54f9
ZE		 1444 if (!cpu_has_vmx_ple())
1445 ple_gap = 0;
1446
6aa8b732
AK		 1447 return alloc_kvm_area();
1448}
1449
1450static __exit void hardware_unsetup(void)
1451{
1452 free_kvm_area();
1453}
1454
6aa8b732
AK		 1455static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1456{
1457 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1458
6af11b9e 1459 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
6aa8b732
AK		 1460 vmcs_write16(sf->selector, save->selector);
1461 vmcs_writel(sf->base, save->base);
1462 vmcs_write32(sf->limit, save->limit);
1463 vmcs_write32(sf->ar_bytes, save->ar);
1464 } else {
1465 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1466 << AR_DPL_SHIFT;
1467 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1468 }
1469}
1470
1471static void enter_pmode(struct kvm_vcpu *vcpu)
1472{
1473 unsigned long flags;
a89a8fb9 1474 struct vcpu_vmx *vmx = to_vmx(vcpu);
6aa8b732 1475
a89a8fb9 1476 vmx->emulation_required = 1;
7ffd92c5 1477 vmx->rmode.vm86_active = 0;
6aa8b732 1478
7ffd92c5
AK		 1479 vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1480 vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1481 vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
6aa8b732
AK		 1482
1483 flags = vmcs_readl(GUEST_RFLAGS);
053de044 1484 flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
7ffd92c5 1485 flags |= (vmx->rmode.save_iopl << IOPL_SHIFT);
6aa8b732
AK		 1486 vmcs_writel(GUEST_RFLAGS, flags);
1487
66aee91a
RR		 1488 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1489 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
6aa8b732
AK		 1490
1491 update_exception_bitmap(vcpu);
1492
a89a8fb9
MG		 1493 if (emulate_invalid_guest_state)
1494 return;
1495
7ffd92c5
AK		 1496 fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1497 fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1498 fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1499 fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
6aa8b732
AK		 1500
1501 vmcs_write16(GUEST_SS_SELECTOR, 0);
1502 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1503
1504 vmcs_write16(GUEST_CS_SELECTOR,
1505 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1506 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1507}
1508
d77c26fc 1509static gva_t rmode_tss_base(struct kvm *kvm)
6aa8b732 1510{
bfc6d222 1511 if (!kvm->arch.tss_addr) {
bc6678a3
MT		 1512 struct kvm_memslots *slots;
1513 gfn_t base_gfn;
1514
1515 slots = rcu_dereference(kvm->memslots);
1516 base_gfn = kvm->memslots->memslots[0].base_gfn +
46a26bf5 1517 kvm->memslots->memslots[0].npages - 3;
cbc94022
IE		 1518 return base_gfn << PAGE_SHIFT;
1519 }
bfc6d222 1520 return kvm->arch.tss_addr;
6aa8b732
AK		 1521}
1522
1523static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1524{
1525 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1526
1527 save->selector = vmcs_read16(sf->selector);
1528 save->base = vmcs_readl(sf->base);
1529 save->limit = vmcs_read32(sf->limit);
1530 save->ar = vmcs_read32(sf->ar_bytes);
15b00f32
JK		 1531 vmcs_write16(sf->selector, save->base >> 4);
1532 vmcs_write32(sf->base, save->base & 0xfffff);
6aa8b732
AK		 1533 vmcs_write32(sf->limit, 0xffff);
1534 vmcs_write32(sf->ar_bytes, 0xf3);
1535}
1536
1537static void enter_rmode(struct kvm_vcpu *vcpu)
1538{
1539 unsigned long flags;
a89a8fb9 1540 struct vcpu_vmx *vmx = to_vmx(vcpu);
6aa8b732 1541
3a624e29
NK		 1542 if (enable_unrestricted_guest)
1543 return;
1544
a89a8fb9 1545 vmx->emulation_required = 1;
7ffd92c5 1546 vmx->rmode.vm86_active = 1;
6aa8b732 1547
7ffd92c5 1548 vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
6aa8b732
AK		 1549 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1550
7ffd92c5 1551 vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
6aa8b732
AK		 1552 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1553
7ffd92c5 1554 vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
6aa8b732
AK		 1555 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1556
1557 flags = vmcs_readl(GUEST_RFLAGS);
7ffd92c5 1558 vmx->rmode.save_iopl
ad312c7c 1559 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
6aa8b732 1560
053de044 1561 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
6aa8b732
AK		 1562
1563 vmcs_writel(GUEST_RFLAGS, flags);
66aee91a 1564 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
6aa8b732
AK		 1565 update_exception_bitmap(vcpu);
1566
a89a8fb9
MG		 1567 if (emulate_invalid_guest_state)
1568 goto continue_rmode;
1569
6aa8b732
AK		 1570 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1571 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1572 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1573
1574 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
abacf8df 1575 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
8cb5b033
AK		 1576 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1577 vmcs_writel(GUEST_CS_BASE, 0xf0000);
6aa8b732
AK		 1578 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1579
7ffd92c5
AK		 1580 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1581 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1582 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1583 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
75880a01 1584
a89a8fb9 1585continue_rmode:
8668a3c4 1586 kvm_mmu_reset_context(vcpu);
b7ebfb05 1587 init_rmode(vcpu->kvm);
6aa8b732
AK		 1588}
1589
401d10de
AS		 1590static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1591{
1592 struct vcpu_vmx *vmx = to_vmx(vcpu);
26bb0981
AK		 1593 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1594
1595 if (!msr)
1596 return;
401d10de 1597
44ea2b17
AK		 1598 /*
1599 * Force kernel_gs_base reloading before EFER changes, as control
1600 * of this msr depends on is_long_mode().
1601 */
1602 vmx_load_host_state(to_vmx(vcpu));
401d10de
AS		 1603 vcpu->arch.shadow_efer = efer;
1604 if (!msr)
1605 return;
1606 if (efer & EFER_LMA) {
1607 vmcs_write32(VM_ENTRY_CONTROLS,
1608 vmcs_read32(VM_ENTRY_CONTROLS) |
1609 VM_ENTRY_IA32E_MODE);
1610 msr->data = efer;
1611 } else {
1612 vmcs_write32(VM_ENTRY_CONTROLS,
1613 vmcs_read32(VM_ENTRY_CONTROLS) &
1614 ~VM_ENTRY_IA32E_MODE);
1615
1616 msr->data = efer & ~EFER_LME;
1617 }
1618 setup_msrs(vmx);
1619}
1620
05b3e0c2 1621#ifdef CONFIG_X86_64
6aa8b732
AK		 1622
1623static void enter_lmode(struct kvm_vcpu *vcpu)
1624{
1625 u32 guest_tr_ar;
1626
1627 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1628 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1629 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
b8688d51 1630 __func__);
6aa8b732
AK		 1631 vmcs_write32(GUEST_TR_AR_BYTES,
1632 (guest_tr_ar & ~AR_TYPE_MASK)
1633 | AR_TYPE_BUSY_64_TSS);
1634 }
ad312c7c 1635 vcpu->arch.shadow_efer |= EFER_LMA;
401d10de 1636 vmx_set_efer(vcpu, vcpu->arch.shadow_efer);
6aa8b732
AK		 1637}
1638
1639static void exit_lmode(struct kvm_vcpu *vcpu)
1640{
ad312c7c 1641 vcpu->arch.shadow_efer &= ~EFER_LMA;
6aa8b732
AK		 1642
1643 vmcs_write32(VM_ENTRY_CONTROLS,
1644 vmcs_read32(VM_ENTRY_CONTROLS)
1e4e6e00 1645 & ~VM_ENTRY_IA32E_MODE);
6aa8b732
AK		 1646}
1647
1648#endif
1649
2384d2b3
SY		 1650static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1651{
1652 vpid_sync_vcpu_all(to_vmx(vcpu));
089d034e 1653 if (enable_ept)
4e1096d2 1654 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
2384d2b3
SY		 1655}
1656
e8467fda
AK		 1657static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1658{
1659 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
1660
1661 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
1662 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
1663}
1664
25c4c276 1665static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
399badf3 1666{
fc78f519
AK		 1667 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
1668
1669 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
1670 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
399badf3
AK		 1671}
1672
1439442c
SY		 1673static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1674{
6de4f3ad
AK		 1675 if (!test_bit(VCPU_EXREG_PDPTR,
1676 (unsigned long *)&vcpu->arch.regs_dirty))
1677 return;
1678
1439442c 1679 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1439442c
SY		 1680 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1681 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1682 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1683 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1684 }
1685}
1686
8f5d549f
AK		 1687static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1688{
1689 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1690 vcpu->arch.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1691 vcpu->arch.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1692 vcpu->arch.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1693 vcpu->arch.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
1694 }
6de4f3ad
AK		 1695
1696 __set_bit(VCPU_EXREG_PDPTR,
1697 (unsigned long *)&vcpu->arch.regs_avail);
1698 __set_bit(VCPU_EXREG_PDPTR,
1699 (unsigned long *)&vcpu->arch.regs_dirty);
8f5d549f
AK		 1700}
1701
1439442c
SY		 1702static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1703
1704static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1705 unsigned long cr0,
1706 struct kvm_vcpu *vcpu)
1707{
1708 if (!(cr0 & X86_CR0_PG)) {
1709 /* From paging/starting to nonpaging */
1710 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
65267ea1 1711 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1439442c
SY		 1712 (CPU_BASED_CR3_LOAD_EXITING |
1713 CPU_BASED_CR3_STORE_EXITING));
1714 vcpu->arch.cr0 = cr0;
fc78f519 1715 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1439442c
SY		 1716 } else if (!is_paging(vcpu)) {
1717 /* From nonpaging to paging */
1718 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
65267ea1 1719 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1439442c
SY		 1720 ~(CPU_BASED_CR3_LOAD_EXITING |
1721 CPU_BASED_CR3_STORE_EXITING));
1722 vcpu->arch.cr0 = cr0;
fc78f519 1723 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1439442c 1724 }
95eb84a7
SY		 1725
1726 if (!(cr0 & X86_CR0_WP))
1727 *hw_cr0 &= ~X86_CR0_WP;
1439442c
SY		 1728}
1729
6aa8b732
AK		 1730static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1731{
7ffd92c5 1732 struct vcpu_vmx *vmx = to_vmx(vcpu);
3a624e29
NK		 1733 unsigned long hw_cr0;
1734
1735 if (enable_unrestricted_guest)
1736 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
1737 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
1738 else
1739 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
1439442c 1740
7ffd92c5 1741 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
6aa8b732
AK		 1742 enter_pmode(vcpu);
1743
7ffd92c5 1744 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
6aa8b732
AK		 1745 enter_rmode(vcpu);
1746
05b3e0c2 1747#ifdef CONFIG_X86_64
ad312c7c 1748 if (vcpu->arch.shadow_efer & EFER_LME) {
707d92fa 1749 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
6aa8b732 1750 enter_lmode(vcpu);
707d92fa 1751 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
6aa8b732
AK		 1752 exit_lmode(vcpu);
1753 }
1754#endif
1755
089d034e 1756 if (enable_ept)
1439442c
SY		 1757 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1758
02daab21
AK		 1759 if (!vcpu->fpu_active)
1760 hw_cr0 |= X86_CR0_TS;
1761
6aa8b732 1762 vmcs_writel(CR0_READ_SHADOW, cr0);
1439442c 1763 vmcs_writel(GUEST_CR0, hw_cr0);
ad312c7c 1764 vcpu->arch.cr0 = cr0;
6aa8b732
AK		 1765}
1766
1439442c
SY		 1767static u64 construct_eptp(unsigned long root_hpa)
1768{
1769 u64 eptp;
1770
1771 /* TODO write the value reading from MSR */
1772 eptp = VMX_EPT_DEFAULT_MT |
1773 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1774 eptp |= (root_hpa & PAGE_MASK);
1775
1776 return eptp;
1777}
1778
6aa8b732
AK		 1779static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1780{
1439442c
SY		 1781 unsigned long guest_cr3;
1782 u64 eptp;
1783
1784 guest_cr3 = cr3;
089d034e 1785 if (enable_ept) {
1439442c
SY		 1786 eptp = construct_eptp(cr3);
1787 vmcs_write64(EPT_POINTER, eptp);
1439442c 1788 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
b927a3ce 1789 vcpu->kvm->arch.ept_identity_map_addr;
7c93be44 1790 ept_load_pdptrs(vcpu);
1439442c
SY		 1791 }
1792
2384d2b3 1793 vmx_flush_tlb(vcpu);
1439442c 1794 vmcs_writel(GUEST_CR3, guest_cr3);
6aa8b732
AK		 1795}
1796
1797static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1798{
7ffd92c5 1799 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
1439442c
SY		 1800 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1801
ad312c7c 1802 vcpu->arch.cr4 = cr4;
bc23008b
AK		 1803 if (enable_ept) {
1804 if (!is_paging(vcpu)) {
1805 hw_cr4 &= ~X86_CR4_PAE;
1806 hw_cr4 |= X86_CR4_PSE;
1807 } else if (!(cr4 & X86_CR4_PAE)) {
1808 hw_cr4 &= ~X86_CR4_PAE;
1809 }
1810 }
1439442c
SY		 1811
1812 vmcs_writel(CR4_READ_SHADOW, cr4);
1813 vmcs_writel(GUEST_CR4, hw_cr4);
6aa8b732
AK		 1814}
1815
6aa8b732
AK		 1816static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1817{
1818 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1819
1820 return vmcs_readl(sf->base);
1821}
1822
1823static void vmx_get_segment(struct kvm_vcpu *vcpu,
1824 struct kvm_segment *var, int seg)
1825{
1826 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1827 u32 ar;
1828
1829 var->base = vmcs_readl(sf->base);
1830 var->limit = vmcs_read32(sf->limit);
1831 var->selector = vmcs_read16(sf->selector);
1832 ar = vmcs_read32(sf->ar_bytes);
9fd4a3b7 1833 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
6aa8b732
AK		 1834 ar = 0;
1835 var->type = ar & 15;
1836 var->s = (ar >> 4) & 1;
1837 var->dpl = (ar >> 5) & 3;
1838 var->present = (ar >> 7) & 1;
1839 var->avl = (ar >> 12) & 1;
1840 var->l = (ar >> 13) & 1;
1841 var->db = (ar >> 14) & 1;
1842 var->g = (ar >> 15) & 1;
1843 var->unusable = (ar >> 16) & 1;
1844}
1845
2e4d2653
IE		 1846static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1847{
4d4ec087 1848 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) /* if real mode */
2e4d2653
IE		 1849 return 0;
1850
1851 if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1852 return 3;
1853
eab4b8aa 1854 return vmcs_read16(GUEST_CS_SELECTOR) & 3;
2e4d2653
IE		 1855}
1856
653e3108 1857static u32 vmx_segment_access_rights(struct kvm_segment *var)
6aa8b732 1858{
6aa8b732
AK		 1859 u32 ar;
1860
653e3108 1861 if (var->unusable)
6aa8b732
AK		 1862 ar = 1 << 16;
1863 else {
1864 ar = var->type & 15;
1865 ar |= (var->s & 1) << 4;
1866 ar |= (var->dpl & 3) << 5;
1867 ar |= (var->present & 1) << 7;
1868 ar |= (var->avl & 1) << 12;
1869 ar |= (var->l & 1) << 13;
1870 ar |= (var->db & 1) << 14;
1871 ar |= (var->g & 1) << 15;
1872 }
f7fbf1fd
UL		 1873 if (ar == 0) /* a 0 value means unusable */
1874 ar = AR_UNUSABLE_MASK;
653e3108
AK		 1875
1876 return ar;
1877}
1878
1879static void vmx_set_segment(struct kvm_vcpu *vcpu,
1880 struct kvm_segment *var, int seg)
1881{
7ffd92c5 1882 struct vcpu_vmx *vmx = to_vmx(vcpu);
653e3108
AK		 1883 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1884 u32 ar;
1885
7ffd92c5
AK		 1886 if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
1887 vmx->rmode.tr.selector = var->selector;
1888 vmx->rmode.tr.base = var->base;
1889 vmx->rmode.tr.limit = var->limit;
1890 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
653e3108
AK		 1891 return;
1892 }
1893 vmcs_writel(sf->base, var->base);
1894 vmcs_write32(sf->limit, var->limit);
1895 vmcs_write16(sf->selector, var->selector);
7ffd92c5 1896 if (vmx->rmode.vm86_active && var->s) {
653e3108
AK		 1897 /*
1898 * Hack real-mode segments into vm86 compatibility.
1899 */
1900 if (var->base == 0xffff0000 && var->selector == 0xf000)
1901 vmcs_writel(sf->base, 0xf0000);
1902 ar = 0xf3;
1903 } else
1904 ar = vmx_segment_access_rights(var);
3a624e29
NK		 1905
1906 /*
1907 * Fix the "Accessed" bit in AR field of segment registers for older
1908 * qemu binaries.
1909 * IA32 arch specifies that at the time of processor reset the
1910 * "Accessed" bit in the AR field of segment registers is 1. And qemu
1911 * is setting it to 0 in the usedland code. This causes invalid guest
1912 * state vmexit when "unrestricted guest" mode is turned on.
1913 * Fix for this setup issue in cpu_reset is being pushed in the qemu
1914 * tree. Newer qemu binaries with that qemu fix would not need this
1915 * kvm hack.
1916 */
1917 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
1918 ar |= 0x1; /* Accessed */
1919
6aa8b732
AK		 1920 vmcs_write32(sf->ar_bytes, ar);
1921}
1922
6aa8b732
AK		 1923static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1924{
1925 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1926
1927 *db = (ar >> 14) & 1;
1928 *l = (ar >> 13) & 1;
1929}
1930
1931static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1932{
1933 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1934 dt->base = vmcs_readl(GUEST_IDTR_BASE);
1935}
1936
1937static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1938{
1939 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1940 vmcs_writel(GUEST_IDTR_BASE, dt->base);
1941}
1942
1943static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1944{
1945 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1946 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1947}
1948
1949static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1950{
1951 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1952 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1953}
1954
648dfaa7
MG		 1955static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
1956{
1957 struct kvm_segment var;
1958 u32 ar;
1959
1960 vmx_get_segment(vcpu, &var, seg);
1961 ar = vmx_segment_access_rights(&var);
1962
1963 if (var.base != (var.selector << 4))
1964 return false;
1965 if (var.limit != 0xffff)
1966 return false;
1967 if (ar != 0xf3)
1968 return false;
1969
1970 return true;
1971}
1972
1973static bool code_segment_valid(struct kvm_vcpu *vcpu)
1974{
1975 struct kvm_segment cs;
1976 unsigned int cs_rpl;
1977
1978 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1979 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
1980
1872a3f4
AK		 1981 if (cs.unusable)
1982 return false;
648dfaa7
MG		 1983 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
1984 return false;
1985 if (!cs.s)
1986 return false;
1872a3f4 1987 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
648dfaa7
MG		 1988 if (cs.dpl > cs_rpl)
1989 return false;
1872a3f4 1990 } else {
648dfaa7
MG		 1991 if (cs.dpl != cs_rpl)
1992 return false;
1993 }
1994 if (!cs.present)
1995 return false;
1996
1997 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
1998 return true;
1999}
2000
2001static bool stack_segment_valid(struct kvm_vcpu *vcpu)
2002{
2003 struct kvm_segment ss;
2004 unsigned int ss_rpl;
2005
2006 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2007 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
2008
1872a3f4
AK		 2009 if (ss.unusable)
2010 return true;
2011 if (ss.type != 3 && ss.type != 7)
648dfaa7
MG		 2012 return false;
2013 if (!ss.s)
2014 return false;
2015 if (ss.dpl != ss_rpl) /* DPL != RPL */
2016 return false;
2017 if (!ss.present)
2018 return false;
2019
2020 return true;
2021}
2022
2023static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
2024{
2025 struct kvm_segment var;
2026 unsigned int rpl;
2027
2028 vmx_get_segment(vcpu, &var, seg);
2029 rpl = var.selector & SELECTOR_RPL_MASK;
2030
1872a3f4
AK		 2031 if (var.unusable)
2032 return true;
648dfaa7
MG		 2033 if (!var.s)
2034 return false;
2035 if (!var.present)
2036 return false;
2037 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
2038 if (var.dpl < rpl) /* DPL < RPL */
2039 return false;
2040 }
2041
2042 /* TODO: Add other members to kvm_segment_field to allow checking for other access
2043 * rights flags
2044 */
2045 return true;
2046}
2047
2048static bool tr_valid(struct kvm_vcpu *vcpu)
2049{
2050 struct kvm_segment tr;
2051
2052 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
2053
1872a3f4
AK		 2054 if (tr.unusable)
2055 return false;
648dfaa7
MG		 2056 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2057 return false;
1872a3f4 2058 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
648dfaa7
MG		 2059 return false;
2060 if (!tr.present)
2061 return false;
2062
2063 return true;
2064}
2065
2066static bool ldtr_valid(struct kvm_vcpu *vcpu)
2067{
2068 struct kvm_segment ldtr;
2069
2070 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
2071
1872a3f4
AK		 2072 if (ldtr.unusable)
2073 return true;
648dfaa7
MG		 2074 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2075 return false;
2076 if (ldtr.type != 2)
2077 return false;
2078 if (!ldtr.present)
2079 return false;
2080
2081 return true;
2082}
2083
2084static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2085{
2086 struct kvm_segment cs, ss;
2087
2088 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2089 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2090
2091 return ((cs.selector & SELECTOR_RPL_MASK) ==
2092 (ss.selector & SELECTOR_RPL_MASK));
2093}
2094
2095/*
2096 * Check if guest state is valid. Returns true if valid, false if
2097 * not.
2098 * We assume that registers are always usable
2099 */
2100static bool guest_state_valid(struct kvm_vcpu *vcpu)
2101{
2102 /* real mode guest state checks */
4d4ec087 2103 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
648dfaa7
MG		 2104 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2105 return false;
2106 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2107 return false;
2108 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2109 return false;
2110 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2111 return false;
2112 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2113 return false;
2114 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2115 return false;
2116 } else {
2117 /* protected mode guest state checks */
2118 if (!cs_ss_rpl_check(vcpu))
2119 return false;
2120 if (!code_segment_valid(vcpu))
2121 return false;
2122 if (!stack_segment_valid(vcpu))
2123 return false;
2124 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2125 return false;
2126 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2127 return false;
2128 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2129 return false;
2130 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2131 return false;
2132 if (!tr_valid(vcpu))
2133 return false;
2134 if (!ldtr_valid(vcpu))
2135 return false;
2136 }
2137 /* TODO:
2138 * - Add checks on RIP
2139 * - Add checks on RFLAGS
2140 */
2141
2142 return true;
2143}
2144
d77c26fc 2145static int init_rmode_tss(struct kvm *kvm)
6aa8b732 2146{
6aa8b732 2147 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
195aefde 2148 u16 data = 0;
10589a46 2149 int ret = 0;
195aefde 2150 int r;
6aa8b732 2151
195aefde
IE		 2152 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2153 if (r < 0)
10589a46 2154 goto out;
195aefde 2155 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
464d17c8
SY		 2156 r = kvm_write_guest_page(kvm, fn++, &data,
2157 TSS_IOPB_BASE_OFFSET, sizeof(u16));
195aefde 2158 if (r < 0)
10589a46 2159 goto out;
195aefde
IE		 2160 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2161 if (r < 0)
10589a46 2162 goto out;
195aefde
IE		 2163 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2164 if (r < 0)
10589a46 2165 goto out;
195aefde 2166 data = ~0;
10589a46
MT		 2167 r = kvm_write_guest_page(kvm, fn, &data,
2168 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2169 sizeof(u8));
195aefde 2170 if (r < 0)
10589a46
MT		 2171 goto out;
2172
2173 ret = 1;
2174out:
10589a46 2175 return ret;
6aa8b732
AK		 2176}
2177
b7ebfb05
SY		 2178static int init_rmode_identity_map(struct kvm *kvm)
2179{
2180 int i, r, ret;
2181 pfn_t identity_map_pfn;
2182 u32 tmp;
2183
089d034e 2184 if (!enable_ept)
b7ebfb05
SY		 2185 return 1;
2186 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2187 printk(KERN_ERR "EPT: identity-mapping pagetable "
2188 "haven't been allocated!\n");
2189 return 0;
2190 }
2191 if (likely(kvm->arch.ept_identity_pagetable_done))
2192 return 1;
2193 ret = 0;
b927a3ce 2194 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
b7ebfb05
SY		 2195 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2196 if (r < 0)
2197 goto out;
2198 /* Set up identity-mapping pagetable for EPT in real mode */
2199 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2200 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2201 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2202 r = kvm_write_guest_page(kvm, identity_map_pfn,
2203 &tmp, i * sizeof(tmp), sizeof(tmp));
2204 if (r < 0)
2205 goto out;
2206 }
2207 kvm->arch.ept_identity_pagetable_done = true;
2208 ret = 1;
2209out:
2210 return ret;
2211}
2212
6aa8b732
AK		 2213static void seg_setup(int seg)
2214{
2215 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3a624e29 2216 unsigned int ar;
6aa8b732
AK		 2217
2218 vmcs_write16(sf->selector, 0);
2219 vmcs_writel(sf->base, 0);
2220 vmcs_write32(sf->limit, 0xffff);
3a624e29
NK		 2221 if (enable_unrestricted_guest) {
2222 ar = 0x93;
2223 if (seg == VCPU_SREG_CS)
2224 ar |= 0x08; /* code segment */
2225 } else
2226 ar = 0xf3;
2227
2228 vmcs_write32(sf->ar_bytes, ar);
6aa8b732
AK		 2229}
2230
f78e0e2e
SY		 2231static int alloc_apic_access_page(struct kvm *kvm)
2232{
2233 struct kvm_userspace_memory_region kvm_userspace_mem;
2234 int r = 0;
2235
79fac95e 2236 mutex_lock(&kvm->slots_lock);
bfc6d222 2237 if (kvm->arch.apic_access_page)
f78e0e2e
SY		 2238 goto out;
2239 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2240 kvm_userspace_mem.flags = 0;
2241 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2242 kvm_userspace_mem.memory_size = PAGE_SIZE;
2243 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2244 if (r)
2245 goto out;
72dc67a6 2246
bfc6d222 2247 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
f78e0e2e 2248out:
79fac95e 2249 mutex_unlock(&kvm->slots_lock);
f78e0e2e
SY		 2250 return r;
2251}
2252
b7ebfb05
SY		 2253static int alloc_identity_pagetable(struct kvm *kvm)
2254{
2255 struct kvm_userspace_memory_region kvm_userspace_mem;
2256 int r = 0;
2257
79fac95e 2258 mutex_lock(&kvm->slots_lock);
b7ebfb05
SY		 2259 if (kvm->arch.ept_identity_pagetable)
2260 goto out;
2261 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2262 kvm_userspace_mem.flags = 0;
b927a3ce
SY		 2263 kvm_userspace_mem.guest_phys_addr =
2264 kvm->arch.ept_identity_map_addr;
b7ebfb05
SY		 2265 kvm_userspace_mem.memory_size = PAGE_SIZE;
2266 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2267 if (r)
2268 goto out;
2269
b7ebfb05 2270 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
b927a3ce 2271 kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
b7ebfb05 2272out:
79fac95e 2273 mutex_unlock(&kvm->slots_lock);
b7ebfb05
SY		 2274 return r;
2275}
2276
2384d2b3
SY		 2277static void allocate_vpid(struct vcpu_vmx *vmx)
2278{
2279 int vpid;
2280
2281 vmx->vpid = 0;
919818ab 2282 if (!enable_vpid)
2384d2b3
SY		 2283 return;
2284 spin_lock(&vmx_vpid_lock);
2285 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2286 if (vpid < VMX_NR_VPIDS) {
2287 vmx->vpid = vpid;
2288 __set_bit(vpid, vmx_vpid_bitmap);
2289 }
2290 spin_unlock(&vmx_vpid_lock);
2291}
2292
5897297b 2293static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
25c5f225 2294{
3e7c73e9 2295 int f = sizeof(unsigned long);
25c5f225
SY		 2296
2297 if (!cpu_has_vmx_msr_bitmap())
2298 return;
2299
2300 /*
2301 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2302 * have the write-low and read-high bitmap offsets the wrong way round.
2303 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2304 */
25c5f225 2305 if (msr <= 0x1fff) {
3e7c73e9
AK		 2306 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2307 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
25c5f225
SY		 2308 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2309 msr &= 0x1fff;
3e7c73e9
AK		 2310 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2311 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
25c5f225 2312 }
25c5f225
SY		 2313}
2314
5897297b
AK		 2315static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2316{
2317 if (!longmode_only)
2318 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2319 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2320}
2321
6aa8b732
AK		 2322/*
2323 * Sets up the vmcs for emulated real mode.
2324 */
8b9cf98c 2325static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
6aa8b732 2326{
468d472f 2327 u32 host_sysenter_cs, msr_low, msr_high;
6aa8b732 2328 u32 junk;
53f658b3 2329 u64 host_pat, tsc_this, tsc_base;
6aa8b732
AK		 2330 unsigned long a;
2331 struct descriptor_table dt;
2332 int i;
cd2276a7 2333 unsigned long kvm_vmx_return;
6e5d865c 2334 u32 exec_control;
6aa8b732 2335
6aa8b732 2336 /* I/O */
3e7c73e9
AK		 2337 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2338 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
6aa8b732 2339
25c5f225 2340 if (cpu_has_vmx_msr_bitmap())
5897297b 2341 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
25c5f225 2342
6aa8b732
AK		 2343 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2344
6aa8b732 2345 /* Control */
1c3d14fe
YS		 2346 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2347 vmcs_config.pin_based_exec_ctrl);
6e5d865c
YS		 2348
2349 exec_control = vmcs_config.cpu_based_exec_ctrl;
2350 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2351 exec_control &= ~CPU_BASED_TPR_SHADOW;
2352#ifdef CONFIG_X86_64
2353 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2354 CPU_BASED_CR8_LOAD_EXITING;
2355#endif
2356 }
089d034e 2357 if (!enable_ept)
d56f546d 2358 exec_control |= CPU_BASED_CR3_STORE_EXITING |
83dbc83a
MT		 2359 CPU_BASED_CR3_LOAD_EXITING |
2360 CPU_BASED_INVLPG_EXITING;
6e5d865c 2361 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
6aa8b732 2362
83ff3b9d
SY		 2363 if (cpu_has_secondary_exec_ctrls()) {
2364 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2365 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2366 exec_control &=
2367 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2384d2b3
SY		 2368 if (vmx->vpid == 0)
2369 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
046d8710 2370 if (!enable_ept) {
d56f546d 2371 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
046d8710
SY		 2372 enable_unrestricted_guest = 0;
2373 }
3a624e29
NK		 2374 if (!enable_unrestricted_guest)
2375 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4b8d54f9
ZE		 2376 if (!ple_gap)
2377 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
83ff3b9d
SY		 2378 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2379 }
f78e0e2e 2380
4b8d54f9
ZE		 2381 if (ple_gap) {
2382 vmcs_write32(PLE_GAP, ple_gap);
2383 vmcs_write32(PLE_WINDOW, ple_window);
2384 }
2385
c7addb90
AK		 2386 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2387 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
6aa8b732
AK		 2388 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
2389
2390 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
2391 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
2392 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2393
2394 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
2395 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2396 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
d6e88aec
AK		 2397 vmcs_write16(HOST_FS_SELECTOR, kvm_read_fs()); /* 22.2.4 */
2398 vmcs_write16(HOST_GS_SELECTOR, kvm_read_gs()); /* 22.2.4 */
6aa8b732 2399 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
05b3e0c2 2400#ifdef CONFIG_X86_64
6aa8b732
AK		 2401 rdmsrl(MSR_FS_BASE, a);
2402 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2403 rdmsrl(MSR_GS_BASE, a);
2404 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2405#else
2406 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2407 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2408#endif
2409
2410 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
2411
d6e88aec 2412 kvm_get_idt(&dt);
6aa8b732
AK		 2413 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
2414
d77c26fc 2415 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
cd2276a7 2416 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2cc51560
ED		 2417 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2418 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2419 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
6aa8b732
AK		 2420
2421 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2422 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2423 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2424 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
2425 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2426 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2427
468d472f
SY		 2428 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2429 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2430 host_pat = msr_low | ((u64) msr_high << 32);
2431 vmcs_write64(HOST_IA32_PAT, host_pat);
2432 }
2433 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2434 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2435 host_pat = msr_low | ((u64) msr_high << 32);
2436 /* Write the default value follow host pat */
2437 vmcs_write64(GUEST_IA32_PAT, host_pat);
2438 /* Keep arch.pat sync with GUEST_IA32_PAT */
2439 vmx->vcpu.arch.pat = host_pat;
2440 }
2441
6aa8b732
AK		 2442 for (i = 0; i < NR_VMX_MSR; ++i) {
2443 u32 index = vmx_msr_index[i];
2444 u32 data_low, data_high;
a2fa3e9f 2445 int j = vmx->nmsrs;
6aa8b732
AK		 2446
2447 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2448 continue;
432bd6cb
AK		 2449 if (wrmsr_safe(index, data_low, data_high) < 0)
2450 continue;
26bb0981
AK		 2451 vmx->guest_msrs[j].index = i;
2452 vmx->guest_msrs[j].data = 0;
d5696725 2453 vmx->guest_msrs[j].mask = -1ull;
a2fa3e9f 2454 ++vmx->nmsrs;
6aa8b732 2455 }
6aa8b732 2456
1c3d14fe 2457 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
6aa8b732
AK		 2458
2459 /* 22.2.1, 20.8.1 */
1c3d14fe
YS		 2460 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2461
e00c8cf2 2462 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4c38609a 2463 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
ce03e4f2
AK		 2464 if (enable_ept)
2465 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4c38609a 2466 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
e00c8cf2 2467
53f658b3
MT		 2468 tsc_base = vmx->vcpu.kvm->arch.vm_init_tsc;
2469 rdtscll(tsc_this);
2470 if (tsc_this < vmx->vcpu.kvm->arch.vm_init_tsc)
2471 tsc_base = tsc_this;
2472
2473 guest_write_tsc(0, tsc_base);
f78e0e2e 2474
e00c8cf2
AK		 2475 return 0;
2476}
2477
b7ebfb05
SY		 2478static int init_rmode(struct kvm *kvm)
2479{
2480 if (!init_rmode_tss(kvm))
2481 return 0;
2482 if (!init_rmode_identity_map(kvm))
2483 return 0;
2484 return 1;
2485}
2486
e00c8cf2
AK		 2487static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2488{
2489 struct vcpu_vmx *vmx = to_vmx(vcpu);
2490 u64 msr;
f656ce01 2491 int ret, idx;
e00c8cf2 2492
5fdbf976 2493 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
f656ce01 2494 idx = srcu_read_lock(&vcpu->kvm->srcu);
b7ebfb05 2495 if (!init_rmode(vmx->vcpu.kvm)) {
e00c8cf2
AK		 2496 ret = -ENOMEM;
2497 goto out;
2498 }
2499
7ffd92c5 2500 vmx->rmode.vm86_active = 0;
e00c8cf2 2501
3b86cd99
JK		 2502 vmx->soft_vnmi_blocked = 0;
2503
ad312c7c 2504 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2d3ad1f4 2505 kvm_set_cr8(&vmx->vcpu, 0);
e00c8cf2 2506 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
c5af89b6 2507 if (kvm_vcpu_is_bsp(&vmx->vcpu))
e00c8cf2
AK		 2508 msr |= MSR_IA32_APICBASE_BSP;
2509 kvm_set_apic_base(&vmx->vcpu, msr);
2510
2511 fx_init(&vmx->vcpu);
2512
5706be0d 2513 seg_setup(VCPU_SREG_CS);
e00c8cf2
AK		 2514 /*
2515 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2516 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2517 */
c5af89b6 2518 if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
e00c8cf2
AK		 2519 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2520 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2521 } else {
ad312c7c
ZX		 2522 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2523 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
e00c8cf2 2524 }
e00c8cf2
AK		 2525
2526 seg_setup(VCPU_SREG_DS);
2527 seg_setup(VCPU_SREG_ES);
2528 seg_setup(VCPU_SREG_FS);
2529 seg_setup(VCPU_SREG_GS);
2530 seg_setup(VCPU_SREG_SS);
2531
2532 vmcs_write16(GUEST_TR_SELECTOR, 0);
2533 vmcs_writel(GUEST_TR_BASE, 0);
2534 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2535 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2536
2537 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2538 vmcs_writel(GUEST_LDTR_BASE, 0);
2539 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2540 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2541
2542 vmcs_write32(GUEST_SYSENTER_CS, 0);
2543 vmcs_writel(GUEST_SYSENTER_ESP, 0);
2544 vmcs_writel(GUEST_SYSENTER_EIP, 0);
2545
2546 vmcs_writel(GUEST_RFLAGS, 0x02);
c5af89b6 2547 if (kvm_vcpu_is_bsp(&vmx->vcpu))
5fdbf976 2548 kvm_rip_write(vcpu, 0xfff0);
e00c8cf2 2549 else
5fdbf976
MT		 2550 kvm_rip_write(vcpu, 0);
2551 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
e00c8cf2 2552
e00c8cf2
AK		 2553 vmcs_writel(GUEST_DR7, 0x400);
2554
2555 vmcs_writel(GUEST_GDTR_BASE, 0);
2556 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2557
2558 vmcs_writel(GUEST_IDTR_BASE, 0);
2559 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2560
2561 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2562 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2563 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2564
e00c8cf2
AK		 2565 /* Special registers */
2566 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2567
2568 setup_msrs(vmx);
2569
6aa8b732
AK		 2570 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
2571
f78e0e2e
SY		 2572 if (cpu_has_vmx_tpr_shadow()) {
2573 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2574 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2575 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
ad312c7c 2576 page_to_phys(vmx->vcpu.arch.apic->regs_page));
f78e0e2e
SY		 2577 vmcs_write32(TPR_THRESHOLD, 0);
2578 }
2579
2580 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2581 vmcs_write64(APIC_ACCESS_ADDR,
bfc6d222 2582 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
6aa8b732 2583
2384d2b3
SY		 2584 if (vmx->vpid != 0)
2585 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2586
fa40052c 2587 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4d4ec087 2588 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
8b9cf98c 2589 vmx_set_cr4(&vmx->vcpu, 0);
8b9cf98c 2590 vmx_set_efer(&vmx->vcpu, 0);
8b9cf98c
RR		 2591 vmx_fpu_activate(&vmx->vcpu);
2592 update_exception_bitmap(&vmx->vcpu);
6aa8b732 2593
2384d2b3
SY		 2594 vpid_sync_vcpu_all(vmx);
2595
3200f405 2596 ret = 0;
6aa8b732 2597
a89a8fb9
MG		 2598 /* HACK: Don't enable emulation on guest boot/reset */
2599 vmx->emulation_required = 0;
2600
6aa8b732 2601out:
f656ce01 2602 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6aa8b732
AK		 2603 return ret;
2604}
2605
3b86cd99
JK		 2606static void enable_irq_window(struct kvm_vcpu *vcpu)
2607{
2608 u32 cpu_based_vm_exec_control;
2609
2610 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2611 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2612 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2613}
2614
2615static void enable_nmi_window(struct kvm_vcpu *vcpu)
2616{
2617 u32 cpu_based_vm_exec_control;
2618
2619 if (!cpu_has_virtual_nmis()) {
2620 enable_irq_window(vcpu);
2621 return;
2622 }
2623
2624 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2625 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2626 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2627}
2628
66fd3f7f 2629static void vmx_inject_irq(struct kvm_vcpu *vcpu)
85f455f7 2630{
9c8cba37 2631 struct vcpu_vmx *vmx = to_vmx(vcpu);
66fd3f7f
GN		 2632 uint32_t intr;
2633 int irq = vcpu->arch.interrupt.nr;
9c8cba37 2634
229456fc 2635 trace_kvm_inj_virq(irq);
2714d1d3 2636
fa89a817 2637 ++vcpu->stat.irq_injections;
7ffd92c5 2638 if (vmx->rmode.vm86_active) {
9c8cba37
AK		 2639 vmx->rmode.irq.pending = true;
2640 vmx->rmode.irq.vector = irq;
5fdbf976 2641 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
ae0bb3e0
GN		 2642 if (vcpu->arch.interrupt.soft)
2643 vmx->rmode.irq.rip +=
2644 vmx->vcpu.arch.event_exit_inst_len;
9c5623e3
AK		 2645 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2646 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2647 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
5fdbf976 2648 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
85f455f7
ED		 2649 return;
2650 }
66fd3f7f
GN		 2651 intr = irq | INTR_INFO_VALID_MASK;
2652 if (vcpu->arch.interrupt.soft) {
2653 intr |= INTR_TYPE_SOFT_INTR;
2654 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2655 vmx->vcpu.arch.event_exit_inst_len);
2656 } else
2657 intr |= INTR_TYPE_EXT_INTR;
2658 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
85f455f7
ED		 2659}
2660
f08864b4
SY		 2661static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2662{
66a5a347
JK		 2663 struct vcpu_vmx *vmx = to_vmx(vcpu);
2664
3b86cd99
JK		 2665 if (!cpu_has_virtual_nmis()) {
2666 /*
2667 * Tracking the NMI-blocked state in software is built upon
2668 * finding the next open IRQ window. This, in turn, depends on
2669 * well-behaving guests: They have to keep IRQs disabled at
2670 * least as long as the NMI handler runs. Otherwise we may
2671 * cause NMI nesting, maybe breaking the guest. But as this is
2672 * highly unlikely, we can live with the residual risk.
2673 */
2674 vmx->soft_vnmi_blocked = 1;
2675 vmx->vnmi_blocked_time = 0;
2676 }
2677
487b391d 2678 ++vcpu->stat.nmi_injections;
7ffd92c5 2679 if (vmx->rmode.vm86_active) {
66a5a347
JK		 2680 vmx->rmode.irq.pending = true;
2681 vmx->rmode.irq.vector = NMI_VECTOR;
2682 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2683 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2684 NMI_VECTOR | INTR_TYPE_SOFT_INTR |
2685 INTR_INFO_VALID_MASK);
2686 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2687 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2688 return;
2689 }
f08864b4
SY		 2690 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2691 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
f08864b4
SY		 2692}
2693
c4282df9 2694static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
33f089ca 2695{
3b86cd99 2696 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
c4282df9 2697 return 0;
33f089ca 2698
c4282df9
GN		 2699 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2700 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS |
2701 GUEST_INTR_STATE_NMI));
33f089ca
JK		 2702}
2703
3cfc3092
JK		 2704static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
2705{
2706 if (!cpu_has_virtual_nmis())
2707 return to_vmx(vcpu)->soft_vnmi_blocked;
2708 else
2709 return !!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2710 GUEST_INTR_STATE_NMI);
2711}
2712
2713static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
2714{
2715 struct vcpu_vmx *vmx = to_vmx(vcpu);
2716
2717 if (!cpu_has_virtual_nmis()) {
2718 if (vmx->soft_vnmi_blocked != masked) {
2719 vmx->soft_vnmi_blocked = masked;
2720 vmx->vnmi_blocked_time = 0;
2721 }
2722 } else {
2723 if (masked)
2724 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
2725 GUEST_INTR_STATE_NMI);
2726 else
2727 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
2728 GUEST_INTR_STATE_NMI);
2729 }
2730}
2731
78646121
GN		 2732static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2733{
c4282df9
GN		 2734 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2735 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2736 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
78646121
GN		 2737}
2738
cbc94022
IE		 2739static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2740{
2741 int ret;
2742 struct kvm_userspace_memory_region tss_mem = {
6fe63979 2743 .slot = TSS_PRIVATE_MEMSLOT,
cbc94022
IE		 2744 .guest_phys_addr = addr,
2745 .memory_size = PAGE_SIZE * 3,
2746 .flags = 0,
2747 };
2748
2749 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2750 if (ret)
2751 return ret;
bfc6d222 2752 kvm->arch.tss_addr = addr;
cbc94022
IE		 2753 return 0;
2754}
2755
6aa8b732
AK		 2756static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2757 int vec, u32 err_code)
2758{
b3f37707
NK		 2759 /*
2760 * Instruction with address size override prefix opcode 0x67
2761 * Cause the #SS fault with 0 error code in VM86 mode.
2762 */
2763 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
851ba692 2764 if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE)
6aa8b732 2765 return 1;
77ab6db0
JK		 2766 /*
2767 * Forward all other exceptions that are valid in real mode.
2768 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2769 * the required debugging infrastructure rework.
2770 */
2771 switch (vec) {
77ab6db0 2772 case DB_VECTOR:
d0bfb940
JK		 2773 if (vcpu->guest_debug &
2774 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2775 return 0;
2776 kvm_queue_exception(vcpu, vec);
2777 return 1;
77ab6db0 2778 case BP_VECTOR:
d0bfb940
JK		 2779 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2780 return 0;
2781 /* fall through */
2782 case DE_VECTOR:
77ab6db0
JK		 2783 case OF_VECTOR:
2784 case BR_VECTOR:
2785 case UD_VECTOR:
2786 case DF_VECTOR:
2787 case SS_VECTOR:
2788 case GP_VECTOR:
2789 case MF_VECTOR:
2790 kvm_queue_exception(vcpu, vec);
2791 return 1;
2792 }
6aa8b732
AK		 2793 return 0;
2794}
2795
a0861c02
AK		 2796/*
2797 * Trigger machine check on the host. We assume all the MSRs are already set up
2798 * by the CPU and that we still run on the same CPU as the MCE occurred on.
2799 * We pass a fake environment to the machine check handler because we want
2800 * the guest to be always treated like user space, no matter what context
2801 * it used internally.
2802 */
2803static void kvm_machine_check(void)
2804{
2805#if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
2806 struct pt_regs regs = {
2807 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
2808 .flags = X86_EFLAGS_IF,
2809 };
2810
2811 do_machine_check(&regs, 0);
2812#endif
2813}
2814
851ba692 2815static int handle_machine_check(struct kvm_vcpu *vcpu)
a0861c02
AK		 2816{
2817 /* already handled by vcpu_run */
2818 return 1;
2819}
2820
851ba692 2821static int handle_exception(struct kvm_vcpu *vcpu)
6aa8b732 2822{
1155f76a 2823 struct vcpu_vmx *vmx = to_vmx(vcpu);
851ba692 2824 struct kvm_run *kvm_run = vcpu->run;
d0bfb940 2825 u32 intr_info, ex_no, error_code;
42dbaa5a 2826 unsigned long cr2, rip, dr6;
6aa8b732
AK		 2827 u32 vect_info;
2828 enum emulation_result er;
2829
1155f76a 2830 vect_info = vmx->idt_vectoring_info;
6aa8b732
AK		 2831 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2832
a0861c02 2833 if (is_machine_check(intr_info))
851ba692 2834 return handle_machine_check(vcpu);
a0861c02 2835
6aa8b732 2836 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
65ac7264
AK		 2837 !is_page_fault(intr_info)) {
2838 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2839 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
2840 vcpu->run->internal.ndata = 2;
2841 vcpu->run->internal.data[0] = vect_info;
2842 vcpu->run->internal.data[1] = intr_info;
2843 return 0;
2844 }
6aa8b732 2845
e4a41889 2846 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
1b6269db 2847 return 1; /* already handled by vmx_vcpu_run() */
2ab455cc
AL		 2848
2849 if (is_no_device(intr_info)) {
5fd86fcf 2850 vmx_fpu_activate(vcpu);
2ab455cc
AL		 2851 return 1;
2852 }
2853
7aa81cc0 2854 if (is_invalid_opcode(intr_info)) {
851ba692 2855 er = emulate_instruction(vcpu, 0, 0, EMULTYPE_TRAP_UD);
7aa81cc0 2856 if (er != EMULATE_DONE)
7ee5d940 2857 kvm_queue_exception(vcpu, UD_VECTOR);
7aa81cc0
AL		 2858 return 1;
2859 }
2860
6aa8b732 2861 error_code = 0;
5fdbf976 2862 rip = kvm_rip_read(vcpu);
2e11384c 2863 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
6aa8b732
AK		 2864 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2865 if (is_page_fault(intr_info)) {
1439442c 2866 /* EPT won't cause page fault directly */
089d034e 2867 if (enable_ept)
1439442c 2868 BUG();
6aa8b732 2869 cr2 = vmcs_readl(EXIT_QUALIFICATION);
229456fc
MT		 2870 trace_kvm_page_fault(cr2, error_code);
2871
3298b75c 2872 if (kvm_event_needs_reinjection(vcpu))
577bdc49 2873 kvm_mmu_unprotect_page_virt(vcpu, cr2);
3067714c 2874 return kvm_mmu_page_fault(vcpu, cr2, error_code);
6aa8b732
AK		 2875 }
2876
7ffd92c5 2877 if (vmx->rmode.vm86_active &&
6aa8b732 2878 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
72d6e5a0 2879 error_code)) {
ad312c7c
ZX		 2880 if (vcpu->arch.halt_request) {
2881 vcpu->arch.halt_request = 0;
72d6e5a0
AK		 2882 return kvm_emulate_halt(vcpu);
2883 }
6aa8b732 2884 return 1;
72d6e5a0 2885 }
6aa8b732 2886
d0bfb940 2887 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
42dbaa5a
JK		 2888 switch (ex_no) {
2889 case DB_VECTOR:
2890 dr6 = vmcs_readl(EXIT_QUALIFICATION);
2891 if (!(vcpu->guest_debug &
2892 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
2893 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
2894 kvm_queue_exception(vcpu, DB_VECTOR);
2895 return 1;
2896 }
2897 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
2898 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
2899 /* fall through */
2900 case BP_VECTOR:
6aa8b732 2901 kvm_run->exit_reason = KVM_EXIT_DEBUG;
d0bfb940
JK		 2902 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
2903 kvm_run->debug.arch.exception = ex_no;
42dbaa5a
JK		 2904 break;
2905 default:
d0bfb940
JK		 2906 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
2907 kvm_run->ex.exception = ex_no;
2908 kvm_run->ex.error_code = error_code;
42dbaa5a 2909 break;
6aa8b732 2910 }
6aa8b732
AK		 2911 return 0;
2912}
2913
851ba692 2914static int handle_external_interrupt(struct kvm_vcpu *vcpu)
6aa8b732 2915{
1165f5fe 2916 ++vcpu->stat.irq_exits;
6aa8b732
AK		 2917 return 1;
2918}
2919
851ba692 2920static int handle_triple_fault(struct kvm_vcpu *vcpu)
988ad74f 2921{
851ba692 2922 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
988ad74f
AK		 2923 return 0;
2924}
6aa8b732 2925
851ba692 2926static int handle_io(struct kvm_vcpu *vcpu)
6aa8b732 2927{
bfdaab09 2928 unsigned long exit_qualification;
34c33d16 2929 int size, in, string;
039576c0 2930 unsigned port;
6aa8b732 2931
1165f5fe 2932 ++vcpu->stat.io_exits;
bfdaab09 2933 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
039576c0 2934 string = (exit_qualification & 16) != 0;
e70669ab
LV		 2935
2936 if (string) {
851ba692 2937 if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DO_MMIO)
e70669ab
LV		 2938 return 0;
2939 return 1;
2940 }
2941
2942 size = (exit_qualification & 7) + 1;
2943 in = (exit_qualification & 8) != 0;
039576c0 2944 port = exit_qualification >> 16;
e70669ab 2945
e93f36bc 2946 skip_emulated_instruction(vcpu);
851ba692 2947 return kvm_emulate_pio(vcpu, in, size, port);
6aa8b732
AK		 2948}
2949
102d8325
IM		 2950static void
2951vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2952{
2953 /*
2954 * Patch in the VMCALL instruction:
2955 */
2956 hypercall[0] = 0x0f;
2957 hypercall[1] = 0x01;
2958 hypercall[2] = 0xc1;
102d8325
IM		 2959}
2960
851ba692 2961static int handle_cr(struct kvm_vcpu *vcpu)
6aa8b732 2962{
229456fc 2963 unsigned long exit_qualification, val;
6aa8b732
AK		 2964 int cr;
2965 int reg;
2966
bfdaab09 2967 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6aa8b732
AK		 2968 cr = exit_qualification & 15;
2969 reg = (exit_qualification >> 8) & 15;
2970 switch ((exit_qualification >> 4) & 3) {
2971 case 0: /* mov to cr */
229456fc
MT		 2972 val = kvm_register_read(vcpu, reg);
2973 trace_kvm_cr_write(cr, val);
6aa8b732
AK		 2974 switch (cr) {
2975 case 0:
229456fc 2976 kvm_set_cr0(vcpu, val);
6aa8b732
AK		 2977 skip_emulated_instruction(vcpu);
2978 return 1;
2979 case 3:
229456fc 2980 kvm_set_cr3(vcpu, val);
6aa8b732
AK		 2981 skip_emulated_instruction(vcpu);
2982 return 1;
2983 case 4:
229456fc 2984 kvm_set_cr4(vcpu, val);
6aa8b732
AK		 2985 skip_emulated_instruction(vcpu);
2986 return 1;
0a5fff19
GN		 2987 case 8: {
2988 u8 cr8_prev = kvm_get_cr8(vcpu);
2989 u8 cr8 = kvm_register_read(vcpu, reg);
2990 kvm_set_cr8(vcpu, cr8);
2991 skip_emulated_instruction(vcpu);
2992 if (irqchip_in_kernel(vcpu->kvm))
2993 return 1;
2994 if (cr8_prev <= cr8)
2995 return 1;
851ba692 2996 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
0a5fff19
GN		 2997 return 0;
2998 }
6aa8b732
AK		 2999 };
3000 break;
25c4c276 3001 case 2: /* clts */
edcafe3c 3002 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4d4ec087 3003 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
25c4c276
AL		 3004 skip_emulated_instruction(vcpu);
3005 return 1;
6aa8b732
AK		 3006 case 1: /*mov from cr*/
3007 switch (cr) {
3008 case 3:
5fdbf976 3009 kvm_register_write(vcpu, reg, vcpu->arch.cr3);
229456fc 3010 trace_kvm_cr_read(cr, vcpu->arch.cr3);
6aa8b732
AK		 3011 skip_emulated_instruction(vcpu);
3012 return 1;
3013 case 8:
229456fc
MT		 3014 val = kvm_get_cr8(vcpu);
3015 kvm_register_write(vcpu, reg, val);
3016 trace_kvm_cr_read(cr, val);
6aa8b732
AK		 3017 skip_emulated_instruction(vcpu);
3018 return 1;
3019 }
3020 break;
3021 case 3: /* lmsw */
a1f83a74 3022 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4d4ec087 3023 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
a1f83a74 3024 kvm_lmsw(vcpu, val);
6aa8b732
AK		 3025
3026 skip_emulated_instruction(vcpu);
3027 return 1;
3028 default:
3029 break;
3030 }
851ba692 3031 vcpu->run->exit_reason = 0;
f0242478 3032 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
6aa8b732
AK		 3033 (int)(exit_qualification >> 4) & 3, cr);
3034 return 0;
3035}
3036
138ac8d8
JK		 3037static int check_dr_alias(struct kvm_vcpu *vcpu)
3038{
3039 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE)) {
3040 kvm_queue_exception(vcpu, UD_VECTOR);
3041 return -1;
3042 }
3043 return 0;
3044}
3045
851ba692 3046static int handle_dr(struct kvm_vcpu *vcpu)
6aa8b732 3047{
bfdaab09 3048 unsigned long exit_qualification;
6aa8b732
AK		 3049 unsigned long val;
3050 int dr, reg;
3051
f2483415 3052 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
0a79b009
AK		 3053 if (!kvm_require_cpl(vcpu, 0))
3054 return 1;
42dbaa5a
JK		 3055 dr = vmcs_readl(GUEST_DR7);
3056 if (dr & DR7_GD) {
3057 /*
3058 * As the vm-exit takes precedence over the debug trap, we
3059 * need to emulate the latter, either for the host or the
3060 * guest debugging itself.
3061 */
3062 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
851ba692
AK		 3063 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
3064 vcpu->run->debug.arch.dr7 = dr;
3065 vcpu->run->debug.arch.pc =
42dbaa5a
JK		 3066 vmcs_readl(GUEST_CS_BASE) +
3067 vmcs_readl(GUEST_RIP);
851ba692
AK		 3068 vcpu->run->debug.arch.exception = DB_VECTOR;
3069 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
42dbaa5a
JK		 3070 return 0;
3071 } else {
3072 vcpu->arch.dr7 &= ~DR7_GD;
3073 vcpu->arch.dr6 |= DR6_BD;
3074 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3075 kvm_queue_exception(vcpu, DB_VECTOR);
3076 return 1;
3077 }
3078 }
3079
bfdaab09 3080 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
42dbaa5a
JK		 3081 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
3082 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
3083 if (exit_qualification & TYPE_MOV_FROM_DR) {
6aa8b732 3084 switch (dr) {
42dbaa5a
JK		 3085 case 0 ... 3:
3086 val = vcpu->arch.db[dr];
3087 break;
138ac8d8
JK		 3088 case 4:
3089 if (check_dr_alias(vcpu) < 0)
3090 return 1;
3091 /* fall through */
6aa8b732 3092 case 6:
42dbaa5a 3093 val = vcpu->arch.dr6;
6aa8b732 3094 break;
138ac8d8
JK		 3095 case 5:
3096 if (check_dr_alias(vcpu) < 0)
3097 return 1;
3098 /* fall through */
3099 default: /* 7 */
42dbaa5a 3100 val = vcpu->arch.dr7;
6aa8b732 3101 break;
6aa8b732 3102 }
5fdbf976 3103 kvm_register_write(vcpu, reg, val);
6aa8b732 3104 } else {
42dbaa5a
JK		 3105 val = vcpu->arch.regs[reg];
3106 switch (dr) {
3107 case 0 ... 3:
3108 vcpu->arch.db[dr] = val;
3109 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
3110 vcpu->arch.eff_db[dr] = val;
3111 break;
138ac8d8
JK		 3112 case 4:
3113 if (check_dr_alias(vcpu) < 0)
f2483415 3114 return 1;
138ac8d8 3115 /* fall through */
42dbaa5a
JK		 3116 case 6:
3117 if (val & 0xffffffff00000000ULL) {
f2483415
JK		 3118 kvm_inject_gp(vcpu, 0);
3119 return 1;
42dbaa5a
JK		 3120 }
3121 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
3122 break;
138ac8d8
JK		 3123 case 5:
3124 if (check_dr_alias(vcpu) < 0)
3125 return 1;
3126 /* fall through */
3127 default: /* 7 */
42dbaa5a 3128 if (val & 0xffffffff00000000ULL) {
f2483415
JK		 3129 kvm_inject_gp(vcpu, 0);
3130 return 1;
42dbaa5a
JK		 3131 }
3132 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
3133 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
3134 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3135 vcpu->arch.switch_db_regs =
3136 (val & DR7_BP_EN_MASK);
3137 }
3138 break;
3139 }
6aa8b732 3140 }
6aa8b732
AK		 3141 skip_emulated_instruction(vcpu);
3142 return 1;
3143}
3144
851ba692 3145static int handle_cpuid(struct kvm_vcpu *vcpu)
6aa8b732 3146{
06465c5a
AK		 3147 kvm_emulate_cpuid(vcpu);
3148 return 1;
6aa8b732
AK		 3149}
3150
851ba692 3151static int handle_rdmsr(struct kvm_vcpu *vcpu)
6aa8b732 3152{
ad312c7c 3153 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6aa8b732
AK		 3154 u64 data;
3155
3156 if (vmx_get_msr(vcpu, ecx, &data)) {
c1a5d4f9 3157 kvm_inject_gp(vcpu, 0);
6aa8b732
AK		 3158 return 1;
3159 }
3160
229456fc 3161 trace_kvm_msr_read(ecx, data);
2714d1d3 3162
6aa8b732 3163 /* FIXME: handling of bits 32:63 of rax, rdx */
ad312c7c
ZX		 3164 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3165 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
6aa8b732
AK		 3166 skip_emulated_instruction(vcpu);
3167 return 1;
3168}
3169
851ba692 3170static int handle_wrmsr(struct kvm_vcpu *vcpu)
6aa8b732 3171{
ad312c7c
ZX		 3172 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3173 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3174 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6aa8b732 3175
229456fc 3176 trace_kvm_msr_write(ecx, data);
2714d1d3 3177
6aa8b732 3178 if (vmx_set_msr(vcpu, ecx, data) != 0) {
c1a5d4f9 3179 kvm_inject_gp(vcpu, 0);
6aa8b732
AK		 3180 return 1;
3181 }
3182
3183 skip_emulated_instruction(vcpu);
3184 return 1;
3185}
3186
851ba692 3187static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6e5d865c
YS		 3188{
3189 return 1;
3190}
3191
851ba692 3192static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6aa8b732 3193{
85f455f7
ED		 3194 u32 cpu_based_vm_exec_control;
3195
3196 /* clear pending irq */
3197 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3198 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3199 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2714d1d3 3200
a26bf12a 3201 ++vcpu->stat.irq_window_exits;
2714d1d3 3202
c1150d8c
DL		 3203 /*
3204 * If the user space waits to inject interrupts, exit as soon as
3205 * possible
3206 */
8061823a 3207 if (!irqchip_in_kernel(vcpu->kvm) &&
851ba692 3208 vcpu->run->request_interrupt_window &&
8061823a 3209 !kvm_cpu_has_interrupt(vcpu)) {
851ba692 3210 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
c1150d8c
DL		 3211 return 0;
3212 }
6aa8b732
AK		 3213 return 1;
3214}
3215
851ba692 3216static int handle_halt(struct kvm_vcpu *vcpu)
6aa8b732
AK		 3217{
3218 skip_emulated_instruction(vcpu);
d3bef15f 3219 return kvm_emulate_halt(vcpu);
6aa8b732
AK		 3220}
3221
851ba692 3222static int handle_vmcall(struct kvm_vcpu *vcpu)
c21415e8 3223{
510043da 3224 skip_emulated_instruction(vcpu);
7aa81cc0
AL		 3225 kvm_emulate_hypercall(vcpu);
3226 return 1;
c21415e8
IM		 3227}
3228
851ba692 3229static int handle_vmx_insn(struct kvm_vcpu *vcpu)
e3c7cb6a
AK		 3230{
3231 kvm_queue_exception(vcpu, UD_VECTOR);
3232 return 1;
3233}
3234
851ba692 3235static int handle_invlpg(struct kvm_vcpu *vcpu)
a7052897 3236{
f9c617f6 3237 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
a7052897
MT		 3238
3239 kvm_mmu_invlpg(vcpu, exit_qualification);
3240 skip_emulated_instruction(vcpu);
3241 return 1;
3242}
3243
851ba692 3244static int handle_wbinvd(struct kvm_vcpu *vcpu)
e5edaa01
ED		 3245{
3246 skip_emulated_instruction(vcpu);
3247 /* TODO: Add support for VT-d/pass-through device */
3248 return 1;
3249}
3250
851ba692 3251static int handle_apic_access(struct kvm_vcpu *vcpu)
f78e0e2e 3252{
f9c617f6 3253 unsigned long exit_qualification;
f78e0e2e
SY		 3254 enum emulation_result er;
3255 unsigned long offset;
3256
f9c617f6 3257 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
f78e0e2e
SY		 3258 offset = exit_qualification & 0xffful;
3259
851ba692 3260 er = emulate_instruction(vcpu, 0, 0, 0);
f78e0e2e
SY		 3261
3262 if (er != EMULATE_DONE) {
3263 printk(KERN_ERR
3264 "Fail to handle apic access vmexit! Offset is 0x%lx\n",
3265 offset);
7f582ab6 3266 return -ENOEXEC;
f78e0e2e
SY		 3267 }
3268 return 1;
3269}
3270
851ba692 3271static int handle_task_switch(struct kvm_vcpu *vcpu)
37817f29 3272{
60637aac 3273 struct vcpu_vmx *vmx = to_vmx(vcpu);
37817f29
IE		 3274 unsigned long exit_qualification;
3275 u16 tss_selector;
64a7ec06
GN		 3276 int reason, type, idt_v;
3277
3278 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3279 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
37817f29
IE		 3280
3281 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3282
3283 reason = (u32)exit_qualification >> 30;
64a7ec06
GN		 3284 if (reason == TASK_SWITCH_GATE && idt_v) {
3285 switch (type) {
3286 case INTR_TYPE_NMI_INTR:
3287 vcpu->arch.nmi_injected = false;
3288 if (cpu_has_virtual_nmis())
3289 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3290 GUEST_INTR_STATE_NMI);
3291 break;
3292 case INTR_TYPE_EXT_INTR:
66fd3f7f 3293 case INTR_TYPE_SOFT_INTR:
64a7ec06
GN		 3294 kvm_clear_interrupt_queue(vcpu);
3295 break;
3296 case INTR_TYPE_HARD_EXCEPTION:
3297 case INTR_TYPE_SOFT_EXCEPTION:
3298 kvm_clear_exception_queue(vcpu);
3299 break;
3300 default:
3301 break;
3302 }
60637aac 3303 }
37817f29
IE		 3304 tss_selector = exit_qualification;
3305
64a7ec06
GN		 3306 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3307 type != INTR_TYPE_EXT_INTR &&
3308 type != INTR_TYPE_NMI_INTR))
3309 skip_emulated_instruction(vcpu);
3310
42dbaa5a
JK		 3311 if (!kvm_task_switch(vcpu, tss_selector, reason))
3312 return 0;
3313
3314 /* clear all local breakpoint enable flags */
3315 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3316
3317 /*
3318 * TODO: What about debug traps on tss switch?
3319 * Are we supposed to inject them and update dr6?
3320 */
3321
3322 return 1;
37817f29
IE		 3323}
3324
851ba692 3325static int handle_ept_violation(struct kvm_vcpu *vcpu)
1439442c 3326{
f9c617f6 3327 unsigned long exit_qualification;
1439442c 3328 gpa_t gpa;
1439442c 3329 int gla_validity;
1439442c 3330
f9c617f6 3331 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
1439442c
SY		 3332
3333 if (exit_qualification & (1 << 6)) {
3334 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
7f582ab6 3335 return -EINVAL;
1439442c
SY		 3336 }
3337
3338 gla_validity = (exit_qualification >> 7) & 0x3;
3339 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3340 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3341 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3342 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
f9c617f6 3343 vmcs_readl(GUEST_LINEAR_ADDRESS));
1439442c
SY		 3344 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3345 (long unsigned int)exit_qualification);
851ba692
AK		 3346 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3347 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
596ae895 3348 return 0;
1439442c
SY		 3349 }
3350
3351 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
229456fc 3352 trace_kvm_page_fault(gpa, exit_qualification);
49cd7d22 3353 return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
1439442c
SY		 3354}
3355
68f89400
MT		 3356static u64 ept_rsvd_mask(u64 spte, int level)
3357{
3358 int i;
3359 u64 mask = 0;
3360
3361 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3362 mask |= (1ULL << i);
3363
3364 if (level > 2)
3365 /* bits 7:3 reserved */
3366 mask |= 0xf8;
3367 else if (level == 2) {
3368 if (spte & (1ULL << 7))
3369 /* 2MB ref, bits 20:12 reserved */
3370 mask |= 0x1ff000;
3371 else
3372 /* bits 6:3 reserved */
3373 mask |= 0x78;
3374 }
3375
3376 return mask;
3377}
3378
3379static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3380 int level)
3381{
3382 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3383
3384 /* 010b (write-only) */
3385 WARN_ON((spte & 0x7) == 0x2);
3386
3387 /* 110b (write/execute) */
3388 WARN_ON((spte & 0x7) == 0x6);
3389
3390 /* 100b (execute-only) and value not supported by logical processor */
3391 if (!cpu_has_vmx_ept_execute_only())
3392 WARN_ON((spte & 0x7) == 0x4);
3393
3394 /* not 000b */
3395 if ((spte & 0x7)) {
3396 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3397
3398 if (rsvd_bits != 0) {
3399 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3400 __func__, rsvd_bits);
3401 WARN_ON(1);
3402 }
3403
3404 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3405 u64 ept_mem_type = (spte & 0x38) >> 3;
3406
3407 if (ept_mem_type == 2 || ept_mem_type == 3 ||
3408 ept_mem_type == 7) {
3409 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3410 __func__, ept_mem_type);
3411 WARN_ON(1);
3412 }
3413 }
3414 }
3415}
3416
851ba692 3417static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
68f89400
MT		 3418{
3419 u64 sptes[4];
3420 int nr_sptes, i;
3421 gpa_t gpa;
3422
3423 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3424
3425 printk(KERN_ERR "EPT: Misconfiguration.\n");
3426 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3427
3428 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3429
3430 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3431 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3432
851ba692
AK		 3433 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3434 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
68f89400
MT		 3435
3436 return 0;
3437}
3438
851ba692 3439static int handle_nmi_window(struct kvm_vcpu *vcpu)
f08864b4
SY		 3440{
3441 u32 cpu_based_vm_exec_control;
3442
3443 /* clear pending NMI */
3444 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3445 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3446 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3447 ++vcpu->stat.nmi_window_exits;
3448
3449 return 1;
3450}
3451
80ced186 3452static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
ea953ef0 3453{
8b3079a5
AK		 3454 struct vcpu_vmx *vmx = to_vmx(vcpu);
3455 enum emulation_result err = EMULATE_DONE;
80ced186 3456 int ret = 1;
ea953ef0
MG		 3457
3458 while (!guest_state_valid(vcpu)) {
851ba692 3459 err = emulate_instruction(vcpu, 0, 0, 0);
ea953ef0 3460
80ced186
MG		 3461 if (err == EMULATE_DO_MMIO) {
3462 ret = 0;
3463 goto out;
3464 }
1d5a4d9b
GT		 3465
3466 if (err != EMULATE_DONE) {
80ced186
MG		 3467 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3468 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
a9c7399d 3469 vcpu->run->internal.ndata = 0;
80ced186
MG		 3470 ret = 0;
3471 goto out;
ea953ef0
MG		 3472 }
3473
3474 if (signal_pending(current))
80ced186 3475 goto out;
ea953ef0
MG		 3476 if (need_resched())
3477 schedule();
3478 }
3479
80ced186
MG		 3480 vmx->emulation_required = 0;
3481out:
3482 return ret;
ea953ef0
MG		 3483}
3484
4b8d54f9
ZE		 3485/*
3486 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3487 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3488 */
9fb41ba8 3489static int handle_pause(struct kvm_vcpu *vcpu)
4b8d54f9
ZE		 3490{
3491 skip_emulated_instruction(vcpu);
3492 kvm_vcpu_on_spin(vcpu);
3493
3494 return 1;
3495}
3496
59708670
SY		 3497static int handle_invalid_op(struct kvm_vcpu *vcpu)
3498{
3499 kvm_queue_exception(vcpu, UD_VECTOR);
3500 return 1;
3501}
3502
6aa8b732
AK		 3503/*
3504 * The exit handlers return 1 if the exit was handled fully and guest execution
3505 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
3506 * to be done to userspace and return 0.
3507 */
851ba692 3508static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
6aa8b732
AK		 3509 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
3510 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
988ad74f 3511 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
f08864b4 3512 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
6aa8b732 3513 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
6aa8b732
AK		 3514 [EXIT_REASON_CR_ACCESS] = handle_cr,
3515 [EXIT_REASON_DR_ACCESS] = handle_dr,
3516 [EXIT_REASON_CPUID] = handle_cpuid,
3517 [EXIT_REASON_MSR_READ] = handle_rdmsr,
3518 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
3519 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
3520 [EXIT_REASON_HLT] = handle_halt,
a7052897 3521 [EXIT_REASON_INVLPG] = handle_invlpg,
c21415e8 3522 [EXIT_REASON_VMCALL] = handle_vmcall,
e3c7cb6a
AK		 3523 [EXIT_REASON_VMCLEAR] = handle_vmx_insn,
3524 [EXIT_REASON_VMLAUNCH] = handle_vmx_insn,
3525 [EXIT_REASON_VMPTRLD] = handle_vmx_insn,
3526 [EXIT_REASON_VMPTRST] = handle_vmx_insn,
3527 [EXIT_REASON_VMREAD] = handle_vmx_insn,
3528 [EXIT_REASON_VMRESUME] = handle_vmx_insn,
3529 [EXIT_REASON_VMWRITE] = handle_vmx_insn,
3530 [EXIT_REASON_VMOFF] = handle_vmx_insn,
3531 [EXIT_REASON_VMON] = handle_vmx_insn,
f78e0e2e
SY		 3532 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
3533 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
e5edaa01 3534 [EXIT_REASON_WBINVD] = handle_wbinvd,
37817f29 3535 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
a0861c02 3536 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
68f89400
MT		 3537 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
3538 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
4b8d54f9 3539 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
59708670
SY		 3540 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_invalid_op,
3541 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_invalid_op,
6aa8b732
AK		 3542};
3543
3544static const int kvm_vmx_max_exit_handlers =
50a3485c 3545 ARRAY_SIZE(kvm_vmx_exit_handlers);
6aa8b732
AK		 3546
3547/*
3548 * The guest has exited. See if we can fix it or if we need userspace
3549 * assistance.
3550 */
851ba692 3551static int vmx_handle_exit(struct kvm_vcpu *vcpu)
6aa8b732 3552{
29bd8a78 3553 struct vcpu_vmx *vmx = to_vmx(vcpu);
a0861c02 3554 u32 exit_reason = vmx->exit_reason;
1155f76a 3555 u32 vectoring_info = vmx->idt_vectoring_info;
29bd8a78 3556
229456fc 3557 trace_kvm_exit(exit_reason, kvm_rip_read(vcpu));
2714d1d3 3558
80ced186
MG		 3559 /* If guest state is invalid, start emulating */
3560 if (vmx->emulation_required && emulate_invalid_guest_state)
3561 return handle_invalid_guest_state(vcpu);
1d5a4d9b 3562
1439442c
SY		 3563 /* Access CR3 don't cause VMExit in paging mode, so we need
3564 * to sync with guest real CR3. */
6de4f3ad 3565 if (enable_ept && is_paging(vcpu))
1439442c 3566 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
1439442c 3567
29bd8a78 3568 if (unlikely(vmx->fail)) {
851ba692
AK		 3569 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3570 vcpu->run->fail_entry.hardware_entry_failure_reason
29bd8a78
AK		 3571 = vmcs_read32(VM_INSTRUCTION_ERROR);
3572 return 0;
3573 }
6aa8b732 3574
d77c26fc 3575 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
1439442c 3576 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
60637aac
JK		 3577 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3578 exit_reason != EXIT_REASON_TASK_SWITCH))
3579 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3580 "(0x%x) and exit reason is 0x%x\n",
3581 __func__, vectoring_info, exit_reason);
3b86cd99
JK		 3582
3583 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
c4282df9 3584 if (vmx_interrupt_allowed(vcpu)) {
3b86cd99 3585 vmx->soft_vnmi_blocked = 0;
3b86cd99 3586 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
4531220b 3587 vcpu->arch.nmi_pending) {
3b86cd99
JK		 3588 /*
3589 * This CPU don't support us in finding the end of an
3590 * NMI-blocked window if the guest runs with IRQs
3591 * disabled. So we pull the trigger after 1 s of
3592 * futile waiting, but inform the user about this.
3593 */
3594 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3595 "state on VCPU %d after 1 s timeout\n",
3596 __func__, vcpu->vcpu_id);
3597 vmx->soft_vnmi_blocked = 0;
3b86cd99 3598 }
3b86cd99
JK		 3599 }
3600
6aa8b732
AK		 3601 if (exit_reason < kvm_vmx_max_exit_handlers
3602 && kvm_vmx_exit_handlers[exit_reason])
851ba692 3603 return kvm_vmx_exit_handlers[exit_reason](vcpu);
6aa8b732 3604 else {
851ba692
AK		 3605 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3606 vcpu->run->hw.hardware_exit_reason = exit_reason;
6aa8b732
AK		 3607 }
3608 return 0;
3609}
3610
95ba8273 3611static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6e5d865c 3612{
95ba8273 3613 if (irr == -1 || tpr < irr) {
6e5d865c
YS		 3614 vmcs_write32(TPR_THRESHOLD, 0);
3615 return;
3616 }
3617
95ba8273 3618 vmcs_write32(TPR_THRESHOLD, irr);
6e5d865c
YS		 3619}
3620
cf393f75
AK		 3621static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3622{
3623 u32 exit_intr_info;
7b4a25cb 3624 u32 idt_vectoring_info = vmx->idt_vectoring_info;
cf393f75
AK		 3625 bool unblock_nmi;
3626 u8 vector;
668f612f
AK		 3627 int type;
3628 bool idtv_info_valid;
cf393f75
AK		 3629
3630 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
20f65983 3631
a0861c02
AK		 3632 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
3633
3634 /* Handle machine checks before interrupts are enabled */
3635 if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3636 || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3637 && is_machine_check(exit_intr_info)))
3638 kvm_machine_check();
3639
20f65983
GN		 3640 /* We need to handle NMIs before interrupts are enabled */
3641 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
229456fc 3642 (exit_intr_info & INTR_INFO_VALID_MASK))
20f65983 3643 asm("int $2");
20f65983
GN		 3644
3645 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3646
cf393f75
AK		 3647 if (cpu_has_virtual_nmis()) {
3648 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3649 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3650 /*
7b4a25cb 3651 * SDM 3: 27.7.1.2 (September 2008)
cf393f75
AK		 3652 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3653 * a guest IRET fault.
7b4a25cb
GN		 3654 * SDM 3: 23.2.2 (September 2008)
3655 * Bit 12 is undefined in any of the following cases:
3656 * If the VM exit sets the valid bit in the IDT-vectoring
3657 * information field.
3658 * If the VM exit is due to a double fault.
cf393f75 3659 */
7b4a25cb
GN		 3660 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3661 vector != DF_VECTOR && !idtv_info_valid)
cf393f75
AK		 3662 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3663 GUEST_INTR_STATE_NMI);
3b86cd99
JK		 3664 } else if (unlikely(vmx->soft_vnmi_blocked))
3665 vmx->vnmi_blocked_time +=
3666 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
668f612f 3667
37b96e98
GN		 3668 vmx->vcpu.arch.nmi_injected = false;
3669 kvm_clear_exception_queue(&vmx->vcpu);
3670 kvm_clear_interrupt_queue(&vmx->vcpu);
3671
3672 if (!idtv_info_valid)
3673 return;
3674
668f612f
AK		 3675 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3676 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
37b96e98 3677
64a7ec06 3678 switch (type) {
37b96e98
GN		 3679 case INTR_TYPE_NMI_INTR:
3680 vmx->vcpu.arch.nmi_injected = true;
668f612f 3681 /*
7b4a25cb 3682 * SDM 3: 27.7.1.2 (September 2008)
37b96e98
GN		 3683 * Clear bit "block by NMI" before VM entry if a NMI
3684 * delivery faulted.
668f612f 3685 */
37b96e98
GN		 3686 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3687 GUEST_INTR_STATE_NMI);
3688 break;
37b96e98 3689 case INTR_TYPE_SOFT_EXCEPTION:
66fd3f7f
GN		 3690 vmx->vcpu.arch.event_exit_inst_len =
3691 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3692 /* fall through */
3693 case INTR_TYPE_HARD_EXCEPTION:
35920a35 3694 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
37b96e98
GN		 3695 u32 err = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3696 kvm_queue_exception_e(&vmx->vcpu, vector, err);
35920a35
AK		 3697 } else
3698 kvm_queue_exception(&vmx->vcpu, vector);
37b96e98 3699 break;
66fd3f7f
GN		 3700 case INTR_TYPE_SOFT_INTR:
3701 vmx->vcpu.arch.event_exit_inst_len =
3702 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3703 /* fall through */
37b96e98 3704 case INTR_TYPE_EXT_INTR:
66fd3f7f
GN		 3705 kvm_queue_interrupt(&vmx->vcpu, vector,
3706 type == INTR_TYPE_SOFT_INTR);
37b96e98
GN		 3707 break;
3708 default:
3709 break;
f7d9238f 3710 }
cf393f75
AK		 3711}
3712
9c8cba37
AK		 3713/*
3714 * Failure to inject an interrupt should give us the information
3715 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
3716 * when fetching the interrupt redirection bitmap in the real-mode
3717 * tss, this doesn't happen. So we do it ourselves.
3718 */
3719static void fixup_rmode_irq(struct vcpu_vmx *vmx)
3720{
3721 vmx->rmode.irq.pending = 0;
5fdbf976 3722 if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
9c8cba37 3723 return;
5fdbf976 3724 kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
9c8cba37
AK		 3725 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
3726 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
3727 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
3728 return;
3729 }
3730 vmx->idt_vectoring_info =
3731 VECTORING_INFO_VALID_MASK
3732 | INTR_TYPE_EXT_INTR
3733 | vmx->rmode.irq.vector;
3734}
3735
c801949d
AK		 3736#ifdef CONFIG_X86_64
3737#define R "r"
3738#define Q "q"
3739#else
3740#define R "e"
3741#define Q "l"
3742#endif
3743
851ba692 3744static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6aa8b732 3745{
a2fa3e9f 3746 struct vcpu_vmx *vmx = to_vmx(vcpu);
e6adf283 3747
3b86cd99
JK		 3748 /* Record the guest's net vcpu time for enforced NMI injections. */
3749 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3750 vmx->entry_time = ktime_get();
3751
80ced186
MG		 3752 /* Don't enter VMX if guest state is invalid, let the exit handler
3753 start emulation until we arrive back to a valid state */
3754 if (vmx->emulation_required && emulate_invalid_guest_state)
a89a8fb9 3755 return;
a89a8fb9 3756
5fdbf976
MT		 3757 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3758 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3759 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3760 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3761
787ff736
GN		 3762 /* When single-stepping over STI and MOV SS, we must clear the
3763 * corresponding interruptibility bits in the guest state. Otherwise
3764 * vmentry fails as it then expects bit 14 (BS) in pending debug
3765 * exceptions being set, but that's not correct for the guest debugging
3766 * case. */
3767 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3768 vmx_set_interrupt_shadow(vcpu, 0);
3769
e6adf283
AK		 3770 /*
3771 * Loading guest fpu may have cleared host cr0.ts
3772 */
3773 vmcs_writel(HOST_CR0, read_cr0());
3774
d77c26fc 3775 asm(
6aa8b732 3776 /* Store host registers */
c801949d
AK		 3777 "push %%"R"dx; push %%"R"bp;"
3778 "push %%"R"cx \n\t"
313dbd49
AK		 3779 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3780 "je 1f \n\t"
3781 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
4ecac3fd 3782 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
313dbd49 3783 "1: \n\t"
d3edefc0
AK		 3784 /* Reload cr2 if changed */
3785 "mov %c[cr2](%0), %%"R"ax \n\t"
3786 "mov %%cr2, %%"R"dx \n\t"
3787 "cmp %%"R"ax, %%"R"dx \n\t"
3788 "je 2f \n\t"
3789 "mov %%"R"ax, %%cr2 \n\t"
3790 "2: \n\t"
6aa8b732 3791 /* Check if vmlaunch of vmresume is needed */
e08aa78a 3792 "cmpl $0, %c[launched](%0) \n\t"
6aa8b732 3793 /* Load guest registers. Don't clobber flags. */
c801949d
AK		 3794 "mov %c[rax](%0), %%"R"ax \n\t"
3795 "mov %c[rbx](%0), %%"R"bx \n\t"
3796 "mov %c[rdx](%0), %%"R"dx \n\t"
3797 "mov %c[rsi](%0), %%"R"si \n\t"
3798 "mov %c[rdi](%0), %%"R"di \n\t"
3799 "mov %c[rbp](%0), %%"R"bp \n\t"
05b3e0c2 3800#ifdef CONFIG_X86_64
e08aa78a
AK		 3801 "mov %c[r8](%0), %%r8 \n\t"
3802 "mov %c[r9](%0), %%r9 \n\t"
3803 "mov %c[r10](%0), %%r10 \n\t"
3804 "mov %c[r11](%0), %%r11 \n\t"
3805 "mov %c[r12](%0), %%r12 \n\t"
3806 "mov %c[r13](%0), %%r13 \n\t"
3807 "mov %c[r14](%0), %%r14 \n\t"
3808 "mov %c[r15](%0), %%r15 \n\t"
6aa8b732 3809#endif
c801949d
AK		 3810 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3811
6aa8b732 3812 /* Enter guest mode */
cd2276a7 3813 "jne .Llaunched \n\t"
4ecac3fd 3814 __ex(ASM_VMX_VMLAUNCH) "\n\t"
cd2276a7 3815 "jmp .Lkvm_vmx_return \n\t"
4ecac3fd 3816 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
cd2276a7 3817 ".Lkvm_vmx_return: "
6aa8b732 3818 /* Save guest registers, load host registers, keep flags */
c801949d
AK		 3819 "xchg %0, (%%"R"sp) \n\t"
3820 "mov %%"R"ax, %c[rax](%0) \n\t"
3821 "mov %%"R"bx, %c[rbx](%0) \n\t"
3822 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3823 "mov %%"R"dx, %c[rdx](%0) \n\t"
3824 "mov %%"R"si, %c[rsi](%0) \n\t"
3825 "mov %%"R"di, %c[rdi](%0) \n\t"
3826 "mov %%"R"bp, %c[rbp](%0) \n\t"
05b3e0c2 3827#ifdef CONFIG_X86_64
e08aa78a
AK		 3828 "mov %%r8, %c[r8](%0) \n\t"
3829 "mov %%r9, %c[r9](%0) \n\t"
3830 "mov %%r10, %c[r10](%0) \n\t"
3831 "mov %%r11, %c[r11](%0) \n\t"
3832 "mov %%r12, %c[r12](%0) \n\t"
3833 "mov %%r13, %c[r13](%0) \n\t"
3834 "mov %%r14, %c[r14](%0) \n\t"
3835 "mov %%r15, %c[r15](%0) \n\t"
6aa8b732 3836#endif
c801949d
AK		 3837 "mov %%cr2, %%"R"ax \n\t"
3838 "mov %%"R"ax, %c[cr2](%0) \n\t"
3839
3840 "pop %%"R"bp; pop %%"R"bp; pop %%"R"dx \n\t"
e08aa78a
AK		 3841 "setbe %c[fail](%0) \n\t"
3842 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
3843 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
3844 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
313dbd49 3845 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
ad312c7c
ZX		 3846 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
3847 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
3848 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
3849 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
3850 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
3851 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
3852 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
05b3e0c2 3853#ifdef CONFIG_X86_64
ad312c7c
ZX		 3854 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
3855 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
3856 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
3857 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
3858 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
3859 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
3860 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
3861 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
6aa8b732 3862#endif
ad312c7c 3863 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
c2036300 3864 : "cc", "memory"
c801949d 3865 , R"bx", R"di", R"si"
c2036300 3866#ifdef CONFIG_X86_64
c2036300
LV		 3867 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
3868#endif
3869 );
6aa8b732 3870
6de4f3ad
AK		 3871 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
3872 | (1 << VCPU_EXREG_PDPTR));
5fdbf976
MT		 3873 vcpu->arch.regs_dirty = 0;
3874
1155f76a 3875 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9c8cba37
AK		 3876 if (vmx->rmode.irq.pending)
3877 fixup_rmode_irq(vmx);
1155f76a 3878
d77c26fc 3879 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
15ad7146 3880 vmx->launched = 1;
1b6269db 3881
cf393f75 3882 vmx_complete_interrupts(vmx);
6aa8b732
AK		 3883}
3884
c801949d
AK		 3885#undef R
3886#undef Q
3887
6aa8b732
AK		 3888static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
3889{
a2fa3e9f
GH		 3890 struct vcpu_vmx *vmx = to_vmx(vcpu);
3891
3892 if (vmx->vmcs) {
543e4243 3893 vcpu_clear(vmx);
a2fa3e9f
GH		 3894 free_vmcs(vmx->vmcs);
3895 vmx->vmcs = NULL;
6aa8b732
AK		 3896 }
3897}
3898
3899static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
3900{
fb3f0f51
RR		 3901 struct vcpu_vmx *vmx = to_vmx(vcpu);
3902
2384d2b3
SY		 3903 spin_lock(&vmx_vpid_lock);
3904 if (vmx->vpid != 0)
3905 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3906 spin_unlock(&vmx_vpid_lock);
6aa8b732 3907 vmx_free_vmcs(vcpu);
fb3f0f51
RR		 3908 kfree(vmx->guest_msrs);
3909 kvm_vcpu_uninit(vcpu);
a4770347 3910 kmem_cache_free(kvm_vcpu_cache, vmx);
6aa8b732
AK		 3911}
3912
fb3f0f51 3913static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6aa8b732 3914{
fb3f0f51 3915 int err;
c16f862d 3916 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
15ad7146 3917 int cpu;
6aa8b732 3918
a2fa3e9f 3919 if (!vmx)
fb3f0f51
RR		 3920 return ERR_PTR(-ENOMEM);
3921
2384d2b3
SY		 3922 allocate_vpid(vmx);
3923
fb3f0f51
RR		 3924 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
3925 if (err)
3926 goto free_vcpu;
965b58a5 3927
a2fa3e9f 3928 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
fb3f0f51
RR		 3929 if (!vmx->guest_msrs) {
3930 err = -ENOMEM;
3931 goto uninit_vcpu;
3932 }
965b58a5 3933
a2fa3e9f
GH		 3934 vmx->vmcs = alloc_vmcs();
3935 if (!vmx->vmcs)
fb3f0f51 3936 goto free_msrs;
a2fa3e9f
GH		 3937
3938 vmcs_clear(vmx->vmcs);
3939
15ad7146
AK		 3940 cpu = get_cpu();
3941 vmx_vcpu_load(&vmx->vcpu, cpu);
8b9cf98c 3942 err = vmx_vcpu_setup(vmx);
fb3f0f51 3943 vmx_vcpu_put(&vmx->vcpu);
15ad7146 3944 put_cpu();
fb3f0f51
RR		 3945 if (err)
3946 goto free_vmcs;
5e4a0b3c
MT		 3947 if (vm_need_virtualize_apic_accesses(kvm))
3948 if (alloc_apic_access_page(kvm) != 0)
3949 goto free_vmcs;
fb3f0f51 3950
b927a3ce
SY		 3951 if (enable_ept) {
3952 if (!kvm->arch.ept_identity_map_addr)
3953 kvm->arch.ept_identity_map_addr =
3954 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
b7ebfb05
SY		 3955 if (alloc_identity_pagetable(kvm) != 0)
3956 goto free_vmcs;
b927a3ce 3957 }
b7ebfb05 3958
fb3f0f51
RR		 3959 return &vmx->vcpu;
3960
3961free_vmcs:
3962 free_vmcs(vmx->vmcs);
3963free_msrs:
fb3f0f51
RR		 3964 kfree(vmx->guest_msrs);
3965uninit_vcpu:
3966 kvm_vcpu_uninit(&vmx->vcpu);
3967free_vcpu:
a4770347 3968 kmem_cache_free(kvm_vcpu_cache, vmx);
fb3f0f51 3969 return ERR_PTR(err);
6aa8b732
AK		 3970}
3971
002c7f7c
YS		 3972static void __init vmx_check_processor_compat(void *rtn)
3973{
3974 struct vmcs_config vmcs_conf;
3975
3976 *(int *)rtn = 0;
3977 if (setup_vmcs_config(&vmcs_conf) < 0)
3978 *(int *)rtn = -EIO;
3979 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
3980 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
3981 smp_processor_id());
3982 *(int *)rtn = -EIO;
3983 }
3984}
3985
67253af5
SY		 3986static int get_ept_level(void)
3987{
3988 return VMX_EPT_DEFAULT_GAW + 1;
3989}
3990
4b12f0de 3991static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
64d4d521 3992{
4b12f0de
SY		 3993 u64 ret;
3994
522c68c4
SY		 3995 /* For VT-d and EPT combination
3996 * 1. MMIO: always map as UC
3997 * 2. EPT with VT-d:
3998 * a. VT-d without snooping control feature: can't guarantee the
3999 * result, try to trust guest.
4000 * b. VT-d with snooping control feature: snooping control feature of
4001 * VT-d engine can guarantee the cache correctness. Just set it
4002 * to WB to keep consistent with host. So the same as item 3.
4003 * 3. EPT without VT-d: always map as WB and set IGMT=1 to keep
4004 * consistent with host MTRR
4005 */
4b12f0de
SY		 4006 if (is_mmio)
4007 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
522c68c4
SY		 4008 else if (vcpu->kvm->arch.iommu_domain &&
4009 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
4010 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
4011 VMX_EPT_MT_EPTE_SHIFT;
4b12f0de 4012 else
522c68c4
SY		 4013 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
4014 | VMX_EPT_IGMT_BIT;
4b12f0de
SY		 4015
4016 return ret;
64d4d521
SY		 4017}
4018
f4c9e87c
AK		 4019#define _ER(x) { EXIT_REASON_##x, #x }
4020
229456fc 4021static const struct trace_print_flags vmx_exit_reasons_str[] = {
f4c9e87c
AK		 4022 _ER(EXCEPTION_NMI),
4023 _ER(EXTERNAL_INTERRUPT),
4024 _ER(TRIPLE_FAULT),
4025 _ER(PENDING_INTERRUPT),
4026 _ER(NMI_WINDOW),
4027 _ER(TASK_SWITCH),
4028 _ER(CPUID),
4029 _ER(HLT),
4030 _ER(INVLPG),
4031 _ER(RDPMC),
4032 _ER(RDTSC),
4033 _ER(VMCALL),
4034 _ER(VMCLEAR),
4035 _ER(VMLAUNCH),
4036 _ER(VMPTRLD),
4037 _ER(VMPTRST),
4038 _ER(VMREAD),
4039 _ER(VMRESUME),
4040 _ER(VMWRITE),
4041 _ER(VMOFF),
4042 _ER(VMON),
4043 _ER(CR_ACCESS),
4044 _ER(DR_ACCESS),
4045 _ER(IO_INSTRUCTION),
4046 _ER(MSR_READ),
4047 _ER(MSR_WRITE),
4048 _ER(MWAIT_INSTRUCTION),
4049 _ER(MONITOR_INSTRUCTION),
4050 _ER(PAUSE_INSTRUCTION),
4051 _ER(MCE_DURING_VMENTRY),
4052 _ER(TPR_BELOW_THRESHOLD),
4053 _ER(APIC_ACCESS),
4054 _ER(EPT_VIOLATION),
4055 _ER(EPT_MISCONFIG),
4056 _ER(WBINVD),
229456fc
MT		 4057 { -1, NULL }
4058};
4059
f4c9e87c
AK		 4060#undef _ER
4061
17cc3935 4062static int vmx_get_lpage_level(void)
344f414f 4063{
878403b7
SY		 4064 if (enable_ept && !cpu_has_vmx_ept_1g_page())
4065 return PT_DIRECTORY_LEVEL;
4066 else
4067 /* For shadow and EPT supported 1GB page */
4068 return PT_PDPE_LEVEL;
344f414f
JR		 4069}
4070
4e47c7a6
SY		 4071static inline u32 bit(int bitno)
4072{
4073 return 1 << (bitno & 31);
4074}
4075
0e851880
SY		 4076static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
4077{
4e47c7a6
SY		 4078 struct kvm_cpuid_entry2 *best;
4079 struct vcpu_vmx *vmx = to_vmx(vcpu);
4080 u32 exec_control;
4081
4082 vmx->rdtscp_enabled = false;
4083 if (vmx_rdtscp_supported()) {
4084 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
4085 if (exec_control & SECONDARY_EXEC_RDTSCP) {
4086 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
4087 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
4088 vmx->rdtscp_enabled = true;
4089 else {
4090 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4091 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4092 exec_control);
4093 }
4094 }
4095 }
0e851880
SY		 4096}
4097
cbdd1bea 4098static struct kvm_x86_ops vmx_x86_ops = {
6aa8b732
AK		 4099 .cpu_has_kvm_support = cpu_has_kvm_support,
4100 .disabled_by_bios = vmx_disabled_by_bios,
4101 .hardware_setup = hardware_setup,
4102 .hardware_unsetup = hardware_unsetup,
002c7f7c 4103 .check_processor_compatibility = vmx_check_processor_compat,
6aa8b732
AK		 4104 .hardware_enable = hardware_enable,
4105 .hardware_disable = hardware_disable,
04547156 4106 .cpu_has_accelerated_tpr = report_flexpriority,
6aa8b732
AK		 4107
4108 .vcpu_create = vmx_create_vcpu,
4109 .vcpu_free = vmx_free_vcpu,
04d2cc77 4110 .vcpu_reset = vmx_vcpu_reset,
6aa8b732 4111
04d2cc77 4112 .prepare_guest_switch = vmx_save_host_state,
6aa8b732
AK		 4113 .vcpu_load = vmx_vcpu_load,
4114 .vcpu_put = vmx_vcpu_put,
4115
4116 .set_guest_debug = set_guest_debug,
4117 .get_msr = vmx_get_msr,
4118 .set_msr = vmx_set_msr,
4119 .get_segment_base = vmx_get_segment_base,
4120 .get_segment = vmx_get_segment,
4121 .set_segment = vmx_set_segment,
2e4d2653 4122 .get_cpl = vmx_get_cpl,
6aa8b732 4123 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
e8467fda 4124 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
25c4c276 4125 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
6aa8b732 4126 .set_cr0 = vmx_set_cr0,
6aa8b732
AK		 4127 .set_cr3 = vmx_set_cr3,
4128 .set_cr4 = vmx_set_cr4,
6aa8b732 4129 .set_efer = vmx_set_efer,
6aa8b732
AK		 4130 .get_idt = vmx_get_idt,
4131 .set_idt = vmx_set_idt,
4132 .get_gdt = vmx_get_gdt,
4133 .set_gdt = vmx_set_gdt,
5fdbf976 4134 .cache_reg = vmx_cache_reg,
6aa8b732
AK		 4135 .get_rflags = vmx_get_rflags,
4136 .set_rflags = vmx_set_rflags,
02daab21 4137 .fpu_deactivate = vmx_fpu_deactivate,
6aa8b732
AK		 4138
4139 .tlb_flush = vmx_flush_tlb,
6aa8b732 4140
6aa8b732 4141 .run = vmx_vcpu_run,
6062d012 4142 .handle_exit = vmx_handle_exit,
6aa8b732 4143 .skip_emulated_instruction = skip_emulated_instruction,
2809f5d2
GC		 4144 .set_interrupt_shadow = vmx_set_interrupt_shadow,
4145 .get_interrupt_shadow = vmx_get_interrupt_shadow,
102d8325 4146 .patch_hypercall = vmx_patch_hypercall,
2a8067f1 4147 .set_irq = vmx_inject_irq,
95ba8273 4148 .set_nmi = vmx_inject_nmi,
298101da 4149 .queue_exception = vmx_queue_exception,
78646121 4150 .interrupt_allowed = vmx_interrupt_allowed,
95ba8273 4151 .nmi_allowed = vmx_nmi_allowed,
3cfc3092
JK		 4152 .get_nmi_mask = vmx_get_nmi_mask,
4153 .set_nmi_mask = vmx_set_nmi_mask,
95ba8273
GN		 4154 .enable_nmi_window = enable_nmi_window,
4155 .enable_irq_window = enable_irq_window,
4156 .update_cr8_intercept = update_cr8_intercept,
95ba8273 4157
cbc94022 4158 .set_tss_addr = vmx_set_tss_addr,
67253af5 4159 .get_tdp_level = get_ept_level,
4b12f0de 4160 .get_mt_mask = vmx_get_mt_mask,
229456fc
MT		 4161
4162 .exit_reasons_str = vmx_exit_reasons_str,
17cc3935 4163 .get_lpage_level = vmx_get_lpage_level,
0e851880
SY		 4164
4165 .cpuid_update = vmx_cpuid_update,
4e47c7a6
SY		 4166
4167 .rdtscp_supported = vmx_rdtscp_supported,
6aa8b732
AK		 4168};
4169
4170static int __init vmx_init(void)
4171{
26bb0981
AK		 4172 int r, i;
4173
4174 rdmsrl_safe(MSR_EFER, &host_efer);
4175
4176 for (i = 0; i < NR_VMX_MSR; ++i)
4177 kvm_define_shared_msr(i, vmx_msr_index[i]);
fdef3ad1 4178
3e7c73e9 4179 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
fdef3ad1
HQ		 4180 if (!vmx_io_bitmap_a)
4181 return -ENOMEM;
4182
3e7c73e9 4183 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
fdef3ad1
HQ		 4184 if (!vmx_io_bitmap_b) {
4185 r = -ENOMEM;
4186 goto out;
4187 }
4188
5897297b
AK		 4189 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
4190 if (!vmx_msr_bitmap_legacy) {
25c5f225
SY		 4191 r = -ENOMEM;
4192 goto out1;
4193 }
4194
5897297b
AK		 4195 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
4196 if (!vmx_msr_bitmap_longmode) {
4197 r = -ENOMEM;
4198 goto out2;
4199 }
4200
fdef3ad1
HQ		 4201 /*
4202 * Allow direct access to the PC debug port (it is often used for I/O
4203 * delays, but the vmexits simply slow things down).
4204 */
3e7c73e9
AK		 4205 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4206 clear_bit(0x80, vmx_io_bitmap_a);
fdef3ad1 4207
3e7c73e9 4208 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
fdef3ad1 4209
5897297b
AK		 4210 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4211 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
25c5f225 4212
2384d2b3
SY		 4213 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4214
cb498ea2 4215 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
fdef3ad1 4216 if (r)
5897297b 4217 goto out3;
25c5f225 4218
5897297b
AK		 4219 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4220 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4221 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4222 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4223 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4224 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
fdef3ad1 4225
089d034e 4226 if (enable_ept) {
1439442c 4227 bypass_guest_pf = 0;
5fdbcb9d 4228 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
2aaf69dc 4229 VMX_EPT_WRITABLE_MASK);
534e38b4 4230 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4b12f0de 4231 VMX_EPT_EXECUTABLE_MASK);
5fdbcb9d
SY		 4232 kvm_enable_tdp();
4233 } else
4234 kvm_disable_tdp();
1439442c 4235
c7addb90
AK		 4236 if (bypass_guest_pf)
4237 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4238
fdef3ad1
HQ		 4239 return 0;
4240
5897297b
AK		 4241out3:
4242 free_page((unsigned long)vmx_msr_bitmap_longmode);
25c5f225 4243out2:
5897297b 4244 free_page((unsigned long)vmx_msr_bitmap_legacy);
fdef3ad1 4245out1:
3e7c73e9 4246 free_page((unsigned long)vmx_io_bitmap_b);
fdef3ad1 4247out:
3e7c73e9 4248 free_page((unsigned long)vmx_io_bitmap_a);
fdef3ad1 4249 return r;
6aa8b732
AK		 4250}
4251
4252static void __exit vmx_exit(void)
4253{
5897297b
AK		 4254 free_page((unsigned long)vmx_msr_bitmap_legacy);
4255 free_page((unsigned long)vmx_msr_bitmap_longmode);
3e7c73e9
AK		 4256 free_page((unsigned long)vmx_io_bitmap_b);
4257 free_page((unsigned long)vmx_io_bitmap_a);
fdef3ad1 4258
cb498ea2 4259 kvm_exit();
6aa8b732
AK		 4260}
4261
4262module_init(vmx_init)
4263module_exit(vmx_exit)
kernel source for telekom puls (alcatel/mediatek)