projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: VMX: Avoid duplicate ept tlb flush when setting cr3
[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>
5fdbf976 28#include "kvm_cache_regs.h"
35920a35 29#include "x86.h"
e495606d 30
6aa8b732 31#include <asm/io.h>
3b3be0d1 32#include <asm/desc.h>
13673a90 33#include <asm/vmx.h>
6210e37b 34#include <asm/virtext.h>
a0861c02 35#include <asm/mce.h>
6aa8b732 36
4ecac3fd
AK		 37#define __ex(x) __kvm_handle_fault_on_reboot(x)
38
6aa8b732
AK		 39MODULE_AUTHOR("Qumranet");
40MODULE_LICENSE("GPL");
41
4462d21a 42static int __read_mostly bypass_guest_pf = 1;
c1f8bc04 43module_param(bypass_guest_pf, bool, S_IRUGO);
c7addb90 44
4462d21a 45static int __read_mostly enable_vpid = 1;
736caefe 46module_param_named(vpid, enable_vpid, bool, 0444);
2384d2b3 47
4462d21a 48static int __read_mostly flexpriority_enabled = 1;
736caefe 49module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
4c9fc8ef 50
4462d21a 51static int __read_mostly enable_ept = 1;
736caefe 52module_param_named(ept, enable_ept, bool, S_IRUGO);
d56f546d 53
4462d21a 54static int __read_mostly emulate_invalid_guest_state = 0;
c1f8bc04 55module_param(emulate_invalid_guest_state, bool, S_IRUGO);
04fa4d32 56
a2fa3e9f
GH		 57struct vmcs {
58 u32 revision_id;
59 u32 abort;
60 char data[0];
61};
62
63struct vcpu_vmx {
fb3f0f51 64 struct kvm_vcpu vcpu;
543e4243 65 struct list_head local_vcpus_link;
313dbd49 66 unsigned long host_rsp;
a2fa3e9f 67 int launched;
29bd8a78 68 u8 fail;
1155f76a 69 u32 idt_vectoring_info;
a2fa3e9f
GH		 70 struct kvm_msr_entry *guest_msrs;
71 struct kvm_msr_entry *host_msrs;
72 int nmsrs;
73 int save_nmsrs;
74 int msr_offset_efer;
75#ifdef CONFIG_X86_64
76 int msr_offset_kernel_gs_base;
77#endif
78 struct vmcs *vmcs;
79 struct {
80 int loaded;
81 u16 fs_sel, gs_sel, ldt_sel;
152d3f2f
LV		 82 int gs_ldt_reload_needed;
83 int fs_reload_needed;
51c6cf66 84 int guest_efer_loaded;
d77c26fc 85 } host_state;
9c8cba37
AK		 86 struct {
87 struct {
88 bool pending;
89 u8 vector;
90 unsigned rip;
91 } irq;
92 } rmode;
2384d2b3 93 int vpid;
04fa4d32 94 bool emulation_required;
8b3079a5 95 enum emulation_result invalid_state_emulation_result;
3b86cd99
JK		 96
97 /* Support for vnmi-less CPUs */
98 int soft_vnmi_blocked;
99 ktime_t entry_time;
100 s64 vnmi_blocked_time;
a0861c02 101 u32 exit_reason;
a2fa3e9f
GH		 102};
103
104static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
105{
fb3f0f51 106 return container_of(vcpu, struct vcpu_vmx, vcpu);
a2fa3e9f
GH		 107}
108
b7ebfb05 109static int init_rmode(struct kvm *kvm);
4e1096d2 110static u64 construct_eptp(unsigned long root_hpa);
75880a01 111
6aa8b732
AK		 112static DEFINE_PER_CPU(struct vmcs *, vmxarea);
113static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
543e4243 114static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
6aa8b732 115
3e7c73e9
AK		 116static unsigned long *vmx_io_bitmap_a;
117static unsigned long *vmx_io_bitmap_b;
5897297b
AK		 118static unsigned long *vmx_msr_bitmap_legacy;
119static unsigned long *vmx_msr_bitmap_longmode;
fdef3ad1 120
2384d2b3
SY		 121static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
122static DEFINE_SPINLOCK(vmx_vpid_lock);
123
1c3d14fe 124static struct vmcs_config {
6aa8b732
AK		 125 int size;
126 int order;
127 u32 revision_id;
1c3d14fe
YS		 128 u32 pin_based_exec_ctrl;
129 u32 cpu_based_exec_ctrl;
f78e0e2e 130 u32 cpu_based_2nd_exec_ctrl;
1c3d14fe
YS		 131 u32 vmexit_ctrl;
132 u32 vmentry_ctrl;
133} vmcs_config;
6aa8b732 134
efff9e53 135static struct vmx_capability {
d56f546d
SY		 136 u32 ept;
137 u32 vpid;
138} vmx_capability;
139
6aa8b732
AK		 140#define VMX_SEGMENT_FIELD(seg) \
141 [VCPU_SREG_##seg] = { \
142 .selector = GUEST_##seg##_SELECTOR, \
143 .base = GUEST_##seg##_BASE, \
144 .limit = GUEST_##seg##_LIMIT, \
145 .ar_bytes = GUEST_##seg##_AR_BYTES, \
146 }
147
148static struct kvm_vmx_segment_field {
149 unsigned selector;
150 unsigned base;
151 unsigned limit;
152 unsigned ar_bytes;
153} kvm_vmx_segment_fields[] = {
154 VMX_SEGMENT_FIELD(CS),
155 VMX_SEGMENT_FIELD(DS),
156 VMX_SEGMENT_FIELD(ES),
157 VMX_SEGMENT_FIELD(FS),
158 VMX_SEGMENT_FIELD(GS),
159 VMX_SEGMENT_FIELD(SS),
160 VMX_SEGMENT_FIELD(TR),
161 VMX_SEGMENT_FIELD(LDTR),
162};
163
4d56c8a7
AK		 164/*
165 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
166 * away by decrementing the array size.
167 */
6aa8b732 168static const u32 vmx_msr_index[] = {
05b3e0c2 169#ifdef CONFIG_X86_64
6aa8b732
AK		 170 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
171#endif
172 MSR_EFER, MSR_K6_STAR,
173};
9d8f549d 174#define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
6aa8b732 175
a2fa3e9f
GH		 176static void load_msrs(struct kvm_msr_entry *e, int n)
177{
178 int i;
179
180 for (i = 0; i < n; ++i)
181 wrmsrl(e[i].index, e[i].data);
182}
183
184static void save_msrs(struct kvm_msr_entry *e, int n)
185{
186 int i;
187
188 for (i = 0; i < n; ++i)
189 rdmsrl(e[i].index, e[i].data);
190}
191
6aa8b732
AK		 192static inline int is_page_fault(u32 intr_info)
193{
194 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
195 INTR_INFO_VALID_MASK)) ==
8ab2d2e2 196 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
6aa8b732
AK		 197}
198
2ab455cc
AL		 199static inline int is_no_device(u32 intr_info)
200{
201 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
202 INTR_INFO_VALID_MASK)) ==
8ab2d2e2 203 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
2ab455cc
AL		 204}
205
7aa81cc0
AL		 206static inline int is_invalid_opcode(u32 intr_info)
207{
208 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
209 INTR_INFO_VALID_MASK)) ==
8ab2d2e2 210 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
7aa81cc0
AL		 211}
212
6aa8b732
AK		 213static inline int is_external_interrupt(u32 intr_info)
214{
215 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
216 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
217}
218
a0861c02
AK		 219static inline int is_machine_check(u32 intr_info)
220{
221 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
222 INTR_INFO_VALID_MASK)) ==
223 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
224}
225
25c5f225
SY		 226static inline int cpu_has_vmx_msr_bitmap(void)
227{
04547156 228 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
25c5f225
SY		 229}
230
6e5d865c
YS		 231static inline int cpu_has_vmx_tpr_shadow(void)
232{
04547156 233 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
6e5d865c
YS		 234}
235
236static inline int vm_need_tpr_shadow(struct kvm *kvm)
237{
04547156 238 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
6e5d865c
YS		 239}
240
f78e0e2e
SY		 241static inline int cpu_has_secondary_exec_ctrls(void)
242{
04547156
SY		 243 return vmcs_config.cpu_based_exec_ctrl &
244 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
f78e0e2e
SY		 245}
246
774ead3a 247static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
f78e0e2e 248{
04547156
SY		 249 return vmcs_config.cpu_based_2nd_exec_ctrl &
250 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
251}
252
253static inline bool cpu_has_vmx_flexpriority(void)
254{
255 return cpu_has_vmx_tpr_shadow() &&
256 cpu_has_vmx_virtualize_apic_accesses();
f78e0e2e
SY		 257}
258
d56f546d
SY		 259static inline int cpu_has_vmx_invept_individual_addr(void)
260{
04547156 261 return !!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT);
d56f546d
SY		 262}
263
264static inline int cpu_has_vmx_invept_context(void)
265{
04547156 266 return !!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT);
d56f546d
SY		 267}
268
269static inline int cpu_has_vmx_invept_global(void)
270{
04547156 271 return !!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT);
d56f546d
SY		 272}
273
274static inline int cpu_has_vmx_ept(void)
275{
04547156
SY		 276 return vmcs_config.cpu_based_2nd_exec_ctrl &
277 SECONDARY_EXEC_ENABLE_EPT;
d56f546d
SY		 278}
279
f78e0e2e
SY		 280static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
281{
04547156
SY		 282 return flexpriority_enabled &&
283 (cpu_has_vmx_virtualize_apic_accesses()) &&
284 (irqchip_in_kernel(kvm));
f78e0e2e
SY		 285}
286
2384d2b3
SY		 287static inline int cpu_has_vmx_vpid(void)
288{
04547156
SY		 289 return vmcs_config.cpu_based_2nd_exec_ctrl &
290 SECONDARY_EXEC_ENABLE_VPID;
2384d2b3
SY		 291}
292
f08864b4
SY		 293static inline int cpu_has_virtual_nmis(void)
294{
295 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
296}
297
04547156
SY		 298static inline bool report_flexpriority(void)
299{
300 return flexpriority_enabled;
301}
302
8b9cf98c 303static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
7725f0ba
AK		 304{
305 int i;
306
a2fa3e9f
GH		 307 for (i = 0; i < vmx->nmsrs; ++i)
308 if (vmx->guest_msrs[i].index == msr)
a75beee6
ED		 309 return i;
310 return -1;
311}
312
2384d2b3
SY		 313static inline void __invvpid(int ext, u16 vpid, gva_t gva)
314{
315 struct {
316 u64 vpid : 16;
317 u64 rsvd : 48;
318 u64 gva;
319 } operand = { vpid, 0, gva };
320
4ecac3fd 321 asm volatile (__ex(ASM_VMX_INVVPID)
2384d2b3
SY		 322 /* CF==1 or ZF==1 --> rc = -1 */
323 "; ja 1f ; ud2 ; 1:"
324 : : "a"(&operand), "c"(ext) : "cc", "memory");
325}
326
1439442c
SY		 327static inline void __invept(int ext, u64 eptp, gpa_t gpa)
328{
329 struct {
330 u64 eptp, gpa;
331 } operand = {eptp, gpa};
332
4ecac3fd 333 asm volatile (__ex(ASM_VMX_INVEPT)
1439442c
SY		 334 /* CF==1 or ZF==1 --> rc = -1 */
335 "; ja 1f ; ud2 ; 1:\n"
336 : : "a" (&operand), "c" (ext) : "cc", "memory");
337}
338
8b9cf98c 339static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
a75beee6
ED		 340{
341 int i;
342
8b9cf98c 343 i = __find_msr_index(vmx, msr);
a75beee6 344 if (i >= 0)
a2fa3e9f 345 return &vmx->guest_msrs[i];
8b6d44c7 346 return NULL;
7725f0ba
AK		 347}
348
6aa8b732
AK		 349static void vmcs_clear(struct vmcs *vmcs)
350{
351 u64 phys_addr = __pa(vmcs);
352 u8 error;
353
4ecac3fd 354 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
6aa8b732
AK		 355 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
356 : "cc", "memory");
357 if (error)
358 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
359 vmcs, phys_addr);
360}
361
362static void __vcpu_clear(void *arg)
363{
8b9cf98c 364 struct vcpu_vmx *vmx = arg;
d3b2c338 365 int cpu = raw_smp_processor_id();
6aa8b732 366
8b9cf98c 367 if (vmx->vcpu.cpu == cpu)
a2fa3e9f
GH		 368 vmcs_clear(vmx->vmcs);
369 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
6aa8b732 370 per_cpu(current_vmcs, cpu) = NULL;
ad312c7c 371 rdtscll(vmx->vcpu.arch.host_tsc);
543e4243
AK		 372 list_del(&vmx->local_vcpus_link);
373 vmx->vcpu.cpu = -1;
374 vmx->launched = 0;
6aa8b732
AK		 375}
376
8b9cf98c 377static void vcpu_clear(struct vcpu_vmx *vmx)
8d0be2b3 378{
eae5ecb5
AK		 379 if (vmx->vcpu.cpu == -1)
380 return;
8691e5a8 381 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
8d0be2b3
AK		 382}
383
2384d2b3
SY		 384static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
385{
386 if (vmx->vpid == 0)
387 return;
388
389 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
390}
391
1439442c
SY		 392static inline void ept_sync_global(void)
393{
394 if (cpu_has_vmx_invept_global())
395 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
396}
397
398static inline void ept_sync_context(u64 eptp)
399{
089d034e 400 if (enable_ept) {
1439442c
SY		 401 if (cpu_has_vmx_invept_context())
402 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
403 else
404 ept_sync_global();
405 }
406}
407
408static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
409{
089d034e 410 if (enable_ept) {
1439442c
SY		 411 if (cpu_has_vmx_invept_individual_addr())
412 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
413 eptp, gpa);
414 else
415 ept_sync_context(eptp);
416 }
417}
418
6aa8b732
AK		 419static unsigned long vmcs_readl(unsigned long field)
420{
421 unsigned long value;
422
4ecac3fd 423 asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
6aa8b732
AK		 424 : "=a"(value) : "d"(field) : "cc");
425 return value;
426}
427
428static u16 vmcs_read16(unsigned long field)
429{
430 return vmcs_readl(field);
431}
432
433static u32 vmcs_read32(unsigned long field)
434{
435 return vmcs_readl(field);
436}
437
438static u64 vmcs_read64(unsigned long field)
439{
05b3e0c2 440#ifdef CONFIG_X86_64
6aa8b732
AK		 441 return vmcs_readl(field);
442#else
443 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
444#endif
445}
446
e52de1b8
AK		 447static noinline void vmwrite_error(unsigned long field, unsigned long value)
448{
449 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
450 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
451 dump_stack();
452}
453
6aa8b732
AK		 454static void vmcs_writel(unsigned long field, unsigned long value)
455{
456 u8 error;
457
4ecac3fd 458 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
d77c26fc 459 : "=q"(error) : "a"(value), "d"(field) : "cc");
e52de1b8
AK		 460 if (unlikely(error))
461 vmwrite_error(field, value);
6aa8b732
AK		 462}
463
464static void vmcs_write16(unsigned long field, u16 value)
465{
466 vmcs_writel(field, value);
467}
468
469static void vmcs_write32(unsigned long field, u32 value)
470{
471 vmcs_writel(field, value);
472}
473
474static void vmcs_write64(unsigned long field, u64 value)
475{
6aa8b732 476 vmcs_writel(field, value);
7682f2d0 477#ifndef CONFIG_X86_64
6aa8b732
AK		 478 asm volatile ("");
479 vmcs_writel(field+1, value >> 32);
480#endif
481}
482
2ab455cc
AL		 483static void vmcs_clear_bits(unsigned long field, u32 mask)
484{
485 vmcs_writel(field, vmcs_readl(field) & ~mask);
486}
487
488static void vmcs_set_bits(unsigned long field, u32 mask)
489{
490 vmcs_writel(field, vmcs_readl(field) | mask);
491}
492
abd3f2d6
AK		 493static void update_exception_bitmap(struct kvm_vcpu *vcpu)
494{
495 u32 eb;
496
a0861c02 497 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR);
abd3f2d6
AK		 498 if (!vcpu->fpu_active)
499 eb |= 1u << NM_VECTOR;
d0bfb940
JK		 500 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
501 if (vcpu->guest_debug &
502 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
503 eb |= 1u << DB_VECTOR;
504 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
505 eb |= 1u << BP_VECTOR;
506 }
56b237e3 507 if (vcpu->arch.rmode.vm86_active)
abd3f2d6 508 eb = ~0;
089d034e 509 if (enable_ept)
1439442c 510 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
abd3f2d6
AK		 511 vmcs_write32(EXCEPTION_BITMAP, eb);
512}
513
33ed6329
AK		 514static void reload_tss(void)
515{
33ed6329
AK		 516 /*
517 * VT restores TR but not its size. Useless.
518 */
519 struct descriptor_table gdt;
a5f61300 520 struct desc_struct *descs;
33ed6329 521
d6e88aec 522 kvm_get_gdt(&gdt);
33ed6329
AK		 523 descs = (void *)gdt.base;
524 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
525 load_TR_desc();
33ed6329
AK		 526}
527
8b9cf98c 528static void load_transition_efer(struct vcpu_vmx *vmx)
2cc51560 529{
a2fa3e9f 530 int efer_offset = vmx->msr_offset_efer;
51c6cf66
AK		 531 u64 host_efer = vmx->host_msrs[efer_offset].data;
532 u64 guest_efer = vmx->guest_msrs[efer_offset].data;
533 u64 ignore_bits;
534
535 if (efer_offset < 0)
536 return;
537 /*
538 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
539 * outside long mode
540 */
541 ignore_bits = EFER_NX | EFER_SCE;
542#ifdef CONFIG_X86_64
543 ignore_bits |= EFER_LMA | EFER_LME;
544 /* SCE is meaningful only in long mode on Intel */
545 if (guest_efer & EFER_LMA)
546 ignore_bits &= ~(u64)EFER_SCE;
547#endif
548 if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
549 return;
2cc51560 550
51c6cf66
AK		 551 vmx->host_state.guest_efer_loaded = 1;
552 guest_efer &= ~ignore_bits;
553 guest_efer |= host_efer & ignore_bits;
554 wrmsrl(MSR_EFER, guest_efer);
8b9cf98c 555 vmx->vcpu.stat.efer_reload++;
2cc51560
ED		 556}
557
51c6cf66
AK		 558static void reload_host_efer(struct vcpu_vmx *vmx)
559{
560 if (vmx->host_state.guest_efer_loaded) {
561 vmx->host_state.guest_efer_loaded = 0;
562 load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1);
563 }
564}
565
04d2cc77 566static void vmx_save_host_state(struct kvm_vcpu *vcpu)
33ed6329 567{
04d2cc77
AK		 568 struct vcpu_vmx *vmx = to_vmx(vcpu);
569
a2fa3e9f 570 if (vmx->host_state.loaded)
33ed6329
AK		 571 return;
572
a2fa3e9f 573 vmx->host_state.loaded = 1;
33ed6329
AK		 574 /*
575 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
576 * allow segment selectors with cpl > 0 or ti == 1.
577 */
d6e88aec 578 vmx->host_state.ldt_sel = kvm_read_ldt();
152d3f2f 579 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
d6e88aec 580 vmx->host_state.fs_sel = kvm_read_fs();
152d3f2f 581 if (!(vmx->host_state.fs_sel & 7)) {
a2fa3e9f 582 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
152d3f2f
LV		 583 vmx->host_state.fs_reload_needed = 0;
584 } else {
33ed6329 585 vmcs_write16(HOST_FS_SELECTOR, 0);
152d3f2f 586 vmx->host_state.fs_reload_needed = 1;
33ed6329 587 }
d6e88aec 588 vmx->host_state.gs_sel = kvm_read_gs();
a2fa3e9f
GH		 589 if (!(vmx->host_state.gs_sel & 7))
590 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
33ed6329
AK		 591 else {
592 vmcs_write16(HOST_GS_SELECTOR, 0);
152d3f2f 593 vmx->host_state.gs_ldt_reload_needed = 1;
33ed6329
AK		 594 }
595
596#ifdef CONFIG_X86_64
597 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
598 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
599#else
a2fa3e9f
GH		 600 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
601 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
33ed6329 602#endif
707c0874
AK		 603
604#ifdef CONFIG_X86_64
d77c26fc 605 if (is_long_mode(&vmx->vcpu))
a2fa3e9f
GH		 606 save_msrs(vmx->host_msrs +
607 vmx->msr_offset_kernel_gs_base, 1);
d77c26fc 608
707c0874 609#endif
a2fa3e9f 610 load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
51c6cf66 611 load_transition_efer(vmx);
33ed6329
AK		 612}
613
a9b21b62 614static void __vmx_load_host_state(struct vcpu_vmx *vmx)
33ed6329 615{
15ad7146 616 unsigned long flags;
33ed6329 617
a2fa3e9f 618 if (!vmx->host_state.loaded)
33ed6329
AK		 619 return;
620
e1beb1d3 621 ++vmx->vcpu.stat.host_state_reload;
a2fa3e9f 622 vmx->host_state.loaded = 0;
152d3f2f 623 if (vmx->host_state.fs_reload_needed)
d6e88aec 624 kvm_load_fs(vmx->host_state.fs_sel);
152d3f2f 625 if (vmx->host_state.gs_ldt_reload_needed) {
d6e88aec 626 kvm_load_ldt(vmx->host_state.ldt_sel);
33ed6329
AK		 627 /*
628 * If we have to reload gs, we must take care to
629 * preserve our gs base.
630 */
15ad7146 631 local_irq_save(flags);
d6e88aec 632 kvm_load_gs(vmx->host_state.gs_sel);
33ed6329
AK		 633#ifdef CONFIG_X86_64
634 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
635#endif
15ad7146 636 local_irq_restore(flags);
33ed6329 637 }
152d3f2f 638 reload_tss();
a2fa3e9f
GH		 639 save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
640 load_msrs(vmx->host_msrs, vmx->save_nmsrs);
51c6cf66 641 reload_host_efer(vmx);
33ed6329
AK		 642}
643
a9b21b62
AK		 644static void vmx_load_host_state(struct vcpu_vmx *vmx)
645{
646 preempt_disable();
647 __vmx_load_host_state(vmx);
648 preempt_enable();
649}
650
6aa8b732
AK		 651/*
652 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
653 * vcpu mutex is already taken.
654 */
15ad7146 655static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
6aa8b732 656{
a2fa3e9f
GH		 657 struct vcpu_vmx *vmx = to_vmx(vcpu);
658 u64 phys_addr = __pa(vmx->vmcs);
019960ae 659 u64 tsc_this, delta, new_offset;
6aa8b732 660
a3d7f85f 661 if (vcpu->cpu != cpu) {
8b9cf98c 662 vcpu_clear(vmx);
2f599714 663 kvm_migrate_timers(vcpu);
2384d2b3 664 vpid_sync_vcpu_all(vmx);
543e4243
AK		 665 local_irq_disable();
666 list_add(&vmx->local_vcpus_link,
667 &per_cpu(vcpus_on_cpu, cpu));
668 local_irq_enable();
a3d7f85f 669 }
6aa8b732 670
a2fa3e9f 671 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
6aa8b732
AK		 672 u8 error;
673
a2fa3e9f 674 per_cpu(current_vmcs, cpu) = vmx->vmcs;
4ecac3fd 675 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
6aa8b732
AK		 676 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
677 : "cc");
678 if (error)
679 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
a2fa3e9f 680 vmx->vmcs, phys_addr);
6aa8b732
AK		 681 }
682
683 if (vcpu->cpu != cpu) {
684 struct descriptor_table dt;
685 unsigned long sysenter_esp;
686
687 vcpu->cpu = cpu;
688 /*
689 * Linux uses per-cpu TSS and GDT, so set these when switching
690 * processors.
691 */
d6e88aec
AK		 692 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
693 kvm_get_gdt(&dt);
6aa8b732
AK		 694 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
695
696 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
697 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
7700270e
AK		 698
699 /*
700 * Make sure the time stamp counter is monotonous.
701 */
702 rdtscll(tsc_this);
019960ae
AK		 703 if (tsc_this < vcpu->arch.host_tsc) {
704 delta = vcpu->arch.host_tsc - tsc_this;
705 new_offset = vmcs_read64(TSC_OFFSET) + delta;
706 vmcs_write64(TSC_OFFSET, new_offset);
707 }
6aa8b732 708 }
6aa8b732
AK		 709}
710
711static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
712{
a9b21b62 713 __vmx_load_host_state(to_vmx(vcpu));
6aa8b732
AK		 714}
715
5fd86fcf
AK		 716static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
717{
718 if (vcpu->fpu_active)
719 return;
720 vcpu->fpu_active = 1;
707d92fa 721 vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
ad312c7c 722 if (vcpu->arch.cr0 & X86_CR0_TS)
707d92fa 723 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
5fd86fcf
AK		 724 update_exception_bitmap(vcpu);
725}
726
727static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
728{
729 if (!vcpu->fpu_active)
730 return;
731 vcpu->fpu_active = 0;
707d92fa 732 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
5fd86fcf
AK		 733 update_exception_bitmap(vcpu);
734}
735
6aa8b732
AK		 736static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
737{
738 return vmcs_readl(GUEST_RFLAGS);
739}
740
741static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
742{
56b237e3 743 if (vcpu->arch.rmode.vm86_active)
053de044 744 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
6aa8b732
AK		 745 vmcs_writel(GUEST_RFLAGS, rflags);
746}
747
2809f5d2
GC		 748static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
749{
750 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
751 int ret = 0;
752
753 if (interruptibility & GUEST_INTR_STATE_STI)
754 ret |= X86_SHADOW_INT_STI;
755 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
756 ret |= X86_SHADOW_INT_MOV_SS;
757
758 return ret & mask;
759}
760
761static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
762{
763 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
764 u32 interruptibility = interruptibility_old;
765
766 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
767
768 if (mask & X86_SHADOW_INT_MOV_SS)
769 interruptibility |= GUEST_INTR_STATE_MOV_SS;
770 if (mask & X86_SHADOW_INT_STI)
771 interruptibility |= GUEST_INTR_STATE_STI;
772
773 if ((interruptibility != interruptibility_old))
774 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
775}
776
6aa8b732
AK		 777static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
778{
779 unsigned long rip;
6aa8b732 780
5fdbf976 781 rip = kvm_rip_read(vcpu);
6aa8b732 782 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5fdbf976 783 kvm_rip_write(vcpu, rip);
6aa8b732 784
2809f5d2
GC		 785 /* skipping an emulated instruction also counts */
786 vmx_set_interrupt_shadow(vcpu, 0);
6aa8b732
AK		 787}
788
298101da
AK		 789static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
790 bool has_error_code, u32 error_code)
791{
77ab6db0 792 struct vcpu_vmx *vmx = to_vmx(vcpu);
8ab2d2e2 793 u32 intr_info = nr | INTR_INFO_VALID_MASK;
77ab6db0 794
8ab2d2e2 795 if (has_error_code) {
77ab6db0 796 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
8ab2d2e2
JK		 797 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
798 }
77ab6db0 799
56b237e3 800 if (vcpu->arch.rmode.vm86_active) {
77ab6db0
JK		 801 vmx->rmode.irq.pending = true;
802 vmx->rmode.irq.vector = nr;
803 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
ae0bb3e0
GN		 804 if (kvm_exception_is_soft(nr))
805 vmx->rmode.irq.rip +=
806 vmx->vcpu.arch.event_exit_inst_len;
8ab2d2e2
JK		 807 intr_info |= INTR_TYPE_SOFT_INTR;
808 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
77ab6db0
JK		 809 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
810 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
811 return;
812 }
813
66fd3f7f
GN		 814 if (kvm_exception_is_soft(nr)) {
815 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
816 vmx->vcpu.arch.event_exit_inst_len);
8ab2d2e2
JK		 817 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
818 } else
819 intr_info |= INTR_TYPE_HARD_EXCEPTION;
820
821 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
298101da
AK		 822}
823
a75beee6
ED		 824/*
825 * Swap MSR entry in host/guest MSR entry array.
826 */
54e11fa1 827#ifdef CONFIG_X86_64
8b9cf98c 828static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
a75beee6 829{
a2fa3e9f
GH		 830 struct kvm_msr_entry tmp;
831
832 tmp = vmx->guest_msrs[to];
833 vmx->guest_msrs[to] = vmx->guest_msrs[from];
834 vmx->guest_msrs[from] = tmp;
835 tmp = vmx->host_msrs[to];
836 vmx->host_msrs[to] = vmx->host_msrs[from];
837 vmx->host_msrs[from] = tmp;
a75beee6 838}
54e11fa1 839#endif
a75beee6 840
e38aea3e
AK		 841/*
842 * Set up the vmcs to automatically save and restore system
843 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
844 * mode, as fiddling with msrs is very expensive.
845 */
8b9cf98c 846static void setup_msrs(struct vcpu_vmx *vmx)
e38aea3e 847{
2cc51560 848 int save_nmsrs;
5897297b 849 unsigned long *msr_bitmap;
e38aea3e 850
33f9c505 851 vmx_load_host_state(vmx);
a75beee6
ED		 852 save_nmsrs = 0;
853#ifdef CONFIG_X86_64
8b9cf98c 854 if (is_long_mode(&vmx->vcpu)) {
2cc51560
ED		 855 int index;
856
8b9cf98c 857 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
a75beee6 858 if (index >= 0)
8b9cf98c
RR		 859 move_msr_up(vmx, index, save_nmsrs++);
860 index = __find_msr_index(vmx, MSR_LSTAR);
a75beee6 861 if (index >= 0)
8b9cf98c
RR		 862 move_msr_up(vmx, index, save_nmsrs++);
863 index = __find_msr_index(vmx, MSR_CSTAR);
a75beee6 864 if (index >= 0)
8b9cf98c
RR		 865 move_msr_up(vmx, index, save_nmsrs++);
866 index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
a75beee6 867 if (index >= 0)
8b9cf98c 868 move_msr_up(vmx, index, save_nmsrs++);
a75beee6
ED		 869 /*
870 * MSR_K6_STAR is only needed on long mode guests, and only
871 * if efer.sce is enabled.
872 */
8b9cf98c 873 index = __find_msr_index(vmx, MSR_K6_STAR);
ad312c7c 874 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
8b9cf98c 875 move_msr_up(vmx, index, save_nmsrs++);
a75beee6
ED		 876 }
877#endif
a2fa3e9f 878 vmx->save_nmsrs = save_nmsrs;
e38aea3e 879
4d56c8a7 880#ifdef CONFIG_X86_64
a2fa3e9f 881 vmx->msr_offset_kernel_gs_base =
8b9cf98c 882 __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
4d56c8a7 883#endif
8b9cf98c 884 vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER);
5897297b
AK		 885
886 if (cpu_has_vmx_msr_bitmap()) {
887 if (is_long_mode(&vmx->vcpu))
888 msr_bitmap = vmx_msr_bitmap_longmode;
889 else
890 msr_bitmap = vmx_msr_bitmap_legacy;
891
892 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
893 }
e38aea3e
AK		 894}
895
6aa8b732
AK		 896/*
897 * reads and returns guest's timestamp counter "register"
898 * guest_tsc = host_tsc + tsc_offset -- 21.3
899 */
900static u64 guest_read_tsc(void)
901{
902 u64 host_tsc, tsc_offset;
903
904 rdtscll(host_tsc);
905 tsc_offset = vmcs_read64(TSC_OFFSET);
906 return host_tsc + tsc_offset;
907}
908
909/*
910 * writes 'guest_tsc' into guest's timestamp counter "register"
911 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
912 */
53f658b3 913static void guest_write_tsc(u64 guest_tsc, u64 host_tsc)
6aa8b732 914{
6aa8b732
AK		 915 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
916}
917
6aa8b732
AK		 918/*
919 * Reads an msr value (of 'msr_index') into 'pdata'.
920 * Returns 0 on success, non-0 otherwise.
921 * Assumes vcpu_load() was already called.
922 */
923static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
924{
925 u64 data;
a2fa3e9f 926 struct kvm_msr_entry *msr;
6aa8b732
AK		 927
928 if (!pdata) {
929 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
930 return -EINVAL;
931 }
932
933 switch (msr_index) {
05b3e0c2 934#ifdef CONFIG_X86_64
6aa8b732
AK		 935 case MSR_FS_BASE:
936 data = vmcs_readl(GUEST_FS_BASE);
937 break;
938 case MSR_GS_BASE:
939 data = vmcs_readl(GUEST_GS_BASE);
940 break;
941 case MSR_EFER:
3bab1f5d 942 return kvm_get_msr_common(vcpu, msr_index, pdata);
6aa8b732 943#endif
af24a4e4 944 case MSR_IA32_TSC:
6aa8b732
AK		 945 data = guest_read_tsc();
946 break;
947 case MSR_IA32_SYSENTER_CS:
948 data = vmcs_read32(GUEST_SYSENTER_CS);
949 break;
950 case MSR_IA32_SYSENTER_EIP:
f5b42c33 951 data = vmcs_readl(GUEST_SYSENTER_EIP);
6aa8b732
AK		 952 break;
953 case MSR_IA32_SYSENTER_ESP:
f5b42c33 954 data = vmcs_readl(GUEST_SYSENTER_ESP);
6aa8b732 955 break;
6aa8b732 956 default:
516a1a7e 957 vmx_load_host_state(to_vmx(vcpu));
8b9cf98c 958 msr = find_msr_entry(to_vmx(vcpu), msr_index);
3bab1f5d
AK		 959 if (msr) {
960 data = msr->data;
961 break;
6aa8b732 962 }
3bab1f5d 963 return kvm_get_msr_common(vcpu, msr_index, pdata);
6aa8b732
AK		 964 }
965
966 *pdata = data;
967 return 0;
968}
969
970/*
971 * Writes msr value into into the appropriate "register".
972 * Returns 0 on success, non-0 otherwise.
973 * Assumes vcpu_load() was already called.
974 */
975static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
976{
a2fa3e9f
GH		 977 struct vcpu_vmx *vmx = to_vmx(vcpu);
978 struct kvm_msr_entry *msr;
53f658b3 979 u64 host_tsc;
2cc51560
ED		 980 int ret = 0;
981
6aa8b732 982 switch (msr_index) {
3bab1f5d 983 case MSR_EFER:
a9b21b62 984 vmx_load_host_state(vmx);
2cc51560 985 ret = kvm_set_msr_common(vcpu, msr_index, data);
2cc51560 986 break;
16175a79 987#ifdef CONFIG_X86_64
6aa8b732
AK		 988 case MSR_FS_BASE:
989 vmcs_writel(GUEST_FS_BASE, data);
990 break;
991 case MSR_GS_BASE:
992 vmcs_writel(GUEST_GS_BASE, data);
993 break;
994#endif
995 case MSR_IA32_SYSENTER_CS:
996 vmcs_write32(GUEST_SYSENTER_CS, data);
997 break;
998 case MSR_IA32_SYSENTER_EIP:
f5b42c33 999 vmcs_writel(GUEST_SYSENTER_EIP, data);
6aa8b732
AK		 1000 break;
1001 case MSR_IA32_SYSENTER_ESP:
f5b42c33 1002 vmcs_writel(GUEST_SYSENTER_ESP, data);
6aa8b732 1003 break;
af24a4e4 1004 case MSR_IA32_TSC:
53f658b3
MT		 1005 rdtscll(host_tsc);
1006 guest_write_tsc(data, host_tsc);
efa67e0d
CL		 1007 break;
1008 case MSR_P6_PERFCTR0:
1009 case MSR_P6_PERFCTR1:
1010 case MSR_P6_EVNTSEL0:
1011 case MSR_P6_EVNTSEL1:
1012 /*
1013 * Just discard all writes to the performance counters; this
1014 * should keep both older linux and windows 64-bit guests
1015 * happy
1016 */
1017 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", msr_index, data);
1018
6aa8b732 1019 break;
468d472f
SY		 1020 case MSR_IA32_CR_PAT:
1021 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1022 vmcs_write64(GUEST_IA32_PAT, data);
1023 vcpu->arch.pat = data;
1024 break;
1025 }
1026 /* Otherwise falls through to kvm_set_msr_common */
6aa8b732 1027 default:
a9b21b62 1028 vmx_load_host_state(vmx);
8b9cf98c 1029 msr = find_msr_entry(vmx, msr_index);
3bab1f5d
AK		 1030 if (msr) {
1031 msr->data = data;
1032 break;
6aa8b732 1033 }
2cc51560 1034 ret = kvm_set_msr_common(vcpu, msr_index, data);
6aa8b732
AK		 1035 }
1036
2cc51560 1037 return ret;
6aa8b732
AK		 1038}
1039
5fdbf976 1040static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
6aa8b732 1041{
5fdbf976
MT		 1042 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1043 switch (reg) {
1044 case VCPU_REGS_RSP:
1045 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1046 break;
1047 case VCPU_REGS_RIP:
1048 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1049 break;
1050 default:
1051 break;
1052 }
6aa8b732
AK		 1053}
1054
d0bfb940 1055static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
6aa8b732 1056{
d0bfb940
JK		 1057 int old_debug = vcpu->guest_debug;
1058 unsigned long flags;
6aa8b732 1059
d0bfb940
JK		 1060 vcpu->guest_debug = dbg->control;
1061 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
1062 vcpu->guest_debug = 0;
6aa8b732 1063
ae675ef0
JK		 1064 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1065 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1066 else
1067 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1068
d0bfb940
JK		 1069 flags = vmcs_readl(GUEST_RFLAGS);
1070 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
1071 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1072 else if (old_debug & KVM_GUESTDBG_SINGLESTEP)
6aa8b732 1073 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
d0bfb940 1074 vmcs_writel(GUEST_RFLAGS, flags);
6aa8b732 1075
abd3f2d6 1076 update_exception_bitmap(vcpu);
6aa8b732
AK		 1077
1078 return 0;
1079}
1080
1081static __init int cpu_has_kvm_support(void)
1082{
6210e37b 1083 return cpu_has_vmx();
6aa8b732
AK		 1084}
1085
1086static __init int vmx_disabled_by_bios(void)
1087{
1088 u64 msr;
1089
1090 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
9ea542fa
SY		 1091 return (msr & (FEATURE_CONTROL_LOCKED |
1092 FEATURE_CONTROL_VMXON_ENABLED))
1093 == FEATURE_CONTROL_LOCKED;
62b3ffb8 1094 /* locked but not enabled */
6aa8b732
AK		 1095}
1096
774c47f1 1097static void hardware_enable(void *garbage)
6aa8b732
AK		 1098{
1099 int cpu = raw_smp_processor_id();
1100 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1101 u64 old;
1102
543e4243 1103 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
6aa8b732 1104 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
9ea542fa
SY		 1105 if ((old & (FEATURE_CONTROL_LOCKED |
1106 FEATURE_CONTROL_VMXON_ENABLED))
1107 != (FEATURE_CONTROL_LOCKED |
1108 FEATURE_CONTROL_VMXON_ENABLED))
6aa8b732 1109 /* enable and lock */
62b3ffb8 1110 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
9ea542fa
SY		 1111 FEATURE_CONTROL_LOCKED |
1112 FEATURE_CONTROL_VMXON_ENABLED);
66aee91a 1113 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
4ecac3fd
AK		 1114 asm volatile (ASM_VMX_VMXON_RAX
1115 : : "a"(&phys_addr), "m"(phys_addr)
6aa8b732
AK		 1116 : "memory", "cc");
1117}
1118
543e4243
AK		 1119static void vmclear_local_vcpus(void)
1120{
1121 int cpu = raw_smp_processor_id();
1122 struct vcpu_vmx *vmx, *n;
1123
1124 list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1125 local_vcpus_link)
1126 __vcpu_clear(vmx);
1127}
1128
710ff4a8
EH		 1129
1130/* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1131 * tricks.
1132 */
1133static void kvm_cpu_vmxoff(void)
6aa8b732 1134{
4ecac3fd 1135 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
e693d71b 1136 write_cr4(read_cr4() & ~X86_CR4_VMXE);
6aa8b732
AK		 1137}
1138
710ff4a8
EH		 1139static void hardware_disable(void *garbage)
1140{
1141 vmclear_local_vcpus();
1142 kvm_cpu_vmxoff();
1143}
1144
1c3d14fe 1145static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
d77c26fc 1146 u32 msr, u32 *result)
1c3d14fe
YS		 1147{
1148 u32 vmx_msr_low, vmx_msr_high;
1149 u32 ctl = ctl_min | ctl_opt;
1150
1151 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1152
1153 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1154 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
1155
1156 /* Ensure minimum (required) set of control bits are supported. */
1157 if (ctl_min & ~ctl)
002c7f7c 1158 return -EIO;
1c3d14fe
YS		 1159
1160 *result = ctl;
1161 return 0;
1162}
1163
002c7f7c 1164static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
6aa8b732
AK		 1165{
1166 u32 vmx_msr_low, vmx_msr_high;
d56f546d 1167 u32 min, opt, min2, opt2;
1c3d14fe
YS		 1168 u32 _pin_based_exec_control = 0;
1169 u32 _cpu_based_exec_control = 0;
f78e0e2e 1170 u32 _cpu_based_2nd_exec_control = 0;
1c3d14fe
YS		 1171 u32 _vmexit_control = 0;
1172 u32 _vmentry_control = 0;
1173
1174 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
f08864b4 1175 opt = PIN_BASED_VIRTUAL_NMIS;
1c3d14fe
YS		 1176 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1177 &_pin_based_exec_control) < 0)
002c7f7c 1178 return -EIO;
1c3d14fe
YS		 1179
1180 min = CPU_BASED_HLT_EXITING |
1181#ifdef CONFIG_X86_64
1182 CPU_BASED_CR8_LOAD_EXITING |
1183 CPU_BASED_CR8_STORE_EXITING |
1184#endif
d56f546d
SY		 1185 CPU_BASED_CR3_LOAD_EXITING |
1186 CPU_BASED_CR3_STORE_EXITING |
1c3d14fe
YS		 1187 CPU_BASED_USE_IO_BITMAPS |
1188 CPU_BASED_MOV_DR_EXITING |
a7052897
MT		 1189 CPU_BASED_USE_TSC_OFFSETING |
1190 CPU_BASED_INVLPG_EXITING;
f78e0e2e 1191 opt = CPU_BASED_TPR_SHADOW |
25c5f225 1192 CPU_BASED_USE_MSR_BITMAPS |
f78e0e2e 1193 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1c3d14fe
YS		 1194 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1195 &_cpu_based_exec_control) < 0)
002c7f7c 1196 return -EIO;
6e5d865c
YS		 1197#ifdef CONFIG_X86_64
1198 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1199 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1200 ~CPU_BASED_CR8_STORE_EXITING;
1201#endif
f78e0e2e 1202 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
d56f546d
SY		 1203 min2 = 0;
1204 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2384d2b3 1205 SECONDARY_EXEC_WBINVD_EXITING |
d56f546d
SY		 1206 SECONDARY_EXEC_ENABLE_VPID |
1207 SECONDARY_EXEC_ENABLE_EPT;
1208 if (adjust_vmx_controls(min2, opt2,
1209 MSR_IA32_VMX_PROCBASED_CTLS2,
f78e0e2e
SY		 1210 &_cpu_based_2nd_exec_control) < 0)
1211 return -EIO;
1212 }
1213#ifndef CONFIG_X86_64
1214 if (!(_cpu_based_2nd_exec_control &
1215 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1216 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1217#endif
d56f546d 1218 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
a7052897
MT		 1219 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1220 enabled */
d56f546d 1221 min &= ~(CPU_BASED_CR3_LOAD_EXITING |
a7052897
MT		 1222 CPU_BASED_CR3_STORE_EXITING |
1223 CPU_BASED_INVLPG_EXITING);
d56f546d
SY		 1224 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1225 &_cpu_based_exec_control) < 0)
1226 return -EIO;
1227 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1228 vmx_capability.ept, vmx_capability.vpid);
1229 }
1c3d14fe
YS		 1230
1231 min = 0;
1232#ifdef CONFIG_X86_64
1233 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1234#endif
468d472f 1235 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1c3d14fe
YS		 1236 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1237 &_vmexit_control) < 0)
002c7f7c 1238 return -EIO;
1c3d14fe 1239
468d472f
SY		 1240 min = 0;
1241 opt = VM_ENTRY_LOAD_IA32_PAT;
1c3d14fe
YS		 1242 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1243 &_vmentry_control) < 0)
002c7f7c 1244 return -EIO;
6aa8b732 1245
c68876fd 1246 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1c3d14fe
YS		 1247
1248 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1249 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
002c7f7c 1250 return -EIO;
1c3d14fe
YS		 1251
1252#ifdef CONFIG_X86_64
1253 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1254 if (vmx_msr_high & (1u<<16))
002c7f7c 1255 return -EIO;
1c3d14fe
YS		 1256#endif
1257
1258 /* Require Write-Back (WB) memory type for VMCS accesses. */
1259 if (((vmx_msr_high >> 18) & 15) != 6)
002c7f7c 1260 return -EIO;
1c3d14fe 1261
002c7f7c
YS		 1262 vmcs_conf->size = vmx_msr_high & 0x1fff;
1263 vmcs_conf->order = get_order(vmcs_config.size);
1264 vmcs_conf->revision_id = vmx_msr_low;
1c3d14fe 1265
002c7f7c
YS		 1266 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1267 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
f78e0e2e 1268 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
002c7f7c
YS		 1269 vmcs_conf->vmexit_ctrl = _vmexit_control;
1270 vmcs_conf->vmentry_ctrl = _vmentry_control;
1c3d14fe
YS		 1271
1272 return 0;
c68876fd 1273}
6aa8b732
AK		 1274
1275static struct vmcs *alloc_vmcs_cpu(int cpu)
1276{
1277 int node = cpu_to_node(cpu);
1278 struct page *pages;
1279 struct vmcs *vmcs;
1280
6484eb3e 1281 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
6aa8b732
AK		 1282 if (!pages)
1283 return NULL;
1284 vmcs = page_address(pages);
1c3d14fe
YS		 1285 memset(vmcs, 0, vmcs_config.size);
1286 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
6aa8b732
AK		 1287 return vmcs;
1288}
1289
1290static struct vmcs *alloc_vmcs(void)
1291{
d3b2c338 1292 return alloc_vmcs_cpu(raw_smp_processor_id());
6aa8b732
AK		 1293}
1294
1295static void free_vmcs(struct vmcs *vmcs)
1296{
1c3d14fe 1297 free_pages((unsigned long)vmcs, vmcs_config.order);
6aa8b732
AK		 1298}
1299
39959588 1300static void free_kvm_area(void)
6aa8b732
AK		 1301{
1302 int cpu;
1303
1304 for_each_online_cpu(cpu)
1305 free_vmcs(per_cpu(vmxarea, cpu));
1306}
1307
6aa8b732
AK		 1308static __init int alloc_kvm_area(void)
1309{
1310 int cpu;
1311
1312 for_each_online_cpu(cpu) {
1313 struct vmcs *vmcs;
1314
1315 vmcs = alloc_vmcs_cpu(cpu);
1316 if (!vmcs) {
1317 free_kvm_area();
1318 return -ENOMEM;
1319 }
1320
1321 per_cpu(vmxarea, cpu) = vmcs;
1322 }
1323 return 0;
1324}
1325
1326static __init int hardware_setup(void)
1327{
002c7f7c
YS		 1328 if (setup_vmcs_config(&vmcs_config) < 0)
1329 return -EIO;
50a37eb4
JR		 1330
1331 if (boot_cpu_has(X86_FEATURE_NX))
1332 kvm_enable_efer_bits(EFER_NX);
1333
93ba03c2
SY		 1334 if (!cpu_has_vmx_vpid())
1335 enable_vpid = 0;
1336
1337 if (!cpu_has_vmx_ept())
1338 enable_ept = 0;
1339
1340 if (!cpu_has_vmx_flexpriority())
1341 flexpriority_enabled = 0;
1342
95ba8273
GN		 1343 if (!cpu_has_vmx_tpr_shadow())
1344 kvm_x86_ops->update_cr8_intercept = NULL;
1345
6aa8b732
AK		 1346 return alloc_kvm_area();
1347}
1348
1349static __exit void hardware_unsetup(void)
1350{
1351 free_kvm_area();
1352}
1353
6aa8b732
AK		 1354static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1355{
1356 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1357
6af11b9e 1358 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
6aa8b732
AK		 1359 vmcs_write16(sf->selector, save->selector);
1360 vmcs_writel(sf->base, save->base);
1361 vmcs_write32(sf->limit, save->limit);
1362 vmcs_write32(sf->ar_bytes, save->ar);
1363 } else {
1364 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1365 << AR_DPL_SHIFT;
1366 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1367 }
1368}
1369
1370static void enter_pmode(struct kvm_vcpu *vcpu)
1371{
1372 unsigned long flags;
a89a8fb9 1373 struct vcpu_vmx *vmx = to_vmx(vcpu);
6aa8b732 1374
a89a8fb9 1375 vmx->emulation_required = 1;
56b237e3 1376 vcpu->arch.rmode.vm86_active = 0;
6aa8b732 1377
ad312c7c
ZX		 1378 vmcs_writel(GUEST_TR_BASE, vcpu->arch.rmode.tr.base);
1379 vmcs_write32(GUEST_TR_LIMIT, vcpu->arch.rmode.tr.limit);
1380 vmcs_write32(GUEST_TR_AR_BYTES, vcpu->arch.rmode.tr.ar);
6aa8b732
AK		 1381
1382 flags = vmcs_readl(GUEST_RFLAGS);
053de044 1383 flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
ad312c7c 1384 flags |= (vcpu->arch.rmode.save_iopl << IOPL_SHIFT);
6aa8b732
AK		 1385 vmcs_writel(GUEST_RFLAGS, flags);
1386
66aee91a
RR		 1387 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1388 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
6aa8b732
AK		 1389
1390 update_exception_bitmap(vcpu);
1391
a89a8fb9
MG		 1392 if (emulate_invalid_guest_state)
1393 return;
1394
ad312c7c
ZX		 1395 fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1396 fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1397 fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1398 fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
6aa8b732
AK		 1399
1400 vmcs_write16(GUEST_SS_SELECTOR, 0);
1401 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1402
1403 vmcs_write16(GUEST_CS_SELECTOR,
1404 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1405 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1406}
1407
d77c26fc 1408static gva_t rmode_tss_base(struct kvm *kvm)
6aa8b732 1409{
bfc6d222 1410 if (!kvm->arch.tss_addr) {
cbc94022
IE		 1411 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1412 kvm->memslots[0].npages - 3;
1413 return base_gfn << PAGE_SHIFT;
1414 }
bfc6d222 1415 return kvm->arch.tss_addr;
6aa8b732
AK		 1416}
1417
1418static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1419{
1420 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1421
1422 save->selector = vmcs_read16(sf->selector);
1423 save->base = vmcs_readl(sf->base);
1424 save->limit = vmcs_read32(sf->limit);
1425 save->ar = vmcs_read32(sf->ar_bytes);
15b00f32
JK		 1426 vmcs_write16(sf->selector, save->base >> 4);
1427 vmcs_write32(sf->base, save->base & 0xfffff);
6aa8b732
AK		 1428 vmcs_write32(sf->limit, 0xffff);
1429 vmcs_write32(sf->ar_bytes, 0xf3);
1430}
1431
1432static void enter_rmode(struct kvm_vcpu *vcpu)
1433{
1434 unsigned long flags;
a89a8fb9 1435 struct vcpu_vmx *vmx = to_vmx(vcpu);
6aa8b732 1436
a89a8fb9 1437 vmx->emulation_required = 1;
56b237e3 1438 vcpu->arch.rmode.vm86_active = 1;
6aa8b732 1439
ad312c7c 1440 vcpu->arch.rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
6aa8b732
AK		 1441 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1442
ad312c7c 1443 vcpu->arch.rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
6aa8b732
AK		 1444 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1445
ad312c7c 1446 vcpu->arch.rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
6aa8b732
AK		 1447 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1448
1449 flags = vmcs_readl(GUEST_RFLAGS);
ad312c7c
ZX		 1450 vcpu->arch.rmode.save_iopl
1451 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
6aa8b732 1452
053de044 1453 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
6aa8b732
AK		 1454
1455 vmcs_writel(GUEST_RFLAGS, flags);
66aee91a 1456 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
6aa8b732
AK		 1457 update_exception_bitmap(vcpu);
1458
a89a8fb9
MG		 1459 if (emulate_invalid_guest_state)
1460 goto continue_rmode;
1461
6aa8b732
AK		 1462 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1463 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1464 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1465
1466 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
abacf8df 1467 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
8cb5b033
AK		 1468 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1469 vmcs_writel(GUEST_CS_BASE, 0xf0000);
6aa8b732
AK		 1470 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1471
ad312c7c
ZX		 1472 fix_rmode_seg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1473 fix_rmode_seg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1474 fix_rmode_seg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1475 fix_rmode_seg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
75880a01 1476
a89a8fb9 1477continue_rmode:
8668a3c4 1478 kvm_mmu_reset_context(vcpu);
b7ebfb05 1479 init_rmode(vcpu->kvm);
6aa8b732
AK		 1480}
1481
401d10de
AS		 1482static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1483{
1484 struct vcpu_vmx *vmx = to_vmx(vcpu);
1485 struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1486
1487 vcpu->arch.shadow_efer = efer;
1488 if (!msr)
1489 return;
1490 if (efer & EFER_LMA) {
1491 vmcs_write32(VM_ENTRY_CONTROLS,
1492 vmcs_read32(VM_ENTRY_CONTROLS) |
1493 VM_ENTRY_IA32E_MODE);
1494 msr->data = efer;
1495 } else {
1496 vmcs_write32(VM_ENTRY_CONTROLS,
1497 vmcs_read32(VM_ENTRY_CONTROLS) &
1498 ~VM_ENTRY_IA32E_MODE);
1499
1500 msr->data = efer & ~EFER_LME;
1501 }
1502 setup_msrs(vmx);
1503}
1504
05b3e0c2 1505#ifdef CONFIG_X86_64
6aa8b732
AK		 1506
1507static void enter_lmode(struct kvm_vcpu *vcpu)
1508{
1509 u32 guest_tr_ar;
1510
1511 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1512 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1513 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
b8688d51 1514 __func__);
6aa8b732
AK		 1515 vmcs_write32(GUEST_TR_AR_BYTES,
1516 (guest_tr_ar & ~AR_TYPE_MASK)
1517 | AR_TYPE_BUSY_64_TSS);
1518 }
ad312c7c 1519 vcpu->arch.shadow_efer |= EFER_LMA;
401d10de 1520 vmx_set_efer(vcpu, vcpu->arch.shadow_efer);
6aa8b732
AK		 1521}
1522
1523static void exit_lmode(struct kvm_vcpu *vcpu)
1524{
ad312c7c 1525 vcpu->arch.shadow_efer &= ~EFER_LMA;
6aa8b732
AK		 1526
1527 vmcs_write32(VM_ENTRY_CONTROLS,
1528 vmcs_read32(VM_ENTRY_CONTROLS)
1e4e6e00 1529 & ~VM_ENTRY_IA32E_MODE);
6aa8b732
AK		 1530}
1531
1532#endif
1533
2384d2b3
SY		 1534static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1535{
1536 vpid_sync_vcpu_all(to_vmx(vcpu));
089d034e 1537 if (enable_ept)
4e1096d2 1538 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
2384d2b3
SY		 1539}
1540
25c4c276 1541static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
399badf3 1542{
ad312c7c
ZX		 1543 vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1544 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
399badf3
AK		 1545}
1546
1439442c
SY		 1547static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1548{
1549 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1550 if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
1551 printk(KERN_ERR "EPT: Fail to load pdptrs!\n");
1552 return;
1553 }
1554 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1555 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1556 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1557 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1558 }
1559}
1560
1561static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1562
1563static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1564 unsigned long cr0,
1565 struct kvm_vcpu *vcpu)
1566{
1567 if (!(cr0 & X86_CR0_PG)) {
1568 /* From paging/starting to nonpaging */
1569 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
65267ea1 1570 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1439442c
SY		 1571 (CPU_BASED_CR3_LOAD_EXITING |
1572 CPU_BASED_CR3_STORE_EXITING));
1573 vcpu->arch.cr0 = cr0;
1574 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1575 *hw_cr0 |= X86_CR0_PE | X86_CR0_PG;
1576 *hw_cr0 &= ~X86_CR0_WP;
1577 } else if (!is_paging(vcpu)) {
1578 /* From nonpaging to paging */
1579 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
65267ea1 1580 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1439442c
SY		 1581 ~(CPU_BASED_CR3_LOAD_EXITING |
1582 CPU_BASED_CR3_STORE_EXITING));
1583 vcpu->arch.cr0 = cr0;
1584 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1585 if (!(vcpu->arch.cr0 & X86_CR0_WP))
1586 *hw_cr0 &= ~X86_CR0_WP;
1587 }
1588}
1589
1590static void ept_update_paging_mode_cr4(unsigned long *hw_cr4,
1591 struct kvm_vcpu *vcpu)
1592{
1593 if (!is_paging(vcpu)) {
1594 *hw_cr4 &= ~X86_CR4_PAE;
1595 *hw_cr4 |= X86_CR4_PSE;
1596 } else if (!(vcpu->arch.cr4 & X86_CR4_PAE))
1597 *hw_cr4 &= ~X86_CR4_PAE;
1598}
1599
6aa8b732
AK		 1600static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1601{
1439442c
SY		 1602 unsigned long hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) |
1603 KVM_VM_CR0_ALWAYS_ON;
1604
5fd86fcf
AK		 1605 vmx_fpu_deactivate(vcpu);
1606
56b237e3 1607 if (vcpu->arch.rmode.vm86_active && (cr0 & X86_CR0_PE))
6aa8b732
AK		 1608 enter_pmode(vcpu);
1609
56b237e3 1610 if (!vcpu->arch.rmode.vm86_active && !(cr0 & X86_CR0_PE))
6aa8b732
AK		 1611 enter_rmode(vcpu);
1612
05b3e0c2 1613#ifdef CONFIG_X86_64
ad312c7c 1614 if (vcpu->arch.shadow_efer & EFER_LME) {
707d92fa 1615 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
6aa8b732 1616 enter_lmode(vcpu);
707d92fa 1617 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
6aa8b732
AK		 1618 exit_lmode(vcpu);
1619 }
1620#endif
1621
089d034e 1622 if (enable_ept)
1439442c
SY		 1623 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1624
6aa8b732 1625 vmcs_writel(CR0_READ_SHADOW, cr0);
1439442c 1626 vmcs_writel(GUEST_CR0, hw_cr0);
ad312c7c 1627 vcpu->arch.cr0 = cr0;
5fd86fcf 1628
707d92fa 1629 if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
5fd86fcf 1630 vmx_fpu_activate(vcpu);
6aa8b732
AK		 1631}
1632
1439442c
SY		 1633static u64 construct_eptp(unsigned long root_hpa)
1634{
1635 u64 eptp;
1636
1637 /* TODO write the value reading from MSR */
1638 eptp = VMX_EPT_DEFAULT_MT |
1639 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1640 eptp |= (root_hpa & PAGE_MASK);
1641
1642 return eptp;
1643}
1644
6aa8b732
AK		 1645static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1646{
1439442c
SY		 1647 unsigned long guest_cr3;
1648 u64 eptp;
1649
1650 guest_cr3 = cr3;
089d034e 1651 if (enable_ept) {
1439442c
SY		 1652 eptp = construct_eptp(cr3);
1653 vmcs_write64(EPT_POINTER, eptp);
1439442c
SY		 1654 ept_load_pdptrs(vcpu);
1655 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1656 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
1657 }
1658
2384d2b3 1659 vmx_flush_tlb(vcpu);
1439442c 1660 vmcs_writel(GUEST_CR3, guest_cr3);
ad312c7c 1661 if (vcpu->arch.cr0 & X86_CR0_PE)
5fd86fcf 1662 vmx_fpu_deactivate(vcpu);
6aa8b732
AK		 1663}
1664
1665static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1666{
56b237e3 1667 unsigned long hw_cr4 = cr4 | (vcpu->arch.rmode.vm86_active ?
1439442c
SY		 1668 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1669
ad312c7c 1670 vcpu->arch.cr4 = cr4;
089d034e 1671 if (enable_ept)
1439442c
SY		 1672 ept_update_paging_mode_cr4(&hw_cr4, vcpu);
1673
1674 vmcs_writel(CR4_READ_SHADOW, cr4);
1675 vmcs_writel(GUEST_CR4, hw_cr4);
6aa8b732
AK		 1676}
1677
6aa8b732
AK		 1678static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1679{
1680 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1681
1682 return vmcs_readl(sf->base);
1683}
1684
1685static void vmx_get_segment(struct kvm_vcpu *vcpu,
1686 struct kvm_segment *var, int seg)
1687{
1688 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1689 u32 ar;
1690
1691 var->base = vmcs_readl(sf->base);
1692 var->limit = vmcs_read32(sf->limit);
1693 var->selector = vmcs_read16(sf->selector);
1694 ar = vmcs_read32(sf->ar_bytes);
9fd4a3b7 1695 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
6aa8b732
AK		 1696 ar = 0;
1697 var->type = ar & 15;
1698 var->s = (ar >> 4) & 1;
1699 var->dpl = (ar >> 5) & 3;
1700 var->present = (ar >> 7) & 1;
1701 var->avl = (ar >> 12) & 1;
1702 var->l = (ar >> 13) & 1;
1703 var->db = (ar >> 14) & 1;
1704 var->g = (ar >> 15) & 1;
1705 var->unusable = (ar >> 16) & 1;
1706}
1707
2e4d2653
IE		 1708static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1709{
1710 struct kvm_segment kvm_seg;
1711
1712 if (!(vcpu->arch.cr0 & X86_CR0_PE)) /* if real mode */
1713 return 0;
1714
1715 if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1716 return 3;
1717
1718 vmx_get_segment(vcpu, &kvm_seg, VCPU_SREG_CS);
1719 return kvm_seg.selector & 3;
1720}
1721
653e3108 1722static u32 vmx_segment_access_rights(struct kvm_segment *var)
6aa8b732 1723{
6aa8b732
AK		 1724 u32 ar;
1725
653e3108 1726 if (var->unusable)
6aa8b732
AK		 1727 ar = 1 << 16;
1728 else {
1729 ar = var->type & 15;
1730 ar |= (var->s & 1) << 4;
1731 ar |= (var->dpl & 3) << 5;
1732 ar |= (var->present & 1) << 7;
1733 ar |= (var->avl & 1) << 12;
1734 ar |= (var->l & 1) << 13;
1735 ar |= (var->db & 1) << 14;
1736 ar |= (var->g & 1) << 15;
1737 }
f7fbf1fd
UL		 1738 if (ar == 0) /* a 0 value means unusable */
1739 ar = AR_UNUSABLE_MASK;
653e3108
AK		 1740
1741 return ar;
1742}
1743
1744static void vmx_set_segment(struct kvm_vcpu *vcpu,
1745 struct kvm_segment *var, int seg)
1746{
1747 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1748 u32 ar;
1749
56b237e3 1750 if (vcpu->arch.rmode.vm86_active && seg == VCPU_SREG_TR) {
ad312c7c
ZX		 1751 vcpu->arch.rmode.tr.selector = var->selector;
1752 vcpu->arch.rmode.tr.base = var->base;
1753 vcpu->arch.rmode.tr.limit = var->limit;
1754 vcpu->arch.rmode.tr.ar = vmx_segment_access_rights(var);
653e3108
AK		 1755 return;
1756 }
1757 vmcs_writel(sf->base, var->base);
1758 vmcs_write32(sf->limit, var->limit);
1759 vmcs_write16(sf->selector, var->selector);
56b237e3 1760 if (vcpu->arch.rmode.vm86_active && var->s) {
653e3108
AK		 1761 /*
1762 * Hack real-mode segments into vm86 compatibility.
1763 */
1764 if (var->base == 0xffff0000 && var->selector == 0xf000)
1765 vmcs_writel(sf->base, 0xf0000);
1766 ar = 0xf3;
1767 } else
1768 ar = vmx_segment_access_rights(var);
6aa8b732
AK		 1769 vmcs_write32(sf->ar_bytes, ar);
1770}
1771
6aa8b732
AK		 1772static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1773{
1774 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1775
1776 *db = (ar >> 14) & 1;
1777 *l = (ar >> 13) & 1;
1778}
1779
1780static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1781{
1782 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1783 dt->base = vmcs_readl(GUEST_IDTR_BASE);
1784}
1785
1786static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1787{
1788 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1789 vmcs_writel(GUEST_IDTR_BASE, dt->base);
1790}
1791
1792static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1793{
1794 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1795 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1796}
1797
1798static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1799{
1800 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1801 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1802}
1803
648dfaa7
MG		 1804static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
1805{
1806 struct kvm_segment var;
1807 u32 ar;
1808
1809 vmx_get_segment(vcpu, &var, seg);
1810 ar = vmx_segment_access_rights(&var);
1811
1812 if (var.base != (var.selector << 4))
1813 return false;
1814 if (var.limit != 0xffff)
1815 return false;
1816 if (ar != 0xf3)
1817 return false;
1818
1819 return true;
1820}
1821
1822static bool code_segment_valid(struct kvm_vcpu *vcpu)
1823{
1824 struct kvm_segment cs;
1825 unsigned int cs_rpl;
1826
1827 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1828 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
1829
1872a3f4
AK		 1830 if (cs.unusable)
1831 return false;
648dfaa7
MG		 1832 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
1833 return false;
1834 if (!cs.s)
1835 return false;
1872a3f4 1836 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
648dfaa7
MG		 1837 if (cs.dpl > cs_rpl)
1838 return false;
1872a3f4 1839 } else {
648dfaa7
MG		 1840 if (cs.dpl != cs_rpl)
1841 return false;
1842 }
1843 if (!cs.present)
1844 return false;
1845
1846 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
1847 return true;
1848}
1849
1850static bool stack_segment_valid(struct kvm_vcpu *vcpu)
1851{
1852 struct kvm_segment ss;
1853 unsigned int ss_rpl;
1854
1855 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1856 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
1857
1872a3f4
AK		 1858 if (ss.unusable)
1859 return true;
1860 if (ss.type != 3 && ss.type != 7)
648dfaa7
MG		 1861 return false;
1862 if (!ss.s)
1863 return false;
1864 if (ss.dpl != ss_rpl) /* DPL != RPL */
1865 return false;
1866 if (!ss.present)
1867 return false;
1868
1869 return true;
1870}
1871
1872static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
1873{
1874 struct kvm_segment var;
1875 unsigned int rpl;
1876
1877 vmx_get_segment(vcpu, &var, seg);
1878 rpl = var.selector & SELECTOR_RPL_MASK;
1879
1872a3f4
AK		 1880 if (var.unusable)
1881 return true;
648dfaa7
MG		 1882 if (!var.s)
1883 return false;
1884 if (!var.present)
1885 return false;
1886 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
1887 if (var.dpl < rpl) /* DPL < RPL */
1888 return false;
1889 }
1890
1891 /* TODO: Add other members to kvm_segment_field to allow checking for other access
1892 * rights flags
1893 */
1894 return true;
1895}
1896
1897static bool tr_valid(struct kvm_vcpu *vcpu)
1898{
1899 struct kvm_segment tr;
1900
1901 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
1902
1872a3f4
AK		 1903 if (tr.unusable)
1904 return false;
648dfaa7
MG		 1905 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
1906 return false;
1872a3f4 1907 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
648dfaa7
MG		 1908 return false;
1909 if (!tr.present)
1910 return false;
1911
1912 return true;
1913}
1914
1915static bool ldtr_valid(struct kvm_vcpu *vcpu)
1916{
1917 struct kvm_segment ldtr;
1918
1919 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
1920
1872a3f4
AK		 1921 if (ldtr.unusable)
1922 return true;
648dfaa7
MG		 1923 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
1924 return false;
1925 if (ldtr.type != 2)
1926 return false;
1927 if (!ldtr.present)
1928 return false;
1929
1930 return true;
1931}
1932
1933static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
1934{
1935 struct kvm_segment cs, ss;
1936
1937 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1938 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1939
1940 return ((cs.selector & SELECTOR_RPL_MASK) ==
1941 (ss.selector & SELECTOR_RPL_MASK));
1942}
1943
1944/*
1945 * Check if guest state is valid. Returns true if valid, false if
1946 * not.
1947 * We assume that registers are always usable
1948 */
1949static bool guest_state_valid(struct kvm_vcpu *vcpu)
1950{
1951 /* real mode guest state checks */
1952 if (!(vcpu->arch.cr0 & X86_CR0_PE)) {
1953 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
1954 return false;
1955 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
1956 return false;
1957 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
1958 return false;
1959 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
1960 return false;
1961 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
1962 return false;
1963 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
1964 return false;
1965 } else {
1966 /* protected mode guest state checks */
1967 if (!cs_ss_rpl_check(vcpu))
1968 return false;
1969 if (!code_segment_valid(vcpu))
1970 return false;
1971 if (!stack_segment_valid(vcpu))
1972 return false;
1973 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
1974 return false;
1975 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
1976 return false;
1977 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
1978 return false;
1979 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
1980 return false;
1981 if (!tr_valid(vcpu))
1982 return false;
1983 if (!ldtr_valid(vcpu))
1984 return false;
1985 }
1986 /* TODO:
1987 * - Add checks on RIP
1988 * - Add checks on RFLAGS
1989 */
1990
1991 return true;
1992}
1993
d77c26fc 1994static int init_rmode_tss(struct kvm *kvm)
6aa8b732 1995{
6aa8b732 1996 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
195aefde 1997 u16 data = 0;
10589a46 1998 int ret = 0;
195aefde 1999 int r;
6aa8b732 2000
195aefde
IE		 2001 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2002 if (r < 0)
10589a46 2003 goto out;
195aefde 2004 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
464d17c8
SY		 2005 r = kvm_write_guest_page(kvm, fn++, &data,
2006 TSS_IOPB_BASE_OFFSET, sizeof(u16));
195aefde 2007 if (r < 0)
10589a46 2008 goto out;
195aefde
IE		 2009 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2010 if (r < 0)
10589a46 2011 goto out;
195aefde
IE		 2012 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2013 if (r < 0)
10589a46 2014 goto out;
195aefde 2015 data = ~0;
10589a46
MT		 2016 r = kvm_write_guest_page(kvm, fn, &data,
2017 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2018 sizeof(u8));
195aefde 2019 if (r < 0)
10589a46
MT		 2020 goto out;
2021
2022 ret = 1;
2023out:
10589a46 2024 return ret;
6aa8b732
AK		 2025}
2026
b7ebfb05
SY		 2027static int init_rmode_identity_map(struct kvm *kvm)
2028{
2029 int i, r, ret;
2030 pfn_t identity_map_pfn;
2031 u32 tmp;
2032
089d034e 2033 if (!enable_ept)
b7ebfb05
SY		 2034 return 1;
2035 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2036 printk(KERN_ERR "EPT: identity-mapping pagetable "
2037 "haven't been allocated!\n");
2038 return 0;
2039 }
2040 if (likely(kvm->arch.ept_identity_pagetable_done))
2041 return 1;
2042 ret = 0;
2043 identity_map_pfn = VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT;
2044 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2045 if (r < 0)
2046 goto out;
2047 /* Set up identity-mapping pagetable for EPT in real mode */
2048 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2049 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2050 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2051 r = kvm_write_guest_page(kvm, identity_map_pfn,
2052 &tmp, i * sizeof(tmp), sizeof(tmp));
2053 if (r < 0)
2054 goto out;
2055 }
2056 kvm->arch.ept_identity_pagetable_done = true;
2057 ret = 1;
2058out:
2059 return ret;
2060}
2061
6aa8b732
AK		 2062static void seg_setup(int seg)
2063{
2064 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2065
2066 vmcs_write16(sf->selector, 0);
2067 vmcs_writel(sf->base, 0);
2068 vmcs_write32(sf->limit, 0xffff);
a16b20da 2069 vmcs_write32(sf->ar_bytes, 0xf3);
6aa8b732
AK		 2070}
2071
f78e0e2e
SY		 2072static int alloc_apic_access_page(struct kvm *kvm)
2073{
2074 struct kvm_userspace_memory_region kvm_userspace_mem;
2075 int r = 0;
2076
72dc67a6 2077 down_write(&kvm->slots_lock);
bfc6d222 2078 if (kvm->arch.apic_access_page)
f78e0e2e
SY		 2079 goto out;
2080 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2081 kvm_userspace_mem.flags = 0;
2082 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2083 kvm_userspace_mem.memory_size = PAGE_SIZE;
2084 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2085 if (r)
2086 goto out;
72dc67a6 2087
bfc6d222 2088 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
f78e0e2e 2089out:
72dc67a6 2090 up_write(&kvm->slots_lock);
f78e0e2e
SY		 2091 return r;
2092}
2093
b7ebfb05
SY		 2094static int alloc_identity_pagetable(struct kvm *kvm)
2095{
2096 struct kvm_userspace_memory_region kvm_userspace_mem;
2097 int r = 0;
2098
2099 down_write(&kvm->slots_lock);
2100 if (kvm->arch.ept_identity_pagetable)
2101 goto out;
2102 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2103 kvm_userspace_mem.flags = 0;
2104 kvm_userspace_mem.guest_phys_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
2105 kvm_userspace_mem.memory_size = PAGE_SIZE;
2106 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2107 if (r)
2108 goto out;
2109
b7ebfb05
SY		 2110 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2111 VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT);
b7ebfb05
SY		 2112out:
2113 up_write(&kvm->slots_lock);
2114 return r;
2115}
2116
2384d2b3
SY		 2117static void allocate_vpid(struct vcpu_vmx *vmx)
2118{
2119 int vpid;
2120
2121 vmx->vpid = 0;
919818ab 2122 if (!enable_vpid)
2384d2b3
SY		 2123 return;
2124 spin_lock(&vmx_vpid_lock);
2125 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2126 if (vpid < VMX_NR_VPIDS) {
2127 vmx->vpid = vpid;
2128 __set_bit(vpid, vmx_vpid_bitmap);
2129 }
2130 spin_unlock(&vmx_vpid_lock);
2131}
2132
5897297b 2133static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
25c5f225 2134{
3e7c73e9 2135 int f = sizeof(unsigned long);
25c5f225
SY		 2136
2137 if (!cpu_has_vmx_msr_bitmap())
2138 return;
2139
2140 /*
2141 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2142 * have the write-low and read-high bitmap offsets the wrong way round.
2143 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2144 */
25c5f225 2145 if (msr <= 0x1fff) {
3e7c73e9
AK		 2146 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2147 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
25c5f225
SY		 2148 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2149 msr &= 0x1fff;
3e7c73e9
AK		 2150 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2151 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
25c5f225 2152 }
25c5f225
SY		 2153}
2154
5897297b
AK		 2155static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2156{
2157 if (!longmode_only)
2158 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2159 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2160}
2161
6aa8b732
AK		 2162/*
2163 * Sets up the vmcs for emulated real mode.
2164 */
8b9cf98c 2165static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
6aa8b732 2166{
468d472f 2167 u32 host_sysenter_cs, msr_low, msr_high;
6aa8b732 2168 u32 junk;
53f658b3 2169 u64 host_pat, tsc_this, tsc_base;
6aa8b732
AK		 2170 unsigned long a;
2171 struct descriptor_table dt;
2172 int i;
cd2276a7 2173 unsigned long kvm_vmx_return;
6e5d865c 2174 u32 exec_control;
6aa8b732 2175
6aa8b732 2176 /* I/O */
3e7c73e9
AK		 2177 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2178 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
6aa8b732 2179
25c5f225 2180 if (cpu_has_vmx_msr_bitmap())
5897297b 2181 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
25c5f225 2182
6aa8b732
AK		 2183 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2184
6aa8b732 2185 /* Control */
1c3d14fe
YS		 2186 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2187 vmcs_config.pin_based_exec_ctrl);
6e5d865c
YS		 2188
2189 exec_control = vmcs_config.cpu_based_exec_ctrl;
2190 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2191 exec_control &= ~CPU_BASED_TPR_SHADOW;
2192#ifdef CONFIG_X86_64
2193 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2194 CPU_BASED_CR8_LOAD_EXITING;
2195#endif
2196 }
089d034e 2197 if (!enable_ept)
d56f546d 2198 exec_control |= CPU_BASED_CR3_STORE_EXITING |
83dbc83a
MT		 2199 CPU_BASED_CR3_LOAD_EXITING |
2200 CPU_BASED_INVLPG_EXITING;
6e5d865c 2201 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
6aa8b732 2202
83ff3b9d
SY		 2203 if (cpu_has_secondary_exec_ctrls()) {
2204 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2205 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2206 exec_control &=
2207 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2384d2b3
SY		 2208 if (vmx->vpid == 0)
2209 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
089d034e 2210 if (!enable_ept)
d56f546d 2211 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
83ff3b9d
SY		 2212 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2213 }
f78e0e2e 2214
c7addb90
AK		 2215 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2216 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
6aa8b732
AK		 2217 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
2218
2219 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
2220 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
2221 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2222
2223 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
2224 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2225 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
d6e88aec
AK		 2226 vmcs_write16(HOST_FS_SELECTOR, kvm_read_fs()); /* 22.2.4 */
2227 vmcs_write16(HOST_GS_SELECTOR, kvm_read_gs()); /* 22.2.4 */
6aa8b732 2228 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
05b3e0c2 2229#ifdef CONFIG_X86_64
6aa8b732
AK		 2230 rdmsrl(MSR_FS_BASE, a);
2231 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2232 rdmsrl(MSR_GS_BASE, a);
2233 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2234#else
2235 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2236 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2237#endif
2238
2239 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
2240
d6e88aec 2241 kvm_get_idt(&dt);
6aa8b732
AK		 2242 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
2243
d77c26fc 2244 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
cd2276a7 2245 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2cc51560
ED		 2246 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2247 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2248 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
6aa8b732
AK		 2249
2250 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2251 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2252 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2253 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
2254 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2255 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2256
468d472f
SY		 2257 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2258 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2259 host_pat = msr_low | ((u64) msr_high << 32);
2260 vmcs_write64(HOST_IA32_PAT, host_pat);
2261 }
2262 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2263 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2264 host_pat = msr_low | ((u64) msr_high << 32);
2265 /* Write the default value follow host pat */
2266 vmcs_write64(GUEST_IA32_PAT, host_pat);
2267 /* Keep arch.pat sync with GUEST_IA32_PAT */
2268 vmx->vcpu.arch.pat = host_pat;
2269 }
2270
6aa8b732
AK		 2271 for (i = 0; i < NR_VMX_MSR; ++i) {
2272 u32 index = vmx_msr_index[i];
2273 u32 data_low, data_high;
2274 u64 data;
a2fa3e9f 2275 int j = vmx->nmsrs;
6aa8b732
AK		 2276
2277 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2278 continue;
432bd6cb
AK		 2279 if (wrmsr_safe(index, data_low, data_high) < 0)
2280 continue;
6aa8b732 2281 data = data_low | ((u64)data_high << 32);
a2fa3e9f
GH		 2282 vmx->host_msrs[j].index = index;
2283 vmx->host_msrs[j].reserved = 0;
2284 vmx->host_msrs[j].data = data;
2285 vmx->guest_msrs[j] = vmx->host_msrs[j];
2286 ++vmx->nmsrs;
6aa8b732 2287 }
6aa8b732 2288
1c3d14fe 2289 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
6aa8b732
AK		 2290
2291 /* 22.2.1, 20.8.1 */
1c3d14fe
YS		 2292 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2293
e00c8cf2
AK		 2294 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2295 vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
2296
53f658b3
MT		 2297 tsc_base = vmx->vcpu.kvm->arch.vm_init_tsc;
2298 rdtscll(tsc_this);
2299 if (tsc_this < vmx->vcpu.kvm->arch.vm_init_tsc)
2300 tsc_base = tsc_this;
2301
2302 guest_write_tsc(0, tsc_base);
f78e0e2e 2303
e00c8cf2
AK		 2304 return 0;
2305}
2306
b7ebfb05
SY		 2307static int init_rmode(struct kvm *kvm)
2308{
2309 if (!init_rmode_tss(kvm))
2310 return 0;
2311 if (!init_rmode_identity_map(kvm))
2312 return 0;
2313 return 1;
2314}
2315
e00c8cf2
AK		 2316static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2317{
2318 struct vcpu_vmx *vmx = to_vmx(vcpu);
2319 u64 msr;
2320 int ret;
2321
5fdbf976 2322 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3200f405 2323 down_read(&vcpu->kvm->slots_lock);
b7ebfb05 2324 if (!init_rmode(vmx->vcpu.kvm)) {
e00c8cf2
AK		 2325 ret = -ENOMEM;
2326 goto out;
2327 }
2328
56b237e3 2329 vmx->vcpu.arch.rmode.vm86_active = 0;
e00c8cf2 2330
3b86cd99
JK		 2331 vmx->soft_vnmi_blocked = 0;
2332
ad312c7c 2333 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2d3ad1f4 2334 kvm_set_cr8(&vmx->vcpu, 0);
e00c8cf2
AK		 2335 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2336 if (vmx->vcpu.vcpu_id == 0)
2337 msr |= MSR_IA32_APICBASE_BSP;
2338 kvm_set_apic_base(&vmx->vcpu, msr);
2339
2340 fx_init(&vmx->vcpu);
2341
5706be0d 2342 seg_setup(VCPU_SREG_CS);
e00c8cf2
AK		 2343 /*
2344 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2345 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2346 */
2347 if (vmx->vcpu.vcpu_id == 0) {
2348 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2349 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2350 } else {
ad312c7c
ZX		 2351 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2352 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
e00c8cf2 2353 }
e00c8cf2
AK		 2354
2355 seg_setup(VCPU_SREG_DS);
2356 seg_setup(VCPU_SREG_ES);
2357 seg_setup(VCPU_SREG_FS);
2358 seg_setup(VCPU_SREG_GS);
2359 seg_setup(VCPU_SREG_SS);
2360
2361 vmcs_write16(GUEST_TR_SELECTOR, 0);
2362 vmcs_writel(GUEST_TR_BASE, 0);
2363 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2364 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2365
2366 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2367 vmcs_writel(GUEST_LDTR_BASE, 0);
2368 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2369 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2370
2371 vmcs_write32(GUEST_SYSENTER_CS, 0);
2372 vmcs_writel(GUEST_SYSENTER_ESP, 0);
2373 vmcs_writel(GUEST_SYSENTER_EIP, 0);
2374
2375 vmcs_writel(GUEST_RFLAGS, 0x02);
2376 if (vmx->vcpu.vcpu_id == 0)
5fdbf976 2377 kvm_rip_write(vcpu, 0xfff0);
e00c8cf2 2378 else
5fdbf976
MT		 2379 kvm_rip_write(vcpu, 0);
2380 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
e00c8cf2 2381
e00c8cf2
AK		 2382 vmcs_writel(GUEST_DR7, 0x400);
2383
2384 vmcs_writel(GUEST_GDTR_BASE, 0);
2385 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2386
2387 vmcs_writel(GUEST_IDTR_BASE, 0);
2388 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2389
2390 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2391 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2392 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2393
e00c8cf2
AK		 2394 /* Special registers */
2395 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2396
2397 setup_msrs(vmx);
2398
6aa8b732
AK		 2399 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
2400
f78e0e2e
SY		 2401 if (cpu_has_vmx_tpr_shadow()) {
2402 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2403 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2404 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
ad312c7c 2405 page_to_phys(vmx->vcpu.arch.apic->regs_page));
f78e0e2e
SY		 2406 vmcs_write32(TPR_THRESHOLD, 0);
2407 }
2408
2409 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2410 vmcs_write64(APIC_ACCESS_ADDR,
bfc6d222 2411 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
6aa8b732 2412
2384d2b3
SY		 2413 if (vmx->vpid != 0)
2414 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2415
ad312c7c
ZX		 2416 vmx->vcpu.arch.cr0 = 0x60000010;
2417 vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
8b9cf98c 2418 vmx_set_cr4(&vmx->vcpu, 0);
8b9cf98c 2419 vmx_set_efer(&vmx->vcpu, 0);
8b9cf98c
RR		 2420 vmx_fpu_activate(&vmx->vcpu);
2421 update_exception_bitmap(&vmx->vcpu);
6aa8b732 2422
2384d2b3
SY		 2423 vpid_sync_vcpu_all(vmx);
2424
3200f405 2425 ret = 0;
6aa8b732 2426
a89a8fb9
MG		 2427 /* HACK: Don't enable emulation on guest boot/reset */
2428 vmx->emulation_required = 0;
2429
6aa8b732 2430out:
3200f405 2431 up_read(&vcpu->kvm->slots_lock);
6aa8b732
AK		 2432 return ret;
2433}
2434
3b86cd99
JK		 2435static void enable_irq_window(struct kvm_vcpu *vcpu)
2436{
2437 u32 cpu_based_vm_exec_control;
2438
2439 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2440 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2441 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2442}
2443
2444static void enable_nmi_window(struct kvm_vcpu *vcpu)
2445{
2446 u32 cpu_based_vm_exec_control;
2447
2448 if (!cpu_has_virtual_nmis()) {
2449 enable_irq_window(vcpu);
2450 return;
2451 }
2452
2453 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2454 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2455 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2456}
2457
66fd3f7f 2458static void vmx_inject_irq(struct kvm_vcpu *vcpu)
85f455f7 2459{
9c8cba37 2460 struct vcpu_vmx *vmx = to_vmx(vcpu);
66fd3f7f
GN		 2461 uint32_t intr;
2462 int irq = vcpu->arch.interrupt.nr;
9c8cba37 2463
2714d1d3
FEL		 2464 KVMTRACE_1D(INJ_VIRQ, vcpu, (u32)irq, handler);
2465
fa89a817 2466 ++vcpu->stat.irq_injections;
56b237e3 2467 if (vcpu->arch.rmode.vm86_active) {
9c8cba37
AK		 2468 vmx->rmode.irq.pending = true;
2469 vmx->rmode.irq.vector = irq;
5fdbf976 2470 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
ae0bb3e0
GN		 2471 if (vcpu->arch.interrupt.soft)
2472 vmx->rmode.irq.rip +=
2473 vmx->vcpu.arch.event_exit_inst_len;
9c5623e3
AK		 2474 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2475 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2476 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
5fdbf976 2477 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
85f455f7
ED		 2478 return;
2479 }
66fd3f7f
GN		 2480 intr = irq | INTR_INFO_VALID_MASK;
2481 if (vcpu->arch.interrupt.soft) {
2482 intr |= INTR_TYPE_SOFT_INTR;
2483 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2484 vmx->vcpu.arch.event_exit_inst_len);
2485 } else
2486 intr |= INTR_TYPE_EXT_INTR;
2487 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
85f455f7
ED		 2488}
2489
f08864b4
SY		 2490static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2491{
66a5a347
JK		 2492 struct vcpu_vmx *vmx = to_vmx(vcpu);
2493
3b86cd99
JK		 2494 if (!cpu_has_virtual_nmis()) {
2495 /*
2496 * Tracking the NMI-blocked state in software is built upon
2497 * finding the next open IRQ window. This, in turn, depends on
2498 * well-behaving guests: They have to keep IRQs disabled at
2499 * least as long as the NMI handler runs. Otherwise we may
2500 * cause NMI nesting, maybe breaking the guest. But as this is
2501 * highly unlikely, we can live with the residual risk.
2502 */
2503 vmx->soft_vnmi_blocked = 1;
2504 vmx->vnmi_blocked_time = 0;
2505 }
2506
487b391d 2507 ++vcpu->stat.nmi_injections;
56b237e3 2508 if (vcpu->arch.rmode.vm86_active) {
66a5a347
JK		 2509 vmx->rmode.irq.pending = true;
2510 vmx->rmode.irq.vector = NMI_VECTOR;
2511 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2512 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2513 NMI_VECTOR | INTR_TYPE_SOFT_INTR |
2514 INTR_INFO_VALID_MASK);
2515 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2516 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2517 return;
2518 }
f08864b4
SY		 2519 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2520 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
f08864b4
SY		 2521}
2522
c4282df9 2523static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
33f089ca 2524{
3b86cd99 2525 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
c4282df9 2526 return 0;
33f089ca 2527
c4282df9
GN		 2528 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2529 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS |
2530 GUEST_INTR_STATE_NMI));
33f089ca
JK		 2531}
2532
78646121
GN		 2533static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2534{
c4282df9
GN		 2535 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2536 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2537 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
78646121
GN		 2538}
2539
cbc94022
IE		 2540static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2541{
2542 int ret;
2543 struct kvm_userspace_memory_region tss_mem = {
6fe63979 2544 .slot = TSS_PRIVATE_MEMSLOT,
cbc94022
IE		 2545 .guest_phys_addr = addr,
2546 .memory_size = PAGE_SIZE * 3,
2547 .flags = 0,
2548 };
2549
2550 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2551 if (ret)
2552 return ret;
bfc6d222 2553 kvm->arch.tss_addr = addr;
cbc94022
IE		 2554 return 0;
2555}
2556
6aa8b732
AK		 2557static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2558 int vec, u32 err_code)
2559{
b3f37707
NK		 2560 /*
2561 * Instruction with address size override prefix opcode 0x67
2562 * Cause the #SS fault with 0 error code in VM86 mode.
2563 */
2564 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
3427318f 2565 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE)
6aa8b732 2566 return 1;
77ab6db0
JK		 2567 /*
2568 * Forward all other exceptions that are valid in real mode.
2569 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2570 * the required debugging infrastructure rework.
2571 */
2572 switch (vec) {
77ab6db0 2573 case DB_VECTOR:
d0bfb940
JK		 2574 if (vcpu->guest_debug &
2575 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2576 return 0;
2577 kvm_queue_exception(vcpu, vec);
2578 return 1;
77ab6db0 2579 case BP_VECTOR:
d0bfb940
JK		 2580 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2581 return 0;
2582 /* fall through */
2583 case DE_VECTOR:
77ab6db0
JK		 2584 case OF_VECTOR:
2585 case BR_VECTOR:
2586 case UD_VECTOR:
2587 case DF_VECTOR:
2588 case SS_VECTOR:
2589 case GP_VECTOR:
2590 case MF_VECTOR:
2591 kvm_queue_exception(vcpu, vec);
2592 return 1;
2593 }
6aa8b732
AK		 2594 return 0;
2595}
2596
a0861c02
AK		 2597/*
2598 * Trigger machine check on the host. We assume all the MSRs are already set up
2599 * by the CPU and that we still run on the same CPU as the MCE occurred on.
2600 * We pass a fake environment to the machine check handler because we want
2601 * the guest to be always treated like user space, no matter what context
2602 * it used internally.
2603 */
2604static void kvm_machine_check(void)
2605{
2606#if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
2607 struct pt_regs regs = {
2608 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
2609 .flags = X86_EFLAGS_IF,
2610 };
2611
2612 do_machine_check(&regs, 0);
2613#endif
2614}
2615
2616static int handle_machine_check(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2617{
2618 /* already handled by vcpu_run */
2619 return 1;
2620}
2621
6aa8b732
AK		 2622static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2623{
1155f76a 2624 struct vcpu_vmx *vmx = to_vmx(vcpu);
d0bfb940 2625 u32 intr_info, ex_no, error_code;
42dbaa5a 2626 unsigned long cr2, rip, dr6;
6aa8b732
AK		 2627 u32 vect_info;
2628 enum emulation_result er;
2629
1155f76a 2630 vect_info = vmx->idt_vectoring_info;
6aa8b732
AK		 2631 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2632
a0861c02
AK		 2633 if (is_machine_check(intr_info))
2634 return handle_machine_check(vcpu, kvm_run);
2635
6aa8b732 2636 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
d77c26fc 2637 !is_page_fault(intr_info))
6aa8b732 2638 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
b8688d51 2639 "intr info 0x%x\n", __func__, vect_info, intr_info);
6aa8b732 2640
e4a41889 2641 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
1b6269db 2642 return 1; /* already handled by vmx_vcpu_run() */
2ab455cc
AL		 2643
2644 if (is_no_device(intr_info)) {
5fd86fcf 2645 vmx_fpu_activate(vcpu);
2ab455cc
AL		 2646 return 1;
2647 }
2648
7aa81cc0 2649 if (is_invalid_opcode(intr_info)) {
571008da 2650 er = emulate_instruction(vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
7aa81cc0 2651 if (er != EMULATE_DONE)
7ee5d940 2652 kvm_queue_exception(vcpu, UD_VECTOR);
7aa81cc0
AL		 2653 return 1;
2654 }
2655
6aa8b732 2656 error_code = 0;
5fdbf976 2657 rip = kvm_rip_read(vcpu);
2e11384c 2658 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
6aa8b732
AK		 2659 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2660 if (is_page_fault(intr_info)) {
1439442c 2661 /* EPT won't cause page fault directly */
089d034e 2662 if (enable_ept)
1439442c 2663 BUG();
6aa8b732 2664 cr2 = vmcs_readl(EXIT_QUALIFICATION);
2714d1d3
FEL		 2665 KVMTRACE_3D(PAGE_FAULT, vcpu, error_code, (u32)cr2,
2666 (u32)((u64)cr2 >> 32), handler);
3298b75c 2667 if (kvm_event_needs_reinjection(vcpu))
577bdc49 2668 kvm_mmu_unprotect_page_virt(vcpu, cr2);
3067714c 2669 return kvm_mmu_page_fault(vcpu, cr2, error_code);
6aa8b732
AK		 2670 }
2671
56b237e3 2672 if (vcpu->arch.rmode.vm86_active &&
6aa8b732 2673 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
72d6e5a0 2674 error_code)) {
ad312c7c
ZX		 2675 if (vcpu->arch.halt_request) {
2676 vcpu->arch.halt_request = 0;
72d6e5a0
AK		 2677 return kvm_emulate_halt(vcpu);
2678 }
6aa8b732 2679 return 1;
72d6e5a0 2680 }
6aa8b732 2681
d0bfb940 2682 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
42dbaa5a
JK		 2683 switch (ex_no) {
2684 case DB_VECTOR:
2685 dr6 = vmcs_readl(EXIT_QUALIFICATION);
2686 if (!(vcpu->guest_debug &
2687 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
2688 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
2689 kvm_queue_exception(vcpu, DB_VECTOR);
2690 return 1;
2691 }
2692 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
2693 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
2694 /* fall through */
2695 case BP_VECTOR:
6aa8b732 2696 kvm_run->exit_reason = KVM_EXIT_DEBUG;
d0bfb940
JK		 2697 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
2698 kvm_run->debug.arch.exception = ex_no;
42dbaa5a
JK		 2699 break;
2700 default:
d0bfb940
JK		 2701 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
2702 kvm_run->ex.exception = ex_no;
2703 kvm_run->ex.error_code = error_code;
42dbaa5a 2704 break;
6aa8b732 2705 }
6aa8b732
AK		 2706 return 0;
2707}
2708
2709static int handle_external_interrupt(struct kvm_vcpu *vcpu,
2710 struct kvm_run *kvm_run)
2711{
1165f5fe 2712 ++vcpu->stat.irq_exits;
2714d1d3 2713 KVMTRACE_1D(INTR, vcpu, vmcs_read32(VM_EXIT_INTR_INFO), handler);
6aa8b732
AK		 2714 return 1;
2715}
2716
988ad74f
AK		 2717static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2718{
2719 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2720 return 0;
2721}
6aa8b732 2722
6aa8b732
AK		 2723static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2724{
bfdaab09 2725 unsigned long exit_qualification;
34c33d16 2726 int size, in, string;
039576c0 2727 unsigned port;
6aa8b732 2728
1165f5fe 2729 ++vcpu->stat.io_exits;
bfdaab09 2730 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
039576c0 2731 string = (exit_qualification & 16) != 0;
e70669ab
LV		 2732
2733 if (string) {
3427318f
LV		 2734 if (emulate_instruction(vcpu,
2735 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
e70669ab
LV		 2736 return 0;
2737 return 1;
2738 }
2739
2740 size = (exit_qualification & 7) + 1;
2741 in = (exit_qualification & 8) != 0;
039576c0 2742 port = exit_qualification >> 16;
e70669ab 2743
e93f36bc 2744 skip_emulated_instruction(vcpu);
3090dd73 2745 return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
6aa8b732
AK		 2746}
2747
102d8325
IM		 2748static void
2749vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2750{
2751 /*
2752 * Patch in the VMCALL instruction:
2753 */
2754 hypercall[0] = 0x0f;
2755 hypercall[1] = 0x01;
2756 hypercall[2] = 0xc1;
102d8325
IM		 2757}
2758
6aa8b732
AK		 2759static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2760{
bfdaab09 2761 unsigned long exit_qualification;
6aa8b732
AK		 2762 int cr;
2763 int reg;
2764
bfdaab09 2765 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6aa8b732
AK		 2766 cr = exit_qualification & 15;
2767 reg = (exit_qualification >> 8) & 15;
2768 switch ((exit_qualification >> 4) & 3) {
2769 case 0: /* mov to cr */
5fdbf976
MT		 2770 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr,
2771 (u32)kvm_register_read(vcpu, reg),
2772 (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2773 handler);
6aa8b732
AK		 2774 switch (cr) {
2775 case 0:
5fdbf976 2776 kvm_set_cr0(vcpu, kvm_register_read(vcpu, reg));
6aa8b732
AK		 2777 skip_emulated_instruction(vcpu);
2778 return 1;
2779 case 3:
5fdbf976 2780 kvm_set_cr3(vcpu, kvm_register_read(vcpu, reg));
6aa8b732
AK		 2781 skip_emulated_instruction(vcpu);
2782 return 1;
2783 case 4:
5fdbf976 2784 kvm_set_cr4(vcpu, kvm_register_read(vcpu, reg));
6aa8b732
AK		 2785 skip_emulated_instruction(vcpu);
2786 return 1;
0a5fff19
GN		 2787 case 8: {
2788 u8 cr8_prev = kvm_get_cr8(vcpu);
2789 u8 cr8 = kvm_register_read(vcpu, reg);
2790 kvm_set_cr8(vcpu, cr8);
2791 skip_emulated_instruction(vcpu);
2792 if (irqchip_in_kernel(vcpu->kvm))
2793 return 1;
2794 if (cr8_prev <= cr8)
2795 return 1;
2796 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2797 return 0;
2798 }
6aa8b732
AK		 2799 };
2800 break;
25c4c276 2801 case 2: /* clts */
5fd86fcf 2802 vmx_fpu_deactivate(vcpu);
ad312c7c
ZX		 2803 vcpu->arch.cr0 &= ~X86_CR0_TS;
2804 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
5fd86fcf 2805 vmx_fpu_activate(vcpu);
2714d1d3 2806 KVMTRACE_0D(CLTS, vcpu, handler);
25c4c276
AL		 2807 skip_emulated_instruction(vcpu);
2808 return 1;
6aa8b732
AK		 2809 case 1: /*mov from cr*/
2810 switch (cr) {
2811 case 3:
5fdbf976 2812 kvm_register_write(vcpu, reg, vcpu->arch.cr3);
2714d1d3 2813 KVMTRACE_3D(CR_READ, vcpu, (u32)cr,
5fdbf976
MT		 2814 (u32)kvm_register_read(vcpu, reg),
2815 (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2714d1d3 2816 handler);
6aa8b732
AK		 2817 skip_emulated_instruction(vcpu);
2818 return 1;
2819 case 8:
5fdbf976 2820 kvm_register_write(vcpu, reg, kvm_get_cr8(vcpu));
2714d1d3 2821 KVMTRACE_2D(CR_READ, vcpu, (u32)cr,
5fdbf976 2822 (u32)kvm_register_read(vcpu, reg), handler);
6aa8b732
AK		 2823 skip_emulated_instruction(vcpu);
2824 return 1;
2825 }
2826 break;
2827 case 3: /* lmsw */
2d3ad1f4 2828 kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
6aa8b732
AK		 2829
2830 skip_emulated_instruction(vcpu);
2831 return 1;
2832 default:
2833 break;
2834 }
2835 kvm_run->exit_reason = 0;
f0242478 2836 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
6aa8b732
AK		 2837 (int)(exit_qualification >> 4) & 3, cr);
2838 return 0;
2839}
2840
2841static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2842{
bfdaab09 2843 unsigned long exit_qualification;
6aa8b732
AK		 2844 unsigned long val;
2845 int dr, reg;
2846
42dbaa5a
JK		 2847 dr = vmcs_readl(GUEST_DR7);
2848 if (dr & DR7_GD) {
2849 /*
2850 * As the vm-exit takes precedence over the debug trap, we
2851 * need to emulate the latter, either for the host or the
2852 * guest debugging itself.
2853 */
2854 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
2855 kvm_run->debug.arch.dr6 = vcpu->arch.dr6;
2856 kvm_run->debug.arch.dr7 = dr;
2857 kvm_run->debug.arch.pc =
2858 vmcs_readl(GUEST_CS_BASE) +
2859 vmcs_readl(GUEST_RIP);
2860 kvm_run->debug.arch.exception = DB_VECTOR;
2861 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2862 return 0;
2863 } else {
2864 vcpu->arch.dr7 &= ~DR7_GD;
2865 vcpu->arch.dr6 |= DR6_BD;
2866 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2867 kvm_queue_exception(vcpu, DB_VECTOR);
2868 return 1;
2869 }
2870 }
2871
bfdaab09 2872 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
42dbaa5a
JK		 2873 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
2874 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
2875 if (exit_qualification & TYPE_MOV_FROM_DR) {
6aa8b732 2876 switch (dr) {
42dbaa5a
JK		 2877 case 0 ... 3:
2878 val = vcpu->arch.db[dr];
2879 break;
6aa8b732 2880 case 6:
42dbaa5a 2881 val = vcpu->arch.dr6;
6aa8b732
AK		 2882 break;
2883 case 7:
42dbaa5a 2884 val = vcpu->arch.dr7;
6aa8b732
AK		 2885 break;
2886 default:
2887 val = 0;
2888 }
5fdbf976 2889 kvm_register_write(vcpu, reg, val);
2714d1d3 2890 KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
6aa8b732 2891 } else {
42dbaa5a
JK		 2892 val = vcpu->arch.regs[reg];
2893 switch (dr) {
2894 case 0 ... 3:
2895 vcpu->arch.db[dr] = val;
2896 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
2897 vcpu->arch.eff_db[dr] = val;
2898 break;
2899 case 4 ... 5:
2900 if (vcpu->arch.cr4 & X86_CR4_DE)
2901 kvm_queue_exception(vcpu, UD_VECTOR);
2902 break;
2903 case 6:
2904 if (val & 0xffffffff00000000ULL) {
2905 kvm_queue_exception(vcpu, GP_VECTOR);
2906 break;
2907 }
2908 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
2909 break;
2910 case 7:
2911 if (val & 0xffffffff00000000ULL) {
2912 kvm_queue_exception(vcpu, GP_VECTOR);
2913 break;
2914 }
2915 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
2916 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
2917 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2918 vcpu->arch.switch_db_regs =
2919 (val & DR7_BP_EN_MASK);
2920 }
2921 break;
2922 }
2923 KVMTRACE_2D(DR_WRITE, vcpu, (u32)dr, (u32)val, handler);
6aa8b732 2924 }
6aa8b732
AK		 2925 skip_emulated_instruction(vcpu);
2926 return 1;
2927}
2928
2929static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2930{
06465c5a
AK		 2931 kvm_emulate_cpuid(vcpu);
2932 return 1;
6aa8b732
AK		 2933}
2934
2935static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2936{
ad312c7c 2937 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6aa8b732
AK		 2938 u64 data;
2939
2940 if (vmx_get_msr(vcpu, ecx, &data)) {
c1a5d4f9 2941 kvm_inject_gp(vcpu, 0);
6aa8b732
AK		 2942 return 1;
2943 }
2944
2714d1d3
FEL		 2945 KVMTRACE_3D(MSR_READ, vcpu, ecx, (u32)data, (u32)(data >> 32),
2946 handler);
2947
6aa8b732 2948 /* FIXME: handling of bits 32:63 of rax, rdx */
ad312c7c
ZX		 2949 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
2950 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
6aa8b732
AK		 2951 skip_emulated_instruction(vcpu);
2952 return 1;
2953}
2954
2955static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2956{
ad312c7c
ZX		 2957 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2958 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
2959 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6aa8b732 2960
2714d1d3
FEL		 2961 KVMTRACE_3D(MSR_WRITE, vcpu, ecx, (u32)data, (u32)(data >> 32),
2962 handler);
2963
6aa8b732 2964 if (vmx_set_msr(vcpu, ecx, data) != 0) {
c1a5d4f9 2965 kvm_inject_gp(vcpu, 0);
6aa8b732
AK		 2966 return 1;
2967 }
2968
2969 skip_emulated_instruction(vcpu);
2970 return 1;
2971}
2972
6e5d865c
YS		 2973static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu,
2974 struct kvm_run *kvm_run)
2975{
2976 return 1;
2977}
2978
6aa8b732
AK		 2979static int handle_interrupt_window(struct kvm_vcpu *vcpu,
2980 struct kvm_run *kvm_run)
2981{
85f455f7
ED		 2982 u32 cpu_based_vm_exec_control;
2983
2984 /* clear pending irq */
2985 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2986 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2987 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2714d1d3
FEL		 2988
2989 KVMTRACE_0D(PEND_INTR, vcpu, handler);
a26bf12a 2990 ++vcpu->stat.irq_window_exits;
2714d1d3 2991
c1150d8c
DL		 2992 /*
2993 * If the user space waits to inject interrupts, exit as soon as
2994 * possible
2995 */
8061823a
GN		 2996 if (!irqchip_in_kernel(vcpu->kvm) &&
2997 kvm_run->request_interrupt_window &&
2998 !kvm_cpu_has_interrupt(vcpu)) {
c1150d8c 2999 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
c1150d8c
DL		 3000 return 0;
3001 }
6aa8b732
AK		 3002 return 1;
3003}
3004
3005static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3006{
3007 skip_emulated_instruction(vcpu);
d3bef15f 3008 return kvm_emulate_halt(vcpu);
6aa8b732
AK		 3009}
3010
c21415e8
IM		 3011static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3012{
510043da 3013 skip_emulated_instruction(vcpu);
7aa81cc0
AL		 3014 kvm_emulate_hypercall(vcpu);
3015 return 1;
c21415e8
IM		 3016}
3017
e3c7cb6a
AK		 3018static int handle_vmx_insn(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3019{
3020 kvm_queue_exception(vcpu, UD_VECTOR);
3021 return 1;
3022}
3023
a7052897
MT		 3024static int handle_invlpg(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3025{
f9c617f6 3026 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
a7052897
MT		 3027
3028 kvm_mmu_invlpg(vcpu, exit_qualification);
3029 skip_emulated_instruction(vcpu);
3030 return 1;
3031}
3032
e5edaa01
ED		 3033static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3034{
3035 skip_emulated_instruction(vcpu);
3036 /* TODO: Add support for VT-d/pass-through device */
3037 return 1;
3038}
3039
f78e0e2e
SY		 3040static int handle_apic_access(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3041{
f9c617f6 3042 unsigned long exit_qualification;
f78e0e2e
SY		 3043 enum emulation_result er;
3044 unsigned long offset;
3045
f9c617f6 3046 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
f78e0e2e
SY		 3047 offset = exit_qualification & 0xffful;
3048
3049 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
3050
3051 if (er != EMULATE_DONE) {
3052 printk(KERN_ERR
3053 "Fail to handle apic access vmexit! Offset is 0x%lx\n",
3054 offset);
3055 return -ENOTSUPP;
3056 }
3057 return 1;
3058}
3059
37817f29
IE		 3060static int handle_task_switch(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3061{
60637aac 3062 struct vcpu_vmx *vmx = to_vmx(vcpu);
37817f29
IE		 3063 unsigned long exit_qualification;
3064 u16 tss_selector;
64a7ec06
GN		 3065 int reason, type, idt_v;
3066
3067 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3068 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
37817f29
IE		 3069
3070 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3071
3072 reason = (u32)exit_qualification >> 30;
64a7ec06
GN		 3073 if (reason == TASK_SWITCH_GATE && idt_v) {
3074 switch (type) {
3075 case INTR_TYPE_NMI_INTR:
3076 vcpu->arch.nmi_injected = false;
3077 if (cpu_has_virtual_nmis())
3078 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3079 GUEST_INTR_STATE_NMI);
3080 break;
3081 case INTR_TYPE_EXT_INTR:
66fd3f7f 3082 case INTR_TYPE_SOFT_INTR:
64a7ec06
GN		 3083 kvm_clear_interrupt_queue(vcpu);
3084 break;
3085 case INTR_TYPE_HARD_EXCEPTION:
3086 case INTR_TYPE_SOFT_EXCEPTION:
3087 kvm_clear_exception_queue(vcpu);
3088 break;
3089 default:
3090 break;
3091 }
60637aac 3092 }
37817f29
IE		 3093 tss_selector = exit_qualification;
3094
64a7ec06
GN		 3095 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3096 type != INTR_TYPE_EXT_INTR &&
3097 type != INTR_TYPE_NMI_INTR))
3098 skip_emulated_instruction(vcpu);
3099
42dbaa5a
JK		 3100 if (!kvm_task_switch(vcpu, tss_selector, reason))
3101 return 0;
3102
3103 /* clear all local breakpoint enable flags */
3104 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3105
3106 /*
3107 * TODO: What about debug traps on tss switch?
3108 * Are we supposed to inject them and update dr6?
3109 */
3110
3111 return 1;
37817f29
IE		 3112}
3113
1439442c
SY		 3114static int handle_ept_violation(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3115{
f9c617f6 3116 unsigned long exit_qualification;
1439442c 3117 gpa_t gpa;
1439442c 3118 int gla_validity;
1439442c 3119
f9c617f6 3120 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
1439442c
SY		 3121
3122 if (exit_qualification & (1 << 6)) {
3123 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3124 return -ENOTSUPP;
3125 }
3126
3127 gla_validity = (exit_qualification >> 7) & 0x3;
3128 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3129 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3130 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3131 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
f9c617f6 3132 vmcs_readl(GUEST_LINEAR_ADDRESS));
1439442c
SY		 3133 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3134 (long unsigned int)exit_qualification);
3135 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3136 kvm_run->hw.hardware_exit_reason = 0;
3137 return -ENOTSUPP;
3138 }
3139
3140 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
49cd7d22 3141 return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
1439442c
SY		 3142}
3143
f08864b4
SY		 3144static int handle_nmi_window(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3145{
3146 u32 cpu_based_vm_exec_control;
3147
3148 /* clear pending NMI */
3149 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3150 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3151 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3152 ++vcpu->stat.nmi_window_exits;
3153
3154 return 1;
3155}
3156
ea953ef0
MG		 3157static void handle_invalid_guest_state(struct kvm_vcpu *vcpu,
3158 struct kvm_run *kvm_run)
3159{
8b3079a5
AK		 3160 struct vcpu_vmx *vmx = to_vmx(vcpu);
3161 enum emulation_result err = EMULATE_DONE;
ea953ef0 3162
ea953ef0 3163 local_irq_enable();
34f0c1ad 3164 preempt_enable();
ea953ef0
MG		 3165
3166 while (!guest_state_valid(vcpu)) {
3167 err = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
3168
1d5a4d9b
GT		 3169 if (err == EMULATE_DO_MMIO)
3170 break;
3171
3172 if (err != EMULATE_DONE) {
3173 kvm_report_emulation_failure(vcpu, "emulation failure");
263799a3 3174 break;
ea953ef0
MG		 3175 }
3176
3177 if (signal_pending(current))
3178 break;
3179 if (need_resched())
3180 schedule();
3181 }
3182
ea953ef0 3183 preempt_disable();
34f0c1ad 3184 local_irq_disable();
8b3079a5
AK		 3185
3186 vmx->invalid_state_emulation_result = err;
ea953ef0
MG		 3187}
3188
6aa8b732
AK		 3189/*
3190 * The exit handlers return 1 if the exit was handled fully and guest execution
3191 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
3192 * to be done to userspace and return 0.
3193 */
3194static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
3195 struct kvm_run *kvm_run) = {
3196 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
3197 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
988ad74f 3198 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
f08864b4 3199 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
6aa8b732 3200 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
6aa8b732
AK		 3201 [EXIT_REASON_CR_ACCESS] = handle_cr,
3202 [EXIT_REASON_DR_ACCESS] = handle_dr,
3203 [EXIT_REASON_CPUID] = handle_cpuid,
3204 [EXIT_REASON_MSR_READ] = handle_rdmsr,
3205 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
3206 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
3207 [EXIT_REASON_HLT] = handle_halt,
a7052897 3208 [EXIT_REASON_INVLPG] = handle_invlpg,
c21415e8 3209 [EXIT_REASON_VMCALL] = handle_vmcall,
e3c7cb6a
AK		 3210 [EXIT_REASON_VMCLEAR] = handle_vmx_insn,
3211 [EXIT_REASON_VMLAUNCH] = handle_vmx_insn,
3212 [EXIT_REASON_VMPTRLD] = handle_vmx_insn,
3213 [EXIT_REASON_VMPTRST] = handle_vmx_insn,
3214 [EXIT_REASON_VMREAD] = handle_vmx_insn,
3215 [EXIT_REASON_VMRESUME] = handle_vmx_insn,
3216 [EXIT_REASON_VMWRITE] = handle_vmx_insn,
3217 [EXIT_REASON_VMOFF] = handle_vmx_insn,
3218 [EXIT_REASON_VMON] = handle_vmx_insn,
f78e0e2e
SY		 3219 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
3220 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
e5edaa01 3221 [EXIT_REASON_WBINVD] = handle_wbinvd,
37817f29 3222 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
1439442c 3223 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
a0861c02 3224 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
6aa8b732
AK		 3225};
3226
3227static const int kvm_vmx_max_exit_handlers =
50a3485c 3228 ARRAY_SIZE(kvm_vmx_exit_handlers);
6aa8b732
AK		 3229
3230/*
3231 * The guest has exited. See if we can fix it or if we need userspace
3232 * assistance.
3233 */
6062d012 3234static int vmx_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
6aa8b732 3235{
29bd8a78 3236 struct vcpu_vmx *vmx = to_vmx(vcpu);
a0861c02 3237 u32 exit_reason = vmx->exit_reason;
1155f76a 3238 u32 vectoring_info = vmx->idt_vectoring_info;
29bd8a78 3239
5fdbf976
MT		 3240 KVMTRACE_3D(VMEXIT, vcpu, exit_reason, (u32)kvm_rip_read(vcpu),
3241 (u32)((u64)kvm_rip_read(vcpu) >> 32), entryexit);
2714d1d3 3242
1d5a4d9b
GT		 3243 /* If we need to emulate an MMIO from handle_invalid_guest_state
3244 * we just return 0 */
10f32d84
AK		 3245 if (vmx->emulation_required && emulate_invalid_guest_state) {
3246 if (guest_state_valid(vcpu))
3247 vmx->emulation_required = 0;
8b3079a5 3248 return vmx->invalid_state_emulation_result != EMULATE_DO_MMIO;
10f32d84 3249 }
1d5a4d9b 3250
1439442c
SY		 3251 /* Access CR3 don't cause VMExit in paging mode, so we need
3252 * to sync with guest real CR3. */
089d034e 3253 if (enable_ept && is_paging(vcpu)) {
1439442c
SY		 3254 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3255 ept_load_pdptrs(vcpu);
3256 }
3257
29bd8a78
AK		 3258 if (unlikely(vmx->fail)) {
3259 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3260 kvm_run->fail_entry.hardware_entry_failure_reason
3261 = vmcs_read32(VM_INSTRUCTION_ERROR);
3262 return 0;
3263 }
6aa8b732 3264
d77c26fc 3265 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
1439442c 3266 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
60637aac
JK		 3267 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3268 exit_reason != EXIT_REASON_TASK_SWITCH))
3269 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3270 "(0x%x) and exit reason is 0x%x\n",
3271 __func__, vectoring_info, exit_reason);
3b86cd99
JK		 3272
3273 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
c4282df9 3274 if (vmx_interrupt_allowed(vcpu)) {
3b86cd99 3275 vmx->soft_vnmi_blocked = 0;
3b86cd99 3276 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
4531220b 3277 vcpu->arch.nmi_pending) {
3b86cd99
JK		 3278 /*
3279 * This CPU don't support us in finding the end of an
3280 * NMI-blocked window if the guest runs with IRQs
3281 * disabled. So we pull the trigger after 1 s of
3282 * futile waiting, but inform the user about this.
3283 */
3284 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3285 "state on VCPU %d after 1 s timeout\n",
3286 __func__, vcpu->vcpu_id);
3287 vmx->soft_vnmi_blocked = 0;
3b86cd99 3288 }
3b86cd99
JK		 3289 }
3290
6aa8b732
AK		 3291 if (exit_reason < kvm_vmx_max_exit_handlers
3292 && kvm_vmx_exit_handlers[exit_reason])
3293 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
3294 else {
3295 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3296 kvm_run->hw.hardware_exit_reason = exit_reason;
3297 }
3298 return 0;
3299}
3300
95ba8273 3301static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6e5d865c 3302{
95ba8273 3303 if (irr == -1 || tpr < irr) {
6e5d865c
YS		 3304 vmcs_write32(TPR_THRESHOLD, 0);
3305 return;
3306 }
3307
95ba8273 3308 vmcs_write32(TPR_THRESHOLD, irr);
6e5d865c
YS		 3309}
3310
cf393f75
AK		 3311static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3312{
3313 u32 exit_intr_info;
7b4a25cb 3314 u32 idt_vectoring_info = vmx->idt_vectoring_info;
cf393f75
AK		 3315 bool unblock_nmi;
3316 u8 vector;
668f612f
AK		 3317 int type;
3318 bool idtv_info_valid;
cf393f75
AK		 3319
3320 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
20f65983 3321
a0861c02
AK		 3322 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
3323
3324 /* Handle machine checks before interrupts are enabled */
3325 if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3326 || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3327 && is_machine_check(exit_intr_info)))
3328 kvm_machine_check();
3329
20f65983
GN		 3330 /* We need to handle NMIs before interrupts are enabled */
3331 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3332 (exit_intr_info & INTR_INFO_VALID_MASK)) {
3333 KVMTRACE_0D(NMI, &vmx->vcpu, handler);
3334 asm("int $2");
3335 }
3336
3337 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3338
cf393f75
AK		 3339 if (cpu_has_virtual_nmis()) {
3340 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3341 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3342 /*
7b4a25cb 3343 * SDM 3: 27.7.1.2 (September 2008)
cf393f75
AK		 3344 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3345 * a guest IRET fault.
7b4a25cb
GN		 3346 * SDM 3: 23.2.2 (September 2008)
3347 * Bit 12 is undefined in any of the following cases:
3348 * If the VM exit sets the valid bit in the IDT-vectoring
3349 * information field.
3350 * If the VM exit is due to a double fault.
cf393f75 3351 */
7b4a25cb
GN		 3352 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3353 vector != DF_VECTOR && !idtv_info_valid)
cf393f75
AK		 3354 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3355 GUEST_INTR_STATE_NMI);
3b86cd99
JK		 3356 } else if (unlikely(vmx->soft_vnmi_blocked))
3357 vmx->vnmi_blocked_time +=
3358 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
668f612f 3359
37b96e98
GN		 3360 vmx->vcpu.arch.nmi_injected = false;
3361 kvm_clear_exception_queue(&vmx->vcpu);
3362 kvm_clear_interrupt_queue(&vmx->vcpu);
3363
3364 if (!idtv_info_valid)
3365 return;
3366
668f612f
AK		 3367 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3368 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
37b96e98 3369
64a7ec06 3370 switch (type) {
37b96e98
GN		 3371 case INTR_TYPE_NMI_INTR:
3372 vmx->vcpu.arch.nmi_injected = true;
668f612f 3373 /*
7b4a25cb 3374 * SDM 3: 27.7.1.2 (September 2008)
37b96e98
GN		 3375 * Clear bit "block by NMI" before VM entry if a NMI
3376 * delivery faulted.
668f612f 3377 */
37b96e98
GN		 3378 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3379 GUEST_INTR_STATE_NMI);
3380 break;
37b96e98 3381 case INTR_TYPE_SOFT_EXCEPTION:
66fd3f7f
GN		 3382 vmx->vcpu.arch.event_exit_inst_len =
3383 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3384 /* fall through */
3385 case INTR_TYPE_HARD_EXCEPTION:
35920a35 3386 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
37b96e98
GN		 3387 u32 err = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3388 kvm_queue_exception_e(&vmx->vcpu, vector, err);
35920a35
AK		 3389 } else
3390 kvm_queue_exception(&vmx->vcpu, vector);
37b96e98 3391 break;
66fd3f7f
GN		 3392 case INTR_TYPE_SOFT_INTR:
3393 vmx->vcpu.arch.event_exit_inst_len =
3394 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3395 /* fall through */
37b96e98 3396 case INTR_TYPE_EXT_INTR:
66fd3f7f
GN		 3397 kvm_queue_interrupt(&vmx->vcpu, vector,
3398 type == INTR_TYPE_SOFT_INTR);
37b96e98
GN		 3399 break;
3400 default:
3401 break;
f7d9238f 3402 }
cf393f75
AK		 3403}
3404
9c8cba37
AK		 3405/*
3406 * Failure to inject an interrupt should give us the information
3407 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
3408 * when fetching the interrupt redirection bitmap in the real-mode
3409 * tss, this doesn't happen. So we do it ourselves.
3410 */
3411static void fixup_rmode_irq(struct vcpu_vmx *vmx)
3412{
3413 vmx->rmode.irq.pending = 0;
5fdbf976 3414 if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
9c8cba37 3415 return;
5fdbf976 3416 kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
9c8cba37
AK		 3417 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
3418 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
3419 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
3420 return;
3421 }
3422 vmx->idt_vectoring_info =
3423 VECTORING_INFO_VALID_MASK
3424 | INTR_TYPE_EXT_INTR
3425 | vmx->rmode.irq.vector;
3426}
3427
c801949d
AK		 3428#ifdef CONFIG_X86_64
3429#define R "r"
3430#define Q "q"
3431#else
3432#define R "e"
3433#define Q "l"
3434#endif
3435
04d2cc77 3436static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6aa8b732 3437{
a2fa3e9f 3438 struct vcpu_vmx *vmx = to_vmx(vcpu);
e6adf283 3439
3b86cd99
JK		 3440 /* Record the guest's net vcpu time for enforced NMI injections. */
3441 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3442 vmx->entry_time = ktime_get();
3443
a89a8fb9
MG		 3444 /* Handle invalid guest state instead of entering VMX */
3445 if (vmx->emulation_required && emulate_invalid_guest_state) {
3446 handle_invalid_guest_state(vcpu, kvm_run);
3447 return;
3448 }
3449
5fdbf976
MT		 3450 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3451 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3452 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3453 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3454
787ff736
GN		 3455 /* When single-stepping over STI and MOV SS, we must clear the
3456 * corresponding interruptibility bits in the guest state. Otherwise
3457 * vmentry fails as it then expects bit 14 (BS) in pending debug
3458 * exceptions being set, but that's not correct for the guest debugging
3459 * case. */
3460 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3461 vmx_set_interrupt_shadow(vcpu, 0);
3462
e6adf283
AK		 3463 /*
3464 * Loading guest fpu may have cleared host cr0.ts
3465 */
3466 vmcs_writel(HOST_CR0, read_cr0());
3467
42dbaa5a
JK		 3468 set_debugreg(vcpu->arch.dr6, 6);
3469
d77c26fc 3470 asm(
6aa8b732 3471 /* Store host registers */
c801949d
AK		 3472 "push %%"R"dx; push %%"R"bp;"
3473 "push %%"R"cx \n\t"
313dbd49
AK		 3474 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3475 "je 1f \n\t"
3476 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
4ecac3fd 3477 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
313dbd49 3478 "1: \n\t"
6aa8b732 3479 /* Check if vmlaunch of vmresume is needed */
e08aa78a 3480 "cmpl $0, %c[launched](%0) \n\t"
6aa8b732 3481 /* Load guest registers. Don't clobber flags. */
c801949d
AK		 3482 "mov %c[cr2](%0), %%"R"ax \n\t"
3483 "mov %%"R"ax, %%cr2 \n\t"
3484 "mov %c[rax](%0), %%"R"ax \n\t"
3485 "mov %c[rbx](%0), %%"R"bx \n\t"
3486 "mov %c[rdx](%0), %%"R"dx \n\t"
3487 "mov %c[rsi](%0), %%"R"si \n\t"
3488 "mov %c[rdi](%0), %%"R"di \n\t"
3489 "mov %c[rbp](%0), %%"R"bp \n\t"
05b3e0c2 3490#ifdef CONFIG_X86_64
e08aa78a
AK		 3491 "mov %c[r8](%0), %%r8 \n\t"
3492 "mov %c[r9](%0), %%r9 \n\t"
3493 "mov %c[r10](%0), %%r10 \n\t"
3494 "mov %c[r11](%0), %%r11 \n\t"
3495 "mov %c[r12](%0), %%r12 \n\t"
3496 "mov %c[r13](%0), %%r13 \n\t"
3497 "mov %c[r14](%0), %%r14 \n\t"
3498 "mov %c[r15](%0), %%r15 \n\t"
6aa8b732 3499#endif
c801949d
AK		 3500 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3501
6aa8b732 3502 /* Enter guest mode */
cd2276a7 3503 "jne .Llaunched \n\t"
4ecac3fd 3504 __ex(ASM_VMX_VMLAUNCH) "\n\t"
cd2276a7 3505 "jmp .Lkvm_vmx_return \n\t"
4ecac3fd 3506 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
cd2276a7 3507 ".Lkvm_vmx_return: "
6aa8b732 3508 /* Save guest registers, load host registers, keep flags */
c801949d
AK		 3509 "xchg %0, (%%"R"sp) \n\t"
3510 "mov %%"R"ax, %c[rax](%0) \n\t"
3511 "mov %%"R"bx, %c[rbx](%0) \n\t"
3512 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3513 "mov %%"R"dx, %c[rdx](%0) \n\t"
3514 "mov %%"R"si, %c[rsi](%0) \n\t"
3515 "mov %%"R"di, %c[rdi](%0) \n\t"
3516 "mov %%"R"bp, %c[rbp](%0) \n\t"
05b3e0c2 3517#ifdef CONFIG_X86_64
e08aa78a
AK		 3518 "mov %%r8, %c[r8](%0) \n\t"
3519 "mov %%r9, %c[r9](%0) \n\t"
3520 "mov %%r10, %c[r10](%0) \n\t"
3521 "mov %%r11, %c[r11](%0) \n\t"
3522 "mov %%r12, %c[r12](%0) \n\t"
3523 "mov %%r13, %c[r13](%0) \n\t"
3524 "mov %%r14, %c[r14](%0) \n\t"
3525 "mov %%r15, %c[r15](%0) \n\t"
6aa8b732 3526#endif
c801949d
AK		 3527 "mov %%cr2, %%"R"ax \n\t"
3528 "mov %%"R"ax, %c[cr2](%0) \n\t"
3529
3530 "pop %%"R"bp; pop %%"R"bp; pop %%"R"dx \n\t"
e08aa78a
AK		 3531 "setbe %c[fail](%0) \n\t"
3532 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
3533 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
3534 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
313dbd49 3535 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
ad312c7c
ZX		 3536 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
3537 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
3538 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
3539 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
3540 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
3541 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
3542 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
05b3e0c2 3543#ifdef CONFIG_X86_64
ad312c7c
ZX		 3544 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
3545 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
3546 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
3547 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
3548 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
3549 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
3550 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
3551 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
6aa8b732 3552#endif
ad312c7c 3553 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
c2036300 3554 : "cc", "memory"
c801949d 3555 , R"bx", R"di", R"si"
c2036300 3556#ifdef CONFIG_X86_64
c2036300
LV		 3557 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
3558#endif
3559 );
6aa8b732 3560
5fdbf976
MT		 3561 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3562 vcpu->arch.regs_dirty = 0;
3563
42dbaa5a
JK		 3564 get_debugreg(vcpu->arch.dr6, 6);
3565
1155f76a 3566 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9c8cba37
AK		 3567 if (vmx->rmode.irq.pending)
3568 fixup_rmode_irq(vmx);
1155f76a 3569
d77c26fc 3570 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
15ad7146 3571 vmx->launched = 1;
1b6269db 3572
cf393f75 3573 vmx_complete_interrupts(vmx);
6aa8b732
AK		 3574}
3575
c801949d
AK		 3576#undef R
3577#undef Q
3578
6aa8b732
AK		 3579static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
3580{
a2fa3e9f
GH		 3581 struct vcpu_vmx *vmx = to_vmx(vcpu);
3582
3583 if (vmx->vmcs) {
543e4243 3584 vcpu_clear(vmx);
a2fa3e9f
GH		 3585 free_vmcs(vmx->vmcs);
3586 vmx->vmcs = NULL;
6aa8b732
AK		 3587 }
3588}
3589
3590static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
3591{
fb3f0f51
RR		 3592 struct vcpu_vmx *vmx = to_vmx(vcpu);
3593
2384d2b3
SY		 3594 spin_lock(&vmx_vpid_lock);
3595 if (vmx->vpid != 0)
3596 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3597 spin_unlock(&vmx_vpid_lock);
6aa8b732 3598 vmx_free_vmcs(vcpu);
fb3f0f51
RR		 3599 kfree(vmx->host_msrs);
3600 kfree(vmx->guest_msrs);
3601 kvm_vcpu_uninit(vcpu);
a4770347 3602 kmem_cache_free(kvm_vcpu_cache, vmx);
6aa8b732
AK		 3603}
3604
fb3f0f51 3605static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6aa8b732 3606{
fb3f0f51 3607 int err;
c16f862d 3608 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
15ad7146 3609 int cpu;
6aa8b732 3610
a2fa3e9f 3611 if (!vmx)
fb3f0f51
RR		 3612 return ERR_PTR(-ENOMEM);
3613
2384d2b3
SY		 3614 allocate_vpid(vmx);
3615
fb3f0f51
RR		 3616 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
3617 if (err)
3618 goto free_vcpu;
965b58a5 3619
a2fa3e9f 3620 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
fb3f0f51
RR		 3621 if (!vmx->guest_msrs) {
3622 err = -ENOMEM;
3623 goto uninit_vcpu;
3624 }
965b58a5 3625
a2fa3e9f
GH		 3626 vmx->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3627 if (!vmx->host_msrs)
fb3f0f51 3628 goto free_guest_msrs;
965b58a5 3629
a2fa3e9f
GH		 3630 vmx->vmcs = alloc_vmcs();
3631 if (!vmx->vmcs)
fb3f0f51 3632 goto free_msrs;
a2fa3e9f
GH		 3633
3634 vmcs_clear(vmx->vmcs);
3635
15ad7146
AK		 3636 cpu = get_cpu();
3637 vmx_vcpu_load(&vmx->vcpu, cpu);
8b9cf98c 3638 err = vmx_vcpu_setup(vmx);
fb3f0f51 3639 vmx_vcpu_put(&vmx->vcpu);
15ad7146 3640 put_cpu();
fb3f0f51
RR		 3641 if (err)
3642 goto free_vmcs;
5e4a0b3c
MT		 3643 if (vm_need_virtualize_apic_accesses(kvm))
3644 if (alloc_apic_access_page(kvm) != 0)
3645 goto free_vmcs;
fb3f0f51 3646
089d034e 3647 if (enable_ept)
b7ebfb05
SY		 3648 if (alloc_identity_pagetable(kvm) != 0)
3649 goto free_vmcs;
3650
fb3f0f51
RR		 3651 return &vmx->vcpu;
3652
3653free_vmcs:
3654 free_vmcs(vmx->vmcs);
3655free_msrs:
3656 kfree(vmx->host_msrs);
3657free_guest_msrs:
3658 kfree(vmx->guest_msrs);
3659uninit_vcpu:
3660 kvm_vcpu_uninit(&vmx->vcpu);
3661free_vcpu:
a4770347 3662 kmem_cache_free(kvm_vcpu_cache, vmx);
fb3f0f51 3663 return ERR_PTR(err);
6aa8b732
AK		 3664}
3665
002c7f7c
YS		 3666static void __init vmx_check_processor_compat(void *rtn)
3667{
3668 struct vmcs_config vmcs_conf;
3669
3670 *(int *)rtn = 0;
3671 if (setup_vmcs_config(&vmcs_conf) < 0)
3672 *(int *)rtn = -EIO;
3673 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
3674 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
3675 smp_processor_id());
3676 *(int *)rtn = -EIO;
3677 }
3678}
3679
67253af5
SY		 3680static int get_ept_level(void)
3681{
3682 return VMX_EPT_DEFAULT_GAW + 1;
3683}
3684
4b12f0de 3685static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
64d4d521 3686{
4b12f0de
SY		 3687 u64 ret;
3688
522c68c4
SY		 3689 /* For VT-d and EPT combination
3690 * 1. MMIO: always map as UC
3691 * 2. EPT with VT-d:
3692 * a. VT-d without snooping control feature: can't guarantee the
3693 * result, try to trust guest.
3694 * b. VT-d with snooping control feature: snooping control feature of
3695 * VT-d engine can guarantee the cache correctness. Just set it
3696 * to WB to keep consistent with host. So the same as item 3.
3697 * 3. EPT without VT-d: always map as WB and set IGMT=1 to keep
3698 * consistent with host MTRR
3699 */
4b12f0de
SY		 3700 if (is_mmio)
3701 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
522c68c4
SY		 3702 else if (vcpu->kvm->arch.iommu_domain &&
3703 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
3704 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
3705 VMX_EPT_MT_EPTE_SHIFT;
4b12f0de 3706 else
522c68c4
SY		 3707 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
3708 | VMX_EPT_IGMT_BIT;
4b12f0de
SY		 3709
3710 return ret;
64d4d521
SY		 3711}
3712
cbdd1bea 3713static struct kvm_x86_ops vmx_x86_ops = {
6aa8b732
AK		 3714 .cpu_has_kvm_support = cpu_has_kvm_support,
3715 .disabled_by_bios = vmx_disabled_by_bios,
3716 .hardware_setup = hardware_setup,
3717 .hardware_unsetup = hardware_unsetup,
002c7f7c 3718 .check_processor_compatibility = vmx_check_processor_compat,
6aa8b732
AK		 3719 .hardware_enable = hardware_enable,
3720 .hardware_disable = hardware_disable,
04547156 3721 .cpu_has_accelerated_tpr = report_flexpriority,
6aa8b732
AK		 3722
3723 .vcpu_create = vmx_create_vcpu,
3724 .vcpu_free = vmx_free_vcpu,
04d2cc77 3725 .vcpu_reset = vmx_vcpu_reset,
6aa8b732 3726
04d2cc77 3727 .prepare_guest_switch = vmx_save_host_state,
6aa8b732
AK		 3728 .vcpu_load = vmx_vcpu_load,
3729 .vcpu_put = vmx_vcpu_put,
3730
3731 .set_guest_debug = set_guest_debug,
3732 .get_msr = vmx_get_msr,
3733 .set_msr = vmx_set_msr,
3734 .get_segment_base = vmx_get_segment_base,
3735 .get_segment = vmx_get_segment,
3736 .set_segment = vmx_set_segment,
2e4d2653 3737 .get_cpl = vmx_get_cpl,
6aa8b732 3738 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
25c4c276 3739 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
6aa8b732 3740 .set_cr0 = vmx_set_cr0,
6aa8b732
AK		 3741 .set_cr3 = vmx_set_cr3,
3742 .set_cr4 = vmx_set_cr4,
6aa8b732 3743 .set_efer = vmx_set_efer,
6aa8b732
AK		 3744 .get_idt = vmx_get_idt,
3745 .set_idt = vmx_set_idt,
3746 .get_gdt = vmx_get_gdt,
3747 .set_gdt = vmx_set_gdt,
5fdbf976 3748 .cache_reg = vmx_cache_reg,
6aa8b732
AK		 3749 .get_rflags = vmx_get_rflags,
3750 .set_rflags = vmx_set_rflags,
3751
3752 .tlb_flush = vmx_flush_tlb,
6aa8b732 3753
6aa8b732 3754 .run = vmx_vcpu_run,
6062d012 3755 .handle_exit = vmx_handle_exit,
6aa8b732 3756 .skip_emulated_instruction = skip_emulated_instruction,
2809f5d2
GC		 3757 .set_interrupt_shadow = vmx_set_interrupt_shadow,
3758 .get_interrupt_shadow = vmx_get_interrupt_shadow,
102d8325 3759 .patch_hypercall = vmx_patch_hypercall,
2a8067f1 3760 .set_irq = vmx_inject_irq,
95ba8273 3761 .set_nmi = vmx_inject_nmi,
298101da 3762 .queue_exception = vmx_queue_exception,
78646121 3763 .interrupt_allowed = vmx_interrupt_allowed,
95ba8273
GN		 3764 .nmi_allowed = vmx_nmi_allowed,
3765 .enable_nmi_window = enable_nmi_window,
3766 .enable_irq_window = enable_irq_window,
3767 .update_cr8_intercept = update_cr8_intercept,
95ba8273 3768
cbc94022 3769 .set_tss_addr = vmx_set_tss_addr,
67253af5 3770 .get_tdp_level = get_ept_level,
4b12f0de 3771 .get_mt_mask = vmx_get_mt_mask,
6aa8b732
AK		 3772};
3773
3774static int __init vmx_init(void)
3775{
fdef3ad1
HQ		 3776 int r;
3777
3e7c73e9 3778 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
fdef3ad1
HQ		 3779 if (!vmx_io_bitmap_a)
3780 return -ENOMEM;
3781
3e7c73e9 3782 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
fdef3ad1
HQ		 3783 if (!vmx_io_bitmap_b) {
3784 r = -ENOMEM;
3785 goto out;
3786 }
3787
5897297b
AK		 3788 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
3789 if (!vmx_msr_bitmap_legacy) {
25c5f225
SY		 3790 r = -ENOMEM;
3791 goto out1;
3792 }
3793
5897297b
AK		 3794 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
3795 if (!vmx_msr_bitmap_longmode) {
3796 r = -ENOMEM;
3797 goto out2;
3798 }
3799
fdef3ad1
HQ		 3800 /*
3801 * Allow direct access to the PC debug port (it is often used for I/O
3802 * delays, but the vmexits simply slow things down).
3803 */
3e7c73e9
AK		 3804 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
3805 clear_bit(0x80, vmx_io_bitmap_a);
fdef3ad1 3806
3e7c73e9 3807 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
fdef3ad1 3808
5897297b
AK		 3809 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
3810 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
25c5f225 3811
2384d2b3
SY		 3812 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
3813
cb498ea2 3814 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
fdef3ad1 3815 if (r)
5897297b 3816 goto out3;
25c5f225 3817
5897297b
AK		 3818 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
3819 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
3820 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
3821 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
3822 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
3823 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
fdef3ad1 3824
089d034e 3825 if (enable_ept) {
1439442c 3826 bypass_guest_pf = 0;
5fdbcb9d 3827 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
2aaf69dc 3828 VMX_EPT_WRITABLE_MASK);
534e38b4 3829 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4b12f0de 3830 VMX_EPT_EXECUTABLE_MASK);
5fdbcb9d
SY		 3831 kvm_enable_tdp();
3832 } else
3833 kvm_disable_tdp();
1439442c 3834
c7addb90
AK		 3835 if (bypass_guest_pf)
3836 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
3837
1439442c
SY		 3838 ept_sync_global();
3839
fdef3ad1
HQ		 3840 return 0;
3841
5897297b
AK		 3842out3:
3843 free_page((unsigned long)vmx_msr_bitmap_longmode);
25c5f225 3844out2:
5897297b 3845 free_page((unsigned long)vmx_msr_bitmap_legacy);
fdef3ad1 3846out1:
3e7c73e9 3847 free_page((unsigned long)vmx_io_bitmap_b);
fdef3ad1 3848out:
3e7c73e9 3849 free_page((unsigned long)vmx_io_bitmap_a);
fdef3ad1 3850 return r;
6aa8b732
AK		 3851}
3852
3853static void __exit vmx_exit(void)
3854{
5897297b
AK		 3855 free_page((unsigned long)vmx_msr_bitmap_legacy);
3856 free_page((unsigned long)vmx_msr_bitmap_longmode);
3e7c73e9
AK		 3857 free_page((unsigned long)vmx_io_bitmap_b);
3858 free_page((unsigned long)vmx_io_bitmap_a);
fdef3ad1 3859
cb498ea2 3860 kvm_exit();
6aa8b732
AK		 3861}
3862
3863module_init(vmx_init)
3864module_exit(vmx_exit)
kernel source for telekom puls (alcatel/mediatek)