]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - arch/x86/kvm/vmx.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: VMX: refactor IRQ and NMI window enabling
[mirror_ubuntu-artful-kernel.git] / arch / x86 / kvm / vmx.c
CommitLineData
6aa8b732
AK		 1/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
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"
6aa8b732 19#include "vmx.h"
1d737c8a 20#include "mmu.h"
e495606d 21
edf88417 22#include <linux/kvm_host.h>
6aa8b732 23#include <linux/module.h>
9d8f549d 24#include <linux/kernel.h>
6aa8b732
AK		 25#include <linux/mm.h>
26#include <linux/highmem.h>
e8edc6e0 27#include <linux/sched.h>
c7addb90 28#include <linux/moduleparam.h>
5fdbf976 29#include "kvm_cache_regs.h"
35920a35 30#include "x86.h"
e495606d 31
6aa8b732 32#include <asm/io.h>
3b3be0d1 33#include <asm/desc.h>
6aa8b732 34
4ecac3fd
AK		 35#define __ex(x) __kvm_handle_fault_on_reboot(x)
36
6aa8b732
AK		 37MODULE_AUTHOR("Qumranet");
38MODULE_LICENSE("GPL");
39
c7addb90
AK		 40static int bypass_guest_pf = 1;
41module_param(bypass_guest_pf, bool, 0);
42
2384d2b3
SY		 43static int enable_vpid = 1;
44module_param(enable_vpid, bool, 0);
45
4c9fc8ef
AK		 46static int flexpriority_enabled = 1;
47module_param(flexpriority_enabled, bool, 0);
48
1439442c 49static int enable_ept = 1;
d56f546d
SY		 50module_param(enable_ept, bool, 0);
51
04fa4d32
MG		 52static int emulate_invalid_guest_state = 0;
53module_param(emulate_invalid_guest_state, bool, 0);
54
a2fa3e9f
GH		 55struct vmcs {
56 u32 revision_id;
57 u32 abort;
58 char data[0];
59};
60
61struct vcpu_vmx {
fb3f0f51 62 struct kvm_vcpu vcpu;
543e4243 63 struct list_head local_vcpus_link;
313dbd49 64 unsigned long host_rsp;
a2fa3e9f 65 int launched;
29bd8a78 66 u8 fail;
1155f76a 67 u32 idt_vectoring_info;
a2fa3e9f
GH		 68 struct kvm_msr_entry *guest_msrs;
69 struct kvm_msr_entry *host_msrs;
70 int nmsrs;
71 int save_nmsrs;
72 int msr_offset_efer;
73#ifdef CONFIG_X86_64
74 int msr_offset_kernel_gs_base;
75#endif
76 struct vmcs *vmcs;
77 struct {
78 int loaded;
79 u16 fs_sel, gs_sel, ldt_sel;
152d3f2f
LV		 80 int gs_ldt_reload_needed;
81 int fs_reload_needed;
51c6cf66 82 int guest_efer_loaded;
d77c26fc 83 } host_state;
9c8cba37
AK		 84 struct {
85 struct {
86 bool pending;
87 u8 vector;
88 unsigned rip;
89 } irq;
90 } rmode;
2384d2b3 91 int vpid;
04fa4d32 92 bool emulation_required;
a2fa3e9f
GH		 93};
94
95static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
96{
fb3f0f51 97 return container_of(vcpu, struct vcpu_vmx, vcpu);
a2fa3e9f
GH		 98}
99
b7ebfb05 100static int init_rmode(struct kvm *kvm);
4e1096d2 101static u64 construct_eptp(unsigned long root_hpa);
75880a01 102
6aa8b732
AK		 103static DEFINE_PER_CPU(struct vmcs *, vmxarea);
104static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
543e4243 105static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
6aa8b732 106
fdef3ad1
HQ		 107static struct page *vmx_io_bitmap_a;
108static struct page *vmx_io_bitmap_b;
25c5f225 109static struct page *vmx_msr_bitmap;
fdef3ad1 110
2384d2b3
SY		 111static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
112static DEFINE_SPINLOCK(vmx_vpid_lock);
113
1c3d14fe 114static struct vmcs_config {
6aa8b732
AK		 115 int size;
116 int order;
117 u32 revision_id;
1c3d14fe
YS		 118 u32 pin_based_exec_ctrl;
119 u32 cpu_based_exec_ctrl;
f78e0e2e 120 u32 cpu_based_2nd_exec_ctrl;
1c3d14fe
YS		 121 u32 vmexit_ctrl;
122 u32 vmentry_ctrl;
123} vmcs_config;
6aa8b732 124
d56f546d
SY		 125struct vmx_capability {
126 u32 ept;
127 u32 vpid;
128} vmx_capability;
129
6aa8b732
AK		 130#define VMX_SEGMENT_FIELD(seg) \
131 [VCPU_SREG_##seg] = { \
132 .selector = GUEST_##seg##_SELECTOR, \
133 .base = GUEST_##seg##_BASE, \
134 .limit = GUEST_##seg##_LIMIT, \
135 .ar_bytes = GUEST_##seg##_AR_BYTES, \
136 }
137
138static struct kvm_vmx_segment_field {
139 unsigned selector;
140 unsigned base;
141 unsigned limit;
142 unsigned ar_bytes;
143} kvm_vmx_segment_fields[] = {
144 VMX_SEGMENT_FIELD(CS),
145 VMX_SEGMENT_FIELD(DS),
146 VMX_SEGMENT_FIELD(ES),
147 VMX_SEGMENT_FIELD(FS),
148 VMX_SEGMENT_FIELD(GS),
149 VMX_SEGMENT_FIELD(SS),
150 VMX_SEGMENT_FIELD(TR),
151 VMX_SEGMENT_FIELD(LDTR),
152};
153
4d56c8a7
AK		 154/*
155 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
156 * away by decrementing the array size.
157 */
6aa8b732 158static const u32 vmx_msr_index[] = {
05b3e0c2 159#ifdef CONFIG_X86_64
6aa8b732
AK		 160 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
161#endif
162 MSR_EFER, MSR_K6_STAR,
163};
9d8f549d 164#define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
6aa8b732 165
a2fa3e9f
GH		 166static void load_msrs(struct kvm_msr_entry *e, int n)
167{
168 int i;
169
170 for (i = 0; i < n; ++i)
171 wrmsrl(e[i].index, e[i].data);
172}
173
174static void save_msrs(struct kvm_msr_entry *e, int n)
175{
176 int i;
177
178 for (i = 0; i < n; ++i)
179 rdmsrl(e[i].index, e[i].data);
180}
181
6aa8b732
AK		 182static inline int is_page_fault(u32 intr_info)
183{
184 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
185 INTR_INFO_VALID_MASK)) ==
186 (INTR_TYPE_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
187}
188
2ab455cc
AL		 189static inline int is_no_device(u32 intr_info)
190{
191 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
192 INTR_INFO_VALID_MASK)) ==
193 (INTR_TYPE_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
194}
195
7aa81cc0
AL		 196static inline int is_invalid_opcode(u32 intr_info)
197{
198 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
199 INTR_INFO_VALID_MASK)) ==
200 (INTR_TYPE_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
201}
202
6aa8b732
AK		 203static inline int is_external_interrupt(u32 intr_info)
204{
205 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
206 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
207}
208
25c5f225
SY		 209static inline int cpu_has_vmx_msr_bitmap(void)
210{
211 return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS);
212}
213
6e5d865c
YS		 214static inline int cpu_has_vmx_tpr_shadow(void)
215{
216 return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW);
217}
218
219static inline int vm_need_tpr_shadow(struct kvm *kvm)
220{
221 return ((cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm)));
222}
223
f78e0e2e
SY		 224static inline int cpu_has_secondary_exec_ctrls(void)
225{
226 return (vmcs_config.cpu_based_exec_ctrl &
227 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS);
228}
229
774ead3a 230static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
f78e0e2e 231{
4c9fc8ef
AK		 232 return flexpriority_enabled
233 && (vmcs_config.cpu_based_2nd_exec_ctrl &
234 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
f78e0e2e
SY		 235}
236
d56f546d
SY		 237static inline int cpu_has_vmx_invept_individual_addr(void)
238{
239 return (!!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT));
240}
241
242static inline int cpu_has_vmx_invept_context(void)
243{
244 return (!!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT));
245}
246
247static inline int cpu_has_vmx_invept_global(void)
248{
249 return (!!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT));
250}
251
252static inline int cpu_has_vmx_ept(void)
253{
254 return (vmcs_config.cpu_based_2nd_exec_ctrl &
255 SECONDARY_EXEC_ENABLE_EPT);
256}
257
258static inline int vm_need_ept(void)
259{
260 return (cpu_has_vmx_ept() && enable_ept);
261}
262
f78e0e2e
SY		 263static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
264{
265 return ((cpu_has_vmx_virtualize_apic_accesses()) &&
266 (irqchip_in_kernel(kvm)));
267}
268
2384d2b3
SY		 269static inline int cpu_has_vmx_vpid(void)
270{
271 return (vmcs_config.cpu_based_2nd_exec_ctrl &
272 SECONDARY_EXEC_ENABLE_VPID);
273}
274
f08864b4
SY		 275static inline int cpu_has_virtual_nmis(void)
276{
277 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
278}
279
8b9cf98c 280static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
7725f0ba
AK		 281{
282 int i;
283
a2fa3e9f
GH		 284 for (i = 0; i < vmx->nmsrs; ++i)
285 if (vmx->guest_msrs[i].index == msr)
a75beee6
ED		 286 return i;
287 return -1;
288}
289
2384d2b3
SY		 290static inline void __invvpid(int ext, u16 vpid, gva_t gva)
291{
292 struct {
293 u64 vpid : 16;
294 u64 rsvd : 48;
295 u64 gva;
296 } operand = { vpid, 0, gva };
297
4ecac3fd 298 asm volatile (__ex(ASM_VMX_INVVPID)
2384d2b3
SY		 299 /* CF==1 or ZF==1 --> rc = -1 */
300 "; ja 1f ; ud2 ; 1:"
301 : : "a"(&operand), "c"(ext) : "cc", "memory");
302}
303
1439442c
SY		 304static inline void __invept(int ext, u64 eptp, gpa_t gpa)
305{
306 struct {
307 u64 eptp, gpa;
308 } operand = {eptp, gpa};
309
4ecac3fd 310 asm volatile (__ex(ASM_VMX_INVEPT)
1439442c
SY		 311 /* CF==1 or ZF==1 --> rc = -1 */
312 "; ja 1f ; ud2 ; 1:\n"
313 : : "a" (&operand), "c" (ext) : "cc", "memory");
314}
315
8b9cf98c 316static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
a75beee6
ED		 317{
318 int i;
319
8b9cf98c 320 i = __find_msr_index(vmx, msr);
a75beee6 321 if (i >= 0)
a2fa3e9f 322 return &vmx->guest_msrs[i];
8b6d44c7 323 return NULL;
7725f0ba
AK		 324}
325
6aa8b732
AK		 326static void vmcs_clear(struct vmcs *vmcs)
327{
328 u64 phys_addr = __pa(vmcs);
329 u8 error;
330
4ecac3fd 331 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
6aa8b732
AK		 332 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
333 : "cc", "memory");
334 if (error)
335 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
336 vmcs, phys_addr);
337}
338
339static void __vcpu_clear(void *arg)
340{
8b9cf98c 341 struct vcpu_vmx *vmx = arg;
d3b2c338 342 int cpu = raw_smp_processor_id();
6aa8b732 343
8b9cf98c 344 if (vmx->vcpu.cpu == cpu)
a2fa3e9f
GH		 345 vmcs_clear(vmx->vmcs);
346 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
6aa8b732 347 per_cpu(current_vmcs, cpu) = NULL;
ad312c7c 348 rdtscll(vmx->vcpu.arch.host_tsc);
543e4243
AK		 349 list_del(&vmx->local_vcpus_link);
350 vmx->vcpu.cpu = -1;
351 vmx->launched = 0;
6aa8b732
AK		 352}
353
8b9cf98c 354static void vcpu_clear(struct vcpu_vmx *vmx)
8d0be2b3 355{
eae5ecb5
AK		 356 if (vmx->vcpu.cpu == -1)
357 return;
8691e5a8 358 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
8d0be2b3
AK		 359}
360
2384d2b3
SY		 361static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
362{
363 if (vmx->vpid == 0)
364 return;
365
366 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
367}
368
1439442c
SY		 369static inline void ept_sync_global(void)
370{
371 if (cpu_has_vmx_invept_global())
372 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
373}
374
375static inline void ept_sync_context(u64 eptp)
376{
377 if (vm_need_ept()) {
378 if (cpu_has_vmx_invept_context())
379 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
380 else
381 ept_sync_global();
382 }
383}
384
385static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
386{
387 if (vm_need_ept()) {
388 if (cpu_has_vmx_invept_individual_addr())
389 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
390 eptp, gpa);
391 else
392 ept_sync_context(eptp);
393 }
394}
395
6aa8b732
AK		 396static unsigned long vmcs_readl(unsigned long field)
397{
398 unsigned long value;
399
4ecac3fd 400 asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
6aa8b732
AK		 401 : "=a"(value) : "d"(field) : "cc");
402 return value;
403}
404
405static u16 vmcs_read16(unsigned long field)
406{
407 return vmcs_readl(field);
408}
409
410static u32 vmcs_read32(unsigned long field)
411{
412 return vmcs_readl(field);
413}
414
415static u64 vmcs_read64(unsigned long field)
416{
05b3e0c2 417#ifdef CONFIG_X86_64
6aa8b732
AK		 418 return vmcs_readl(field);
419#else
420 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
421#endif
422}
423
e52de1b8
AK		 424static noinline void vmwrite_error(unsigned long field, unsigned long value)
425{
426 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
427 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
428 dump_stack();
429}
430
6aa8b732
AK		 431static void vmcs_writel(unsigned long field, unsigned long value)
432{
433 u8 error;
434
4ecac3fd 435 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
d77c26fc 436 : "=q"(error) : "a"(value), "d"(field) : "cc");
e52de1b8
AK		 437 if (unlikely(error))
438 vmwrite_error(field, value);
6aa8b732
AK		 439}
440
441static void vmcs_write16(unsigned long field, u16 value)
442{
443 vmcs_writel(field, value);
444}
445
446static void vmcs_write32(unsigned long field, u32 value)
447{
448 vmcs_writel(field, value);
449}
450
451static void vmcs_write64(unsigned long field, u64 value)
452{
6aa8b732 453 vmcs_writel(field, value);
7682f2d0 454#ifndef CONFIG_X86_64
6aa8b732
AK		 455 asm volatile ("");
456 vmcs_writel(field+1, value >> 32);
457#endif
458}
459
2ab455cc
AL		 460static void vmcs_clear_bits(unsigned long field, u32 mask)
461{
462 vmcs_writel(field, vmcs_readl(field) & ~mask);
463}
464
465static void vmcs_set_bits(unsigned long field, u32 mask)
466{
467 vmcs_writel(field, vmcs_readl(field) | mask);
468}
469
abd3f2d6
AK		 470static void update_exception_bitmap(struct kvm_vcpu *vcpu)
471{
472 u32 eb;
473
7aa81cc0 474 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR);
abd3f2d6
AK		 475 if (!vcpu->fpu_active)
476 eb |= 1u << NM_VECTOR;
477 if (vcpu->guest_debug.enabled)
19bd8afd 478 eb |= 1u << DB_VECTOR;
ad312c7c 479 if (vcpu->arch.rmode.active)
abd3f2d6 480 eb = ~0;
1439442c
SY		 481 if (vm_need_ept())
482 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
abd3f2d6
AK		 483 vmcs_write32(EXCEPTION_BITMAP, eb);
484}
485
33ed6329
AK		 486static void reload_tss(void)
487{
33ed6329
AK		 488 /*
489 * VT restores TR but not its size. Useless.
490 */
491 struct descriptor_table gdt;
a5f61300 492 struct desc_struct *descs;
33ed6329 493
d6e88aec 494 kvm_get_gdt(&gdt);
33ed6329
AK		 495 descs = (void *)gdt.base;
496 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
497 load_TR_desc();
33ed6329
AK		 498}
499
8b9cf98c 500static void load_transition_efer(struct vcpu_vmx *vmx)
2cc51560 501{
a2fa3e9f 502 int efer_offset = vmx->msr_offset_efer;
51c6cf66
AK		 503 u64 host_efer = vmx->host_msrs[efer_offset].data;
504 u64 guest_efer = vmx->guest_msrs[efer_offset].data;
505 u64 ignore_bits;
506
507 if (efer_offset < 0)
508 return;
509 /*
510 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
511 * outside long mode
512 */
513 ignore_bits = EFER_NX | EFER_SCE;
514#ifdef CONFIG_X86_64
515 ignore_bits |= EFER_LMA | EFER_LME;
516 /* SCE is meaningful only in long mode on Intel */
517 if (guest_efer & EFER_LMA)
518 ignore_bits &= ~(u64)EFER_SCE;
519#endif
520 if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
521 return;
2cc51560 522
51c6cf66
AK		 523 vmx->host_state.guest_efer_loaded = 1;
524 guest_efer &= ~ignore_bits;
525 guest_efer |= host_efer & ignore_bits;
526 wrmsrl(MSR_EFER, guest_efer);
8b9cf98c 527 vmx->vcpu.stat.efer_reload++;
2cc51560
ED		 528}
529
51c6cf66
AK		 530static void reload_host_efer(struct vcpu_vmx *vmx)
531{
532 if (vmx->host_state.guest_efer_loaded) {
533 vmx->host_state.guest_efer_loaded = 0;
534 load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1);
535 }
536}
537
04d2cc77 538static void vmx_save_host_state(struct kvm_vcpu *vcpu)
33ed6329 539{
04d2cc77
AK		 540 struct vcpu_vmx *vmx = to_vmx(vcpu);
541
a2fa3e9f 542 if (vmx->host_state.loaded)
33ed6329
AK		 543 return;
544
a2fa3e9f 545 vmx->host_state.loaded = 1;
33ed6329
AK		 546 /*
547 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
548 * allow segment selectors with cpl > 0 or ti == 1.
549 */
d6e88aec 550 vmx->host_state.ldt_sel = kvm_read_ldt();
152d3f2f 551 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
d6e88aec 552 vmx->host_state.fs_sel = kvm_read_fs();
152d3f2f 553 if (!(vmx->host_state.fs_sel & 7)) {
a2fa3e9f 554 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
152d3f2f
LV		 555 vmx->host_state.fs_reload_needed = 0;
556 } else {
33ed6329 557 vmcs_write16(HOST_FS_SELECTOR, 0);
152d3f2f 558 vmx->host_state.fs_reload_needed = 1;
33ed6329 559 }
d6e88aec 560 vmx->host_state.gs_sel = kvm_read_gs();
a2fa3e9f
GH		 561 if (!(vmx->host_state.gs_sel & 7))
562 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
33ed6329
AK		 563 else {
564 vmcs_write16(HOST_GS_SELECTOR, 0);
152d3f2f 565 vmx->host_state.gs_ldt_reload_needed = 1;
33ed6329
AK		 566 }
567
568#ifdef CONFIG_X86_64
569 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
570 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
571#else
a2fa3e9f
GH		 572 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
573 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
33ed6329 574#endif
707c0874
AK		 575
576#ifdef CONFIG_X86_64
d77c26fc 577 if (is_long_mode(&vmx->vcpu))
a2fa3e9f
GH		 578 save_msrs(vmx->host_msrs +
579 vmx->msr_offset_kernel_gs_base, 1);
d77c26fc 580
707c0874 581#endif
a2fa3e9f 582 load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
51c6cf66 583 load_transition_efer(vmx);
33ed6329
AK		 584}
585
a9b21b62 586static void __vmx_load_host_state(struct vcpu_vmx *vmx)
33ed6329 587{
15ad7146 588 unsigned long flags;
33ed6329 589
a2fa3e9f 590 if (!vmx->host_state.loaded)
33ed6329
AK		 591 return;
592
e1beb1d3 593 ++vmx->vcpu.stat.host_state_reload;
a2fa3e9f 594 vmx->host_state.loaded = 0;
152d3f2f 595 if (vmx->host_state.fs_reload_needed)
d6e88aec 596 kvm_load_fs(vmx->host_state.fs_sel);
152d3f2f 597 if (vmx->host_state.gs_ldt_reload_needed) {
d6e88aec 598 kvm_load_ldt(vmx->host_state.ldt_sel);
33ed6329
AK		 599 /*
600 * If we have to reload gs, we must take care to
601 * preserve our gs base.
602 */
15ad7146 603 local_irq_save(flags);
d6e88aec 604 kvm_load_gs(vmx->host_state.gs_sel);
33ed6329
AK		 605#ifdef CONFIG_X86_64
606 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
607#endif
15ad7146 608 local_irq_restore(flags);
33ed6329 609 }
152d3f2f 610 reload_tss();
a2fa3e9f
GH		 611 save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
612 load_msrs(vmx->host_msrs, vmx->save_nmsrs);
51c6cf66 613 reload_host_efer(vmx);
33ed6329
AK		 614}
615
a9b21b62
AK		 616static void vmx_load_host_state(struct vcpu_vmx *vmx)
617{
618 preempt_disable();
619 __vmx_load_host_state(vmx);
620 preempt_enable();
621}
622
6aa8b732
AK		 623/*
624 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
625 * vcpu mutex is already taken.
626 */
15ad7146 627static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
6aa8b732 628{
a2fa3e9f
GH		 629 struct vcpu_vmx *vmx = to_vmx(vcpu);
630 u64 phys_addr = __pa(vmx->vmcs);
019960ae 631 u64 tsc_this, delta, new_offset;
6aa8b732 632
a3d7f85f 633 if (vcpu->cpu != cpu) {
8b9cf98c 634 vcpu_clear(vmx);
2f599714 635 kvm_migrate_timers(vcpu);
2384d2b3 636 vpid_sync_vcpu_all(vmx);
543e4243
AK		 637 local_irq_disable();
638 list_add(&vmx->local_vcpus_link,
639 &per_cpu(vcpus_on_cpu, cpu));
640 local_irq_enable();
a3d7f85f 641 }
6aa8b732 642
a2fa3e9f 643 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
6aa8b732
AK		 644 u8 error;
645
a2fa3e9f 646 per_cpu(current_vmcs, cpu) = vmx->vmcs;
4ecac3fd 647 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
6aa8b732
AK		 648 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
649 : "cc");
650 if (error)
651 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
a2fa3e9f 652 vmx->vmcs, phys_addr);
6aa8b732
AK		 653 }
654
655 if (vcpu->cpu != cpu) {
656 struct descriptor_table dt;
657 unsigned long sysenter_esp;
658
659 vcpu->cpu = cpu;
660 /*
661 * Linux uses per-cpu TSS and GDT, so set these when switching
662 * processors.
663 */
d6e88aec
AK		 664 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
665 kvm_get_gdt(&dt);
6aa8b732
AK		 666 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
667
668 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
669 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
7700270e
AK		 670
671 /*
672 * Make sure the time stamp counter is monotonous.
673 */
674 rdtscll(tsc_this);
019960ae
AK		 675 if (tsc_this < vcpu->arch.host_tsc) {
676 delta = vcpu->arch.host_tsc - tsc_this;
677 new_offset = vmcs_read64(TSC_OFFSET) + delta;
678 vmcs_write64(TSC_OFFSET, new_offset);
679 }
6aa8b732 680 }
6aa8b732
AK		 681}
682
683static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
684{
a9b21b62 685 __vmx_load_host_state(to_vmx(vcpu));
6aa8b732
AK		 686}
687
5fd86fcf
AK		 688static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
689{
690 if (vcpu->fpu_active)
691 return;
692 vcpu->fpu_active = 1;
707d92fa 693 vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
ad312c7c 694 if (vcpu->arch.cr0 & X86_CR0_TS)
707d92fa 695 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
5fd86fcf
AK		 696 update_exception_bitmap(vcpu);
697}
698
699static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
700{
701 if (!vcpu->fpu_active)
702 return;
703 vcpu->fpu_active = 0;
707d92fa 704 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
5fd86fcf
AK		 705 update_exception_bitmap(vcpu);
706}
707
6aa8b732
AK		 708static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
709{
710 return vmcs_readl(GUEST_RFLAGS);
711}
712
713static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
714{
ad312c7c 715 if (vcpu->arch.rmode.active)
053de044 716 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
6aa8b732
AK		 717 vmcs_writel(GUEST_RFLAGS, rflags);
718}
719
720static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
721{
722 unsigned long rip;
723 u32 interruptibility;
724
5fdbf976 725 rip = kvm_rip_read(vcpu);
6aa8b732 726 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5fdbf976 727 kvm_rip_write(vcpu, rip);
6aa8b732
AK		 728
729 /*
730 * We emulated an instruction, so temporary interrupt blocking
731 * should be removed, if set.
732 */
733 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
734 if (interruptibility & 3)
735 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
736 interruptibility & ~3);
ad312c7c 737 vcpu->arch.interrupt_window_open = 1;
6aa8b732
AK		 738}
739
298101da
AK		 740static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
741 bool has_error_code, u32 error_code)
742{
77ab6db0
JK		 743 struct vcpu_vmx *vmx = to_vmx(vcpu);
744
745 if (has_error_code)
746 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
747
748 if (vcpu->arch.rmode.active) {
749 vmx->rmode.irq.pending = true;
750 vmx->rmode.irq.vector = nr;
751 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
752 if (nr == BP_VECTOR)
753 vmx->rmode.irq.rip++;
754 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
755 nr | INTR_TYPE_SOFT_INTR
756 | (has_error_code ? INTR_INFO_DELIVER_CODE_MASK : 0)
757 | INTR_INFO_VALID_MASK);
758 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
759 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
760 return;
761 }
762
298101da
AK		 763 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
764 nr | INTR_TYPE_EXCEPTION
2e11384c 765 | (has_error_code ? INTR_INFO_DELIVER_CODE_MASK : 0)
298101da 766 | INTR_INFO_VALID_MASK);
298101da
AK		 767}
768
769static bool vmx_exception_injected(struct kvm_vcpu *vcpu)
770{
35920a35 771 return false;
298101da
AK		 772}
773
a75beee6
ED		 774/*
775 * Swap MSR entry in host/guest MSR entry array.
776 */
54e11fa1 777#ifdef CONFIG_X86_64
8b9cf98c 778static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
a75beee6 779{
a2fa3e9f
GH		 780 struct kvm_msr_entry tmp;
781
782 tmp = vmx->guest_msrs[to];
783 vmx->guest_msrs[to] = vmx->guest_msrs[from];
784 vmx->guest_msrs[from] = tmp;
785 tmp = vmx->host_msrs[to];
786 vmx->host_msrs[to] = vmx->host_msrs[from];
787 vmx->host_msrs[from] = tmp;
a75beee6 788}
54e11fa1 789#endif
a75beee6 790
e38aea3e
AK		 791/*
792 * Set up the vmcs to automatically save and restore system
793 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
794 * mode, as fiddling with msrs is very expensive.
795 */
8b9cf98c 796static void setup_msrs(struct vcpu_vmx *vmx)
e38aea3e 797{
2cc51560 798 int save_nmsrs;
e38aea3e 799
33f9c505 800 vmx_load_host_state(vmx);
a75beee6
ED		 801 save_nmsrs = 0;
802#ifdef CONFIG_X86_64
8b9cf98c 803 if (is_long_mode(&vmx->vcpu)) {
2cc51560
ED		 804 int index;
805
8b9cf98c 806 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
a75beee6 807 if (index >= 0)
8b9cf98c
RR		 808 move_msr_up(vmx, index, save_nmsrs++);
809 index = __find_msr_index(vmx, MSR_LSTAR);
a75beee6 810 if (index >= 0)
8b9cf98c
RR		 811 move_msr_up(vmx, index, save_nmsrs++);
812 index = __find_msr_index(vmx, MSR_CSTAR);
a75beee6 813 if (index >= 0)
8b9cf98c
RR		 814 move_msr_up(vmx, index, save_nmsrs++);
815 index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
a75beee6 816 if (index >= 0)
8b9cf98c 817 move_msr_up(vmx, index, save_nmsrs++);
a75beee6
ED		 818 /*
819 * MSR_K6_STAR is only needed on long mode guests, and only
820 * if efer.sce is enabled.
821 */
8b9cf98c 822 index = __find_msr_index(vmx, MSR_K6_STAR);
ad312c7c 823 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
8b9cf98c 824 move_msr_up(vmx, index, save_nmsrs++);
a75beee6
ED		 825 }
826#endif
a2fa3e9f 827 vmx->save_nmsrs = save_nmsrs;
e38aea3e 828
4d56c8a7 829#ifdef CONFIG_X86_64
a2fa3e9f 830 vmx->msr_offset_kernel_gs_base =
8b9cf98c 831 __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
4d56c8a7 832#endif
8b9cf98c 833 vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER);
e38aea3e
AK		 834}
835
6aa8b732
AK		 836/*
837 * reads and returns guest's timestamp counter "register"
838 * guest_tsc = host_tsc + tsc_offset -- 21.3
839 */
840static u64 guest_read_tsc(void)
841{
842 u64 host_tsc, tsc_offset;
843
844 rdtscll(host_tsc);
845 tsc_offset = vmcs_read64(TSC_OFFSET);
846 return host_tsc + tsc_offset;
847}
848
849/*
850 * writes 'guest_tsc' into guest's timestamp counter "register"
851 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
852 */
853static void guest_write_tsc(u64 guest_tsc)
854{
855 u64 host_tsc;
856
857 rdtscll(host_tsc);
858 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
859}
860
6aa8b732
AK		 861/*
862 * Reads an msr value (of 'msr_index') into 'pdata'.
863 * Returns 0 on success, non-0 otherwise.
864 * Assumes vcpu_load() was already called.
865 */
866static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
867{
868 u64 data;
a2fa3e9f 869 struct kvm_msr_entry *msr;
6aa8b732
AK		 870
871 if (!pdata) {
872 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
873 return -EINVAL;
874 }
875
876 switch (msr_index) {
05b3e0c2 877#ifdef CONFIG_X86_64
6aa8b732
AK		 878 case MSR_FS_BASE:
879 data = vmcs_readl(GUEST_FS_BASE);
880 break;
881 case MSR_GS_BASE:
882 data = vmcs_readl(GUEST_GS_BASE);
883 break;
884 case MSR_EFER:
3bab1f5d 885 return kvm_get_msr_common(vcpu, msr_index, pdata);
6aa8b732
AK		 886#endif
887 case MSR_IA32_TIME_STAMP_COUNTER:
888 data = guest_read_tsc();
889 break;
890 case MSR_IA32_SYSENTER_CS:
891 data = vmcs_read32(GUEST_SYSENTER_CS);
892 break;
893 case MSR_IA32_SYSENTER_EIP:
f5b42c33 894 data = vmcs_readl(GUEST_SYSENTER_EIP);
6aa8b732
AK		 895 break;
896 case MSR_IA32_SYSENTER_ESP:
f5b42c33 897 data = vmcs_readl(GUEST_SYSENTER_ESP);
6aa8b732 898 break;
6aa8b732 899 default:
8b9cf98c 900 msr = find_msr_entry(to_vmx(vcpu), msr_index);
3bab1f5d
AK		 901 if (msr) {
902 data = msr->data;
903 break;
6aa8b732 904 }
3bab1f5d 905 return kvm_get_msr_common(vcpu, msr_index, pdata);
6aa8b732
AK		 906 }
907
908 *pdata = data;
909 return 0;
910}
911
912/*
913 * Writes msr value into into the appropriate "register".
914 * Returns 0 on success, non-0 otherwise.
915 * Assumes vcpu_load() was already called.
916 */
917static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
918{
a2fa3e9f
GH		 919 struct vcpu_vmx *vmx = to_vmx(vcpu);
920 struct kvm_msr_entry *msr;
2cc51560
ED		 921 int ret = 0;
922
6aa8b732 923 switch (msr_index) {
05b3e0c2 924#ifdef CONFIG_X86_64
3bab1f5d 925 case MSR_EFER:
a9b21b62 926 vmx_load_host_state(vmx);
2cc51560 927 ret = kvm_set_msr_common(vcpu, msr_index, data);
2cc51560 928 break;
6aa8b732
AK		 929 case MSR_FS_BASE:
930 vmcs_writel(GUEST_FS_BASE, data);
931 break;
932 case MSR_GS_BASE:
933 vmcs_writel(GUEST_GS_BASE, data);
934 break;
935#endif
936 case MSR_IA32_SYSENTER_CS:
937 vmcs_write32(GUEST_SYSENTER_CS, data);
938 break;
939 case MSR_IA32_SYSENTER_EIP:
f5b42c33 940 vmcs_writel(GUEST_SYSENTER_EIP, data);
6aa8b732
AK		 941 break;
942 case MSR_IA32_SYSENTER_ESP:
f5b42c33 943 vmcs_writel(GUEST_SYSENTER_ESP, data);
6aa8b732 944 break;
d27d4aca 945 case MSR_IA32_TIME_STAMP_COUNTER:
6aa8b732 946 guest_write_tsc(data);
efa67e0d
CL		 947 break;
948 case MSR_P6_PERFCTR0:
949 case MSR_P6_PERFCTR1:
950 case MSR_P6_EVNTSEL0:
951 case MSR_P6_EVNTSEL1:
952 /*
953 * Just discard all writes to the performance counters; this
954 * should keep both older linux and windows 64-bit guests
955 * happy
956 */
957 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", msr_index, data);
958
6aa8b732 959 break;
6aa8b732 960 default:
a9b21b62 961 vmx_load_host_state(vmx);
8b9cf98c 962 msr = find_msr_entry(vmx, msr_index);
3bab1f5d
AK		 963 if (msr) {
964 msr->data = data;
965 break;
6aa8b732 966 }
2cc51560 967 ret = kvm_set_msr_common(vcpu, msr_index, data);
6aa8b732
AK		 968 }
969
2cc51560 970 return ret;
6aa8b732
AK		 971}
972
5fdbf976 973static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
6aa8b732 974{
5fdbf976
MT		 975 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
976 switch (reg) {
977 case VCPU_REGS_RSP:
978 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
979 break;
980 case VCPU_REGS_RIP:
981 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
982 break;
983 default:
984 break;
985 }
6aa8b732
AK		 986}
987
988static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
989{
990 unsigned long dr7 = 0x400;
6aa8b732
AK		 991 int old_singlestep;
992
6aa8b732
AK		 993 old_singlestep = vcpu->guest_debug.singlestep;
994
995 vcpu->guest_debug.enabled = dbg->enabled;
996 if (vcpu->guest_debug.enabled) {
997 int i;
998
999 dr7 |= 0x200; /* exact */
1000 for (i = 0; i < 4; ++i) {
1001 if (!dbg->breakpoints[i].enabled)
1002 continue;
1003 vcpu->guest_debug.bp[i] = dbg->breakpoints[i].address;
1004 dr7 |= 2 << (i*2); /* global enable */
1005 dr7 |= 0 << (i*4+16); /* execution breakpoint */
1006 }
1007
6aa8b732 1008 vcpu->guest_debug.singlestep = dbg->singlestep;
abd3f2d6 1009 } else
6aa8b732 1010 vcpu->guest_debug.singlestep = 0;
6aa8b732
AK		 1011
1012 if (old_singlestep && !vcpu->guest_debug.singlestep) {
1013 unsigned long flags;
1014
1015 flags = vmcs_readl(GUEST_RFLAGS);
1016 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1017 vmcs_writel(GUEST_RFLAGS, flags);
1018 }
1019
abd3f2d6 1020 update_exception_bitmap(vcpu);
6aa8b732
AK		 1021 vmcs_writel(GUEST_DR7, dr7);
1022
1023 return 0;
1024}
1025
2a8067f1
ED		 1026static int vmx_get_irq(struct kvm_vcpu *vcpu)
1027{
f7d9238f
AK		 1028 if (!vcpu->arch.interrupt.pending)
1029 return -1;
1030 return vcpu->arch.interrupt.nr;
2a8067f1
ED		 1031}
1032
6aa8b732
AK		 1033static __init int cpu_has_kvm_support(void)
1034{
1035 unsigned long ecx = cpuid_ecx(1);
1036 return test_bit(5, &ecx); /* CPUID.1:ECX.VMX[bit 5] -> VT */
1037}
1038
1039static __init int vmx_disabled_by_bios(void)
1040{
1041 u64 msr;
1042
1043 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
9ea542fa
SY		 1044 return (msr & (FEATURE_CONTROL_LOCKED |
1045 FEATURE_CONTROL_VMXON_ENABLED))
1046 == FEATURE_CONTROL_LOCKED;
62b3ffb8 1047 /* locked but not enabled */
6aa8b732
AK		 1048}
1049
774c47f1 1050static void hardware_enable(void *garbage)
6aa8b732
AK		 1051{
1052 int cpu = raw_smp_processor_id();
1053 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1054 u64 old;
1055
543e4243 1056 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
6aa8b732 1057 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
9ea542fa
SY		 1058 if ((old & (FEATURE_CONTROL_LOCKED |
1059 FEATURE_CONTROL_VMXON_ENABLED))
1060 != (FEATURE_CONTROL_LOCKED |
1061 FEATURE_CONTROL_VMXON_ENABLED))
6aa8b732 1062 /* enable and lock */
62b3ffb8 1063 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
9ea542fa
SY		 1064 FEATURE_CONTROL_LOCKED |
1065 FEATURE_CONTROL_VMXON_ENABLED);
66aee91a 1066 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
4ecac3fd
AK		 1067 asm volatile (ASM_VMX_VMXON_RAX
1068 : : "a"(&phys_addr), "m"(phys_addr)
6aa8b732
AK		 1069 : "memory", "cc");
1070}
1071
543e4243
AK		 1072static void vmclear_local_vcpus(void)
1073{
1074 int cpu = raw_smp_processor_id();
1075 struct vcpu_vmx *vmx, *n;
1076
1077 list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1078 local_vcpus_link)
1079 __vcpu_clear(vmx);
1080}
1081
6aa8b732
AK		 1082static void hardware_disable(void *garbage)
1083{
543e4243 1084 vmclear_local_vcpus();
4ecac3fd 1085 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
e693d71b 1086 write_cr4(read_cr4() & ~X86_CR4_VMXE);
6aa8b732
AK		 1087}
1088
1c3d14fe 1089static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
d77c26fc 1090 u32 msr, u32 *result)
1c3d14fe
YS		 1091{
1092 u32 vmx_msr_low, vmx_msr_high;
1093 u32 ctl = ctl_min | ctl_opt;
1094
1095 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1096
1097 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1098 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
1099
1100 /* Ensure minimum (required) set of control bits are supported. */
1101 if (ctl_min & ~ctl)
002c7f7c 1102 return -EIO;
1c3d14fe
YS		 1103
1104 *result = ctl;
1105 return 0;
1106}
1107
002c7f7c 1108static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
6aa8b732
AK		 1109{
1110 u32 vmx_msr_low, vmx_msr_high;
d56f546d 1111 u32 min, opt, min2, opt2;
1c3d14fe
YS		 1112 u32 _pin_based_exec_control = 0;
1113 u32 _cpu_based_exec_control = 0;
f78e0e2e 1114 u32 _cpu_based_2nd_exec_control = 0;
1c3d14fe
YS		 1115 u32 _vmexit_control = 0;
1116 u32 _vmentry_control = 0;
1117
1118 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
f08864b4 1119 opt = PIN_BASED_VIRTUAL_NMIS;
1c3d14fe
YS		 1120 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1121 &_pin_based_exec_control) < 0)
002c7f7c 1122 return -EIO;
1c3d14fe
YS		 1123
1124 min = CPU_BASED_HLT_EXITING |
1125#ifdef CONFIG_X86_64
1126 CPU_BASED_CR8_LOAD_EXITING |
1127 CPU_BASED_CR8_STORE_EXITING |
1128#endif
d56f546d
SY		 1129 CPU_BASED_CR3_LOAD_EXITING |
1130 CPU_BASED_CR3_STORE_EXITING |
1c3d14fe
YS		 1131 CPU_BASED_USE_IO_BITMAPS |
1132 CPU_BASED_MOV_DR_EXITING |
a7052897
MT		 1133 CPU_BASED_USE_TSC_OFFSETING |
1134 CPU_BASED_INVLPG_EXITING;
f78e0e2e 1135 opt = CPU_BASED_TPR_SHADOW |
25c5f225 1136 CPU_BASED_USE_MSR_BITMAPS |
f78e0e2e 1137 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1c3d14fe
YS		 1138 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1139 &_cpu_based_exec_control) < 0)
002c7f7c 1140 return -EIO;
6e5d865c
YS		 1141#ifdef CONFIG_X86_64
1142 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1143 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1144 ~CPU_BASED_CR8_STORE_EXITING;
1145#endif
f78e0e2e 1146 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
d56f546d
SY		 1147 min2 = 0;
1148 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2384d2b3 1149 SECONDARY_EXEC_WBINVD_EXITING |
d56f546d
SY		 1150 SECONDARY_EXEC_ENABLE_VPID |
1151 SECONDARY_EXEC_ENABLE_EPT;
1152 if (adjust_vmx_controls(min2, opt2,
1153 MSR_IA32_VMX_PROCBASED_CTLS2,
f78e0e2e
SY		 1154 &_cpu_based_2nd_exec_control) < 0)
1155 return -EIO;
1156 }
1157#ifndef CONFIG_X86_64
1158 if (!(_cpu_based_2nd_exec_control &
1159 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1160 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1161#endif
d56f546d 1162 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
a7052897
MT		 1163 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1164 enabled */
d56f546d 1165 min &= ~(CPU_BASED_CR3_LOAD_EXITING |
a7052897
MT		 1166 CPU_BASED_CR3_STORE_EXITING |
1167 CPU_BASED_INVLPG_EXITING);
d56f546d
SY		 1168 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1169 &_cpu_based_exec_control) < 0)
1170 return -EIO;
1171 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1172 vmx_capability.ept, vmx_capability.vpid);
1173 }
1c3d14fe
YS		 1174
1175 min = 0;
1176#ifdef CONFIG_X86_64
1177 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1178#endif
1179 opt = 0;
1180 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1181 &_vmexit_control) < 0)
002c7f7c 1182 return -EIO;
1c3d14fe
YS		 1183
1184 min = opt = 0;
1185 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1186 &_vmentry_control) < 0)
002c7f7c 1187 return -EIO;
6aa8b732 1188
c68876fd 1189 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1c3d14fe
YS		 1190
1191 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1192 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
002c7f7c 1193 return -EIO;
1c3d14fe
YS		 1194
1195#ifdef CONFIG_X86_64
1196 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1197 if (vmx_msr_high & (1u<<16))
002c7f7c 1198 return -EIO;
1c3d14fe
YS		 1199#endif
1200
1201 /* Require Write-Back (WB) memory type for VMCS accesses. */
1202 if (((vmx_msr_high >> 18) & 15) != 6)
002c7f7c 1203 return -EIO;
1c3d14fe 1204
002c7f7c
YS		 1205 vmcs_conf->size = vmx_msr_high & 0x1fff;
1206 vmcs_conf->order = get_order(vmcs_config.size);
1207 vmcs_conf->revision_id = vmx_msr_low;
1c3d14fe 1208
002c7f7c
YS		 1209 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1210 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
f78e0e2e 1211 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
002c7f7c
YS		 1212 vmcs_conf->vmexit_ctrl = _vmexit_control;
1213 vmcs_conf->vmentry_ctrl = _vmentry_control;
1c3d14fe
YS		 1214
1215 return 0;
c68876fd 1216}
6aa8b732
AK		 1217
1218static struct vmcs *alloc_vmcs_cpu(int cpu)
1219{
1220 int node = cpu_to_node(cpu);
1221 struct page *pages;
1222 struct vmcs *vmcs;
1223
1c3d14fe 1224 pages = alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
6aa8b732
AK		 1225 if (!pages)
1226 return NULL;
1227 vmcs = page_address(pages);
1c3d14fe
YS		 1228 memset(vmcs, 0, vmcs_config.size);
1229 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
6aa8b732
AK		 1230 return vmcs;
1231}
1232
1233static struct vmcs *alloc_vmcs(void)
1234{
d3b2c338 1235 return alloc_vmcs_cpu(raw_smp_processor_id());
6aa8b732
AK		 1236}
1237
1238static void free_vmcs(struct vmcs *vmcs)
1239{
1c3d14fe 1240 free_pages((unsigned long)vmcs, vmcs_config.order);
6aa8b732
AK		 1241}
1242
39959588 1243static void free_kvm_area(void)
6aa8b732
AK		 1244{
1245 int cpu;
1246
1247 for_each_online_cpu(cpu)
1248 free_vmcs(per_cpu(vmxarea, cpu));
1249}
1250
6aa8b732
AK		 1251static __init int alloc_kvm_area(void)
1252{
1253 int cpu;
1254
1255 for_each_online_cpu(cpu) {
1256 struct vmcs *vmcs;
1257
1258 vmcs = alloc_vmcs_cpu(cpu);
1259 if (!vmcs) {
1260 free_kvm_area();
1261 return -ENOMEM;
1262 }
1263
1264 per_cpu(vmxarea, cpu) = vmcs;
1265 }
1266 return 0;
1267}
1268
1269static __init int hardware_setup(void)
1270{
002c7f7c
YS		 1271 if (setup_vmcs_config(&vmcs_config) < 0)
1272 return -EIO;
50a37eb4
JR		 1273
1274 if (boot_cpu_has(X86_FEATURE_NX))
1275 kvm_enable_efer_bits(EFER_NX);
1276
6aa8b732
AK		 1277 return alloc_kvm_area();
1278}
1279
1280static __exit void hardware_unsetup(void)
1281{
1282 free_kvm_area();
1283}
1284
6aa8b732
AK		 1285static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1286{
1287 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1288
6af11b9e 1289 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
6aa8b732
AK		 1290 vmcs_write16(sf->selector, save->selector);
1291 vmcs_writel(sf->base, save->base);
1292 vmcs_write32(sf->limit, save->limit);
1293 vmcs_write32(sf->ar_bytes, save->ar);
1294 } else {
1295 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1296 << AR_DPL_SHIFT;
1297 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1298 }
1299}
1300
1301static void enter_pmode(struct kvm_vcpu *vcpu)
1302{
1303 unsigned long flags;
a89a8fb9 1304 struct vcpu_vmx *vmx = to_vmx(vcpu);
6aa8b732 1305
a89a8fb9 1306 vmx->emulation_required = 1;
ad312c7c 1307 vcpu->arch.rmode.active = 0;
6aa8b732 1308
ad312c7c
ZX		 1309 vmcs_writel(GUEST_TR_BASE, vcpu->arch.rmode.tr.base);
1310 vmcs_write32(GUEST_TR_LIMIT, vcpu->arch.rmode.tr.limit);
1311 vmcs_write32(GUEST_TR_AR_BYTES, vcpu->arch.rmode.tr.ar);
6aa8b732
AK		 1312
1313 flags = vmcs_readl(GUEST_RFLAGS);
053de044 1314 flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
ad312c7c 1315 flags |= (vcpu->arch.rmode.save_iopl << IOPL_SHIFT);
6aa8b732
AK		 1316 vmcs_writel(GUEST_RFLAGS, flags);
1317
66aee91a
RR		 1318 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1319 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
6aa8b732
AK		 1320
1321 update_exception_bitmap(vcpu);
1322
a89a8fb9
MG		 1323 if (emulate_invalid_guest_state)
1324 return;
1325
ad312c7c
ZX		 1326 fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1327 fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1328 fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1329 fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
6aa8b732
AK		 1330
1331 vmcs_write16(GUEST_SS_SELECTOR, 0);
1332 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1333
1334 vmcs_write16(GUEST_CS_SELECTOR,
1335 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1336 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1337}
1338
d77c26fc 1339static gva_t rmode_tss_base(struct kvm *kvm)
6aa8b732 1340{
bfc6d222 1341 if (!kvm->arch.tss_addr) {
cbc94022
IE		 1342 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1343 kvm->memslots[0].npages - 3;
1344 return base_gfn << PAGE_SHIFT;
1345 }
bfc6d222 1346 return kvm->arch.tss_addr;
6aa8b732
AK		 1347}
1348
1349static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1350{
1351 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1352
1353 save->selector = vmcs_read16(sf->selector);
1354 save->base = vmcs_readl(sf->base);
1355 save->limit = vmcs_read32(sf->limit);
1356 save->ar = vmcs_read32(sf->ar_bytes);
15b00f32
JK		 1357 vmcs_write16(sf->selector, save->base >> 4);
1358 vmcs_write32(sf->base, save->base & 0xfffff);
6aa8b732
AK		 1359 vmcs_write32(sf->limit, 0xffff);
1360 vmcs_write32(sf->ar_bytes, 0xf3);
1361}
1362
1363static void enter_rmode(struct kvm_vcpu *vcpu)
1364{
1365 unsigned long flags;
a89a8fb9 1366 struct vcpu_vmx *vmx = to_vmx(vcpu);
6aa8b732 1367
a89a8fb9 1368 vmx->emulation_required = 1;
ad312c7c 1369 vcpu->arch.rmode.active = 1;
6aa8b732 1370
ad312c7c 1371 vcpu->arch.rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
6aa8b732
AK		 1372 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1373
ad312c7c 1374 vcpu->arch.rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
6aa8b732
AK		 1375 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1376
ad312c7c 1377 vcpu->arch.rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
6aa8b732
AK		 1378 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1379
1380 flags = vmcs_readl(GUEST_RFLAGS);
ad312c7c
ZX		 1381 vcpu->arch.rmode.save_iopl
1382 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
6aa8b732 1383
053de044 1384 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
6aa8b732
AK		 1385
1386 vmcs_writel(GUEST_RFLAGS, flags);
66aee91a 1387 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
6aa8b732
AK		 1388 update_exception_bitmap(vcpu);
1389
a89a8fb9
MG		 1390 if (emulate_invalid_guest_state)
1391 goto continue_rmode;
1392
6aa8b732
AK		 1393 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1394 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1395 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1396
1397 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
abacf8df 1398 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
8cb5b033
AK		 1399 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1400 vmcs_writel(GUEST_CS_BASE, 0xf0000);
6aa8b732
AK		 1401 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1402
ad312c7c
ZX		 1403 fix_rmode_seg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1404 fix_rmode_seg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1405 fix_rmode_seg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1406 fix_rmode_seg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
75880a01 1407
a89a8fb9 1408continue_rmode:
8668a3c4 1409 kvm_mmu_reset_context(vcpu);
b7ebfb05 1410 init_rmode(vcpu->kvm);
6aa8b732
AK		 1411}
1412
05b3e0c2 1413#ifdef CONFIG_X86_64
6aa8b732
AK		 1414
1415static void enter_lmode(struct kvm_vcpu *vcpu)
1416{
1417 u32 guest_tr_ar;
1418
1419 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1420 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1421 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
b8688d51 1422 __func__);
6aa8b732
AK		 1423 vmcs_write32(GUEST_TR_AR_BYTES,
1424 (guest_tr_ar & ~AR_TYPE_MASK)
1425 | AR_TYPE_BUSY_64_TSS);
1426 }
1427
ad312c7c 1428 vcpu->arch.shadow_efer |= EFER_LMA;
6aa8b732 1429
8b9cf98c 1430 find_msr_entry(to_vmx(vcpu), MSR_EFER)->data |= EFER_LMA | EFER_LME;
6aa8b732
AK		 1431 vmcs_write32(VM_ENTRY_CONTROLS,
1432 vmcs_read32(VM_ENTRY_CONTROLS)
1e4e6e00 1433 | VM_ENTRY_IA32E_MODE);
6aa8b732
AK		 1434}
1435
1436static void exit_lmode(struct kvm_vcpu *vcpu)
1437{
ad312c7c 1438 vcpu->arch.shadow_efer &= ~EFER_LMA;
6aa8b732
AK		 1439
1440 vmcs_write32(VM_ENTRY_CONTROLS,
1441 vmcs_read32(VM_ENTRY_CONTROLS)
1e4e6e00 1442 & ~VM_ENTRY_IA32E_MODE);
6aa8b732
AK		 1443}
1444
1445#endif
1446
2384d2b3
SY		 1447static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1448{
1449 vpid_sync_vcpu_all(to_vmx(vcpu));
4e1096d2
SY		 1450 if (vm_need_ept())
1451 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
2384d2b3
SY		 1452}
1453
25c4c276 1454static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
399badf3 1455{
ad312c7c
ZX		 1456 vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1457 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
399badf3
AK		 1458}
1459
1439442c
SY		 1460static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1461{
1462 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1463 if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
1464 printk(KERN_ERR "EPT: Fail to load pdptrs!\n");
1465 return;
1466 }
1467 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1468 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1469 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1470 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1471 }
1472}
1473
1474static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1475
1476static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1477 unsigned long cr0,
1478 struct kvm_vcpu *vcpu)
1479{
1480 if (!(cr0 & X86_CR0_PG)) {
1481 /* From paging/starting to nonpaging */
1482 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
65267ea1 1483 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1439442c
SY		 1484 (CPU_BASED_CR3_LOAD_EXITING |
1485 CPU_BASED_CR3_STORE_EXITING));
1486 vcpu->arch.cr0 = cr0;
1487 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1488 *hw_cr0 |= X86_CR0_PE | X86_CR0_PG;
1489 *hw_cr0 &= ~X86_CR0_WP;
1490 } else if (!is_paging(vcpu)) {
1491 /* From nonpaging to paging */
1492 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
65267ea1 1493 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1439442c
SY		 1494 ~(CPU_BASED_CR3_LOAD_EXITING |
1495 CPU_BASED_CR3_STORE_EXITING));
1496 vcpu->arch.cr0 = cr0;
1497 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1498 if (!(vcpu->arch.cr0 & X86_CR0_WP))
1499 *hw_cr0 &= ~X86_CR0_WP;
1500 }
1501}
1502
1503static void ept_update_paging_mode_cr4(unsigned long *hw_cr4,
1504 struct kvm_vcpu *vcpu)
1505{
1506 if (!is_paging(vcpu)) {
1507 *hw_cr4 &= ~X86_CR4_PAE;
1508 *hw_cr4 |= X86_CR4_PSE;
1509 } else if (!(vcpu->arch.cr4 & X86_CR4_PAE))
1510 *hw_cr4 &= ~X86_CR4_PAE;
1511}
1512
6aa8b732
AK		 1513static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1514{
1439442c
SY		 1515 unsigned long hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) |
1516 KVM_VM_CR0_ALWAYS_ON;
1517
5fd86fcf
AK		 1518 vmx_fpu_deactivate(vcpu);
1519
ad312c7c 1520 if (vcpu->arch.rmode.active && (cr0 & X86_CR0_PE))
6aa8b732
AK		 1521 enter_pmode(vcpu);
1522
ad312c7c 1523 if (!vcpu->arch.rmode.active && !(cr0 & X86_CR0_PE))
6aa8b732
AK		 1524 enter_rmode(vcpu);
1525
05b3e0c2 1526#ifdef CONFIG_X86_64
ad312c7c 1527 if (vcpu->arch.shadow_efer & EFER_LME) {
707d92fa 1528 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
6aa8b732 1529 enter_lmode(vcpu);
707d92fa 1530 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
6aa8b732
AK		 1531 exit_lmode(vcpu);
1532 }
1533#endif
1534
1439442c
SY		 1535 if (vm_need_ept())
1536 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1537
6aa8b732 1538 vmcs_writel(CR0_READ_SHADOW, cr0);
1439442c 1539 vmcs_writel(GUEST_CR0, hw_cr0);
ad312c7c 1540 vcpu->arch.cr0 = cr0;
5fd86fcf 1541
707d92fa 1542 if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
5fd86fcf 1543 vmx_fpu_activate(vcpu);
6aa8b732
AK		 1544}
1545
1439442c
SY		 1546static u64 construct_eptp(unsigned long root_hpa)
1547{
1548 u64 eptp;
1549
1550 /* TODO write the value reading from MSR */
1551 eptp = VMX_EPT_DEFAULT_MT |
1552 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1553 eptp |= (root_hpa & PAGE_MASK);
1554
1555 return eptp;
1556}
1557
6aa8b732
AK		 1558static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1559{
1439442c
SY		 1560 unsigned long guest_cr3;
1561 u64 eptp;
1562
1563 guest_cr3 = cr3;
1564 if (vm_need_ept()) {
1565 eptp = construct_eptp(cr3);
1566 vmcs_write64(EPT_POINTER, eptp);
1567 ept_sync_context(eptp);
1568 ept_load_pdptrs(vcpu);
1569 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1570 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
1571 }
1572
2384d2b3 1573 vmx_flush_tlb(vcpu);
1439442c 1574 vmcs_writel(GUEST_CR3, guest_cr3);
ad312c7c 1575 if (vcpu->arch.cr0 & X86_CR0_PE)
5fd86fcf 1576 vmx_fpu_deactivate(vcpu);
6aa8b732
AK		 1577}
1578
1579static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1580{
1439442c
SY		 1581 unsigned long hw_cr4 = cr4 | (vcpu->arch.rmode.active ?
1582 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1583
ad312c7c 1584 vcpu->arch.cr4 = cr4;
1439442c
SY		 1585 if (vm_need_ept())
1586 ept_update_paging_mode_cr4(&hw_cr4, vcpu);
1587
1588 vmcs_writel(CR4_READ_SHADOW, cr4);
1589 vmcs_writel(GUEST_CR4, hw_cr4);
6aa8b732
AK		 1590}
1591
6aa8b732
AK		 1592static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1593{
8b9cf98c
RR		 1594 struct vcpu_vmx *vmx = to_vmx(vcpu);
1595 struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
6aa8b732 1596
ad312c7c 1597 vcpu->arch.shadow_efer = efer;
9f62e19a
JR		 1598 if (!msr)
1599 return;
6aa8b732
AK		 1600 if (efer & EFER_LMA) {
1601 vmcs_write32(VM_ENTRY_CONTROLS,
1602 vmcs_read32(VM_ENTRY_CONTROLS) |
1e4e6e00 1603 VM_ENTRY_IA32E_MODE);
6aa8b732
AK		 1604 msr->data = efer;
1605
1606 } else {
1607 vmcs_write32(VM_ENTRY_CONTROLS,
1608 vmcs_read32(VM_ENTRY_CONTROLS) &
1e4e6e00 1609 ~VM_ENTRY_IA32E_MODE);
6aa8b732
AK		 1610
1611 msr->data = efer & ~EFER_LME;
1612 }
8b9cf98c 1613 setup_msrs(vmx);
6aa8b732
AK		 1614}
1615
6aa8b732
AK		 1616static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1617{
1618 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1619
1620 return vmcs_readl(sf->base);
1621}
1622
1623static void vmx_get_segment(struct kvm_vcpu *vcpu,
1624 struct kvm_segment *var, int seg)
1625{
1626 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1627 u32 ar;
1628
1629 var->base = vmcs_readl(sf->base);
1630 var->limit = vmcs_read32(sf->limit);
1631 var->selector = vmcs_read16(sf->selector);
1632 ar = vmcs_read32(sf->ar_bytes);
1633 if (ar & AR_UNUSABLE_MASK)
1634 ar = 0;
1635 var->type = ar & 15;
1636 var->s = (ar >> 4) & 1;
1637 var->dpl = (ar >> 5) & 3;
1638 var->present = (ar >> 7) & 1;
1639 var->avl = (ar >> 12) & 1;
1640 var->l = (ar >> 13) & 1;
1641 var->db = (ar >> 14) & 1;
1642 var->g = (ar >> 15) & 1;
1643 var->unusable = (ar >> 16) & 1;
1644}
1645
2e4d2653
IE		 1646static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1647{
1648 struct kvm_segment kvm_seg;
1649
1650 if (!(vcpu->arch.cr0 & X86_CR0_PE)) /* if real mode */
1651 return 0;
1652
1653 if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1654 return 3;
1655
1656 vmx_get_segment(vcpu, &kvm_seg, VCPU_SREG_CS);
1657 return kvm_seg.selector & 3;
1658}
1659
653e3108 1660static u32 vmx_segment_access_rights(struct kvm_segment *var)
6aa8b732 1661{
6aa8b732
AK		 1662 u32 ar;
1663
653e3108 1664 if (var->unusable)
6aa8b732
AK		 1665 ar = 1 << 16;
1666 else {
1667 ar = var->type & 15;
1668 ar |= (var->s & 1) << 4;
1669 ar |= (var->dpl & 3) << 5;
1670 ar |= (var->present & 1) << 7;
1671 ar |= (var->avl & 1) << 12;
1672 ar |= (var->l & 1) << 13;
1673 ar |= (var->db & 1) << 14;
1674 ar |= (var->g & 1) << 15;
1675 }
f7fbf1fd
UL		 1676 if (ar == 0) /* a 0 value means unusable */
1677 ar = AR_UNUSABLE_MASK;
653e3108
AK		 1678
1679 return ar;
1680}
1681
1682static void vmx_set_segment(struct kvm_vcpu *vcpu,
1683 struct kvm_segment *var, int seg)
1684{
1685 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1686 u32 ar;
1687
ad312c7c
ZX		 1688 if (vcpu->arch.rmode.active && seg == VCPU_SREG_TR) {
1689 vcpu->arch.rmode.tr.selector = var->selector;
1690 vcpu->arch.rmode.tr.base = var->base;
1691 vcpu->arch.rmode.tr.limit = var->limit;
1692 vcpu->arch.rmode.tr.ar = vmx_segment_access_rights(var);
653e3108
AK		 1693 return;
1694 }
1695 vmcs_writel(sf->base, var->base);
1696 vmcs_write32(sf->limit, var->limit);
1697 vmcs_write16(sf->selector, var->selector);
ad312c7c 1698 if (vcpu->arch.rmode.active && var->s) {
653e3108
AK		 1699 /*
1700 * Hack real-mode segments into vm86 compatibility.
1701 */
1702 if (var->base == 0xffff0000 && var->selector == 0xf000)
1703 vmcs_writel(sf->base, 0xf0000);
1704 ar = 0xf3;
1705 } else
1706 ar = vmx_segment_access_rights(var);
6aa8b732
AK		 1707 vmcs_write32(sf->ar_bytes, ar);
1708}
1709
6aa8b732
AK		 1710static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1711{
1712 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1713
1714 *db = (ar >> 14) & 1;
1715 *l = (ar >> 13) & 1;
1716}
1717
1718static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1719{
1720 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1721 dt->base = vmcs_readl(GUEST_IDTR_BASE);
1722}
1723
1724static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1725{
1726 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1727 vmcs_writel(GUEST_IDTR_BASE, dt->base);
1728}
1729
1730static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1731{
1732 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1733 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1734}
1735
1736static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1737{
1738 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1739 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1740}
1741
648dfaa7
MG		 1742static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
1743{
1744 struct kvm_segment var;
1745 u32 ar;
1746
1747 vmx_get_segment(vcpu, &var, seg);
1748 ar = vmx_segment_access_rights(&var);
1749
1750 if (var.base != (var.selector << 4))
1751 return false;
1752 if (var.limit != 0xffff)
1753 return false;
1754 if (ar != 0xf3)
1755 return false;
1756
1757 return true;
1758}
1759
1760static bool code_segment_valid(struct kvm_vcpu *vcpu)
1761{
1762 struct kvm_segment cs;
1763 unsigned int cs_rpl;
1764
1765 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1766 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
1767
1768 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
1769 return false;
1770 if (!cs.s)
1771 return false;
1772 if (!(~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK))) {
1773 if (cs.dpl > cs_rpl)
1774 return false;
1775 } else if (cs.type & AR_TYPE_CODE_MASK) {
1776 if (cs.dpl != cs_rpl)
1777 return false;
1778 }
1779 if (!cs.present)
1780 return false;
1781
1782 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
1783 return true;
1784}
1785
1786static bool stack_segment_valid(struct kvm_vcpu *vcpu)
1787{
1788 struct kvm_segment ss;
1789 unsigned int ss_rpl;
1790
1791 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1792 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
1793
1794 if ((ss.type != 3) || (ss.type != 7))
1795 return false;
1796 if (!ss.s)
1797 return false;
1798 if (ss.dpl != ss_rpl) /* DPL != RPL */
1799 return false;
1800 if (!ss.present)
1801 return false;
1802
1803 return true;
1804}
1805
1806static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
1807{
1808 struct kvm_segment var;
1809 unsigned int rpl;
1810
1811 vmx_get_segment(vcpu, &var, seg);
1812 rpl = var.selector & SELECTOR_RPL_MASK;
1813
1814 if (!var.s)
1815 return false;
1816 if (!var.present)
1817 return false;
1818 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
1819 if (var.dpl < rpl) /* DPL < RPL */
1820 return false;
1821 }
1822
1823 /* TODO: Add other members to kvm_segment_field to allow checking for other access
1824 * rights flags
1825 */
1826 return true;
1827}
1828
1829static bool tr_valid(struct kvm_vcpu *vcpu)
1830{
1831 struct kvm_segment tr;
1832
1833 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
1834
1835 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
1836 return false;
1837 if ((tr.type != 3) || (tr.type != 11)) /* TODO: Check if guest is in IA32e mode */
1838 return false;
1839 if (!tr.present)
1840 return false;
1841
1842 return true;
1843}
1844
1845static bool ldtr_valid(struct kvm_vcpu *vcpu)
1846{
1847 struct kvm_segment ldtr;
1848
1849 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
1850
1851 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
1852 return false;
1853 if (ldtr.type != 2)
1854 return false;
1855 if (!ldtr.present)
1856 return false;
1857
1858 return true;
1859}
1860
1861static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
1862{
1863 struct kvm_segment cs, ss;
1864
1865 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1866 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1867
1868 return ((cs.selector & SELECTOR_RPL_MASK) ==
1869 (ss.selector & SELECTOR_RPL_MASK));
1870}
1871
1872/*
1873 * Check if guest state is valid. Returns true if valid, false if
1874 * not.
1875 * We assume that registers are always usable
1876 */
1877static bool guest_state_valid(struct kvm_vcpu *vcpu)
1878{
1879 /* real mode guest state checks */
1880 if (!(vcpu->arch.cr0 & X86_CR0_PE)) {
1881 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
1882 return false;
1883 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
1884 return false;
1885 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
1886 return false;
1887 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
1888 return false;
1889 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
1890 return false;
1891 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
1892 return false;
1893 } else {
1894 /* protected mode guest state checks */
1895 if (!cs_ss_rpl_check(vcpu))
1896 return false;
1897 if (!code_segment_valid(vcpu))
1898 return false;
1899 if (!stack_segment_valid(vcpu))
1900 return false;
1901 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
1902 return false;
1903 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
1904 return false;
1905 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
1906 return false;
1907 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
1908 return false;
1909 if (!tr_valid(vcpu))
1910 return false;
1911 if (!ldtr_valid(vcpu))
1912 return false;
1913 }
1914 /* TODO:
1915 * - Add checks on RIP
1916 * - Add checks on RFLAGS
1917 */
1918
1919 return true;
1920}
1921
d77c26fc 1922static int init_rmode_tss(struct kvm *kvm)
6aa8b732 1923{
6aa8b732 1924 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
195aefde 1925 u16 data = 0;
10589a46 1926 int ret = 0;
195aefde 1927 int r;
6aa8b732 1928
195aefde
IE		 1929 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1930 if (r < 0)
10589a46 1931 goto out;
195aefde 1932 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
464d17c8
SY		 1933 r = kvm_write_guest_page(kvm, fn++, &data,
1934 TSS_IOPB_BASE_OFFSET, sizeof(u16));
195aefde 1935 if (r < 0)
10589a46 1936 goto out;
195aefde
IE		 1937 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
1938 if (r < 0)
10589a46 1939 goto out;
195aefde
IE		 1940 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1941 if (r < 0)
10589a46 1942 goto out;
195aefde 1943 data = ~0;
10589a46
MT		 1944 r = kvm_write_guest_page(kvm, fn, &data,
1945 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
1946 sizeof(u8));
195aefde 1947 if (r < 0)
10589a46
MT		 1948 goto out;
1949
1950 ret = 1;
1951out:
10589a46 1952 return ret;
6aa8b732
AK		 1953}
1954
b7ebfb05
SY		 1955static int init_rmode_identity_map(struct kvm *kvm)
1956{
1957 int i, r, ret;
1958 pfn_t identity_map_pfn;
1959 u32 tmp;
1960
1961 if (!vm_need_ept())
1962 return 1;
1963 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
1964 printk(KERN_ERR "EPT: identity-mapping pagetable "
1965 "haven't been allocated!\n");
1966 return 0;
1967 }
1968 if (likely(kvm->arch.ept_identity_pagetable_done))
1969 return 1;
1970 ret = 0;
1971 identity_map_pfn = VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT;
1972 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
1973 if (r < 0)
1974 goto out;
1975 /* Set up identity-mapping pagetable for EPT in real mode */
1976 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
1977 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
1978 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
1979 r = kvm_write_guest_page(kvm, identity_map_pfn,
1980 &tmp, i * sizeof(tmp), sizeof(tmp));
1981 if (r < 0)
1982 goto out;
1983 }
1984 kvm->arch.ept_identity_pagetable_done = true;
1985 ret = 1;
1986out:
1987 return ret;
1988}
1989
6aa8b732
AK		 1990static void seg_setup(int seg)
1991{
1992 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1993
1994 vmcs_write16(sf->selector, 0);
1995 vmcs_writel(sf->base, 0);
1996 vmcs_write32(sf->limit, 0xffff);
a16b20da 1997 vmcs_write32(sf->ar_bytes, 0xf3);
6aa8b732
AK		 1998}
1999
f78e0e2e
SY		 2000static int alloc_apic_access_page(struct kvm *kvm)
2001{
2002 struct kvm_userspace_memory_region kvm_userspace_mem;
2003 int r = 0;
2004
72dc67a6 2005 down_write(&kvm->slots_lock);
bfc6d222 2006 if (kvm->arch.apic_access_page)
f78e0e2e
SY		 2007 goto out;
2008 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2009 kvm_userspace_mem.flags = 0;
2010 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2011 kvm_userspace_mem.memory_size = PAGE_SIZE;
2012 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2013 if (r)
2014 goto out;
72dc67a6 2015
bfc6d222 2016 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
f78e0e2e 2017out:
72dc67a6 2018 up_write(&kvm->slots_lock);
f78e0e2e
SY		 2019 return r;
2020}
2021
b7ebfb05
SY		 2022static int alloc_identity_pagetable(struct kvm *kvm)
2023{
2024 struct kvm_userspace_memory_region kvm_userspace_mem;
2025 int r = 0;
2026
2027 down_write(&kvm->slots_lock);
2028 if (kvm->arch.ept_identity_pagetable)
2029 goto out;
2030 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2031 kvm_userspace_mem.flags = 0;
2032 kvm_userspace_mem.guest_phys_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
2033 kvm_userspace_mem.memory_size = PAGE_SIZE;
2034 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2035 if (r)
2036 goto out;
2037
b7ebfb05
SY		 2038 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2039 VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT);
b7ebfb05
SY		 2040out:
2041 up_write(&kvm->slots_lock);
2042 return r;
2043}
2044
2384d2b3
SY		 2045static void allocate_vpid(struct vcpu_vmx *vmx)
2046{
2047 int vpid;
2048
2049 vmx->vpid = 0;
2050 if (!enable_vpid || !cpu_has_vmx_vpid())
2051 return;
2052 spin_lock(&vmx_vpid_lock);
2053 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2054 if (vpid < VMX_NR_VPIDS) {
2055 vmx->vpid = vpid;
2056 __set_bit(vpid, vmx_vpid_bitmap);
2057 }
2058 spin_unlock(&vmx_vpid_lock);
2059}
2060
8b2cf73c 2061static void vmx_disable_intercept_for_msr(struct page *msr_bitmap, u32 msr)
25c5f225
SY		 2062{
2063 void *va;
2064
2065 if (!cpu_has_vmx_msr_bitmap())
2066 return;
2067
2068 /*
2069 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2070 * have the write-low and read-high bitmap offsets the wrong way round.
2071 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2072 */
2073 va = kmap(msr_bitmap);
2074 if (msr <= 0x1fff) {
2075 __clear_bit(msr, va + 0x000); /* read-low */
2076 __clear_bit(msr, va + 0x800); /* write-low */
2077 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2078 msr &= 0x1fff;
2079 __clear_bit(msr, va + 0x400); /* read-high */
2080 __clear_bit(msr, va + 0xc00); /* write-high */
2081 }
2082 kunmap(msr_bitmap);
2083}
2084
6aa8b732
AK		 2085/*
2086 * Sets up the vmcs for emulated real mode.
2087 */
8b9cf98c 2088static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
6aa8b732
AK		 2089{
2090 u32 host_sysenter_cs;
2091 u32 junk;
2092 unsigned long a;
2093 struct descriptor_table dt;
2094 int i;
cd2276a7 2095 unsigned long kvm_vmx_return;
6e5d865c 2096 u32 exec_control;
6aa8b732 2097
6aa8b732 2098 /* I/O */
fdef3ad1
HQ		 2099 vmcs_write64(IO_BITMAP_A, page_to_phys(vmx_io_bitmap_a));
2100 vmcs_write64(IO_BITMAP_B, page_to_phys(vmx_io_bitmap_b));
6aa8b732 2101
25c5f225
SY		 2102 if (cpu_has_vmx_msr_bitmap())
2103 vmcs_write64(MSR_BITMAP, page_to_phys(vmx_msr_bitmap));
2104
6aa8b732
AK		 2105 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2106
6aa8b732 2107 /* Control */
1c3d14fe
YS		 2108 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2109 vmcs_config.pin_based_exec_ctrl);
6e5d865c
YS		 2110
2111 exec_control = vmcs_config.cpu_based_exec_ctrl;
2112 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2113 exec_control &= ~CPU_BASED_TPR_SHADOW;
2114#ifdef CONFIG_X86_64
2115 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2116 CPU_BASED_CR8_LOAD_EXITING;
2117#endif
2118 }
d56f546d
SY		 2119 if (!vm_need_ept())
2120 exec_control |= CPU_BASED_CR3_STORE_EXITING |
83dbc83a
MT		 2121 CPU_BASED_CR3_LOAD_EXITING |
2122 CPU_BASED_INVLPG_EXITING;
6e5d865c 2123 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
6aa8b732 2124
83ff3b9d
SY		 2125 if (cpu_has_secondary_exec_ctrls()) {
2126 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2127 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2128 exec_control &=
2129 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2384d2b3
SY		 2130 if (vmx->vpid == 0)
2131 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
d56f546d
SY		 2132 if (!vm_need_ept())
2133 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
83ff3b9d
SY		 2134 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2135 }
f78e0e2e 2136
c7addb90
AK		 2137 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2138 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
6aa8b732
AK		 2139 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
2140
2141 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
2142 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
2143 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2144
2145 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
2146 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2147 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
d6e88aec
AK		 2148 vmcs_write16(HOST_FS_SELECTOR, kvm_read_fs()); /* 22.2.4 */
2149 vmcs_write16(HOST_GS_SELECTOR, kvm_read_gs()); /* 22.2.4 */
6aa8b732 2150 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
05b3e0c2 2151#ifdef CONFIG_X86_64
6aa8b732
AK		 2152 rdmsrl(MSR_FS_BASE, a);
2153 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2154 rdmsrl(MSR_GS_BASE, a);
2155 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2156#else
2157 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2158 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2159#endif
2160
2161 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
2162
d6e88aec 2163 kvm_get_idt(&dt);
6aa8b732
AK		 2164 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
2165
d77c26fc 2166 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
cd2276a7 2167 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2cc51560
ED		 2168 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2169 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2170 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
6aa8b732
AK		 2171
2172 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2173 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2174 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2175 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
2176 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2177 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2178
6aa8b732
AK		 2179 for (i = 0; i < NR_VMX_MSR; ++i) {
2180 u32 index = vmx_msr_index[i];
2181 u32 data_low, data_high;
2182 u64 data;
a2fa3e9f 2183 int j = vmx->nmsrs;
6aa8b732
AK		 2184
2185 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2186 continue;
432bd6cb
AK		 2187 if (wrmsr_safe(index, data_low, data_high) < 0)
2188 continue;
6aa8b732 2189 data = data_low | ((u64)data_high << 32);
a2fa3e9f
GH		 2190 vmx->host_msrs[j].index = index;
2191 vmx->host_msrs[j].reserved = 0;
2192 vmx->host_msrs[j].data = data;
2193 vmx->guest_msrs[j] = vmx->host_msrs[j];
2194 ++vmx->nmsrs;
6aa8b732 2195 }
6aa8b732 2196
1c3d14fe 2197 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
6aa8b732
AK		 2198
2199 /* 22.2.1, 20.8.1 */
1c3d14fe
YS		 2200 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2201
e00c8cf2
AK		 2202 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2203 vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
2204
f78e0e2e 2205
e00c8cf2
AK		 2206 return 0;
2207}
2208
b7ebfb05
SY		 2209static int init_rmode(struct kvm *kvm)
2210{
2211 if (!init_rmode_tss(kvm))
2212 return 0;
2213 if (!init_rmode_identity_map(kvm))
2214 return 0;
2215 return 1;
2216}
2217
e00c8cf2
AK		 2218static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2219{
2220 struct vcpu_vmx *vmx = to_vmx(vcpu);
2221 u64 msr;
2222 int ret;
2223
5fdbf976 2224 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3200f405 2225 down_read(&vcpu->kvm->slots_lock);
b7ebfb05 2226 if (!init_rmode(vmx->vcpu.kvm)) {
e00c8cf2
AK		 2227 ret = -ENOMEM;
2228 goto out;
2229 }
2230
ad312c7c 2231 vmx->vcpu.arch.rmode.active = 0;
e00c8cf2 2232
ad312c7c 2233 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2d3ad1f4 2234 kvm_set_cr8(&vmx->vcpu, 0);
e00c8cf2
AK		 2235 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2236 if (vmx->vcpu.vcpu_id == 0)
2237 msr |= MSR_IA32_APICBASE_BSP;
2238 kvm_set_apic_base(&vmx->vcpu, msr);
2239
2240 fx_init(&vmx->vcpu);
2241
5706be0d 2242 seg_setup(VCPU_SREG_CS);
e00c8cf2
AK		 2243 /*
2244 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2245 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2246 */
2247 if (vmx->vcpu.vcpu_id == 0) {
2248 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2249 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2250 } else {
ad312c7c
ZX		 2251 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2252 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
e00c8cf2 2253 }
e00c8cf2
AK		 2254
2255 seg_setup(VCPU_SREG_DS);
2256 seg_setup(VCPU_SREG_ES);
2257 seg_setup(VCPU_SREG_FS);
2258 seg_setup(VCPU_SREG_GS);
2259 seg_setup(VCPU_SREG_SS);
2260
2261 vmcs_write16(GUEST_TR_SELECTOR, 0);
2262 vmcs_writel(GUEST_TR_BASE, 0);
2263 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2264 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2265
2266 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2267 vmcs_writel(GUEST_LDTR_BASE, 0);
2268 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2269 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2270
2271 vmcs_write32(GUEST_SYSENTER_CS, 0);
2272 vmcs_writel(GUEST_SYSENTER_ESP, 0);
2273 vmcs_writel(GUEST_SYSENTER_EIP, 0);
2274
2275 vmcs_writel(GUEST_RFLAGS, 0x02);
2276 if (vmx->vcpu.vcpu_id == 0)
5fdbf976 2277 kvm_rip_write(vcpu, 0xfff0);
e00c8cf2 2278 else
5fdbf976
MT		 2279 kvm_rip_write(vcpu, 0);
2280 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
e00c8cf2
AK		 2281
2282 /* todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0 */
2283 vmcs_writel(GUEST_DR7, 0x400);
2284
2285 vmcs_writel(GUEST_GDTR_BASE, 0);
2286 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2287
2288 vmcs_writel(GUEST_IDTR_BASE, 0);
2289 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2290
2291 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2292 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2293 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2294
2295 guest_write_tsc(0);
2296
2297 /* Special registers */
2298 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2299
2300 setup_msrs(vmx);
2301
6aa8b732
AK		 2302 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
2303
f78e0e2e
SY		 2304 if (cpu_has_vmx_tpr_shadow()) {
2305 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2306 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2307 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
ad312c7c 2308 page_to_phys(vmx->vcpu.arch.apic->regs_page));
f78e0e2e
SY		 2309 vmcs_write32(TPR_THRESHOLD, 0);
2310 }
2311
2312 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2313 vmcs_write64(APIC_ACCESS_ADDR,
bfc6d222 2314 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
6aa8b732 2315
2384d2b3
SY		 2316 if (vmx->vpid != 0)
2317 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2318
ad312c7c
ZX		 2319 vmx->vcpu.arch.cr0 = 0x60000010;
2320 vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
8b9cf98c 2321 vmx_set_cr4(&vmx->vcpu, 0);
8b9cf98c 2322 vmx_set_efer(&vmx->vcpu, 0);
8b9cf98c
RR		 2323 vmx_fpu_activate(&vmx->vcpu);
2324 update_exception_bitmap(&vmx->vcpu);
6aa8b732 2325
2384d2b3
SY		 2326 vpid_sync_vcpu_all(vmx);
2327
3200f405 2328 ret = 0;
6aa8b732 2329
a89a8fb9
MG		 2330 /* HACK: Don't enable emulation on guest boot/reset */
2331 vmx->emulation_required = 0;
2332
6aa8b732 2333out:
3200f405 2334 up_read(&vcpu->kvm->slots_lock);
6aa8b732
AK		 2335 return ret;
2336}
2337
85f455f7
ED		 2338static void vmx_inject_irq(struct kvm_vcpu *vcpu, int irq)
2339{
9c8cba37
AK		 2340 struct vcpu_vmx *vmx = to_vmx(vcpu);
2341
2714d1d3
FEL		 2342 KVMTRACE_1D(INJ_VIRQ, vcpu, (u32)irq, handler);
2343
fa89a817 2344 ++vcpu->stat.irq_injections;
ad312c7c 2345 if (vcpu->arch.rmode.active) {
9c8cba37
AK		 2346 vmx->rmode.irq.pending = true;
2347 vmx->rmode.irq.vector = irq;
5fdbf976 2348 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
9c5623e3
AK		 2349 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2350 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2351 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
5fdbf976 2352 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
85f455f7
ED		 2353 return;
2354 }
2355 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2356 irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
2357}
2358
f08864b4
SY		 2359static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2360{
2361 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2362 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
f08864b4
SY		 2363}
2364
33f089ca
JK		 2365static void vmx_update_window_states(struct kvm_vcpu *vcpu)
2366{
2367 u32 guest_intr = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2368
2369 vcpu->arch.nmi_window_open =
2370 !(guest_intr & (GUEST_INTR_STATE_STI |
2371 GUEST_INTR_STATE_MOV_SS |
2372 GUEST_INTR_STATE_NMI));
2373
2374 vcpu->arch.interrupt_window_open =
2375 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2376 !(guest_intr & (GUEST_INTR_STATE_STI |
2377 GUEST_INTR_STATE_MOV_SS)));
2378}
2379
6aa8b732
AK		 2380static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
2381{
ad312c7c
ZX		 2382 int word_index = __ffs(vcpu->arch.irq_summary);
2383 int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
6aa8b732
AK		 2384 int irq = word_index * BITS_PER_LONG + bit_index;
2385
ad312c7c
ZX		 2386 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
2387 if (!vcpu->arch.irq_pending[word_index])
2388 clear_bit(word_index, &vcpu->arch.irq_summary);
ecfc79c7 2389 kvm_queue_interrupt(vcpu, irq);
6aa8b732
AK		 2390}
2391
f460ee43 2392static void enable_irq_window(struct kvm_vcpu *vcpu)
6aa8b732 2393{
c1150d8c
DL		 2394 u32 cpu_based_vm_exec_control;
2395
f460ee43
JK		 2396 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2397 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2398 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2399}
c1150d8c 2400
f460ee43
JK		 2401static void enable_nmi_window(struct kvm_vcpu *vcpu)
2402{
2403 u32 cpu_based_vm_exec_control;
c1150d8c 2404
f460ee43
JK		 2405 if (!cpu_has_virtual_nmis())
2406 return;
ecfc79c7 2407
c1150d8c 2408 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
f460ee43
JK		 2409 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2410 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2411}
2412
2413static void do_interrupt_requests(struct kvm_vcpu *vcpu,
2414 struct kvm_run *kvm_run)
2415{
2416 vmx_update_window_states(vcpu);
2417
2418 if (vcpu->arch.interrupt_window_open) {
2419 if (vcpu->arch.irq_summary && !vcpu->arch.interrupt.pending)
2420 kvm_do_inject_irq(vcpu);
2421
2422 if (vcpu->arch.interrupt.pending)
2423 vmx_inject_irq(vcpu, vcpu->arch.interrupt.nr);
2424 }
ad312c7c
ZX		 2425 if (!vcpu->arch.interrupt_window_open &&
2426 (vcpu->arch.irq_summary || kvm_run->request_interrupt_window))
f460ee43 2427 enable_irq_window(vcpu);
6aa8b732
AK		 2428}
2429
cbc94022
IE		 2430static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2431{
2432 int ret;
2433 struct kvm_userspace_memory_region tss_mem = {
2434 .slot = 8,
2435 .guest_phys_addr = addr,
2436 .memory_size = PAGE_SIZE * 3,
2437 .flags = 0,
2438 };
2439
2440 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2441 if (ret)
2442 return ret;
bfc6d222 2443 kvm->arch.tss_addr = addr;
cbc94022
IE		 2444 return 0;
2445}
2446
6aa8b732
AK		 2447static void kvm_guest_debug_pre(struct kvm_vcpu *vcpu)
2448{
2449 struct kvm_guest_debug *dbg = &vcpu->guest_debug;
2450
2451 set_debugreg(dbg->bp[0], 0);
2452 set_debugreg(dbg->bp[1], 1);
2453 set_debugreg(dbg->bp[2], 2);
2454 set_debugreg(dbg->bp[3], 3);
2455
2456 if (dbg->singlestep) {
2457 unsigned long flags;
2458
2459 flags = vmcs_readl(GUEST_RFLAGS);
2460 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
2461 vmcs_writel(GUEST_RFLAGS, flags);
2462 }
2463}
2464
2465static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2466 int vec, u32 err_code)
2467{
b3f37707
NK		 2468 /*
2469 * Instruction with address size override prefix opcode 0x67
2470 * Cause the #SS fault with 0 error code in VM86 mode.
2471 */
2472 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
3427318f 2473 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE)
6aa8b732 2474 return 1;
77ab6db0
JK		 2475 /*
2476 * Forward all other exceptions that are valid in real mode.
2477 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2478 * the required debugging infrastructure rework.
2479 */
2480 switch (vec) {
2481 case DE_VECTOR:
2482 case DB_VECTOR:
2483 case BP_VECTOR:
2484 case OF_VECTOR:
2485 case BR_VECTOR:
2486 case UD_VECTOR:
2487 case DF_VECTOR:
2488 case SS_VECTOR:
2489 case GP_VECTOR:
2490 case MF_VECTOR:
2491 kvm_queue_exception(vcpu, vec);
2492 return 1;
2493 }
6aa8b732
AK		 2494 return 0;
2495}
2496
2497static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2498{
1155f76a 2499 struct vcpu_vmx *vmx = to_vmx(vcpu);
6aa8b732
AK		 2500 u32 intr_info, error_code;
2501 unsigned long cr2, rip;
2502 u32 vect_info;
2503 enum emulation_result er;
2504
1155f76a 2505 vect_info = vmx->idt_vectoring_info;
6aa8b732
AK		 2506 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2507
2508 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
d77c26fc 2509 !is_page_fault(intr_info))
6aa8b732 2510 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
b8688d51 2511 "intr info 0x%x\n", __func__, vect_info, intr_info);
6aa8b732 2512
85f455f7 2513 if (!irqchip_in_kernel(vcpu->kvm) && is_external_interrupt(vect_info)) {
6aa8b732 2514 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
ad312c7c
ZX		 2515 set_bit(irq, vcpu->arch.irq_pending);
2516 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
6aa8b732
AK		 2517 }
2518
e4a41889 2519 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
1b6269db 2520 return 1; /* already handled by vmx_vcpu_run() */
2ab455cc
AL		 2521
2522 if (is_no_device(intr_info)) {
5fd86fcf 2523 vmx_fpu_activate(vcpu);
2ab455cc
AL		 2524 return 1;
2525 }
2526
7aa81cc0 2527 if (is_invalid_opcode(intr_info)) {
571008da 2528 er = emulate_instruction(vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
7aa81cc0 2529 if (er != EMULATE_DONE)
7ee5d940 2530 kvm_queue_exception(vcpu, UD_VECTOR);
7aa81cc0
AL		 2531 return 1;
2532 }
2533
6aa8b732 2534 error_code = 0;
5fdbf976 2535 rip = kvm_rip_read(vcpu);
2e11384c 2536 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
6aa8b732
AK		 2537 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2538 if (is_page_fault(intr_info)) {
1439442c
SY		 2539 /* EPT won't cause page fault directly */
2540 if (vm_need_ept())
2541 BUG();
6aa8b732 2542 cr2 = vmcs_readl(EXIT_QUALIFICATION);
2714d1d3
FEL		 2543 KVMTRACE_3D(PAGE_FAULT, vcpu, error_code, (u32)cr2,
2544 (u32)((u64)cr2 >> 32), handler);
f7d9238f 2545 if (vcpu->arch.interrupt.pending || vcpu->arch.exception.pending)
577bdc49 2546 kvm_mmu_unprotect_page_virt(vcpu, cr2);
3067714c 2547 return kvm_mmu_page_fault(vcpu, cr2, error_code);
6aa8b732
AK		 2548 }
2549
ad312c7c 2550 if (vcpu->arch.rmode.active &&
6aa8b732 2551 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
72d6e5a0 2552 error_code)) {
ad312c7c
ZX		 2553 if (vcpu->arch.halt_request) {
2554 vcpu->arch.halt_request = 0;
72d6e5a0
AK		 2555 return kvm_emulate_halt(vcpu);
2556 }
6aa8b732 2557 return 1;
72d6e5a0 2558 }
6aa8b732 2559
d77c26fc
MD		 2560 if ((intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK)) ==
2561 (INTR_TYPE_EXCEPTION | 1)) {
6aa8b732
AK		 2562 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2563 return 0;
2564 }
2565 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
2566 kvm_run->ex.exception = intr_info & INTR_INFO_VECTOR_MASK;
2567 kvm_run->ex.error_code = error_code;
2568 return 0;
2569}
2570
2571static int handle_external_interrupt(struct kvm_vcpu *vcpu,
2572 struct kvm_run *kvm_run)
2573{
1165f5fe 2574 ++vcpu->stat.irq_exits;
2714d1d3 2575 KVMTRACE_1D(INTR, vcpu, vmcs_read32(VM_EXIT_INTR_INFO), handler);
6aa8b732
AK		 2576 return 1;
2577}
2578
988ad74f
AK		 2579static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2580{
2581 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2582 return 0;
2583}
6aa8b732 2584
6aa8b732
AK		 2585static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2586{
bfdaab09 2587 unsigned long exit_qualification;
039576c0
AK		 2588 int size, down, in, string, rep;
2589 unsigned port;
6aa8b732 2590
1165f5fe 2591 ++vcpu->stat.io_exits;
bfdaab09 2592 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
039576c0 2593 string = (exit_qualification & 16) != 0;
e70669ab
LV		 2594
2595 if (string) {
3427318f
LV		 2596 if (emulate_instruction(vcpu,
2597 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
e70669ab
LV		 2598 return 0;
2599 return 1;
2600 }
2601
2602 size = (exit_qualification & 7) + 1;
2603 in = (exit_qualification & 8) != 0;
039576c0 2604 down = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
039576c0
AK		 2605 rep = (exit_qualification & 32) != 0;
2606 port = exit_qualification >> 16;
e70669ab 2607
3090dd73 2608 return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
6aa8b732
AK		 2609}
2610
102d8325
IM		 2611static void
2612vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2613{
2614 /*
2615 * Patch in the VMCALL instruction:
2616 */
2617 hypercall[0] = 0x0f;
2618 hypercall[1] = 0x01;
2619 hypercall[2] = 0xc1;
102d8325
IM		 2620}
2621
6aa8b732
AK		 2622static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2623{
bfdaab09 2624 unsigned long exit_qualification;
6aa8b732
AK		 2625 int cr;
2626 int reg;
2627
bfdaab09 2628 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6aa8b732
AK		 2629 cr = exit_qualification & 15;
2630 reg = (exit_qualification >> 8) & 15;
2631 switch ((exit_qualification >> 4) & 3) {
2632 case 0: /* mov to cr */
5fdbf976
MT		 2633 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr,
2634 (u32)kvm_register_read(vcpu, reg),
2635 (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2636 handler);
6aa8b732
AK		 2637 switch (cr) {
2638 case 0:
5fdbf976 2639 kvm_set_cr0(vcpu, kvm_register_read(vcpu, reg));
6aa8b732
AK		 2640 skip_emulated_instruction(vcpu);
2641 return 1;
2642 case 3:
5fdbf976 2643 kvm_set_cr3(vcpu, kvm_register_read(vcpu, reg));
6aa8b732
AK		 2644 skip_emulated_instruction(vcpu);
2645 return 1;
2646 case 4:
5fdbf976 2647 kvm_set_cr4(vcpu, kvm_register_read(vcpu, reg));
6aa8b732
AK		 2648 skip_emulated_instruction(vcpu);
2649 return 1;
2650 case 8:
5fdbf976 2651 kvm_set_cr8(vcpu, kvm_register_read(vcpu, reg));
6aa8b732 2652 skip_emulated_instruction(vcpu);
e5314067
AK		 2653 if (irqchip_in_kernel(vcpu->kvm))
2654 return 1;
253abdee
YS		 2655 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2656 return 0;
6aa8b732
AK		 2657 };
2658 break;
25c4c276 2659 case 2: /* clts */
5fd86fcf 2660 vmx_fpu_deactivate(vcpu);
ad312c7c
ZX		 2661 vcpu->arch.cr0 &= ~X86_CR0_TS;
2662 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
5fd86fcf 2663 vmx_fpu_activate(vcpu);
2714d1d3 2664 KVMTRACE_0D(CLTS, vcpu, handler);
25c4c276
AL		 2665 skip_emulated_instruction(vcpu);
2666 return 1;
6aa8b732
AK		 2667 case 1: /*mov from cr*/
2668 switch (cr) {
2669 case 3:
5fdbf976 2670 kvm_register_write(vcpu, reg, vcpu->arch.cr3);
2714d1d3 2671 KVMTRACE_3D(CR_READ, vcpu, (u32)cr,
5fdbf976
MT		 2672 (u32)kvm_register_read(vcpu, reg),
2673 (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2714d1d3 2674 handler);
6aa8b732
AK		 2675 skip_emulated_instruction(vcpu);
2676 return 1;
2677 case 8:
5fdbf976 2678 kvm_register_write(vcpu, reg, kvm_get_cr8(vcpu));
2714d1d3 2679 KVMTRACE_2D(CR_READ, vcpu, (u32)cr,
5fdbf976 2680 (u32)kvm_register_read(vcpu, reg), handler);
6aa8b732
AK		 2681 skip_emulated_instruction(vcpu);
2682 return 1;
2683 }
2684 break;
2685 case 3: /* lmsw */
2d3ad1f4 2686 kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
6aa8b732
AK		 2687
2688 skip_emulated_instruction(vcpu);
2689 return 1;
2690 default:
2691 break;
2692 }
2693 kvm_run->exit_reason = 0;
f0242478 2694 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
6aa8b732
AK		 2695 (int)(exit_qualification >> 4) & 3, cr);
2696 return 0;
2697}
2698
2699static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2700{
bfdaab09 2701 unsigned long exit_qualification;
6aa8b732
AK		 2702 unsigned long val;
2703 int dr, reg;
2704
2705 /*
2706 * FIXME: this code assumes the host is debugging the guest.
2707 * need to deal with guest debugging itself too.
2708 */
bfdaab09 2709 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6aa8b732
AK		 2710 dr = exit_qualification & 7;
2711 reg = (exit_qualification >> 8) & 15;
6aa8b732
AK		 2712 if (exit_qualification & 16) {
2713 /* mov from dr */
2714 switch (dr) {
2715 case 6:
2716 val = 0xffff0ff0;
2717 break;
2718 case 7:
2719 val = 0x400;
2720 break;
2721 default:
2722 val = 0;
2723 }
5fdbf976 2724 kvm_register_write(vcpu, reg, val);
2714d1d3 2725 KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
6aa8b732
AK		 2726 } else {
2727 /* mov to dr */
2728 }
6aa8b732
AK		 2729 skip_emulated_instruction(vcpu);
2730 return 1;
2731}
2732
2733static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2734{
06465c5a
AK		 2735 kvm_emulate_cpuid(vcpu);
2736 return 1;
6aa8b732
AK		 2737}
2738
2739static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2740{
ad312c7c 2741 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6aa8b732
AK		 2742 u64 data;
2743
2744 if (vmx_get_msr(vcpu, ecx, &data)) {
c1a5d4f9 2745 kvm_inject_gp(vcpu, 0);
6aa8b732
AK		 2746 return 1;
2747 }
2748
2714d1d3
FEL		 2749 KVMTRACE_3D(MSR_READ, vcpu, ecx, (u32)data, (u32)(data >> 32),
2750 handler);
2751
6aa8b732 2752 /* FIXME: handling of bits 32:63 of rax, rdx */
ad312c7c
ZX		 2753 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
2754 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
6aa8b732
AK		 2755 skip_emulated_instruction(vcpu);
2756 return 1;
2757}
2758
2759static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2760{
ad312c7c
ZX		 2761 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2762 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
2763 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6aa8b732 2764
2714d1d3
FEL		 2765 KVMTRACE_3D(MSR_WRITE, vcpu, ecx, (u32)data, (u32)(data >> 32),
2766 handler);
2767
6aa8b732 2768 if (vmx_set_msr(vcpu, ecx, data) != 0) {
c1a5d4f9 2769 kvm_inject_gp(vcpu, 0);
6aa8b732
AK		 2770 return 1;
2771 }
2772
2773 skip_emulated_instruction(vcpu);
2774 return 1;
2775}
2776
6e5d865c
YS		 2777static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu,
2778 struct kvm_run *kvm_run)
2779{
2780 return 1;
2781}
2782
6aa8b732
AK		 2783static int handle_interrupt_window(struct kvm_vcpu *vcpu,
2784 struct kvm_run *kvm_run)
2785{
85f455f7
ED		 2786 u32 cpu_based_vm_exec_control;
2787
2788 /* clear pending irq */
2789 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2790 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2791 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2714d1d3
FEL		 2792
2793 KVMTRACE_0D(PEND_INTR, vcpu, handler);
a26bf12a 2794 ++vcpu->stat.irq_window_exits;
2714d1d3 2795
c1150d8c
DL		 2796 /*
2797 * If the user space waits to inject interrupts, exit as soon as
2798 * possible
2799 */
2800 if (kvm_run->request_interrupt_window &&
ad312c7c 2801 !vcpu->arch.irq_summary) {
c1150d8c 2802 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
c1150d8c
DL		 2803 return 0;
2804 }
6aa8b732
AK		 2805 return 1;
2806}
2807
2808static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2809{
2810 skip_emulated_instruction(vcpu);
d3bef15f 2811 return kvm_emulate_halt(vcpu);
6aa8b732
AK		 2812}
2813
c21415e8
IM		 2814static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2815{
510043da 2816 skip_emulated_instruction(vcpu);
7aa81cc0
AL		 2817 kvm_emulate_hypercall(vcpu);
2818 return 1;
c21415e8
IM		 2819}
2820
a7052897
MT		 2821static int handle_invlpg(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2822{
2823 u64 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
2824
2825 kvm_mmu_invlpg(vcpu, exit_qualification);
2826 skip_emulated_instruction(vcpu);
2827 return 1;
2828}
2829
e5edaa01
ED		 2830static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2831{
2832 skip_emulated_instruction(vcpu);
2833 /* TODO: Add support for VT-d/pass-through device */
2834 return 1;
2835}
2836
f78e0e2e
SY		 2837static int handle_apic_access(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2838{
2839 u64 exit_qualification;
2840 enum emulation_result er;
2841 unsigned long offset;
2842
2843 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
2844 offset = exit_qualification & 0xffful;
2845
2846 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
2847
2848 if (er != EMULATE_DONE) {
2849 printk(KERN_ERR
2850 "Fail to handle apic access vmexit! Offset is 0x%lx\n",
2851 offset);
2852 return -ENOTSUPP;
2853 }
2854 return 1;
2855}
2856
37817f29
IE		 2857static int handle_task_switch(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2858{
60637aac 2859 struct vcpu_vmx *vmx = to_vmx(vcpu);
37817f29
IE		 2860 unsigned long exit_qualification;
2861 u16 tss_selector;
2862 int reason;
2863
2864 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2865
2866 reason = (u32)exit_qualification >> 30;
60637aac
JK		 2867 if (reason == TASK_SWITCH_GATE && vmx->vcpu.arch.nmi_injected &&
2868 (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
2869 (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK)
2870 == INTR_TYPE_NMI_INTR) {
2871 vcpu->arch.nmi_injected = false;
2872 if (cpu_has_virtual_nmis())
2873 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
2874 GUEST_INTR_STATE_NMI);
2875 }
37817f29
IE		 2876 tss_selector = exit_qualification;
2877
2878 return kvm_task_switch(vcpu, tss_selector, reason);
2879}
2880
1439442c
SY		 2881static int handle_ept_violation(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2882{
2883 u64 exit_qualification;
2884 enum emulation_result er;
2885 gpa_t gpa;
2886 unsigned long hva;
2887 int gla_validity;
2888 int r;
2889
2890 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
2891
2892 if (exit_qualification & (1 << 6)) {
2893 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
2894 return -ENOTSUPP;
2895 }
2896
2897 gla_validity = (exit_qualification >> 7) & 0x3;
2898 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
2899 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
2900 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
2901 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
2902 (long unsigned int)vmcs_read64(GUEST_LINEAR_ADDRESS));
2903 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
2904 (long unsigned int)exit_qualification);
2905 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2906 kvm_run->hw.hardware_exit_reason = 0;
2907 return -ENOTSUPP;
2908 }
2909
2910 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
2911 hva = gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT);
2912 if (!kvm_is_error_hva(hva)) {
2913 r = kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
2914 if (r < 0) {
2915 printk(KERN_ERR "EPT: Not enough memory!\n");
2916 return -ENOMEM;
2917 }
2918 return 1;
2919 } else {
2920 /* must be MMIO */
2921 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
2922
2923 if (er == EMULATE_FAIL) {
2924 printk(KERN_ERR
2925 "EPT: Fail to handle EPT violation vmexit!er is %d\n",
2926 er);
2927 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
2928 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
2929 (long unsigned int)vmcs_read64(GUEST_LINEAR_ADDRESS));
2930 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
2931 (long unsigned int)exit_qualification);
2932 return -ENOTSUPP;
2933 } else if (er == EMULATE_DO_MMIO)
2934 return 0;
2935 }
2936 return 1;
2937}
2938
f08864b4
SY		 2939static int handle_nmi_window(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2940{
2941 u32 cpu_based_vm_exec_control;
2942
2943 /* clear pending NMI */
2944 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2945 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
2946 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2947 ++vcpu->stat.nmi_window_exits;
2948
2949 return 1;
2950}
2951
ea953ef0
MG		 2952static void handle_invalid_guest_state(struct kvm_vcpu *vcpu,
2953 struct kvm_run *kvm_run)
2954{
2955 struct vcpu_vmx *vmx = to_vmx(vcpu);
2956 int err;
2957
2958 preempt_enable();
2959 local_irq_enable();
2960
2961 while (!guest_state_valid(vcpu)) {
2962 err = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
2963
2964 switch (err) {
2965 case EMULATE_DONE:
2966 break;
2967 case EMULATE_DO_MMIO:
2968 kvm_report_emulation_failure(vcpu, "mmio");
2969 /* TODO: Handle MMIO */
2970 return;
2971 default:
2972 kvm_report_emulation_failure(vcpu, "emulation failure");
2973 return;
2974 }
2975
2976 if (signal_pending(current))
2977 break;
2978 if (need_resched())
2979 schedule();
2980 }
2981
2982 local_irq_disable();
2983 preempt_disable();
2984
2985 /* Guest state should be valid now, no more emulation should be needed */
2986 vmx->emulation_required = 0;
2987}
2988
6aa8b732
AK		 2989/*
2990 * The exit handlers return 1 if the exit was handled fully and guest execution
2991 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
2992 * to be done to userspace and return 0.
2993 */
2994static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
2995 struct kvm_run *kvm_run) = {
2996 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
2997 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
988ad74f 2998 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
f08864b4 2999 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
6aa8b732 3000 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
6aa8b732
AK		 3001 [EXIT_REASON_CR_ACCESS] = handle_cr,
3002 [EXIT_REASON_DR_ACCESS] = handle_dr,
3003 [EXIT_REASON_CPUID] = handle_cpuid,
3004 [EXIT_REASON_MSR_READ] = handle_rdmsr,
3005 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
3006 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
3007 [EXIT_REASON_HLT] = handle_halt,
a7052897 3008 [EXIT_REASON_INVLPG] = handle_invlpg,
c21415e8 3009 [EXIT_REASON_VMCALL] = handle_vmcall,
f78e0e2e
SY		 3010 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
3011 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
e5edaa01 3012 [EXIT_REASON_WBINVD] = handle_wbinvd,
37817f29 3013 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
1439442c 3014 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
6aa8b732
AK		 3015};
3016
3017static const int kvm_vmx_max_exit_handlers =
50a3485c 3018 ARRAY_SIZE(kvm_vmx_exit_handlers);
6aa8b732
AK		 3019
3020/*
3021 * The guest has exited. See if we can fix it or if we need userspace
3022 * assistance.
3023 */
3024static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3025{
6aa8b732 3026 u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
29bd8a78 3027 struct vcpu_vmx *vmx = to_vmx(vcpu);
1155f76a 3028 u32 vectoring_info = vmx->idt_vectoring_info;
29bd8a78 3029
5fdbf976
MT		 3030 KVMTRACE_3D(VMEXIT, vcpu, exit_reason, (u32)kvm_rip_read(vcpu),
3031 (u32)((u64)kvm_rip_read(vcpu) >> 32), entryexit);
2714d1d3 3032
1439442c
SY		 3033 /* Access CR3 don't cause VMExit in paging mode, so we need
3034 * to sync with guest real CR3. */
3035 if (vm_need_ept() && is_paging(vcpu)) {
3036 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3037 ept_load_pdptrs(vcpu);
3038 }
3039
29bd8a78
AK		 3040 if (unlikely(vmx->fail)) {
3041 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3042 kvm_run->fail_entry.hardware_entry_failure_reason
3043 = vmcs_read32(VM_INSTRUCTION_ERROR);
3044 return 0;
3045 }
6aa8b732 3046
d77c26fc 3047 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
1439442c 3048 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
60637aac
JK		 3049 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3050 exit_reason != EXIT_REASON_TASK_SWITCH))
3051 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3052 "(0x%x) and exit reason is 0x%x\n",
3053 __func__, vectoring_info, exit_reason);
6aa8b732
AK		 3054 if (exit_reason < kvm_vmx_max_exit_handlers
3055 && kvm_vmx_exit_handlers[exit_reason])
3056 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
3057 else {
3058 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3059 kvm_run->hw.hardware_exit_reason = exit_reason;
3060 }
3061 return 0;
3062}
3063
6e5d865c
YS		 3064static void update_tpr_threshold(struct kvm_vcpu *vcpu)
3065{
3066 int max_irr, tpr;
3067
3068 if (!vm_need_tpr_shadow(vcpu->kvm))
3069 return;
3070
3071 if (!kvm_lapic_enabled(vcpu) ||
3072 ((max_irr = kvm_lapic_find_highest_irr(vcpu)) == -1)) {
3073 vmcs_write32(TPR_THRESHOLD, 0);
3074 return;
3075 }
3076
3077 tpr = (kvm_lapic_get_cr8(vcpu) & 0x0f) << 4;
3078 vmcs_write32(TPR_THRESHOLD, (max_irr > tpr) ? tpr >> 4 : max_irr >> 4);
3079}
3080
cf393f75
AK		 3081static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3082{
3083 u32 exit_intr_info;
668f612f 3084 u32 idt_vectoring_info;
cf393f75
AK		 3085 bool unblock_nmi;
3086 u8 vector;
668f612f
AK		 3087 int type;
3088 bool idtv_info_valid;
35920a35 3089 u32 error;
cf393f75
AK		 3090
3091 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3092 if (cpu_has_virtual_nmis()) {
3093 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3094 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3095 /*
3096 * SDM 3: 25.7.1.2
3097 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3098 * a guest IRET fault.
3099 */
3100 if (unblock_nmi && vector != DF_VECTOR)
3101 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3102 GUEST_INTR_STATE_NMI);
3103 }
668f612f
AK		 3104
3105 idt_vectoring_info = vmx->idt_vectoring_info;
3106 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3107 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3108 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3109 if (vmx->vcpu.arch.nmi_injected) {
3110 /*
3111 * SDM 3: 25.7.1.2
3112 * Clear bit "block by NMI" before VM entry if a NMI delivery
3113 * faulted.
3114 */
3115 if (idtv_info_valid && type == INTR_TYPE_NMI_INTR)
3116 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3117 GUEST_INTR_STATE_NMI);
3118 else
3119 vmx->vcpu.arch.nmi_injected = false;
3120 }
35920a35
AK		 3121 kvm_clear_exception_queue(&vmx->vcpu);
3122 if (idtv_info_valid && type == INTR_TYPE_EXCEPTION) {
3123 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3124 error = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3125 kvm_queue_exception_e(&vmx->vcpu, vector, error);
3126 } else
3127 kvm_queue_exception(&vmx->vcpu, vector);
3128 vmx->idt_vectoring_info = 0;
3129 }
f7d9238f
AK		 3130 kvm_clear_interrupt_queue(&vmx->vcpu);
3131 if (idtv_info_valid && type == INTR_TYPE_EXT_INTR) {
3132 kvm_queue_interrupt(&vmx->vcpu, vector);
3133 vmx->idt_vectoring_info = 0;
3134 }
cf393f75
AK		 3135}
3136
85f455f7
ED		 3137static void vmx_intr_assist(struct kvm_vcpu *vcpu)
3138{
6e5d865c
YS		 3139 update_tpr_threshold(vcpu);
3140
33f089ca
JK		 3141 vmx_update_window_states(vcpu);
3142
f08864b4 3143 if (cpu_has_virtual_nmis()) {
668f612f 3144 if (vcpu->arch.nmi_pending && !vcpu->arch.nmi_injected) {
bd2b3ca7
AK		 3145 if (vcpu->arch.interrupt.pending) {
3146 enable_nmi_window(vcpu);
33f089ca 3147 } else if (vcpu->arch.nmi_window_open) {
668f612f
AK		 3148 vcpu->arch.nmi_pending = false;
3149 vcpu->arch.nmi_injected = true;
3150 } else {
f460ee43 3151 enable_nmi_window(vcpu);
668f612f
AK		 3152 return;
3153 }
3154 }
3155 if (vcpu->arch.nmi_injected) {
3156 vmx_inject_nmi(vcpu);
f460ee43
JK		 3157 if (vcpu->arch.nmi_pending)
3158 enable_nmi_window(vcpu);
3159 else if (kvm_cpu_has_interrupt(vcpu))
3160 enable_irq_window(vcpu);
f08864b4
SY		 3161 return;
3162 }
f08864b4 3163 }
f7d9238f 3164 if (!vcpu->arch.interrupt.pending && kvm_cpu_has_interrupt(vcpu)) {
33f089ca 3165 if (vcpu->arch.interrupt_window_open)
f7d9238f
AK		 3166 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu));
3167 else
3168 enable_irq_window(vcpu);
3169 }
3170 if (vcpu->arch.interrupt.pending) {
3171 vmx_inject_irq(vcpu, vcpu->arch.interrupt.nr);
3172 kvm_timer_intr_post(vcpu, vcpu->arch.interrupt.nr);
3173 }
85f455f7
ED		 3174}
3175
9c8cba37
AK		 3176/*
3177 * Failure to inject an interrupt should give us the information
3178 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
3179 * when fetching the interrupt redirection bitmap in the real-mode
3180 * tss, this doesn't happen. So we do it ourselves.
3181 */
3182static void fixup_rmode_irq(struct vcpu_vmx *vmx)
3183{
3184 vmx->rmode.irq.pending = 0;
5fdbf976 3185 if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
9c8cba37 3186 return;
5fdbf976 3187 kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
9c8cba37
AK		 3188 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
3189 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
3190 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
3191 return;
3192 }
3193 vmx->idt_vectoring_info =
3194 VECTORING_INFO_VALID_MASK
3195 | INTR_TYPE_EXT_INTR
3196 | vmx->rmode.irq.vector;
3197}
3198
c801949d
AK		 3199#ifdef CONFIG_X86_64
3200#define R "r"
3201#define Q "q"
3202#else
3203#define R "e"
3204#define Q "l"
3205#endif
3206
04d2cc77 3207static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6aa8b732 3208{
a2fa3e9f 3209 struct vcpu_vmx *vmx = to_vmx(vcpu);
1b6269db 3210 u32 intr_info;
e6adf283 3211
a89a8fb9
MG		 3212 /* Handle invalid guest state instead of entering VMX */
3213 if (vmx->emulation_required && emulate_invalid_guest_state) {
3214 handle_invalid_guest_state(vcpu, kvm_run);
3215 return;
3216 }
3217
5fdbf976
MT		 3218 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3219 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3220 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3221 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3222
e6adf283
AK		 3223 /*
3224 * Loading guest fpu may have cleared host cr0.ts
3225 */
3226 vmcs_writel(HOST_CR0, read_cr0());
3227
d77c26fc 3228 asm(
6aa8b732 3229 /* Store host registers */
c801949d
AK		 3230 "push %%"R"dx; push %%"R"bp;"
3231 "push %%"R"cx \n\t"
313dbd49
AK		 3232 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3233 "je 1f \n\t"
3234 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
4ecac3fd 3235 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
313dbd49 3236 "1: \n\t"
6aa8b732 3237 /* Check if vmlaunch of vmresume is needed */
e08aa78a 3238 "cmpl $0, %c[launched](%0) \n\t"
6aa8b732 3239 /* Load guest registers. Don't clobber flags. */
c801949d
AK		 3240 "mov %c[cr2](%0), %%"R"ax \n\t"
3241 "mov %%"R"ax, %%cr2 \n\t"
3242 "mov %c[rax](%0), %%"R"ax \n\t"
3243 "mov %c[rbx](%0), %%"R"bx \n\t"
3244 "mov %c[rdx](%0), %%"R"dx \n\t"
3245 "mov %c[rsi](%0), %%"R"si \n\t"
3246 "mov %c[rdi](%0), %%"R"di \n\t"
3247 "mov %c[rbp](%0), %%"R"bp \n\t"
05b3e0c2 3248#ifdef CONFIG_X86_64
e08aa78a
AK		 3249 "mov %c[r8](%0), %%r8 \n\t"
3250 "mov %c[r9](%0), %%r9 \n\t"
3251 "mov %c[r10](%0), %%r10 \n\t"
3252 "mov %c[r11](%0), %%r11 \n\t"
3253 "mov %c[r12](%0), %%r12 \n\t"
3254 "mov %c[r13](%0), %%r13 \n\t"
3255 "mov %c[r14](%0), %%r14 \n\t"
3256 "mov %c[r15](%0), %%r15 \n\t"
6aa8b732 3257#endif
c801949d
AK		 3258 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3259
6aa8b732 3260 /* Enter guest mode */
cd2276a7 3261 "jne .Llaunched \n\t"
4ecac3fd 3262 __ex(ASM_VMX_VMLAUNCH) "\n\t"
cd2276a7 3263 "jmp .Lkvm_vmx_return \n\t"
4ecac3fd 3264 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
cd2276a7 3265 ".Lkvm_vmx_return: "
6aa8b732 3266 /* Save guest registers, load host registers, keep flags */
c801949d
AK		 3267 "xchg %0, (%%"R"sp) \n\t"
3268 "mov %%"R"ax, %c[rax](%0) \n\t"
3269 "mov %%"R"bx, %c[rbx](%0) \n\t"
3270 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3271 "mov %%"R"dx, %c[rdx](%0) \n\t"
3272 "mov %%"R"si, %c[rsi](%0) \n\t"
3273 "mov %%"R"di, %c[rdi](%0) \n\t"
3274 "mov %%"R"bp, %c[rbp](%0) \n\t"
05b3e0c2 3275#ifdef CONFIG_X86_64
e08aa78a
AK		 3276 "mov %%r8, %c[r8](%0) \n\t"
3277 "mov %%r9, %c[r9](%0) \n\t"
3278 "mov %%r10, %c[r10](%0) \n\t"
3279 "mov %%r11, %c[r11](%0) \n\t"
3280 "mov %%r12, %c[r12](%0) \n\t"
3281 "mov %%r13, %c[r13](%0) \n\t"
3282 "mov %%r14, %c[r14](%0) \n\t"
3283 "mov %%r15, %c[r15](%0) \n\t"
6aa8b732 3284#endif
c801949d
AK		 3285 "mov %%cr2, %%"R"ax \n\t"
3286 "mov %%"R"ax, %c[cr2](%0) \n\t"
3287
3288 "pop %%"R"bp; pop %%"R"bp; pop %%"R"dx \n\t"
e08aa78a
AK		 3289 "setbe %c[fail](%0) \n\t"
3290 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
3291 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
3292 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
313dbd49 3293 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
ad312c7c
ZX		 3294 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
3295 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
3296 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
3297 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
3298 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
3299 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
3300 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
05b3e0c2 3301#ifdef CONFIG_X86_64
ad312c7c
ZX		 3302 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
3303 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
3304 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
3305 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
3306 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
3307 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
3308 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
3309 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
6aa8b732 3310#endif
ad312c7c 3311 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
c2036300 3312 : "cc", "memory"
c801949d 3313 , R"bx", R"di", R"si"
c2036300 3314#ifdef CONFIG_X86_64
c2036300
LV		 3315 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
3316#endif
3317 );
6aa8b732 3318
5fdbf976
MT		 3319 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3320 vcpu->arch.regs_dirty = 0;
3321
1155f76a 3322 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9c8cba37
AK		 3323 if (vmx->rmode.irq.pending)
3324 fixup_rmode_irq(vmx);
1155f76a 3325
33f089ca 3326 vmx_update_window_states(vcpu);
6aa8b732 3327
d77c26fc 3328 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
15ad7146 3329 vmx->launched = 1;
1b6269db
AK		 3330
3331 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3332
3333 /* We need to handle NMIs before interrupts are enabled */
e4a41889 3334 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
f08864b4 3335 (intr_info & INTR_INFO_VALID_MASK)) {
2714d1d3 3336 KVMTRACE_0D(NMI, vcpu, handler);
1b6269db 3337 asm("int $2");
2714d1d3 3338 }
cf393f75
AK		 3339
3340 vmx_complete_interrupts(vmx);
6aa8b732
AK		 3341}
3342
c801949d
AK		 3343#undef R
3344#undef Q
3345
6aa8b732
AK		 3346static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
3347{
a2fa3e9f
GH		 3348 struct vcpu_vmx *vmx = to_vmx(vcpu);
3349
3350 if (vmx->vmcs) {
543e4243 3351 vcpu_clear(vmx);
a2fa3e9f
GH		 3352 free_vmcs(vmx->vmcs);
3353 vmx->vmcs = NULL;
6aa8b732
AK		 3354 }
3355}
3356
3357static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
3358{
fb3f0f51
RR		 3359 struct vcpu_vmx *vmx = to_vmx(vcpu);
3360
2384d2b3
SY		 3361 spin_lock(&vmx_vpid_lock);
3362 if (vmx->vpid != 0)
3363 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3364 spin_unlock(&vmx_vpid_lock);
6aa8b732 3365 vmx_free_vmcs(vcpu);
fb3f0f51
RR		 3366 kfree(vmx->host_msrs);
3367 kfree(vmx->guest_msrs);
3368 kvm_vcpu_uninit(vcpu);
a4770347 3369 kmem_cache_free(kvm_vcpu_cache, vmx);
6aa8b732
AK		 3370}
3371
fb3f0f51 3372static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6aa8b732 3373{
fb3f0f51 3374 int err;
c16f862d 3375 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
15ad7146 3376 int cpu;
6aa8b732 3377
a2fa3e9f 3378 if (!vmx)
fb3f0f51
RR		 3379 return ERR_PTR(-ENOMEM);
3380
2384d2b3
SY		 3381 allocate_vpid(vmx);
3382
fb3f0f51
RR		 3383 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
3384 if (err)
3385 goto free_vcpu;
965b58a5 3386
a2fa3e9f 3387 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
fb3f0f51
RR		 3388 if (!vmx->guest_msrs) {
3389 err = -ENOMEM;
3390 goto uninit_vcpu;
3391 }
965b58a5 3392
a2fa3e9f
GH		 3393 vmx->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3394 if (!vmx->host_msrs)
fb3f0f51 3395 goto free_guest_msrs;
965b58a5 3396
a2fa3e9f
GH		 3397 vmx->vmcs = alloc_vmcs();
3398 if (!vmx->vmcs)
fb3f0f51 3399 goto free_msrs;
a2fa3e9f
GH		 3400
3401 vmcs_clear(vmx->vmcs);
3402
15ad7146
AK		 3403 cpu = get_cpu();
3404 vmx_vcpu_load(&vmx->vcpu, cpu);
8b9cf98c 3405 err = vmx_vcpu_setup(vmx);
fb3f0f51 3406 vmx_vcpu_put(&vmx->vcpu);
15ad7146 3407 put_cpu();
fb3f0f51
RR		 3408 if (err)
3409 goto free_vmcs;
5e4a0b3c
MT		 3410 if (vm_need_virtualize_apic_accesses(kvm))
3411 if (alloc_apic_access_page(kvm) != 0)
3412 goto free_vmcs;
fb3f0f51 3413
b7ebfb05
SY		 3414 if (vm_need_ept())
3415 if (alloc_identity_pagetable(kvm) != 0)
3416 goto free_vmcs;
3417
fb3f0f51
RR		 3418 return &vmx->vcpu;
3419
3420free_vmcs:
3421 free_vmcs(vmx->vmcs);
3422free_msrs:
3423 kfree(vmx->host_msrs);
3424free_guest_msrs:
3425 kfree(vmx->guest_msrs);
3426uninit_vcpu:
3427 kvm_vcpu_uninit(&vmx->vcpu);
3428free_vcpu:
a4770347 3429 kmem_cache_free(kvm_vcpu_cache, vmx);
fb3f0f51 3430 return ERR_PTR(err);
6aa8b732
AK		 3431}
3432
002c7f7c
YS		 3433static void __init vmx_check_processor_compat(void *rtn)
3434{
3435 struct vmcs_config vmcs_conf;
3436
3437 *(int *)rtn = 0;
3438 if (setup_vmcs_config(&vmcs_conf) < 0)
3439 *(int *)rtn = -EIO;
3440 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
3441 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
3442 smp_processor_id());
3443 *(int *)rtn = -EIO;
3444 }
3445}
3446
67253af5
SY		 3447static int get_ept_level(void)
3448{
3449 return VMX_EPT_DEFAULT_GAW + 1;
3450}
3451
cbdd1bea 3452static struct kvm_x86_ops vmx_x86_ops = {
6aa8b732
AK		 3453 .cpu_has_kvm_support = cpu_has_kvm_support,
3454 .disabled_by_bios = vmx_disabled_by_bios,
3455 .hardware_setup = hardware_setup,
3456 .hardware_unsetup = hardware_unsetup,
002c7f7c 3457 .check_processor_compatibility = vmx_check_processor_compat,
6aa8b732
AK		 3458 .hardware_enable = hardware_enable,
3459 .hardware_disable = hardware_disable,
774ead3a 3460 .cpu_has_accelerated_tpr = cpu_has_vmx_virtualize_apic_accesses,
6aa8b732
AK		 3461
3462 .vcpu_create = vmx_create_vcpu,
3463 .vcpu_free = vmx_free_vcpu,
04d2cc77 3464 .vcpu_reset = vmx_vcpu_reset,
6aa8b732 3465
04d2cc77 3466 .prepare_guest_switch = vmx_save_host_state,
6aa8b732
AK		 3467 .vcpu_load = vmx_vcpu_load,
3468 .vcpu_put = vmx_vcpu_put,
3469
3470 .set_guest_debug = set_guest_debug,
04d2cc77 3471 .guest_debug_pre = kvm_guest_debug_pre,
6aa8b732
AK		 3472 .get_msr = vmx_get_msr,
3473 .set_msr = vmx_set_msr,
3474 .get_segment_base = vmx_get_segment_base,
3475 .get_segment = vmx_get_segment,
3476 .set_segment = vmx_set_segment,
2e4d2653 3477 .get_cpl = vmx_get_cpl,
6aa8b732 3478 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
25c4c276 3479 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
6aa8b732 3480 .set_cr0 = vmx_set_cr0,
6aa8b732
AK		 3481 .set_cr3 = vmx_set_cr3,
3482 .set_cr4 = vmx_set_cr4,
6aa8b732 3483 .set_efer = vmx_set_efer,
6aa8b732
AK		 3484 .get_idt = vmx_get_idt,
3485 .set_idt = vmx_set_idt,
3486 .get_gdt = vmx_get_gdt,
3487 .set_gdt = vmx_set_gdt,
5fdbf976 3488 .cache_reg = vmx_cache_reg,
6aa8b732
AK		 3489 .get_rflags = vmx_get_rflags,
3490 .set_rflags = vmx_set_rflags,
3491
3492 .tlb_flush = vmx_flush_tlb,
6aa8b732 3493
6aa8b732 3494 .run = vmx_vcpu_run,
04d2cc77 3495 .handle_exit = kvm_handle_exit,
6aa8b732 3496 .skip_emulated_instruction = skip_emulated_instruction,
102d8325 3497 .patch_hypercall = vmx_patch_hypercall,
2a8067f1
ED		 3498 .get_irq = vmx_get_irq,
3499 .set_irq = vmx_inject_irq,
298101da
AK		 3500 .queue_exception = vmx_queue_exception,
3501 .exception_injected = vmx_exception_injected,
04d2cc77
AK		 3502 .inject_pending_irq = vmx_intr_assist,
3503 .inject_pending_vectors = do_interrupt_requests,
cbc94022
IE		 3504
3505 .set_tss_addr = vmx_set_tss_addr,
67253af5 3506 .get_tdp_level = get_ept_level,
6aa8b732
AK		 3507};
3508
3509static int __init vmx_init(void)
3510{
25c5f225 3511 void *va;
fdef3ad1
HQ		 3512 int r;
3513
3514 vmx_io_bitmap_a = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
3515 if (!vmx_io_bitmap_a)
3516 return -ENOMEM;
3517
3518 vmx_io_bitmap_b = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
3519 if (!vmx_io_bitmap_b) {
3520 r = -ENOMEM;
3521 goto out;
3522 }
3523
25c5f225
SY		 3524 vmx_msr_bitmap = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
3525 if (!vmx_msr_bitmap) {
3526 r = -ENOMEM;
3527 goto out1;
3528 }
3529
fdef3ad1
HQ		 3530 /*
3531 * Allow direct access to the PC debug port (it is often used for I/O
3532 * delays, but the vmexits simply slow things down).
3533 */
25c5f225
SY		 3534 va = kmap(vmx_io_bitmap_a);
3535 memset(va, 0xff, PAGE_SIZE);
3536 clear_bit(0x80, va);
cd0536d7 3537 kunmap(vmx_io_bitmap_a);
fdef3ad1 3538
25c5f225
SY		 3539 va = kmap(vmx_io_bitmap_b);
3540 memset(va, 0xff, PAGE_SIZE);
cd0536d7 3541 kunmap(vmx_io_bitmap_b);
fdef3ad1 3542
25c5f225
SY		 3543 va = kmap(vmx_msr_bitmap);
3544 memset(va, 0xff, PAGE_SIZE);
3545 kunmap(vmx_msr_bitmap);
3546
2384d2b3
SY		 3547 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
3548
cb498ea2 3549 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
fdef3ad1 3550 if (r)
25c5f225
SY		 3551 goto out2;
3552
3553 vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_FS_BASE);
3554 vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_GS_BASE);
3555 vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_IA32_SYSENTER_CS);
3556 vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_IA32_SYSENTER_ESP);
3557 vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_IA32_SYSENTER_EIP);
fdef3ad1 3558
5fdbcb9d 3559 if (vm_need_ept()) {
1439442c 3560 bypass_guest_pf = 0;
5fdbcb9d
SY		 3561 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
3562 VMX_EPT_WRITABLE_MASK |
928d4bf7
SY		 3563 VMX_EPT_DEFAULT_MT << VMX_EPT_MT_EPTE_SHIFT |
3564 VMX_EPT_IGMT_BIT);
534e38b4 3565 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
5fdbcb9d
SY		 3566 VMX_EPT_EXECUTABLE_MASK);
3567 kvm_enable_tdp();
3568 } else
3569 kvm_disable_tdp();
1439442c 3570
c7addb90
AK		 3571 if (bypass_guest_pf)
3572 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
3573
1439442c
SY		 3574 ept_sync_global();
3575
fdef3ad1
HQ		 3576 return 0;
3577
25c5f225
SY		 3578out2:
3579 __free_page(vmx_msr_bitmap);
fdef3ad1
HQ		 3580out1:
3581 __free_page(vmx_io_bitmap_b);
3582out:
3583 __free_page(vmx_io_bitmap_a);
3584 return r;
6aa8b732
AK		 3585}
3586
3587static void __exit vmx_exit(void)
3588{
25c5f225 3589 __free_page(vmx_msr_bitmap);
fdef3ad1
HQ		 3590 __free_page(vmx_io_bitmap_b);
3591 __free_page(vmx_io_bitmap_a);
3592
cb498ea2 3593 kvm_exit();
6aa8b732
AK		 3594}
3595
3596module_init(vmx_init)
3597module_exit(vmx_exit)
Ubuntu Artful kernel mirror